Sie sind auf Seite 1von 409

Equity Research Industry Update

January 17, 2007


Sector Weighting:

Semiconductors & Components

Market Weight

Semiconductors: Technology and Market Primer 4.0


A Comprehensive Guide to the Dynamic and Complex Semiconductor Industry

CIBC's Technology and Market Primer 4.0 is an updated version of our comprehensive "one-stop-shop" resource for semiconductor research. It is designed for investors new to the space as well as for those seeking a better understanding of key technological and market elements. Inside you will find plain-English discussions of the technology behind the semiconductor industry. We discuss the major drivers of the sector, its cyclicality, and emerging trends. Finally, we dive into individual product and end market segments, including forecasts and competitive positioning.
All figures in US dollars, unless otherwise stated.
07-70310 2007

Allan Mishan 1 (212) 667-7756


Allan.Mishan@us.cibc.com

Rick Schafer 1 (720) 554-1119


Richard.Schafer@us.cibc.com

Daniel M. Gelbtuch 1 (212) 667-8108


Daniel.Gelbtuch@us.cibc.com

Sam Dubinsky 1 (212) 667-7348


sam.dubinsky@us.cibc.com

CIBC World Markets does and seeks to do business with companies covered in its research reports. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of this report. Investors should consider this report as only a single factor in making their investment decision. See "Important Disclosures" section at the end of this report for important required disclosures, including potential conflicts of interest. See "Price Target Calculation" and "Key Risks to Price Target" sections at the end of this report, or at the end of each section hereof, where applicable.
CIBC World Markets Corp., 300 Madison Avenue, New York, NY 10017-6204 (212) 667-7000 (800) 999-6726 CIBC World Markets Inc., P.O. Box 500, 161 Bay Street, BCE Place, Toronto, Canada M5J 2S8 (416) 594-7000

Daniel Morris 1 (720) 554-1124


daniel.morris@us.cibc.com

Hugh Cunningham 1 (212) 667-7082


hugh.cunningham@us.cibc.com

Find CIBC research on Bloomberg, Reuters, firstcall.com and cibcwm.com

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Bon Apptit
This CIBC Semiconductors: Technology and Market Primer 4.0 is an updated version of the comprehensive one-stop-shop resource for the dynamic and complex semiconductor industry we first issued in June 2003 and updated in October 2004 and December 2005. The report is targeted to those investors new to the sector as well as those looking for a comprehensive resource to help them better understand key technological or market elements within the industry. We also suggest it as a desk reference for more experienced investors, as it has lots of forecast and market share data as well as in-depth discussions of many of the important trends affecting the semiconductor industry. We start with basic semiconductor definitions and a simple review of manufacturing processes. We then discuss the semiconductor cycle and key fundamental indicators, and introduce some important concepts. We follow with a discussion of the major semiconductor product groups, including revenue forecasts and market share data. We then take a deep dive into the most important end markets served by the semiconductor industry, highlighting key players and company market shares for each application. Product summaries cover analog, digital processors, logic, memory and discrete devices. End market summaries cover computing, networking, telecom/ datacom, wireless, digital consumer and automotive semiconductors. Weve also included a section on emerging technologies, where we briefly discuss ten technologies that are just getting off the ground and that will be exciting to watch in 2007 and beyond. Any comments that will help us make this a more successful document are welcomed. Bon apptit and enjoy. Allan Mishan Executive Director Sam Dubinsky Director Rick Schafer Managing Director Dan Morris Director Daniel Gelbtuch Executive Director Hugh Cunningham Director

CIBC World Markets Semiconductor Equity Research Note on this version: Version 4.0 is a complete upgrade to last years Primer. All historical data on companies is current as of 3Q06; and industry sales, unit, and utilization data is also current up to September 2006. Market share data has been updated to reflect the most recent third-party sources and our internal estimates. All forecasts have been updated as well, and were also extended to 2010 (from 3.0s 2009). Naturally, there were also changes to formatting and the report structure to make it more readable. Note that we also added two new end market sections: printers and WiMAX. This brings the total number of end markets discussed to 25, plus ten emerging technologies at the back of the report. Also, dont miss the company index at the very back of the report. The semiconductor industry supports hundreds of public and private companies, and its often hard to keep track of where they all participate.

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Acknowledgments
We would like to acknowledge the contributions of our research colleagues at CIBC whose help was invaluable in the preparation of this report. Semiconductor Capital Equipment Gary Hsueh Optical Components Jeff Osborne Adam Hinckley Communication Technology Ittai Kidron Glen Anderson George Iwanyc Electronics Manufacturing Services Todd Coupland Sean Peasgood Enterprise Software Brad Reback Renaud DeVreker Brian Denyeau Entertainment Software Brendan McCabe Brad Reback Brian Denyeau Infrastructure Software Shaul Eyal Yair Reiner Manish Hemrajani Telecom Services Timothy Horan Srinivas Anantha William Maina Ned Baramov

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Table of Contents
Section I: Semiconductor Basics .........................5 Semiconductor Definitions .................................. 6 Semiconductor Device Structure .................... 8 Semiconductor Devices in Systems................. 9 Semiconductor Manufacturing ........................... 12 CMOS ....................................................... 19 Lithography ............................................... 20 Wafer Size................................................. 23 Manufacturing Strategies ............................ 25 Geographic Centers .................................... 29 Semiconductor Capital Equipment ................ 30 Section II: The Semiconductor Market ..............32 Industry Basics................................................ 33 The Semiconductor Cycle.................................. 38 Fundamentals ................................................. 46 Section III: Market Segments and Competitors.53 Semiconductor Device Types ............................. 54 Key Competitors .............................................. 60 Section IV: Product Summary ...........................64 Analog............................................................ 67 Analog SLICs ............................................. 68 Analog ASSPs ............................................ 70 Digital Processors ............................................ 72 Microprocessors ......................................... 73 Microcontrollers ......................................... 76 Digital Signal Processors ............................. 79 Digital Logic .................................................... 81 Special Purpose Logic ................................. 82 Display Drivers .......................................... 84 General Purpose Logic ................................ 85 ASICs ....................................................... 86 FPGAs/Programmable Logic Devices ............. 88 Memory .......................................................... 90 DRAM ....................................................... 91 SRAM ....................................................... 94 Flash ........................................................ 96 Mask ROM, EPROM, Other NV Memory .......... 99 Discretes and Optoelectronics.......................... 101 Discretes................................................. 102 Sensors .................................................. 104 Optoelectronics ........................................ 105 Section V: End Market Summary..................... 107 Computing .................................................... 110 PCs and Servers....................................... 111 PC Displays ............................................. 129 Hard Disk Drives ...................................... 135 Optical Drives .......................................... 149 Printers and Multi-Function Peripherals ....... 154 Networking ................................................... 160 Ethernet.................................................. 168 Wireless LAN (802.11) .............................. 179 Bluetooth ................................................ 191 Storage................................................... 197 Telecom/Datacom.......................................... 211 Central Office Line Cards ........................... 223 Modems .................................................. 226 PON........................................................ 241 Communications Infrastructure .................. 248 Voice-over-IP........................................... 279 Wireless ....................................................... 293 Wireless Handsets .................................... 298 Wireless Infrastructure ............................. 310 WiMAX .................................................... 317 Consumer Devices ......................................... 325 Digital Set-Top Boxes ............................... 326 Digital TV ................................................ 333 DVD Players and Recorders ....................... 343 Digital Cameras and Camcorders................ 350 MP3 and Portable Media Players ................. 356 Video Game Consoles ............................... 361 Flash Memory Cards ................................. 370 Automotive ................................................... 374 Emerging Technologies................................... 379 NAND Cache in PCs .................................. 380 Ultra Wide Band/Wireless USB ................... 381 Zigbee .................................................... 383 Infiniband ............................................... 384 iSCSI ...................................................... 385 Next-Generation Backplanes...................... 386 Mobile TV ................................................ 387 GPS ........................................................ 388 IPTV ....................................................... 389 Coax and Phone Line Networking ............... 390 Company Index ................................................ 392

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Section I: Semiconductor Basics Section I: Semiconductor Basics

This section deals with the basics of semiconductors. Topics include: Semiconductor definitions
- Semiconductor manufacturing process - CMOS - Lithography & Moores Law - Wafer Size - Manufacturing Strategies - Geographic Trends - Semiconductor Equipment

Manufacturing

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Semiconductor Definitions Semiconductor Definitions


A semiconductor is a solid-state substance that is halfway between a conductor and an insulator. When charged, the substance becomes conductive; when the charge is eliminated, it loses its conductive status. By combining conductive material, semiconductor material, and insulators in a pre-determined pattern, the movement of electricity can be precisely controlled. Semiconductors are therefore ideal for building devices that control the operation of electronic equipment. A transistor is the basic element used in building semiconductor devices. A transistor is fashioned from semiconductor material and acts as an on/off switch, which opens and closes when electrically activated.

Source: Computer Desktop Encyclopedia, CIBC World Markets Corp.

A semiconductor is a solid-state substance that is halfway between a conductor and an insulator. When charged with electricity or light, the substance becomes conductive. When that charge is eliminated, it loses its conductive status. By combining conductive material, semiconductor material, and insulators in a pre-determined pattern, the movement of electricity can be precisely controlled. Semiconductors are therefore ideal for building devices that control the operation of electronic equipment. The basic element used in constructing semiconductor devices is the transistor, which is simply a tiny on/off switch fashioned from semiconductor material that opens and closes when electrically activated. The transistor is composed of semiconductor material; thus, when it is in the off stage it is not conductive, and does not allow current to flow through it. When a voltage is applied to the transistor, it moves to the on stage, becoming conductive and allowing current to flow through it.

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Semiconductor Definitions Semiconductor Definitions


The simplest semiconductor devices are comprised of a single transistor; these are called discretes and are usually used to control the flow of signals and power within a larger electronic device. More complex semiconductor devices are built by combining multiple transistors and conductive interconnect material to form logic gates. These logic gates are arranged in a pre-defined pattern to perform more complex processing or storage functions. These devices are called integrated circuits (ICs).

The simplest semiconductor devices are comprised of a single transistor. These devices are referred to as discretes, and they are used in all types of electronic equipment used to control the flow of signals and power within a larger electronic system. When many transistors are combined, an integrated circuit is created that can be used to process or store data signals in an electrical format. Engineers design pathways using transistors and conductive interconnect material set in logical arrangements (called logic gates) to perform specific functions, and these pathways become the circuit architecture etched onto the surface of the chip. Industry participants will often use the terms semiconductor, integrated circuit, chip, or microchip interchangeably; usually they are referring to any microelectronic device built around transistors. All electronic devices are built around semiconductorstransistors make up logic gates, logic gates make up circuits and circuits make up electrical systems. Key end applications are computing, telecommunications, data networking, wireless communications, consumer electronics, automobiles, industrial equipment, and aerospace and military.

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Semiconductor Device Structure Semiconductor Device Structure


Semiconductor devices contain a transistor array, which is etched onto a rectangular piece of silicon called a die. The die is encased in a plastic or ceramic package, and tiny wires called wire bonding are used to connect the input/output gates on the chip to the leads on the outside of the package. Typically, the cost split for a fully packaged IC is roughly 90% for the silicon and 10% for the package.
Plastic/Ceramic Package and Substrate Wire Bonding Back View

Silicon Die Package Leads

Front View
Source: CIBC World Markets Corp.

Back View

Interior View

Typically, semiconductors are sold as stand-alone, fully functional packaged products that are ready to be implemented in an electrical system. The transistor array, which is etched onto a rectangular piece of silicon (called a die) during the wafer fabrication process, is housed in a plastic or ceramic package, and tiny wires called wire bonding are used to connect the input/output gates on the die to the leads on the package. The semiconductor device is snapped or soldered onto a printed circuit board, with the leads attached to conductive pathways along the board. For most semiconductor devices, the split in manufacturing value for a fully packaged semiconductor device is roughly 90% for the silicon and 10% for the package, although this can vary with the type of package and complexity of the device. Lower end devices such as discretes can have 30% or more of its value in the package given the maturity of the transistor process used at the trailing edge. On the other end of the spectrum, high-end devices such as microprocessors sometimes need more complex packages to deal with the speed and heat dissipation of the device, and therefore derive a higher value from the package as well. Note that in many cases, IC vendors offer the same die in a variety of packages to accommodate different feature sets, power requirements, device characteristics, platforms or customers. Depending on the design, the type of package used can have a dramatic impact on the speed, power consumption, heat dissipation and footprint of the device.

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Semiconductor Devices In Systems Semiconductor Devices In Systems


Getting Designed In Semiconductor devices are used as components in larger electrical systems designed by an original equipment manufacturer (OEM). Chip vendors will sample their devices to OEMs, and are awarded design wins as the OEM designs the part into their systems. The chip vendor will then provide production versions of the device, which will be qualified by the OEM before going to full production. Semiconductor devices can be custom designed for a specific customer and platform; these are called ASICs and are usually designed in cooperation with the OEM. Merchant devices called ASSPs are not designed for specific customers and can therefore be used by multiple OEMs. For the most generic components, OEMs will often buy off-the-shelf components, many of which are sold through the distribution channel.

Semiconductor devices are used as components in larger electrical systems designed by an original equipment manufacturer (OEM). Chip vendors supply samples of their devices to OEMs, sometimes based on specifications dictated by the customer. They are then are awarded design wins as the OEM designs the part into their systems. The chip vendor will then provide production versions of the device, which will be qualified by the OEM before going to full production. This entire process can be as short as several weeks or as long as a year or even more. Semiconductor devices can be custom designed for a specific customer and platform; these are called ASICs and are usually designed in cooperation with the OEM. ASICs are generally used when performance is key (e.g., a core router) or when volume on a single device is significant enough (e.g., a game console). Merchant devices called ASSPs or simply standard products are not designed for specific customers and can therefore be designed in by multiple OEMs. ASSPs are generally used when time-to-market and costs are key (e.g., PCs, MP3 players). For the most generic components, OEMs will often buy off-the-shelf components, many of which are sold through the distribution channel. This includes not only memory but also standard logic, analog and discrete devices. Customer engagements take many forms. Sometimes the OEM issues straight purchase orders; other times they will have a deeper engagement with a formal supply contract or even a technology partnership. The deeper the engagement, the more customization the OEM will typically expect from the chip designer. Note that design wins are awarded not only based on the performance of the device; factors such as pricing, software, service, support, existing relationship, track record, strength of the roadmap, scalability and other factors are often just as important.

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Semiconductor Devices In Systems Semiconductor Devices In Systems


System IP and Standards For more complex electronic systems, semiconductor vendors will incorporate system IP (intellectual property) directly into their ICs. Often, they will try to integrate as many logic functions as possible into a single SoC (system-on-a-chip) or chip set in order to reduce the OEMs bill of materials. The IC vendors system knowledge will be critical in winning the design. Chip vendors design their devices to conform to the needs of their target customer set. Sometimes, the OEM will supply specifications directly to their chip vendors. In mature, high-volume markets, the devices will often adhere to specifications as defined by a standard, set forth by a standards body. This allows multiple chip and equipment vendors to compete more easily, speeding time-to-market and lowering the cost of the technology implementation.
When dealing with more complex systems, semiconductor vendors will incorporate system intellectual property (IP) directly into the device, either in the form of logic gates or in software or firmware that runs on top of the device. This is especially true in the case of application specific standard products, which get sold into multiple platforms at multiple vendors. By incorporating the system IP, the chip vendor lowers the design cost for the OEM and also speeds time-to-market, two factors that often matter more than simple performance or device pricing. IC designers will usually try to integrate as many logic functions as possible into a single IC in order to reduce the OEMs bill of materials. Sometimes, it will prove too difficult or too costly to integrate certain functions, and the IC vendor will offer a chip set, either internally or with a partner. In any case, the IC vendors system knowledge will be critical to winning the design. Chip vendors will design their devices to conform to the needs of their target customer set. Sometimes, the OEM will supply specifications directly to its chip vendors, either in an ASIC arrangement or when multiple vendors compete for a standard product win. In the more mature, high-volume markets, devices will often adhere to specifications as defined by a standard, set by a standards body such as International Organization for Standardization (ISO) or the Institute of Electrical and Electronics Engineers (IEEE). This allows multiple vendors to compete, speeds time-tomarket, and lowers the cost of the technology implementation across the supply chain. Chip vendors maintain seats on the standards bodies alongside their OEM customers in order to influence the outcome of standards negotiations.

10

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Semiconductor Devices In Systems Semiconductor Devices In Systems


The Evolving Electronics Supply Chain The electronic equipment industry supply chain has undergone significant changes in the last decade, as OEMs increasingly outsource aspects of their production process. Electronics manufacturing services (EMS) companies build products for their OEM customers. Original device manufacturers (ODM) companies go a step further, taking over portions of the design and procurement process. The PC industry also has a large and mature motherboard sector that serves both PC OEMs and the distribution channel. Semiconductor vendors must maintain relationships with their OEM customers outsourcing partners. Component decisions are increasingly being pushed toward the ODMs and EMS, favoring companies with strong channel relationships.

The electronic equipment industry supply chain has undergone some significant changes in the last decade, as OEMs have shifted their business models away from chasing hardware margins and increasingly toward cultivating software and service businesses. This has driven hardware prices lower and forced sharper attention on costs. In addition to shifting production to low-cost regions like Asia, OEMs are increasingly looking to outsource some aspects of the production process. Electronics manufacturing services (EMS) companies are now responsible for a big piece of PC, wireline, and wireless equipment production. Original device manufacturer (ODM) companies take it a step further, taking over some aspects of the design process. The PC industry also has a large and mature motherboard sector that serves both PC OEMs and the distribution channel. In order to keep pace with the evolving supply chain, semiconductor vendors must maintain relationships with their OEM customers outsource partners. Component decisions are increasingly being pushed toward the ODMs and EMS, favoring companies with strong channel relationships. In the PC market, for example, servicing the motherboard makers is just as important as winning designs with OEMs back in the U.S. and Japan.

11

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Semiconductor Manufacturing Semiconductor Manufacturing


Semiconductor devices are manufactured in specialized factories called wafer fabs using a process known as wafer fabrication. Circular wafers of silicon are put through a cycle of chemical processes in order to etch an ion-charged transistor array as patterned on a set of masks. On top of the transistor array, layers of metal interconnect form pathways between the transistors; the layers are insulated by a dieletric material. After wafer processing, the finished wafer is put through a dicing process, where individual die are separated. These are sent to a back end facility for packaging and assembly and final test.
Source: IBM

Semiconductor devices begin the manufacturing process in specialized factories called wafer fabs using a process known as wafer processing or wafer fabrication. In this set of processes, known as front end manufacturing, circular wafers of silicon are put through a cycle of chemical processes in order to etch a transistor array pattern on the wafer (multiple units are etched onto each wafer; they are later separated into individual devices). During this process, a stepper will image the circuit pattern from a set of masks that contain the device design. Through stages of deposition, masking, etching and implantation using advanced photolithography, a series of intricate transistor arrays are formed on the surface of the wafer. Once the transistor array is complete, the interconnect layers will be built on top of the silicon transistors. These, too, are patterned on a set of masks and are created through multiple cycles of deposition, masking and etching using photolithography. Interconnect is applied layer by layer, with an insulating dieletric material deposited in between. The interconnect material is usually aluminum or copper, while a variety of non-conductive materials are used for the dieletric. Once wafer processing is complete, the wafer is sliced into individual die with a diamond drill through a process called dicing. The individual die are then transferred to a back end facility for final packaging and assembly and final test.

12

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Semiconductor Manufacturing Semiconductor Manufacturing


Semiconductor Manufacturing Flow Diagram
Wafer Manufacturing Wafer Processing Dicing Assembly & Packaging

Crystal Pulling Chemical Vapor Deposition Crystal Slicing Coating Wafer Lapping and Polishing Align and Expose Cleaning Develop and Bake Oxidation Etch and Strip Resist Ion Implantation Physical Vapor Deposition Chemical Mechanical Planarization (CMP) Passivation Wafer Probe/Test
Source: CIBC World Markets Corp.

Key Steps:

Key Steps:

Key Steps:
Dicing

Die Attach Wire Bond Encapsulation and Finishing Final Test

Key Steps:

Semiconductor manufacturing can be divided into four relatively discrete stages. These include wafer production, wafer processing, dicing, and back end assembly and test. Wafer production: In this stage, silicon wafers are created from raw silicon. Separate wafer companies perform this process; few semiconductor device makers produce their own wafers. Wafer processing: Wafers are run through a series of chemical and lithographical processes to etch the transistor array and interconnects. Also called front end processing, this is the longest, most complex and most costly stage of semiconductor manufacturing. Fabbed semiconductor producers and foundry suppliers perform this process. Dicing: The processed wafer is chopped into individual die using a diamond drill. This is done at the beginning of the back end process. Assembly and test: Individual die are placed into a plastic or ceramic package and tiny wire bonding is used to connect the input/output gates on the chip to the leads on the outside of the package. The finished device is then tested. Assembly is done either by fabbed semiconductor producers or by third-party outsourcers; test is usually done in-house by both fabbed and fabless chip producers but can be outsourced.

The steps are discussed in further detail on pages 14-18.

13

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Wafer Manufacturing Wafer Manufacturing


Silicon wafers are produced by heating a mixture of silica and carbon in a furnace, creating wafer-grade silicon. A seed is then dipped into the molten silicon and is slowly twisted and pulled out. This creates a cylindrical ingot several feet long, which is ground to an appropriate diameter (200mm, 300mm, etc). The ingot is then sliced into thin wafers for shipment to IDMs and foundries.

Source: MEMC, CIBC World Markets Corp.

The first step in building a semiconductor device is the manufacture of silicon wafers. A mixture of silica and carbon is heated in a furnace, creating a molten mixture of wafer-grade silicon. A silicon seed is dipped into the melt and slowly pulled out. This creates a cylindrical ingot several feet long, which is ground to the proper diameter (200 mm, 300 mm, etc.). The ingot is then sliced into thin wafers for shipment to IDMs and foundries. The pictures below display crystals, ingots and finished wafers.

Source (all pictures): MEMC, IBM, Texas Instruments, Computer Desktop Encyclopedia.

14

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Wafer Processing: Pre-Metal Wafer Processing: Pre-Metal


In the pre-metal stage, wafers are put through an intense cycle of chemical processes in order to etch a transistor array patterned on a set of masks. Through cycles of deposition/oxidation, photolithography, etching, and ion implantation, the transistor array is created. Charged ions are then deposited to enable transistor functionality.
Oxidation/ Deposition Coating
Light Light Light

Lithography

Chemical Bath Etching Stripping


Ions Ions

Implantation
Silicon Wafer Deposited Material Photoresist Soft Photoresist Mask Ion Charged- Silicon

Source: CIBC World Markets Corp.

The first half of the front end process is called the pre-metal stage. In this stage the transistor array is etched onto the wafer and ions are deposited in selected areas beneath the surface of the silicon. Oxidation/deposition: In this first step, oxide material is deposited onto the surface of the wafer. This can be done through oxidation, which involves the heating of the wafer in a furnace filled with oxide gas, or through physical (PVD) or chemical (CVD) deposition of the material onto the surface of the wafer. Photolithography: The wafer is then coated with a layer of photoresista material that is sensitive to light. A wafer stepper then focuses an intense beam of light through a pre-formatted mask, softening the photoresist material in certain areas of the wafer according to the pattern of the mask in front of the light source. The wafer is then sent through a chemical bath to dissolve the soft photoresist. Etching: Etching tools then remove the oxide material that is not still covered by photoresist. Once etched, the remaining photoresist material is stripped away using special chemicals. Ion implantation: At various stages throughout the process, implantation equipment creates charged regions within the silicon wafer to enable transistor functionality. Ionic materials known as dopants are shot into the wafer, where they remain under the surface of the silicon. Both positive and negative dopants are used.

15

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Wafer Processing: Interconnect Wafer Processing: Interconnect


After the transistors are created, they are connected together to form the logic gates using an interconnect material, usually aluminum or copper. The metal layers are built with stages of deposition, lithography, and etching, similar to the pre-metal stage but using separate equipment and masks. Dieletric material is deposited between the layers to insulate them from one another.
Passivation Layer Metal (Aluminum)
Tungsten plug Tungsten plug Tungsten plug Insulating Material (Dielectric) 3rd Metal Layer 2nd Metal Layer

Metal (Aluminum)
Dielectric

Metal (Al)
Gate Gate +
Source Dielectric

1st Metal Layer

Gate Drain

+
Drain

Source

N-well (negatively charged)

Positively charged silicon wafer

Source: CIBC World Markets Corp.

After the transistor array is created, the wafer is moved to the metal area of the fab, where interconnect layers are deposited and etched to create the logic gates. The metal interconnect layers are created through cycles of deposition, photolithography and etching, similar to the pre-metal stage but with separate equipment. In between each of the metal layers, dieletric material is deposited to insulate the metal layers from one another. A final passivation layer is added above the top metal layer to protect the circuitry. Most devices use aluminum or copper for the interconnect layers. Aluminum tends to resist the flow of electricity as wires are made thinner and narrower; therefore, most device manufacturers shifted to higher performance copper interconnects when they began 0.13-micron production. The number of metal layers varies by device; the highest performance devices can have as many as nine metal layers.

The picture below displays an IBM-fabricated SRAM cell with the insulating oxide films removed.

Source: IBM.

16

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Dicing Dicing
After the wafer is processed, a diamond drill is used to slice the wafer into individual die. Each die is tested before being sent to the back end facility.
Semiconductor Dicing Diamond Drill Finished Wafer and Individual Die

Source: IBM

Source: LSI Logic, Denali Software

Once the wafer is fully processed and all pre-metal, metal and inter-metal layers are built, a diamond circular saw is used to separate the individual die. The die are then tested for defects and the good die are sent to a back-end line for packaging and final test.

17

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Back End: Assembly & Test Back End: Assembly & Test
The individual die are sent to a back end assembly facility, where they are attached to a package, wire bonded, encapsulated, and run through final test.
Die Attach Wire Bonding
Back View

Encapsulation & Finishing

Final Test

Front View
Source: CIBC World Markets Corp.

Once diced, the individual die are then packaged and tested. Each die is first attached to a package, and wire bonding is used to connect the input/output gates on the die to the leads on the outside of the package. The device is then encapsulated, sealed and markings are affixed. A final test is conducted before shipment. Note that many devices are now using a process called flip chip, where the die is flipped during packaging. This increases the performance of the device but is also more costly. The pictures to the right displays an IBM semiconductor package, with a close-up displaying wire bonding.

Source: IBM.

18

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Manufacturing: CMOS Manufacturing: CMOS


CMOS (Complementary Metal Oxide Semiconductor) is the most widely used type of semiconductor design and manufacturing process. CMOS processes use standard silicon wafers and combine both positive and negative transistors.

In general, devices fabricated in CMOS will be cheaper to manufacture and will consume less power than other devices. Designs done in CMOS are also easier to scale to smaller transistor sizes. Most semiconductor designers and manufacturers will use standard CMOS wherever possible. However, certain high-performance or specialized applications require the use of bipolar or compound semiconductor manufacturing processes.

CMOS (pronounced see-mos), which stands for complementary metal oxide semiconductor, is the most widely used type of semiconductor design process in production today. The CMOS process uses standard silicon wafers and combines PMOS (positive) and NMOS (negative) transistors in specific ways, so that the final product consumes less power than PMOS-only or NMOS-only circuits. Devices fabricated in CMOS are cheaper to manufacture and consume less power than devices fabricated using other processes, and CMOS is therefore generally used whenever possible. Other types of manufacturing involve certain compounds more suited to specific types of applications, primarily for high-speed communications or power management and amplification. Silicon germanium (SiGe) is a commonly used compound for high-speed physical layer devices in wireline communications. Gallium arsenide (GaAs) is used in wireless handsets and set-top boxes, as it is better suited to high-voltage RF (radio frequency) applications. It is also used in optoelectronic devices for wireline communications. Other exotic semiconductor processes include silicon bipolar (BiCMOS), indium phosphide (InP), indium gallium phosphide (InGaP) and indium gallium arsenide (InGaAs). In general, these specialized processes are more costly to design and manufacture than standard CMOS and are more difficult to scale down to smaller process technologies (see section on lithography starting on the next page). Therefore, wherever possible, designers tend to choose CMOS over these processes. Although CMOS should represent the bulk of manufacturing in the near future, we note that several large IDMs, including Intel, IBM, AMD and Texas Instruments, have developed a process known as strained silicon, which allows for the use of silicon germanium transistors alongside standard CMOS transistors.

19

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Manufacturing: Lithography Manufacturing: Lithography


Lithography, also known as geometry or simply process technology, describes the level of smallness the manufacturing process can achieve, which determines the size of the transistors on the die.

Smaller transistors mean: 1) electrons move quicker between them faster devices 2) more transistors per square inch smaller die sizes cheaper devices 3) less electricity needed to power them lower power consumption

Lithography, as mentioned in the section on wafer processing, is a general term used for the set of processes that transfer the transistor array and interconnect design from the mask onto the silicon wafer. The level of smallness the manufacturing process can achieve is therefore determined by the lithography equipment; this in turn determines the size of the transistors on the die. The industry continually shrinks transistor size, also known as feature size, by moving to more advanced lithography equipment. On average, the industry moves to a new process geometry node every two years. The advantages of moving to smaller lithography are threefold. First and foremost, smaller transistors mean faster devices. The smaller the transistor, the less time it takes for the electrical charge to move across it. Second, smaller transistors mean smaller overall die size, which means more devices per wafer, which translates into lower cost per device. The ability to integrate additional logic, analog, discrete and memory content more cost-effectively stems from this as well. Third, smaller transistors mean lower power consumption and therefore cooler running devices. In order to achieve smaller geometries, manufacturers must invest in new lithography equipment and must alter their designs to take advantage of the new transistors. Certain technology challenges also need to be addressed (in addition to the normal yield issues that result from any new manufacturing process), such as current leakages, increased source and drain resistance and difficulty in sending thinner laser light through glass. Most recently, the industry has required innovations such as copper interconnect and low-K dialetrics in order to meet these challenges; future innovations currently under development include high-K dieletric and extreme ultra-violet (EUV) lithography.

20

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Manufacturing: Lithography Manufacturing: Lithography


Moores Law Advancement towards finer lithography, which enables smaller and smaller transistors, is the primary driver behind the dramatic improvement in semiconductor device performance over the past few decades. Moores Law states that the number of transistors on a chip doubles about every two years. The phenomenon was first noticed by Intel founder Gordon Moore, and has held basically true for the past 40 years or so. In addition to the doubling of transistor density, each new lithography generation usually brings 0.7X minimum feature scaling, 1.5X faster transistor switching speed, reduced chip power, and reduced chip cost.

The continuing advancement toward more advanced lithography, which enables smaller and smaller transistors, is the primary driver behind the dramatic improvement in semiconductor device performance over the past few decades. These advancements drive Moores Law, which states that the number of transistors on a chip doubles about every two years. Intel co-founder and industry legend Gordon Moore first noted the trend in DRAM densities and the law has held true for the past 40 years or so. In addition to the doubling of transistor density, each new lithography generation usually brings 0.7X minimum feature scaling, 1.5X faster transistor switching speed, reduced chip power, and reduced chip cost. The diagram to the right displays the number of transistors on each successive generation of Intel microprocessors.
Intel Corporation

21

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Manufacturing: Lithography Manufacturing: Lithography


Transistor sizes are measured in microns or nanometers (1 micron = 1,000 nm). About 35% of wafer production in 2005 was at geometries of 0.5-micron and above, with another 15% in 0.25- and 0.35; these are mostly discretes and analog ICs. More complex logic uses 0.18- and 0.13micron, while the most advanced devices use 90-nm (0.09-micron) or bleeding edge 65-nm (0.065-micron).
2005 Production By Process Technology
65-nm 2% 90-nm 19%
100% >500-nm 250/350-nm 80% 180-nm 130-nm 60% 90-nm 65-nm 40%

Monthly Production By Process Technology

>500-nm 35%

130-nm 19%

20%

0%

250/350-nm 15%

180-nm 10%

Note: Data displayed in MSI (millions of square inches of wafer) Source: VLSI, CIBC World Markets Corp.

Note: Data displayed in MSI (millions of square inches of wafer) Source: VLSI, CIBC World Markets Corp.

Transistor size is measured in microns, or one millionth of a meter, although it is increasingly being quoted in nanometers (abbreviated nm). Most devices built today use deep sub-micron transistors, with prevalent geometries being 0.130-micron (130 nm) and below. The majority of advanced ICs in production today use 90-nm (0.09-micron) processes, though IDMs are ramping to volume on devices using 65-nm (0.065-micron), which we estimate will represent 5%-10% of shipments in 2006. On average, the industry moves to a new process geometry node every two years and 45-nm (0.045-micron) products are expected to be commercially available in mid 2007 to early 2008. Traditionally, high-volume processor and logic (usually DSP and ASIC) manufacturers lead the charge at each transistor node; these are the manufacturers that are ramping 65-nm in volume today. High volume commodity DRAM and NAND flash providers are a half-step behind, though memory processes are typically much simpler (fewer metal layers, etc.) than processors and logic, so at times memory vendors may actually pull ahead of logic vendors. Just behind the logic and memory manufacturers are the leading edge foundries, which typically load their most advanced fabs with graphics and PLD chips as well as DSPs. The advanced foundries today concentrate on 0.13micron and 90-nm process, and plan to ramp 65-nm in 2007. The foundries are followed by SRAM and NOR flash memory producers, though many of these have been moving to an outsourcing model in the last few years. Sitting at the trailing edge are discrete device manufacturers, today mostly at 0.25-micron and above. Analog semiconductor production generally takes place on separate processes, typically several generations behind.

22

Mar-99 Jun-99 Sep-99 Dec-99 Mar-00 Jun-00 Sep-00 Dec-00 Mar-01 Jun-01 Sep-01 Dec-01 Mar-02 Jun-02 Sep-02 Dec-02 Mar-03 Jun-03 Sep-03 Dec-03 Mar-04 Jun-04 Sep-04 Dec-04 Mar-05 Jun-05 Sep-05 Dec-05 Mar-06 Jun-06 Sep-06

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Manufacturing: Wafer Size Manufacturing: Wafer Size


Wafer size is the diameter of the wafer used in manufacturing.
Larger wafers more die per wafer lower cost per die

Most fabs today use 150 mm (6-inch) or 200 mm (8inch) wafers. 300 mm (12-inch) fabs are in volume production now; the larger wafers yield more than twice the die per wafer of a 200 mm equivalent.

Source: VLSI Research, Intel.

Source: Promos

Wafer size refers to the diameter of the silicon wafer used in manufacturing. The majority of manufacturers use 150 mm (6-inch) or 200 mm (8-inch) wafers, with 300 mm (12-inch) reserved for the most advanced, highest volume production. The advantages of using larger wafers are purely lower cost: the larger the wafer size, the more die can fit on a single wafer with a less than proportional increase in cost per wafer. No additional chip performance is attained, just more die are produced at lower average cost per die (once the process reaches volume production and start-up costs are amortized). Presently, the industry is moving from 200 mm wafers to 300 mm wafers. Note that the 50% increase in wafer diameter translates to a 2.25X increase in usable wafer area (remember that the area of a circle is determined by r2; some additional gains are realized on the periphery of the wafer, where the larger circumference yields smoother curves). For example, a 200 mm process running a die size of 12 cm x 12 cm would yield 180 gross die per wafer; a 300 mm process running the same die would yield 432 gross die. In order to move to larger wafer size, most equipment in the fab must be replaced, and in many cases, an entirely new facility must be constructed. Consequently, moves to larger wafers occur once every few years and the time between moves has been steadily rising (Intel estimates that its next move, to 450 mm (18-inch) wafers, will not happen until at least 2015).

23

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Manufacturing: Wafer Size Manufacturing: Wafer Size


The ramp of 300 mm continues at a steady pace, with new fabs coming on-line from logic and memory manufacturers as well as foundries. 300 mm wafers represented roughly 18% of total industry production in 2005, and wafer shipments are likely to increase over 70% in 2006. Most manufacturers have paired 300 mm wafers with their most advanced lithographies.
2005 Production By Wafer Size
<150 mm 5% 150 mm 19%
100%

Monthly Production By Wafer Size


80%

300 mm 18%

60%

40% <150 mm 150 mm 20% 200 mm 300 mm

0% Sep-99 Sep-00 Sep-01 Sep-02 Sep-03 Sep-04 Sep-05 May-99 May-00 May-01 May-02 May-03 May-04 May-05

200 mm 58% Note: Data displayed in MSI (millions of square inches of wafer) Source: VLSI, CIBC World Markets Corp.

Note: Data displayed in MSI (millions of square inches of wafer) Source: VLSI, CIBC World Markets Corp.

The ramp of 300 mm continues at a steady pace in 2006. Three hundred mm wafer production increased 58% in 2005 and represented 18% of the industrys overall production. In 2006, 300 mm production should increase over 70%, representing a mid-twenties percentage of total industry production. Note that 300 mm wafers represented an even larger percentage of the leading edge production, as most manufacturers have paired 300 mm wafers with their most advanced lithographies. In general, the companies that have led the charge are both manufacturers of the highest-volume products and companies for whom manufacturing is a key competitive advantage. The most aggressive 300 mm manufacturers include Intel, Texas Instruments, Samsung, TSMC, Hitachi, and Elpida.

24

May-06

Sep-06

Jan-00

Jan-01

Jan-02

Jan-03

Jan-04

Jan-05

Jan-06

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Manufacturing: Strategies Manufacturing: Strategies


Integrated Device Manufacturers (IDMs) manufacture their own devices. Typically, these are mature vendors dealing in high volume products. IDMs have lots of operating and cyclical leverage.
Examples: Intel, Texas Instruments, Samsung, Toshiba, Renesas

Fabless companies design devices themselves but contract the manufacturing to others. These companies are more focused on design, and enjoy more margin and earnings stability as well as lower capital requirements.
Examples: Broadcom, Qualcomm, NVIDIA, Xilinx, Marvell, MediaTek

Foundries and Assembly and Testers are specialized third-party manufacturers that perform wafer fabrication or back-end processing for others on a contract basis.
Examples (foundry): Taiwan Semiconductor, UMC, SMIC, Chartered Examples (assembly and test): ASE, Amkor, Siliconware Precision

Semiconductor companies can be divided along manufacturing lines into three segments: IDMs, fabless and foundries. IDMs design and manufacture their own devices, fabless companies design their devices but dont manufacture them and foundries perform manufacturing but no design. Increasingly, IDMs are moving toward hybrid strategies, retaining some manufacturing in-house while contracting out for some additional capacity. These hybrid strategies offer more production flexibility while minimizing the level of critical capacity investments. Others have given up on manufacturing altogether and are going fabless. Front End Manufacturing Strategies Integrated Device Manufacturer Integrated device manufacturers, or IDMs, design and manufacture their own devices. Typically, these will be the more mature semiconductor manufacturers that deal in very high-volume products. These vendors also typically consider manufacturing expertise a key competitive advantage and a core competency for their businesses, either for technology reasons (e.g., they want to be first to market with next generation lithography, or they have specialized processes, etc.), to leverage economies of scale, to control their supply, or for a host of other reasons. IDMs enjoy high margins in the boom part of the cycle as their fixed costs do not change, but suffer low margins in the trough of the cycle as capacity goes unused. This effect should not be underestimated, as fixed costs (including

25

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

depreciation and fixed material and labor costs) can run as high as 80% of semiconductor manufacturing costs. IDMs must also maintain large capital expenditure budgets to stay ahead of the competition in order to ensure that they do not suffer performance or cost disadvantages; consequently, their ongoing cash needs are greater. Key examples of IDMs include Intel, Samsung, Micron, as well as nearly all the Japanese semiconductor suppliers. We also consider partial-outsourcers like Texas Instruments, STMicroelectronics, Infineon, and Freescale to be IDMs, as manufacturing is still a big part of what these companies do. Fabless Fabless companies are those that design devices themselves but contract the manufacturing to third parties. These vendors typically focus on design, leaving manufacturing competencies to others. The benefits of a fabless model include a lack of fixed costs, which translates into lower capital investment requirements, more consistent margins and better earnings predictability. Fabless companies have the opportunity to gain share vs. IDMs in the downturn (since wafer costs for them will decline while unit costs for IDMs rise with declining output) and will lose share at the peak of the cycle (because of the opposite effect). This effect can be neutralized if all direct competitors are fabless. Key fabless vendors include Broadcom, Qualcomm, Marvell, Xilinx, Altera, NVIDIA, Conexant, and PMC-Sierra. The list of fabless companies has been growing nicely, as the vast majority of semiconductor start-ups are fabless. Also, a number of IDMs have gone fabless in the past few years, including Agere, LSI Logic, AMCC, and Vitesse. Foundry Foundries are specialized third-party manufacturers that perform wafer fabrication for others on a contract basis. Foundries do not design any devices themselves, rather they manufacture the designs of fabless companies and IDMs. Key foundries include Taiwan Semiconductor (TSMC) and United Microelectronics (UMC) in Taiwan, Chartered Semiconductor in Singapore, SMIC in China, DongbuAnam in Korea and IBM Microelectronics in the U.S. Hybrid Strategies As mentioned above, many vendors use a hybrid manufacturing strategy. Most IDMs use external foundries for at least some of their production, usually when in-house capacity becomes tight or for specific leading edge devices. This trend has accelerated in the last cycle, as IDMs found themselves burned by too much capacity in the 2001 downturn. As mentioned above, big-name IDMs like Texas Instruments, STMicroelectronics, Infineon, and Freescale all outsource a significant portion of their production. Others use a fab-light strategy, outsourcing most of their manufacturing but keeping some in-house (usually, the inhouse facility will be used either for older, more mature manufacturing or for specialized processes). Another hybrid strategy, though less common, is where IDMs manufacture their own designs and also sell the excess capacity to fabless companies as a foundry for additional revenue.

Back End Manufacturing Strategies On the assembly and test side, the same three types of companies exist: those with internal facilities, those that outsource and specialty assembly and test outsourcers. We note that very few IDMs will do 100% of the packaging themselves, as packaging types in the industry number in the hundreds and most IDMs are unwilling to invest in all the necessary equipment. Further, many IDMs (particularly high performance IC manufacturers) will focus on leading edge packaging technologies in-house, but will leverage low-cost outsourcers in Asia for the mainstream processes. Still others (particularly low cost producers) will do most of the bulk processes in-house but will contract out some of the higher performance packages. Test functions are often conducted in the same facility as packaging and assembly. However, many fabless companies will outsource the packaging of their devices but will perform testing in-house; this allows them to more quickly identify design flaws. The outsourced assembly and test market is less consolidated than the foundry market, with key vendors being ASE Test, Amkor Technology, Siliconware Precision, and STATS-ChipPAC. Almost all test activity is conducted in the Far East, as it is more labor-intensive than front-end processes and also carries less stringent cleanroom requirements.

26

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Manufacturing: Strategies Manufacturing: Strategies


Foundries The rising costs of semiconductor factories have made it prohibitively expensive for all but the largest vendors to do their own manufacturing. This has accelerated the trend towards outsourcing. The major foundriesTSMC, UMC, SMIC, and Charteredhave therefore nicely outpaced industry revenue growth for the past several years.
Foundry vs. Semiconductor Revenue
$20 Foundry Revenue Semiconductor Revenue $16
Semiconductor Industry Revenue ($B)

2005 Semiconductor Foundry Market Share


$400 $350

Others 16%
Hua Hong NEC 2% DongbuAnam 2% Vanguard 2% Magnachip 2%

$300
Foundry Sector Revenue ($B)

$12

$250

$200

TSMC 45%

$8

$150

IBM 4% Chartered 6% SMIC 6%

$100

$4
$50

$0
20 02 20 04 20 05 19 99 20 00 20 01 20 03

$0

UMC 15%

Source: Dataquest, WSTS, CIBC World Markets Corp.

Source: Dataquest, CIBC World Markets Corp.

The rising costs of semiconductor factories and equipment over the past few years have made it prohibitively expensive for all semiconductor vendors to do their own manufacturing in house. This has forced many high-volume players to search for partners to do their wafer fabrication. Fabless designers have also grown in importance, and even control certain sub-segments of the semiconductor industry (graphics, PLDs, etc.); and since they are 100% outsourced, the foundries have had to grow to meet their supply needs. At the same time, the major foundries have done an excellent job of ramping leading edge capacity with world-class yields. The traditional top-three foundriesTSMC, UMC, and Charteredwere all very aggressive in getting to 0.13micron and have carried that through to 90-nanometer as well. Shanghai start-up foundry SMIC has also gained a lot of ground with logic and memory producers, and even overtook Chartered in 2005 in terms of revenue. Revenue growth trends tell the story well: in foundry, revenue growth has outpaced semiconductor industry growth in four of the last six years, and over that period has risen by 145% vs. the industrys 52%. Going forward, we expect foundry revenue growth to continue to outpace the overall industry, though the rate will likely decelerate somewhat. In terms of market share, TSMC was the leading foundry provider in 2005, with 45% overall revenue share. UMC followed with 15%. SMIC and Chartered each held 6%, and IBM held 4%. Other vendors include Magnachip (which spun out from Hynix), Vanguard International, DongbuAnam, Hua Hong NEC, X-Fab, compound semiconductor foundry Jazz Semiconductor, and Japanese IDMs Fujitsu and Toshiba.

27

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Manufacturing: Strategies Manufacturing: Strategies


Assembly and Test Services The trend toward outsourcing has also benefited back end providers, though not as dramatically as the foundries. IDMs often keep back end production in-house because they can realize lower costs, though fabless companies usually outsource. Top assembly and test houses include ASE, Amkor, Siliconware Precision, and STATS-ChipPAC.
Assembly and Test vs. Semiconductor Revenue
$16

2005 Assembly and Test Market Share


$300

Assembly and Test Revenue


Outsourced Assembly and Test Services Sector Revenue ($B)

Semiconductor Revenue
$250

ASE Group 17%


Semiconductor Industry Revenue ($B)

$12

$200

Others 41%

$8 $150

Amkor 14%

$100 $4 $50

$0
20 00 20 01 20 05 19 99 20 02 20 03 20 04

$0

KYEC 2% Carsem 2%

Siliconware Precision 9%
UTAC ChipMOS 2% PowerTech 3% 2%

STATS ChipPAC 8%

Source: Dataquest, WSTS, CIBC World Markets Corp.

Source: Dataquest, CIBC World Markets Corp.

The trend toward outsourcing has also benefited back end providers, though not as dramatically as the foundries. IDMs have tended to keep more of their back end production in-house because they can realize lower costs and the capital investment required is not nearly as large. As mentioned above, however, very few IDMs will do 100% of the packaging themselves, as packaging types in the industry number in the hundreds and most IDMs are unwilling to invest in all the necessary equipment. Further, many IDMs (particularly high performance IC manufacturers) will focus on leading-edge packaging technologies in-house, but will leverage low-cost outsourcers in Asia for the mainstream processes. Still others (particularly low-cost producers) will do most of the bulk processes in-house but will contract out some of the higher performance packages. The outsourced assembly and test market is less consolidated than the foundry market. Almost all assembly and test activity is conducted in the Far East, with the top two vendors in 2005 being ASE (Taiwan) and Amkor (Korea) , with 17% share and 14% share, respectively. Other key vendors include Siliconware Precision, STATS-ChipPAC, ChipMOS, PowerTech, UTAC, Carsem, KYEC, and Shinko Electric.

28

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Manufacturing: Geographic Centers Manufacturing: Geographic Centers


The largest geographic centers for semiconductor front end production are Japan and Asia, each about 1/3 of total production, with the final 1/3 split roughly evenly between North America and Europe. Asia has grown in importance over the past two decades: Korea is a major center for memory, displays, and commodities, and Taiwan hosts the largest foundries as well as some memory producers. China is also ramping up with several fabs under constructions.
Japan Renesas Toshiba NEC Electronics Sharp Spansion Matsushita Sony Toshiba Seiko-Epson Fujitsu Microcontrollers Discretes Memory Logic, ASSPs Analog North America Intel Texas Instruments Freescale IBM Micron Analog Devices Europe and Israel Infineon STMicro NXP/Philips AMD Intel Robert Bosch Taiwan Winbond Nanya Powerchip Promos Korea Samsung Hynix Magnachip China

Key IDMs

Major Foundries

IBM Micro Jazz Semi

X-Fab Tower Semi

TSMC UMC

Magnachip DongbuAnam

SMIC ASMC Grace Memory Logic

Key Products

MPUs, MCUs, DSPs Logic, ASSPs Analog Memory Discretes

Logic, ASSPs Analog MPUs Discretes Memory

Logic, ASSPs Memory Display Drivers

Memory Logic, ASSPs Display Drivers

The largest geographic centers for semiconductor front end production are Japan and Asia. Japanese electronics OEMs have historically been vertically integrated and thus retain fabs for their own semiconductor production. They are also large suppliers of microcontrollers, discretes, general purpose logic, and analog ICs to the broader market. DRAM, a big priority in Japan in the 1980s, has shifted away from Japan for the most part, but flash has ramped up for both NOR and NAND. Still, Japan accounts for about 34% of semiconductor capacity. Asia ex-Japan has clearly grown in importance over the last two decades, now equal to Japan at 34% of capacity. Korea (11% of total) is a major center for memory, displays and commodities and electronics giant Samsung continues to invest and grow in importance. Taiwan (11% of total) hosts the largest foundries as well as some memory producers. China (9% of total) is also ramping up, mostly with foundry production for memory and logic. The U.S. (17% of total) is still the most important region for high-end logic, with most CPU and DSP production as well as microcontrollers and ASSPs. Custom ASICs, a wide array of memories, analog ICs and commodities also have a large presence. Europe (15% of total, including fabs there owned by U.S. companies) has a similar mix; Infineon, STMicroelectronics, and NXP/Philips are large diversified players that compete across these categories, and Intel and AMD each run large microprocessor fabs in the region (Intel in Ireland and Israel, AMD in Germany). Note that back end factories are more weighted toward Asia due to the large labor component. Back end facilities are spread throughout the region, not only in Taiwan and Korea but also in China, Malaysia, Singapore, and elsewhere.

29

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Semiconductor Equipment Semiconductor Equipment


The semiconductor industry has a large and relatively independent supporting industry that supplies equipment for front and back end processing. Top equipment suppliers include Applied Materials, Tokyo Electron, ASM Lithography, Advantest, KLATencor, Nikon, Lam Research, and Novellus.
Applied Materials 14%
Others 41%

Front End
Tokyo Electron 11%

Automation and Process Control


KLA-Tencor Hitachi Brooks Automation Asyst Applied Materials Daifuku HP

ASML 8% Advantest 6%
Hitachi 2% 3%

Applied Materials Tokyo Electron ASML Nikon Lam Research Novellus Dainippon

Packaging & Assembly


Tokyo Seimitsu ASM Intl. Tokyo Electron Kulicke & Soffa Shinkawa BE Semiconductor

Automated Test
Advantest Teradyne Agilent Credence Yokogawa LTX

KLA-Tencor Nikon 5% Novellus Dainippon 3% Lam Research 4%


3%

Source: Dataquest, CIBC World Markets Corp.

The semiconductor industry has a large and relatively independent supporting industry that supplies wafer fabrication, assembly and test equipment as well as factory automation and process control software and hardware. Few IDMs or foundries make their own equipment, though some of the larger manufacturers will partner with equipment suppliers to help define the functionality of nextgeneration tools. In 2005, the semiconductor capital equipment industry totaled $34.5 billion in sales. Approximately 61% of this total was for front-end fab equipment, which includes CVD and sputtering deposition, diffusion/oxidation and RTP, lithography and ion implantation. Another 5% was for factory automation and 11% was for process control. About 23% of equipment spending was for back-end equipment, including packaging, assembly and automated test equipment. The exhibit to the right displays this breakdown.

Semiconductor Equipment Spending, 2005


Automated Test 9% Packaging and Assembly 10% Automation 5% Process Control 9% Deposition 13% Diffusion, RTP, Implant 6%

Lithography 17%

Other Front End 15%

Etch, Clean, Planarization 16%

Source: Dataquest, CIBC World Markets Corp.

30

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Top front end equipment suppliers include Applied Materials, strong in deposition, RTP (rapid thermal processing), CMP (chemical mechanical planarization), and etch; Tokyo Electron, prominent in thermal diffusion, photoresist processing, and etch; ASML Holding, dominant in lithography scanners; Nikon, also in lithography scanners; Lam Research in etch; Novellus in deposition; and Dainippon Screen in etch, clean, and planarization. Together, these vendors account for just under 75% of total front end equipment. On the process control front and automation front, top vendors include KLA-Tencor, strong in wafer and reticle inspection/metrology; Hitachi, also strong in wafer metrology; Brooks Automation, dominant in tool automation hardware and software; Asyst Technologies in fab automation hardware; Applied Materials mostly in defect review; Daifuku in fab automation hardware; and HP in automation software. Back-end equipment suppliers include assembly and packaging equipment makers Tokyo Seimitsu, ASM International, Tokyo Electron, Kulicke & Soffa, Shinkawa, BE Semiconductor, and Advantest. On the test side, the key vendors are Advantest, Teradyne, Agilent, Credence, Yokogawa, and LTX. The exhibits below display fab equipment market shares for front end and back end equipment as well as automation and process control. 2005 Front End Equipment Market Share
Others 21%

2005 Automation and Process Control Market Share

Applied Materials 22%

Other 30 %

KLA-Tencor 30 %

Hitachi 1% Canon 4% Dainippon 5%

Novellus 5% Lam Research 5%

Tokyo Electron 17%

IBM 3% HP 3% Daifuku 4%

Hitachi 10 % Applied Materials 6% Brooks Automation 7%

Nikon 7%

ASML 13%

Asyst 7%

2005 Packaging and Assembly Market Share


Tokyo Seimitsu 11% ASM Intl. 8% Tokyo Electron 7% Kulicke & Soffa 6% Shinkawa 5%
BE Semiconductor Advantest Unaxis 4% Delta Design 4% 3% 4%

2005 Automated Test Equipment Market Share


LTX 3% Others 4%

Yokogawa 5% Credence 7%

Others 48%

Agilent 14 %

Advantest 50 %

Teradyne 17 %

Source: Dataquest, CIBC World Markets Corp.

31

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Section II: The Semiconductor Market Section II: The Semiconductor Market

This section discusses the semiconductor market, industry fundamentals, and the semiconductor cycle. Topics include: Industry Basics The Semiconductor Cycle Fundamentals

32

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Industry Basics: 40 Years of Growth Industry Basics: 40 Years of Growth


As the enablers of electronic hardware, semiconductors have been at the heart of the technology revolution. The market has grown tremendously over the past 40 years, topping $227 billion in 2005.
$250 75%

$200
Semiconductor Revenue ($M)

50%

$150

25%

$100

0%

$50

Semiconductor Sales Year-over-Year Growth

-25%

$0
19 63 19 65 19 67 19 69 19 71 19 73 19 75 19 77 19 79 19 81 19 83 19 85 19 87 19 89 19 91 19 93 19 95 19 97 19 99 20 01 20 03 20 05

-50%

Source: WSTS, CIBC World Markets Corp.

As the enablers of electronic hardware, semiconductors have been at the heart of the technology sector. The market has grown tremendously in the past 40 years, sustaining a CAGR of 14% over that period of time. Growth has been slowing, however, as the industry has grown and matured; with the 20-year CAGR at 12%, the 10-year CAGR at 5%, and the 5-year CAGR at 2% (note that this 2% figure is based off of the 2000 bubble year and is therefore less significant). Looking at it another way, we have plotted a 5-year moving average of the industrys 5-year CAGR on the graph to the right. After hovering in the mid-to-high teens for more than two decades, we see the average dropping in the post-bubble period. We would expect the CAGR to drift upward as we roll off the difficult comps of the bubble, but for growth to remain in the single digits.

5-Year Moving Average of 5-Year CAGR


25%

20%

15%

10%

5%

5-year rolling average of 5-year industry CAGR

0%

Year-Over-Year Growth
9

3 20 0

19 7

19 8

19 8

19 8

19 9

19 9

19 9

19 9

19 9

20 0

Source: WSTS, CIBC World Markets Corp.

33

20 0

19 7

19 8

19 8

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Industry Basics: Demand Industry Basics: Demand


Semiconductor demand is closely tied to the demand for electronics. The largest consumers of semiconductors are the computing/data processing, communications, and consumer electronics markets. Semiconductor revenue has also been growing as a percentage of electronics revenue as chipmakers have absorbed more and more of the system IP into their devices.
Electronics Revenue By End Market
$1400 Military/Civil Aerospace Electronics Automotive Electronics Consumer Electronics $1000 Communications Electronics Data Processing Electronics
($ in billions) Semiconductors as a % of Electronics Factory Revenue

Semi Revenue as a Percentage of Electronics


25% All Semiconductors All Integrated Circuits Integrated Circuits ex. Memory 20%

$1200

Industrial Electronics

$800

15%

$600

10%

$400

$200

5%

$
8 9 2 3 9 0 1 2 3 4 5 6 7 8 0 1 4 19 8 19 9 20 0 20 0 20 0 19 8 19 9 19 9 19 9 19 9 19 9 19 9 19 9 19 9 19 9 20 0 20 0 20 0 5

0% 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Source: Dataquest, CIBC World Markets Corp.

Note: Data displayed is five-year average to smooth price volatility. Source: Dataquest, WSTS, CIBC World Markets Corp.

Semiconductor companies support the large and diverse electronics industry. Semiconductor demand is thus a function of the demand for electronic equipment and the percentage of electronic equipment bill-of-materials devoted to semiconductors. Electronics Demand The electronics industry totaled about $1.26 trillion in 2005, up 9% from 2004. Note that electronics revenue in 2005 exceeded the 2000 bubble year by 12%, with much healthier fundamentals behind it. By end market, the most important applications are computing/data processing, communications and consumer electronics. In 2005, 29% of electronics revenue was derived from data processing, which includes PCs and servers, PC peripherals, handheld devices, displays and storage. Twenty-three percent was derived from communications, including cellular handsets, networking equipment, telecom and wireless infrastructure, and voice and data access equipment. Twenty-two percent was derived from consumer electronics, including analog and digital TVs, audio/visual equipment, game consoles, set-top boxes, appliances and white goods. Another 12% was derived from industrial electronics, 7% from military and aerospace and 7% from automotive. The exhibits on the following page detail electronics revenue by application.

34

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

2005 Data Processing Equipment Revenue (29%)

2005 Communications Equipment Revenue (23%)


Other Comm 13% Legacy Phones 2% Wireless Infrastructure 8% Broadband 3%

Other 17% Flash Cards/Drives 3% PDAs 1% Storage 7% PCs 44%

Handsets 48%

Displays 8% PC Peripherals 7%

Telecom Infrastructure 17% Servers 13% Networking 9%

2005 Consumer Electronics Revenue (22%)


Analog CRT TV 5%

2005 Other Electronic Equipment Revenue (26%)

Other Consumer 22%

Digital TV 16%

Military/Civil Aerospace Electronics 26%

Automotive Electronics 26%

Analog A/V 5%

Appliances 31%
Game Console 3%

Digital A/V 15% Digital SetTop Box 3% Industrial Electronics 48%

Source: Dataquest, CIBC World Markets Corp.

Semiconductors as a Percentage of Electronics On top of electronics demand, semiconductor revenue has been growing as a percentage of the bill-of-materials and of total electronics revenue. In 2005, this ratio topped 16% (using a five-year average, intended to smooth semiconductor price volatility), having climbed steadily from the 10% range in the early 1990s. This increase has been due to a number of factors, including: Chipmakers have absorbed more and more system IP into their devices. Chip prices increasingly include software components and other costs. Outside of the memory market, chipmakers have for the most part retained pricing power in their devices, whereas OEMs have been subject to tough competition, pressuring their pricing. OEMs have been able to take costs out of the box through outsourcing, just-in-time inventory, etc. Chipmakers have benefited from Moores Law, but fab costs have continued to rise, as have R&D costs. OEMs have increasingly diversified their revenue streams by charging their customers for software, services, maintenance contracts and other non-equipment products. This has allowed them to charge lower margins on equipment. Semiconductor vendors, in contrast, continue to charge only for actual physical product (though they provide some software, this is usually embedded in the price of the IC).

35

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Industry Basics: Supply Industry Basics: Supply


Semiconductor supply is driven by wafer fab capacity. Long construction and equipment installation cycles as well as high cost of plant and equipment mean that capacity plans must be put into place far in advance, causing inevitable periods of surplus and shortage. While IDMs and foundries try to match supply to demand, wafer capacity moves very slowly in both directions, and capital expenditure budgets can be volatile.
Wafer Fab Capacity
600
Monthly Wafer Capcity (Millions of Square Inches of Wafer)

Semiconductor Cap-Ex
$70,000 35% Capital Expenditures Cap-ex As a Percentage of Sales $60,000 30%
Capital Expenditures As a Percentage of Semiconductor Sales

500

65-nm 90-nm 130-nm 180-nm 250/350-nm >500-nm


Capital Expenditures ($M)

$50,000

25%

400

$40,000

20%

300

$30,000

15%

200

$20,000 100

10%

$10,000

5%

Mar-99 Jun-99 Sep-99 Dec-99 Mar-00 Jun-00 Sep-00 Dec-00 Mar-01 Jun-01 Sep-01 Dec-01 Mar-02 Jun-02 Sep-02 Dec-02 Mar-03 Jun-03 Sep-03 Dec-03 Mar-04 Jun-04 Sep-04 Dec-04 Mar-05 Jun-05 Sep-05 Dec-05 Mar-06 Jun-06 Sep-06

$0

0%

8 19 8

Source: VLSI Research, CIBC World Markets Corp.

Source: WSTS, CIBC World Markets Corp.

Semiconductor supply is primarily a function of wafer fab capacity. Long construction and equipment installation cycles as well as high cost of plant and equipment mean that capacity plans must be put into place far in advance, causing inevitable periods of surplus and shortage. While IDMs and foundries try to match supply to demand, wafer capacity moves very slowly in both directions, much slower than changes in demand. Capital expenditure budgets can therefore be very volatile as chip suppliers adjust to the changes in demand. Note that the costs of a leading edge fab have risen dramatically over the past few years, leading to a consolidation of wafer supply in the hands of just a few players. Gone are the days when all semiconductor companies did their own manufacturing; now a company will only choose to do their own manufacturing if: 1) Access to leading edge technology is core to their business model (e.g., Intel, AMD, TI, Samsung, Micron). 2) They run such high volume that they can absorb the fixed costs (e.g., Samsung, Intel, TI, Freescale, Micron). 3) They utilize niche processes (e.g., Maxim, Linear, RF Micro Devices, Anadigics). 4) They have a low-cost strategy that involves manufacturing (e.g., Fairchild, On Semiconductor, IDT, Cypress). The industry has sustained a long-term cap-ex to sales ratio of roughly 20%-25%. When cap-ex rises above (or below) this range for too long, oversupply (or shortage) usually results. Note that due to improvements in capital efficiency, the cap-ex rate has drifted lower, from 25-30% in 1980s to mid-1990s to around 20% today.

36

9 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05

19 8

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Industry Basics: Seasonality Industry Basics: Seasonality


The semiconductor industry is somewhat seasonal, based on demand patterns for electronics. Typically, 48% of sales fall in the first half of the year, while 52% fall in the second half. On a quarter-to-quarter basis, sales typically fall in Q1, are flat +/- in Q2 and rise in Q3 and Q4. In general, devices sold into consumer channels (CE, handsets, PCs) are typically more seasonal than those sold to industrial, enterprise, or service provider customers.
Percentage Of Yearly Sales By Quarter
27% Last 20 years Last 15 years 26% Last 10 years Last 5 years
3%
% of Yearly Sales

Historical Quarterly Sequential Growth


6% 5% 4%

25%

% of Yearly Sales

2% 1% 0% 1Q -1% 2Q 3Q 4Q

24%

23%

Last 20 years -2% -3% Last 15 years Last 10 years Last 5 years

22% 1Q 2Q 3Q 4Q

-4%

Source: WSTS, CIBC World Markets Corp.

Source: WSTS, CIBC World Markets Corp.

The semiconductor industry is somewhat seasonal, based on demand patterns for electronics but with a slight lead due to the time it takes to build, test and ship the end product (semiconductors will usually go on the board in the early stages of production). These demand patterns tend to favor Q4, which sees a bump in demand for the holiday selling season; and also for Q3, which sees the holiday build as well as a somewhat less significant back to school season. This is especially true for devices sold into the consumer channel, including consumer electronics, handsets and PCs. Demand for semiconductors sold into enterprise and service provider will see smoother demand, though there can sometimes be a budget flush at the end of the year. Typically, 48% of sales fall in the first half of the year, while 52% fall in the second half. On a quarter-to-quarter basis, sales typically fall in Q1, are flat +/- in Q2, and rise in Q3 and Q4. Note that we have witnessed a slight shift in seasonality from Q4 into Q1 as electronics consumption in Asia becomes more meaningful; Asian consumers have a holiday selling season around Chinese new year, which usually falls in late January or early February.

37

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The Semiconductor Cycle The Semiconductor Cycle


The semiconductor industry is very cyclical, and identifying the phase of the cycle is critical to investment success.
Phase One "The Downturn" Phase Two "The Recovery" Fundamentals
Macroeconomics Economic activity Pricing Logic Analog Commodities Lead times Analog/logic Commodities Capital additions Production line additions New capacity additions OEM/distributors Orders Double ordering Order cancellations Inventory levels Verticle end markets Sell-through Forward outlook Slowdown Picking up Expansion Stable growth

Phase Three "The Expansion"

Phase Four "The Peak"

Decreasing Decreasing Decreasing rapidly

Decreasing to stable Decreasing Decreasing

Increasing Stable to increasing Increasing

Stable Increasing Peaking

Contracting Contracting rapidly

Stable Contracting to stable

Stable to expanding Expanding

Expanding to peaking Peaking - on allocation

No (over-capacity) No (over-capacity)

Some None

Yes Yes

Yes (too many) Yes (too many)

Falling off a cliff No Accelerating Too high

Picking up No Falling Low

Strong No Very few, if any Too low

Frenzied Yes Limited Building

Even slower Cautionary

Growing Cautionary

Strong Cautiously optimistic

Slowing Optimistic

Investment Implications
Group stance Which stocks to buy Underweight Sell across the board; if exposure to the group is mandatory, stick with franchise names Overweight Large cap, franchise companies Overweight to market weight Smaller cap stocks with greater upside potential (the rising tide will fift all boats) Market weight Need to be selective; choose companies that will continue to take market share

Source: CIBC World Markets Corp.

The semiconductor sector is highly cyclical. A typical cycle can generally be broken down into four key phases: Downturn: This phase follows the peak in revenues of the previous cycle and is usually sparked by slowing economic activity. Orders dry up and cancellations begin to happen. OEMs start to work through too-high inventories. Pricing begins to fall as sales fall and capacity built during the boom period pressures pricing. IDMs and foundries cancel pilot line additions and may even close down or consolidate trailing-edge fabs. Recovery: The recovery is a period of stabilization following the free-fall of the downturn. In this phase, prices stabilize but remain at low levels and unit demand starts to pick up with increased economic activity. Orders pick up and cancellations decrease and inventories at IC vendors, distribution and at OEMs subsist at low levels after a long burn-off. In general, capacity is not being added and may even be decreasing. Expansion: Following the recovery comes the new expansion. Economic activity drives end demand, which causes stronger orders and few cancellations. Prices begin to rise and IDMs begin to cautiously add capacity to meet increasing output forecasts. Lead times stretch out and order visibility improves. Peak: Eventually, the expansion phase leads to a peak as economic activity enters a period of stable growth. Prices remain strong as capacity additions from the expansion phase do not yet flow through to hit pricing. Orders continue to arrive at a frenzied pace with limited cancellations. Double ordering begins and inventory builds in the channel. IDMs and foundries continue to add capacity, until a downturn begins.

38

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The Semi Cycle: Sales The Semi Cycle: Sales


Monthly year-over-year sales growth has been volatile over the past few years and has cycled as high as 42% and as low as -2%. The market was in a recovery/upturn from 2002-2004 and has since vacillated between a downturn and a recovery.
$24.0 60% 50% $20.0
Monthly Semiconductor Revenue ($B)

40% 30%
Year-Over-Year Growth

$16.0

20% 10%

$12.0

0% -10%

$8.0

-20% -30%

$4.0

YoY Growth Three-Month Average Sales

-40% -50% -60%

$0.0 Jan-00

Jan-01

Jan-02

Jan-03

Jan-04

Jan-05

Jan-06

Source: WSTS, CIBC World Markets Corp.

The breakout growth of electronic hardware has driven dramatic growth in semiconductor sales over the long term. Still, the cyclical nature of the semiconductor industry has caused phenomenal boom periods and destructive downturns. Fundamental industry characteristics such as long product production cycles, the quick pace of innovation (funded by increasingly techcentric capital markets), volatile pricing, inventory effects and capacity builds have caused wild sales volatility yearto-year. The data reflects this quite clearly: while the mean yearover-year revenue growth rate has been in the low- to mid-teens over any long-term period we look at, the standard deviation for this growth rate has approached or topped 20% in every instance (and in fact spiked in the bubble and post-bubble period). The exhibit to the right displays this data.

Semiconductor Sales Long Term CAGR, YoY Growth Averages and Standard Deviation

40 years 30 years 20 years 15 years 10 years 5 years

2005 CAGR 14% 14% 12% 11% 5% 2%

Mean YoY Gr. 16% 14% 13% 12% 10% 10%

Median YoY Gr. 18% 18% 10% 9% 7% 13%

Standard Deviation 19% 18% 19% 19% 22% 24%

Source: WSTS, CIBC World Markets Corp.

39

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The Semi Cycle: Units and ASPs The Semi Cycle: Units and ASPs
Unit growth has been slightly less volatile than sales; it has has cycled as high as 33% and as low as -3%. Long production cycles and double-ordering cause inventory builds and corrections through each cycle. ASPs have remained volatile and generally add to the boom during periods of strong unit demand (and when inventories build) and exacerbate the downturn when demand softens (and inventories decline).
Unit Shipments
14.0 50% 40% 12.0 30% 10.0
Monthly IC Unit Shipments (B)

Average Selling Price


$1.00 $0.90 $2.50 $0.80 $0.70 $2.00
Monthly IC ASP

$3.00

Monthly Semiconductor ASP

20%
Year-Over-Year Growth

8.0

10% 0%

$0.60 $0.50 $0.40 $1.00 $0.30 $0.20 Three-Month Average Semiconductor ASP $0.50 $0.10 $1.50

6.0

-10% -20% -30%

4.0

2.0

YoY Growth Three-Month Average IC Units -40% -50%

Three-Month Average IC ASP $0.00

0.0 Jan-96 Jan-97 Jan-98 Jan-99 Jan-00 Jan-01 Jan-02 Jan-03 Jan-04 Jan-05 Jan-06

$0.00 Jan-96 Jan-97 Jan-98 Jan-99 Jan-00 Jan-01 Jan-02 Jan-03 Jan-04 Jan-05 Jan-06

Source: WSTS, CIBC World Markets Corp.

Source: WSTS, CIBC World Markets Corp.

Semiconductor unit shipments have also proved volatile, though they generally show more stability than revenues. While the long-term trend has been up, shipments are affected by periods of double-ordering and subsequent inventory corrections, which exaggerate more muted upturns and downturns in end demand. This volatility can be seen best in the last major boom/bust of the late 1990s/2000s, where year-over-year growth in unit shipments ranged from the 40% level on the high side down to -35% on the low side. This range is about 40% smaller than the range for revenue growth, but is still plenty of volatility for such a large industry. The market has been quite resilient since the last major downturn. Since 2002, year-over-year growth in unit shipments ranged from as high as 33% to as low as -3%. This is likely attributable to more effective management of supply chains by both semiconductor vendors and customers; recent inventory builds and corrections over the past few years have been relatively mild. Pricing has also moved around over time, adding to the boom during periods of strong unit demand and exacerbating the downturn when demand decelerates. The commoditization of memory products like DRAM and flash has added to this phenomenon as it represents such a large percentage of industry sales. Note that the trend of semiconductor companies outsourcing production to foundries has mitigated ASP volatility to some extent.

40

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The Semi Cycle: Capacity Utilization The Semi Cycle: Capacity Utilization
Capacity utilization is the link between units and ASPs. As unit production rises, capacity utilization also rises, and producers slow down price decreases (or even raise prices). Eventually, new capacity comes on line to meet demand. In the downturn, capacity becomes a liability as IDMs and foundries try to fill up their fabs by offering price concessions. Eventually, restructuring removes excess capacity and prices stabilize until unit demand returns.
Capacity vs. Production
600 100% 500 90%
Millions of Square Wafer Inches

Capacity Utilization

400 80% 300 70% 200

100

Chip Capacity Chip Production

60%

Three-Month Average W.W. Front End Utilization Three-Month Average W.W. Test Utilization Three-Month Average W.W. Assembly Utilization

Jan-00 Jan-01 Jan-02 Jan-03 Jan-04 Jan-05 Jan-06

50% Jan-00

Jan-01

Jan-02

Jan-03

Jan-04

Jan-05

Jan-06

Source: VLSI, CIBC World Markets Corp.

Source: VLSI, CIBC World Markets Corp.

Capacity utilization provides the link between units and ASPs. During cyclical expansions, rising unit demand pushes utilization higher and IDMs become less willing to offer price concessions to customers. Foundries also slow price declines for wafers to their customers, which in turn prompts fabless competitors to react with their own pricing adjustments. Thus, when utilization rates run high, price curves for high-volume products slow down or in some cases prices actually rise, especially for commodities. Since the positive pricing effects happen at the same time as increasing unit shipments, revenue is doubly affected. Note that lead times also stretch out during this period, as OEMs are forced to order further in advance to guarantee component supply. Rising utilization and prices generally spark IDMs and foundries to bring on more capacity to meet the demand. As this capacity ramps, prices usually stabilize at the higher output level. When unit demand slows, this additional capacity becomes a liability and IDMs begin to offer lower prices to win competitive business in order to fill up their fabs. Foundries also offer lower wafer prices to their fabless customers, which in turn adjust their pricing schemes as they seek to win market share by dropping price. Prices begin to fall; as this happens at the same as unit demand declines (or slows), revenue is doubly affected. Lead times usually contract, as component availability improves and OEMs begin to place orders closer to their needs to take advantage of falling prices. As the downturn progresses, IDMs will take steps to restructure their manufacturing, first by cutting heads, then consolidating facilities or even disposing of old equipment. Eventually, unit demand returns to absorb the lower capacity, prices improve, lead times stretch out, and the cycle begins again.

41

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The Semi Cycle: Capital Spending The Semi Cycle: Capital Spending
Capital spending is the primary source of the semiconductor industrys cyclicality: pricing is dependent on capacity utilization, which in turn is dependent on the interaction between unit demand and capital spending.
During the boom, IDMs increase spending to meet rising unit demand. They inevitably spend too much, sparking price deflation when demand dries up. They then spend too little in the downturn, sparking a cyclical expansion.
Semiconductor Cap-ex
$70 Semiconductor Capital Spending $60 $200 $50
Dollars in Billions
Dollars in Billions

Sales and Cap-ex


$250 Semiconductor Capital Spending Semiconductor Sales Cap-ex to Sales 30%
% of Semiconductor Sales

35%

$150

25%

$40

$30

$100

20%

$20 $50 $10 15%

$0 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

$0 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

10%

Source: VLSI, CIBC World Markets Corp.

Source: VLSI, WSTS, CIBC World Markets Corp.

Though we tend to measure the cycle in terms of semiconductor industry revenue, capital spending holds the key to the industrys cyclicality. Semiconductor industry revenue is, at its core, a function of unit demand and pricing. While unit demand is based on end demand for components (though some elasticity does exist, allowing pricing to influence end demand), pricing is based on capacity utilization, as IDMs and foundries will generally be flexible with pricing so long as they have excess capacity and the fab economics make sense. Capacity utilization, in turn, is dependent on the interaction between wafer demand and capital spending. A typical cycle (peak-to-peak) will begin with an overspend/overcapacity situation. During the expansion preceding the peak, IDMs and foundries increase spending on capacity to meet rising demand. End demand for electronics is difficult to predict, and fab projects take years to complete; consequently they inevitably overspend at the peak of the cycle. This sparks price deflation as demand growth begins to slow, at the same time that new capacity comes online. The downturn brings a wave of restructuring, and IDMs look to slash capital spending to conserve cash and reduce depreciation expenses. The length of the downturn is difficult to predict, and further, the deeper the downturn, the less available incremental funds become for new capital projects. Thus, IDMs inevitably spend too little in the downturn, sparking a new cyclical expansion as prices firm up, demand increases and margins begin to improve. In general, we view 23%-25% of sales to be a normal level of capital spending for the semiconductor industry. When the ratio sits outside this range for any length of time, over/undercapacity usually results.

42

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Past Cycles Past Cycles


$5,000 $4,500 Semiconductor Monthly Revenue ($M) $4,000 $3,500 $3,000 $2,500 $2,000 $1,500 $1,000 2,000 $500 $0 Oct-84 Oct-85 Oct-86 Oct-87 Oct-88 Oct-89 Jun-84 Jun-85 Jun-86 Jun-87 Jun-88 Jun-89 Feb-85 Feb-86 Feb-87 Feb-88 Feb-89 Feb-90 Jun-90 Oct-90 Feb-91 0 6,000

1984-1990
3-Monthly Average Sales 3-Monthly Average Units ASP (on Unit Axis) Cyclical Peak

14,000

$14,000

1990-1995
3-Monthly Average Sales 3-Monthly Average Units ASP (on Unit Axis) Cyclical Peak

25,000

Semiconductor Monthly Unit Shipments (M)

12,000

$12,000
Semiconductor Monthly Revenue ($M)

10,000

$10,000 15,000

8,000

$8,000

$6,000

10,000

4,000

$4,000 5,000 $2,000

$0 Feb-91 Feb-92 Feb-93 Feb-94 Feb-95 Feb-96 Jun-90 Jun-91 Jun-92 Jun-93 Jun-94 Jun-95 Oct-90 Oct-91 Oct-92 Oct-93 Oct-94 Oct-95

1995-1997
$14,000 25,000 $20,000 $18,000 $12,000
Semiconductor Monthly Unit Shipments (M)

1997-2000
40,000 3-Monthly Average Sales 3-Monthly Average Units ASP (on Unit Axis) Cyclical Peak 35,000 30,000 $14,000 $12,000 $10,000 $8,000 $6,000 10,000 $4,000 $2,000 0 $0 Apr-98 Apr-99 Aug-97 Aug-98 Aug-99 Apr-00 Aug-00 Feb-98 Feb-99 Feb-00 Dec-97 Dec-98 Dec-99 Dec-00 Feb-01 Jun-97 Jun-98 Jun-99 Jun-00 Oct-97 Oct-98 Oct-99 Oct-00 5,000 0 25,000 20,000 15,000
Semiconductor Monthly Unit Shipments (M)

20,000
Semiconductor Monthly Revenue ($M)

$16,000
Semiconductor Monthly Revenue ($M)

$10,000 15,000

$8,000

$6,000

10,000

$4,000 3-Monthly Average Sales 3-Monthly Average Units ASP (on Unit Axis) Cyclical Peak Sep-95 Sep-96 May-96 May-97 Sep-97 Mar-96 Mar-97 Nov-95 Nov-96 Nov-97 Mar-98 Jan-96 Jan-97 Jan-98 Jul-95 Jul-96 Jul-97 5,000

$2,000

$0

Source: WSTS, CIBC World Markets Corp.

We depict above four historical semiconductor cycles, showing revenue, units and ASPs (we define a cycle as the period of time between revenue peaks on a three-month average basis; we also display the six-month period before and after each cycle). There are a number of consistent trends that can be seen in these charts: Each cyclical downturn in sales begins with a decline in units followed by a more severe drop in ASPs. Units typically recover faster than ASPs, if ASPs recover at all. Units tend to grow at a faster rate than revenue during a given cycle; however, when units decline, they rarely decline more than revenue. This reflects the elasticity of demand and OEMs trying to anticipate future changes in prices for their components. Cycles can last as little as two years or as long as six years. There is little consistency between cycles in the growth rate over the course of the cycle, as this is dependent more on macro demand trends.

43

Semiconductor Monthly Unit Shipments (M)

20,000

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The Last Cycle: A Long Recovery The Last Cycle: A Long Recovery
The 2001 downturn was the steepest and longest in industry history. Demand contracted following Y2K and the telecom bubble, and the recovery took several quarters as excess supply was digested. The sector enjoyed a nice expansion in 2003 and 1H04 before peaking later in 2004.
$20,000 $18,000 $16,000
Semiconductor Monthly Revenue ($M)

45,000 40,000 35,000 30,000 25,000


Semiconductor Monthly Unit Shipments (M)

$14,000 $12,000 $10,000

20,000 $8,000 $6,000 $4,000 $2,000 $0 May-00 May-01 May-02 May-03 May-04 Feb-01 Feb-02 Feb-03 Feb-04 Aug-00 Nov-00 Aug-01 Nov-01 Aug-02 Nov-02 Aug-03 Nov-03 Aug-04 Nov-04 Feb-05

Downturn

Recovery

Expansion

Peak

15,000 10,000 5,000 0

3-Monthly Average Sales 3-Monthly Average Units ASP (on Unit Axis) Cyclical Peak

Source: WSTS, CIBC World Markets Corp.

The last semiconductor cycle began in late 2000 and ended in late 2004. The year 2001 proved to be the industrys worst year ever, with sales falling 32% year over year. The severity of the downturn was a direct consequence of the magnitude of the preceding boom period: surging demand for telecom, Internet and IT infrastructure dried up, eliminating a major driver of end demand. Inventory built up from months and months of double-ordering remained in the channel, further exacerbating unit shipments. At the same time, capacity that had been built to support unfulfilled growth forecasts came online to slaughter pricing. Note that unlike other downturns, the 2001 downturn was driven by a sharp, sustained falloff in end demand; thus, the supply-side pricing lever only kicked in after units had declined. Note that the first down month was November 2000 and sales finally hit bottom in January 2002, at 54% of peak levels. The recovery phase was slower and longer than in past cycles, not surprising given the sharpness of the decline. Monthly sales at the peak only slightly exceeded the 2000 peak, and it took nearly four years to get there. The industry went through some pretty dramatic change over the course of the cycle, with the trend toward outsourcing sharply accelerating as many IDMs sought to reduce the systematic risk inherent in their business models by leveraging outside supply. Some went as far as turning fabless (e.g. Agere, AMCC, Vitesse), while many adopted strategies of outsourcing a percentage of their supply to help smooth peaks and valleys (e.g., Texas Instruments, STMicroelectronics, Infineon, Freescale, etc.).

44

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The Current Cycle: Stalled Momentum The Current Cycle: Stalled Momentum
The current cycle has hovered between a downturn and a recovery for the past two years. In 2005, oversupply muted growth prospects, though end demand actually held up quite well. In 2006, supply tightened but weaker demand sparked a mid-cycle inventory correction.
$25,000 60,000

40,000 $15,000 30,000 $10,000 20,000

$5,000

3-Monthly Average Sales 3-Monthly Average Units ASP (on Unit Axis) Cyclical Peak

10,000

$0 Feb-05 Feb-06 Apr-05 Aug-04 Dec-04 Aug-05 Dec-05 Apr-06 Aug-06 Oct-04 Jun-04 Jun-05 Oct-05 Jun-06

Source: WSTS, CIBC World Markets Corp.

The current cycle has hovered between a downturn and a recovery for the past two years. The cycle began in 2005 with a moderate oversupply situation, characterized by excess inventory at customers, declining lead times, and over capacity at foundries and IDMs. Still, end demand actually held up and inventory issues were largely worked through exiting the year; it appeared we were headed for a recovery in 2006. The demand environment moderated in 2006 for the major semiconductor verticals. Though inventories entered the year at lean levels, they built quickly in 1H06 as back end tightness sparked double ordering. As the back end eased in 2H06 along with more moderate demand, an inventory correction began that is expected to last at least through year end and possibly into early 2007. We still believe there is a chance for the start of an upturn in 2007, though this will be largely dependent on the demand environment and the health of overall economy.

45

Semiconductor Monthly Unit Shipments (M)

$20,000
Semiconductor Monthly Revenue ($M)

50,000

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Fundamentals: Gross Margin Fundamentals: Gross Margin


Gross margins for the industry are generally in the mid40% range, though they compress during periods of low utilization or difficult pricing. Outsourcing trends have resulted in less margin volatility since the 2000 bubble.
60% Gross Margin Last ten years average Last five years average Last three years average

50%

40%

30%

20%

10%

0%
D ec M -98 a Ju r-99 n Se -99 pD 99 ec M -99 a Ju r-00 n Se -00 p D -00 ec M -00 a Ju r-01 n Se -01 pD 01 ec M -01 a Ju r-02 n Se -02 pDe 0 2 c M -02 ar Ju -03 n Se -03 p D -03 ec M -03 a Ju r-04 n Se -04 pD 04 ec M -04 a Ju r-05 n Se -05 p D -05 ec M -05 a Ju r-06 n Se -06 p06
Source: Bloomberg, CIBC World Markets Corp.

Gross margins for semiconductor companies have generally hovered around the 40%-50% range. A hefty portion of COGS is driven by depreciation of plant and equipment, with the remainder in fixed and variable labor and material costs. In general, companies whose products are more differentiated and proprietary tend to have higher margins; this would include vendors of analog and mixed signal ICs, PC microprocessors, DSPs, FPGAs, ASICs and special purpose logic (ASSPs). Companies whose products are more commoditized and thus differentiate themselves based on their manufacturing competencies tend to have lower margins; this would include vendors of memory, microcontrollers, discretes, optoelectronics and general purpose logic. Gross margins tend to compress rapidly during periods of difficult pricing or low utilization and expand quickly when prices firm and utilization is on the rise. The effect can be exaggerated by industry accounting practices employed by many IDMs, whereby fixed costs are allocated to inventory each quarter at an estimated standard utilization rate and excess fixed costs (positive or negative) are charged to the cost of goods line. In periods of low utilization, actual production trails the estimate, hurting gross margin in the current period. In periods of high utilization, actual production leads the estimate, leading to negative charges to cost of goods and a boost to gross margin. Note that outsourcing trends have mitigated margin volatility since the 2001 downturn, and now only a handful of semiconductor vendors (mostly memory vendors) do not outsource at least 10% of their supply needs. This eliminates much of the margin leverage we are used to seeing in IDM business models. Even amid challenging conditions in 2005 and 2006, industry gross margins still held above 40%.

46

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Fundamentals: Operating Margins Fundamentals: Operating Margins


Operating margins have historically averaged near 15%. Margins turned negative in 2001, and were slow to recover as many vendors were unwilling to restructure. Since 2004, more consistent demand and a greater focus on profitability have generated more consistent margins.
30% 25% 20% 15% 10% 5% 0%
D ec M -98 a Ju r-99 n Se -99 pD 99 ec M -99 a Ju r-00 n Se -00 pD 00 ec M -00 a Ju r-01 n Se -01 p D -01 ec M -01 a Ju r-02 n Se -02 pDe 0 2 c M -02 ar Ju -03 n Se -03 p D -03 ec M -03 a Ju r-04 n Se -04 pD 04 ec M -04 a Ju r-05 n Se -05 p D -05 ec M -05 a Ju r-06 n Se -06 p06

Operating Margin Last ten years average Last five years average Last three years average

-5%

-10% -15% -20%


Source: Bloomberg, CIBC World Markets Corp.

Operating margins for semiconductor companies have historically averaged around 15%, with the average company spending about 20%-25% of sales on R&D and SG&A expenses. Operating margins above 20% are not uncommon, especially for best-in-breed companies; or even for laggards in the expansion and peak phases of the cycle. Operating margin turned negative during the 2001 downturn and only climbed back to historical levels as the expansion was getting under way in late 2003. Operating margin did not, however, exceed 20% for any length of time even at the peak of the cycle, reflecting the fact that many vendors were reluctant to take deep enough restructuring actions during the downturn, especially on the R&D line. The technology boom of the late 1990s has left many semiconductor vendors flush with cash, be they with profitable business models or not. In the current cycle, operating margins have held steady in the mid-teens, despite the challenging environment. Many semiconductor vendors have been focused on maintaining profitability and have undertaken some level of restructuring, including the move to more flexible outsourcing models, headcount reductions, and strategy shifts. Demand, too, has been relatively consistent, with just shallow cyclicality even during the 2004 and 2006 inventory corrections. Note that average operating margins have not dipped below 14% since 2004.

47

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Fundamentals: Receivables Fundamentals: Receivables


Receivables at semiconductor vendors have historically held 45-50 days. Receivable days sometimes spike during demand downturns as OEMs conserve cash, stretching payment cycles, but they usually correct after a couple of quarters.
70 DSOs Last ten years average Last five years average Last three years average 60

50

40

30

20

10

D ec M -98 a Ju r-99 n Se -99 p D -99 ec M -99 a Ju r-00 n Se -00 p D -00 ec M -00 a Ju r-01 n Se -01 p D -01 ec M -01 a Ju r-02 n Se -02 p De - 0 2 c M -02 ar Ju -03 n Se -03 p D -03 ec M -03 a Ju r-04 n Se -04 p D -04 ec M -04 a Ju r-05 n Se -05 p D -05 ec M -05 a Ju r-06 n Se -06 p06
Source: Bloomberg, CIBC World Markets Corp.

Turning to the balance sheet, receivables at semiconductor vendors have historically held 45-50 days and are seasonally affected. During downturns, receivables can spike as OEMs look to conserve cash by stretching payment cycles. The DSO ratio can also be affected by the quick drop in the sales numerator.

48

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Fundamentals: Inventories Fundamentals: Inventories


Inventories at semiconductor vendors have trended up over the past few years. OEMs are reluctant to hold inventories except in the tightest supply environment. At the same time, the rise in outsourcing allows chip vendors to hold more inventory with less risk.
90 80 70 60 50 40 30 20 10 D ec M -98 a Ju r-99 n Se -99 p D -99 ec M -99 a Ju r-00 n Se -00 p D -00 ec M -00 a Ju r-01 n Se -01 p D -01 ec M -01 a Ju r-02 n Se -02 p De - 0 2 c M -02 ar Ju -03 n Se -03 p D -03 ec M -03 a Ju r-04 n Se -04 p D -04 ec M -04 a Ju r-05 n Se -05 p D -05 ec M -05 a Ju r-06 n Se -06 p06
Source: Bloomberg, CIBC World Markets Corp.

Inventory Days Last ten years average Last five years average Last three years average

Inventory days at semiconductor vendors have varied considerably over the past few cycles. Inventory days rose dramatically during the 2001 downturn (peaked at 80 days), as the cost of sales numerator declined rapidly and while OEMs worked through their own excess inventories. Inventories began to trend down in 2002 and 2003 as vendors held production firm as the demand environment improved, but then rose in 2004 as IC vendors put stocks in place to support more optimistic forecasts in a period of tight supply. As supply eased, inventory levels peaked once again at 78 days in 1Q05 and took most of the year to work through. Inventories gradually crept up through most of 2006 and now sit near historical highs at 79 days as the demand environment proved softer than many had anticipated. Though inventory levels are highly tied to the semiconductor cycle, the historical increase in inventory days is also attributable to OEMs reluctance to hold excess inventory. As a result, semiconductor vendors have shown increased focus on developing an inventory strategy to meet their customers needs, which may mean relatively high inventory stocks for the foreseeable future. The rise of outsourcing supports this increased inventory with less risk. Note that historical averages have been increasing over time (10-year average is 67 days, 5-year average is 72 days, and the 3-year average is 74 days).

49

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Fundamentals: Inventories Fundamentals: Inventories


Even more important than inventories at semiconductor vendors themselves are inventories downstream, at the OEMs, EMS providers, and distributors. These are a key variable exacerbating the cyclicality of the industry. When these inventories run too low, OEMs tend to double-book in order to guarantee supply to meet demand forecasts. When they run too high, OEMs tend to cancel and push out orders until end demand returns.
PC & Server OEMs
50 45 40 35 30 25 20 15 10 5 5 ec M -98 ar Ju 99 n Se -99 pD 99 ec M -99 ar Ju 00 n S -00 ep D -00 ec M -00 ar Ju 01 n Se -01 p De -01 c M -01 ar Ju 02 n Se -02 pDe 02 c M -02 ar Ju 03 nSe 03 pD 03 ec M -03 ar Ju 04 n Se -04 pD 04 ec M -04 ar Ju 05 n Se -05 pDe 05 c M -05 ar Ju 06 n S -06 ep -0 6 D

Hard Disk Drive OEMs


35 30 Hard Disk Drive Inventory Days Last ten years average Last five years average Last three years average

PC and Server OEM Inventory Days Last ten years average Last five years average Last three years average

25

20

15

10

ec M -98 ar Ju 99 n Se -99 pD 99 ec M -99 ar Ju 00 n Se -00 pD 00 ec M -00 ar Ju 01 n Se -01 pDe 01 c M -01 ar Ju 02 nSe 02 pD 02 ec M -02 ar Ju 03 n Se -03 pD 03 ec M -03 ar Ju 04 n Se -04 pD 04 ec M -04 ar Ju 05 n Se -05 pDe 05 c M -05 ar Ju 06 n Se -06 p06 D

Source: Bloomberg, CIBC World Markets Corp.

Source: Bloomberg, CIBC World Markets Corp.

Continued...

Even more important than inventories at semiconductor vendors themselves are inventories downstream, at the various classes of OEMs and EMS providers, which are a key variable in exacerbating the cyclicality of the industry. When these inventories run too low, OEMs tend to double-book in order to guarantee supply to meet demand forecasts. When they run too high, OEMs tend to cancel and push out orders until end demand returns. We summarize below (and on the next page) the state of inventories by end market: PC and Server OEMs: The decline in inventory days at PC and server OEMs has been nothing short of remarkable, with inventory days dropping from 65 days to 20-25 days during the past 10 years. This drop has been critical in reducing PC price points and retaining profit margins, especially since the cost of microprocessors and DRAM two major components in the PCare among the fastest dropping in the industry (when measured on a per-MHz or per-bit basis). The move to just-in-time inventory has also played a huge role, and the impact by Dell alonewith its 5 days of inventory on handhave affected the industry figure significantly. At present, PC and server hardware OEM inventories are at slightly elevated levels. Hard Disk Drive: HDD drive inventory days tend to be relatively lowabout 20-25 daysbut can fluctuate considerably as OEMs contend with choppy demand and manage the transition to larger densities and more integrated designsevents that do not always coincide with the ramp of new PC models. Most recently, HDD inventory sits at 30 days, near an all time high, as the industry copes with consolidation (Seagate/Maxtor) and a soft PC demand environment.

50

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Fundamentals: Inventories Fundamentals: Inventories


Wireline Comm OEMs
140 Wireline OEM Inventory Days 120 Last ten years average Last five years average Last three years average 100 60 80 50 40 30 40 20 20 10 ec M -98 ar Ju 99 n Se -99 pD 99 ec M -99 ar Ju 00 n S -00 ep D -00 ec M -00 ar Ju 01 n Se -01 pDe 01 c M -01 ar Ju 02 nSe 02 pD 02 ec M -02 ar Ju 03 n Se -03 pD 03 ec M -03 ar Ju 04 n Se -04 pD 04 ec M -04 ar Ju 05 n Se -05 p D -05 ec M -05 ar Ju 06 n Se -06 p06 D

Wireless Comm OEMs


90 Wireless OEM Inventory Days 80 70 Last ten years average Last five years average Last three years average

60

D ec M -98 ar Ju 99 n Se -99 pD 99 ec M -99 ar Ju 00 n Se -00 pD 00 ec M -00 ar Ju 01 n Se -01 pD 01 ec M -01 ar Ju 02 n Se -02 pD 02 ec M -02 ar Ju 03 n Se -03 pD 03 ec M -03 ar Ju 04 n Se -04 pD 04 ec M -04 ar Ju 05 n Se -05 pD 05 ec M -05 ar Ju 06 n Se -06 p06

EMS Providers
90 EMS Inventory Days 80 70 60 50 40 30 20 10 10 De cM 98 ar Ju 99 nSe 99 pDe 99 cM 99 ar Ju 00 nSe 00 pDe 00 cM 00 ar Ju 01 nSe 01 pDe 01 cM 01 ar Ju 02 nSe 02 pDe 02 cM 02 ar Ju 03 nSe 03 pDe 03 cM 03 ar Ju 04 nSe 04 pDe 04 cM 04 ar Ju 05 nSe 05 pDe 05 cM 05 ar Ju 0 6 nSe 06 p06

Distributors
50

Last ten years average Last five years average Last three years average

40

30

20 Distributors Inventory Days Last ten years average Last five years average Last three years average

D ec M -98 ar Ju 99 n Se -99 p D -99 ec M -99 ar Ju 00 n Se -00 pD 00 ec M -00 ar Ju 01 n Se -01 pD 01 ec M -01 ar Ju 02 n Se -02 pDe 02 c M -02 ar Ju 03 n Se -03 pD 03 ec M -03 ar Ju 04 n S -04 ep D -04 ec M -04 ar Ju 05 n Se -05 p D -05 ec M -05 ar Ju 06 n Se -06 p06

Source: Bloomberg, CIBC World Markets Corp.

Wireline OEMs: Given the long product design and life cycles associated with wireline infrastructure equipment, wireline OEMs have traditionally held higher levels of inventory (60-80 days range) than many tech peers. Note that inventory levels were at their worst in 2000/2001 due to the telecom meltdown (> 100 days), but have since been managed much more prudently. OEMs have adjusted purchasing behavior and have taken advantage of shorter lead times from suppliers. Even after a modest 2004 build, inventory still sat at 70 days, way below the historical peak. OEM inventories are now at slightly elevated levels. Wireless OEMs: Wireless OEMs have historically ranged around 40-60 days and are declining over time. They tend to be choppy given the pace of new standard introductions, and there is some seasonality, too inventories usually jump in Q1, stabilize and decline sharply in Q4. Inventories are now at slightly elevated levels at 41 days. EMS Providers: EMS inventories have historically ranged around 40-50 days. Inventory was healthy through most of 2004 and 2005, but has been building recently and are now at elevated levels at 49 days. Note that some of the variability in EMS inventory days is due to OEMs shifting inventories back and forth. Distributors: Distributors build their businesses around efficient inventory management and have been able to limit inventories historically to 35-40 days. Inventory last spiked in 2000/2001 (near 50 days) due to double-ordering. Since then, distributor inventories have been quite healthy and currently sit below 35 days.

51

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Fundamentals: Debt Fundamentals: Debt


Given the long-term nature of manufacturing assets, the industry maintains some level of debt. The willingness of investors to fund companies with equity capital, however, has kept the debt/equity ratio for the industry to less than 20% for the most part.
28% Debt to Equity Last ten years average Last five years average Last three years average 24%

20%

16%

12%

8%

4%

0%
D ec M -98 a Ju r-99 n Se -99 p D -99 ec M -99 a Ju r-00 n Se -00 p D -00 ec M -00 a Ju r-01 n Se -01 p D -01 ec M -01 a Ju r-02 n Se -02 p De - 0 2 c M -02 ar Ju -03 n Se -03 p D -03 ec M -03 a Ju r-04 n Se -04 p D -04 ec M -04 a Ju r-05 n Se -05 p D -05 ec M -05 a Ju r-06 n Se -06 p06
Source: Bloomberg, CIBC World Markets Corp.

Given the long-term nature of semiconductor manufacturing assets, the industry maintains some level of debt, primarily (though not exclusively) concentrated in the hands of IDMs and outsourcers. The willingness of investors to fund semiconductor companies (given the correct perception that semiconductor companies are high-tech companies but that they have real assets) with equity capital has limited the need for debt, however, and debt/equity for the industry (excluding Japan and Taiwan, which have different funding pools) has held at or below 20%. In 1999 and 2000, the debt/equity ratio penetrated 15% as fabless and IDM semiconductors alike took advantage of inflated multiples in the equity market to raise capital. As valuations fell, however, the attractiveness of issuing further equity diminished at the same time that interest rates for corporate, convertible and even high-yield debt have plummeted. The debt/equity ratio thus rose to over 23% in 2003, its highest level in ten years. We note that the ratio was somewhat affected by writedowns of the equity denominator during the 2001 downturn; this included not only inventories and manufacturing assets but a huge amount of intangible assets (primarily acquisition-related), as well as investments in public and private affiliates and even deferred tax assets. More recently, the ratio has held steady below 20%. This is due largely to the higher profitability levels at semiconductor vendors, which have driven up the equity line. Some limited consolidation has also created some book equity as acquirers pay premiums vs. book for acquired assets. Note that some private equity activity in 2006 is likely understating the debt position of the industry at present.

52

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Section III: Market Segments Section III: Market Segments

This section summarizes the major product and end market segments of the industry as well as the competitors within each. Topics include: Semiconductor Device Types
- SIA Framework: What A Device IS - CIBC Framework: What A Device DOES

Key Competitors
- By Product - By End Market

53

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Semiconductor Device Types Semiconductor Device Types


The SIA Framework
What a device IS
Discretes, Optoelectonics, and Sensors - Includes all non-integrated circuit semiconductor devices. A discrete is a single transistor in a package. Sensors are discrete devices that measure real-world input. Optoelectronics are discrete devices that produce or measure light. Analog - Devices used to process real-world signals using electronic voltage patterns that represent the original. Includes SLICs (standard linear components) and ASSPs (application-specific analog processors). MOS Microcomponents - All digital processors, including microprocessors (MPUs), microcontrollers (MCUs), and digital signal processors (DSPs). MOS Logic - Includes all non-microcomponent digital logic, including ASICs (custom logic), ASSPs (standard specialty logic products), FPGAs (programmable logic), and general purpose logic. MOS Memory - Memory devices are used to store data either for short periods of time or permanently. Includes volatile (DRAM, SRAM) and non-volatile (flash, ROM) memory.

The Semiconductor Industry Association (SIA), the industry body that reports monthly semiconductor industry sales, divides the industry into five broad categories and generally focuses on identifying the type of device sold. When we discuss industry segmentation in terms of the type of semiconductor product, we will generally use the SIA framework. Discretes, Optoelectronics and Sensors: Includes all nonintegrated circuit semiconductors. Discretes are devices that contain just a single transistor in a package. These include diodes, small signal and switching transistors, power transistors, rectifiers, thyristors and other discretes. Optoelectronics include displays, lamps, couplers, laser devices, image sensors, infrared and other optoelectronics. Sensors include those devices that measure temperature, pressure, displacement, velocity, acceleration, stress, strain, or any other physical, chemical, or biological property. They also include actuators, which are responsible for movement.

The SIA Monthly Blue Book, Which Tracks Monthly Billings Data

September, 2006

SEMICONDUCTOR INDUSTRY BLUE BOOK


NET BILLINGS BY PRODUCTS & REGION

Administered by WSTS, Inc. Compiled by Mohler, Nixon & Williams in Campbell PricewaterhouseCoopers LLP in Zurich and ASG Audit Corporation in Tokyo

54

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Analog: Analog circuits are those used to process real world analog signals, using electronic voltage patterns within the device to represent the original signal (the word analog is related to analogy). Analog circuits are divided into two broad categories: SLICs and ASSPs. Standard linear integrated circuits (SLICs) are standard analog components used in multiple applications, usually purchased off the shelf, often through distribution. Key types are amplifiers, interface circuits, voltage regulators, data converters and comparators. Application specific standard products (ASSPs) are more specializedthey are targeted at individual applications (though not individual customers or platformsthat would be classified as a custom ASIC). Analog ASSPs are segmented by application, specifically: automotive, consumer, computer, telecom, industrial and other.

MOS Microcomponents: This includes digital processors, including microprocessors, microcontrollers and DSPs. Digital devices do not deal in the specific voltages of electrons, and instead deal in 1s and 0s1 being the presence of a voltage and 0 being the absence of a voltage. The 1s and 0s can be used to represent just about anything, so long as they conform to some predefined pattern or code as dictated by software. Note that since digital processors deal only in digital signals, any real world input must be converted to digital before it can be processed by a digital device. The microcomponent segment includes the more generic digital IC devicesthose that can be used in a variety of applications for computation or signal processing. Microprocessors (CPUs or MPUs) include large, complex processors found in computing and other compute-intensive applications and typically form the brains of an entire electrical system. Microcontrollers (MCUs) are smaller stand-alone processors that perform dedicated or embedded computer functions within a system and are usually designed with on-chip RAM, ROM and logic in order to minimize support chip count. Digital signal processors (DSPs) are specialized digital processors typically used to process real-time data. We note that, until 2001, the microcomponents category also included microperipherals, which are dedicated logic devices used to augment MPU or MCU system performance, such as graphics controllers, PC chipsets, Ethernet and modem controllers and hard disk drive controllers; these devices are now classified in the special purpose logic category.

MOS Logic: This segment includes all non-microcomponent related digital logic circuits, including gate array and standard cells (ASICs), field programmable logic devices (PLDs or FPLDs), display drivers, general purpose logic and special purpose logic/microperipherals (which are somewhat synonymous with ASSPs). MOS Memory: Memory devices are used to store data in silicon, either for short periods of time or permanently. Volatile memory includes those memory devices that lose stored information when the power is lost. These include DRAM (Dynamic Random Access Memory) and SRAM (Static Random Access Memory). Non-volatile memory retains the stored information when the system is powered off. It includes NOR and NAND flash, mask programmable ROM, EPROM and EEPROM.

55

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Semiconductor Device Types Semiconductor Device Types


The CIBC Framework
What a device DOES
Analog Processing - Devices used to process real-world signals. Includes SLICs (standard analog components) and ASSPs (applicationspecific ICs). Digital Processing - All digital devices used to process signal converted from the analog world. Includes digital signal processors (DSPs), custom and standard digital logic and micro-peripherals, some microcontrollers, and any other digital device not responsible for system control or memory storage. Processors - All processors that control the electronic device at the system level, including all microprocessors (MPUs) and most microcontrollers (MCUs). Memory - All volatile (DRAM, SRAM) and non-volatile (EPROM, flash) memory. Identical to SIA category. System Enabling Devices - analog, digital, and discrete devices used as enabling factors on the board, including power circuits, glue logic, switches, clocks, and discretes.

While the SIA framework is an accurate segmentation of the industry by device type, we often look at the semiconductor industry by end market, with an eye toward system-level design. For this purpose, we prefer to use a simpler, more functional segmentation to help investors understand the various devices and how they fit together in an electrical system. In this framework, we segment the market into five types of devices; based on the function they perform within the system. Note that we display an icon next to each type of device, which we will consistently use in all block diagrams: Analog Processing: This includes all analog devices used to process real world signals. Analog ICs deal with complex input from the real world, including sound, light, video, radio waves, electrical signals, temperature, pressure, displacement, velocity, acceleration, stress, strain, or any other physical, chemical, or biological property. They also convert them to digital signals for processing by the digital processors or the system CPU. This category would include most of what is found in the SIA product category for analog devices, but would exclude some enabling analog devices such as power management and timing circuits, whose function is secondary to the function of the electrical system.

56

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Digital Processing: This includes all digital devices used to perform specialized processing of signals converted from the analog world. This includes DSPs used in wireless and broadband, specialized logic used in datacom and telecom, audio/video processors in consumer electronics, or any other digital processor that does not perform overall system control. This category would generally include most of the SIAs MOS logic category (excluding general purpose logic), along with most DSPs and some MCUs. Processors: This includes processors that perform system control. They are usually highly programmable and run the system software, where applicable. Examples are PC microprocessors, application processors in cellular phones, control plane processors in networking equipment, or any other microcontroller that controls the majority of system functions. This category would include all microprocessors, most microcontrollers and some DSPs and logic components, depending on how they are used in the system. Memory: Identical to the MOS memory category described in the SIA framework, this category includes all volatile (DRAM, SRAM) and nonvolatile (flash, EPROM, etc.) memory. System Enabling Devices: This includes all discrete, analog and digital devices that are used on the board as enabling devices, rather than for processing or storage. Typical functions including switching signals between ICs on the board, stepping up or stepping down voltages, performing system timing and sensing outside input. Examples of devices include discretes, optoelectronics and sensors, as well as power circuits, timing devices, glue logic and some programmable logic devices.

57

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Example: A Basic Electrical System Example: A Basic Electrical System


Provides temporary or permanent storage of data used by the system processor Controls power flow into the system as well as battery management in portable devices

Processes the signal at the highest level, implements system software, performs overall system control functions

Performs timing functions to enable system performance Senses real world input

Memory Memory Clock

Microprocessor/ Microcontoller/ Application Processor

Power Supply
Aids in system functioning by controlling signal flow, power delivery, etc.

Real World Input

Sensor

Analog Processor

Digital Processor

Digital Processor

Analog Processor

Real World Output

Source: CIBC World Markets Receives real-world input, performs analog processing, converts it to digital Performs digital processing on the incoming signal Performs digital processing on the outgoing signal Converts the digital signal to analog and produces real-world output

The diagram above depicts an electrical system and identifies the devices on the board using the CIBC framework. In the diagram, some real world input is received, interpreted by a sensor (an enabling device) and processed in an analog format by an analog processor. The analog processor converts the signal to digital and passes it to a digital processor, which performs some more intensive digital processing on the signal. The digital processor then passes it to the microprocessor/microcontroller/application processor, which, using software, decides what to do with the signal and can make any changes to it if needed. At some point, the processor sends out the signal to another digital processor, which performs some digital processing and passes the signal on to an analog processor, which converts the signal to analog and produces some real world output. Devices not directly in the signal chain include memory, which is used for short-term or long-term storage of signals, and other enabling devices, such as power management, discretes, timing devices, switches, etc. A simple example would be a digital camera. When the shutter button is activated, an image sensor (enabling device) receives the visual input, which is processed by an analog front end (analog processor) and converted to a digital format. A DSP or an ASIC (digital processor) then encodes and compresses the signal and tells a microcontroller (microprocessor/control processor) that a picture has arrived. The picture file is then stored in flash memory (memory), and the microcontroller tells an LCD controller (digital processor) to produce an image on the LCD screen. The LCD controller then directs signals from row and column drivers to produce the image on the screen. The table on the following page displays additional examples.

58

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

End Market Application Computing Desktop PC

Analog Processors

Digital Processors

MPU/Control Processor

Memory

System Enabling Devices

Graphics interface, sound Chipset, graphics Microprocessor interface, Ethernet PHY, processor, sound processor modem line driver Graphic/sound interface, Ethernet PHY, modem, WLAN/Bluetooth radio, power management Read channel, preamp, motor driver Microprocessor Chipset, graphics processor, sound processor, WLAN/Bluetooth baseband/MAC Servo ASIC, hard disk controller Microcontroller (embedded in hard disk controller)

DRAM, ROM, Power ICs, clocks, flash switches, discretes DRAM, ROM, Power ICs, clocks, flash switches, discretes, LCD drivers DRAM, SRAM, flash Magnetic sensors, power ICs, clocks, switches, discretes

Notebook PC

Hard disk drive

Communications Wireless handset Power amp, radio (RF/IF), Digital baseband (DSP), Bluetooth baseband/MAC audio codec, Bluetooth radio, analog baseband Ethernet switch Ethernet PHY Ethernet MAC, switch chipset, CAM/NSE, programmable logic Applications processor (can be embedded in DSP) MIPS/PowerPC processor Flash, SRAM, Sensors, power ICs, low-power clocks, switches, DRAM discretes, LCD drivers DRAM, SRAM, FIFO/ multi-port SRAM Power ICs, clocks, switches, discretes

Wireless LAN access point

Radio (RF/IF), power amp Baseband/MAC Framer, FEC, network processor, traffic manager DOCSIS MAC/processor (cable modem), data pump DSP (DSL) Fibre Channel controller ASIC

MIPS/ARM/PowerPC DRAM, SRAM Power ICs, clocks, processor switches, discretes MIPS/PowerPC processor Communications processor FIFOs/Multiports, DRAM, SRAM DRAM, SRAM, flash Optoelectronics, power ICs, clocks, switches, discretes Sensors, power ICs, clocks, switches, discretes Power ICs, clocks, switches, discretes

SONET/SDH line PMD, CDR, SONET card transceiver/PHY, backplane SerDes Cable/ADSL modem Tuner/line driver, analog front end

Fibre Channel Fibre Channel Host Bus Adapter SerDes/PHY (HBA) Consumer Digital set-top box/DVR Tuner, analog front end (mux/demux)

Microprocessor (can SRAM, flash be integrated into FC controller)

MPEG decoder/encoder, graphics processor

Microcontroller

DRAM, flash, ROM

Power ICs, clocks, switches, discretes CCD/CMOS image sensor, power ICs, clocks, switches, discretes, LCD drivers Power ICs, clocks, switches, discretes, LCD drivers Sensors, power ICs, clocks, switches, discretes Sensors, power ICs, clocks, switches, discretes, LCD drivers (handheld)

Digital still camera A/D converter, Bluetooth/ Digital image processor, WLAN radio ASICs, Bluetooth/WLAN baseband/MAC

Microcontroller, Flash, DRAM application processor

PMP/MP3 player

Audio driver, D/A converter Laser supply driver, spindle driver, servo driver, D/A converter

DSP, audio/video processor/ASSP Front end decoder, servo processor, CD/DVD read channel DSP, MPEG decoder, audio decoder

Microcontroller

Flash

DVD player

Microcontroller

DRAM, SRAM, flash

Video game console

Audio driver, video driver, Graphics processor, ASICs, Microprocessor Bluetooth/WLAN radio, Bluetooth/WLAN Ethernet PHY baseband/MAC, Ethernet MAC

DRAM, SRAM, flash

Source: CIBC World Markets Corp.

59

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Key Semiconductor Competitors Key Semiconductor Competitors


Semiconductor Market Top 10 Competitors
Intel (US)The largest semiconductor manufacturer in the world at about 15% of the market, Intel dominates the PC microprocessor segment (with about 75%-80% unit share) and is also a top supplier of PC chipsets, NOR flash memory, embedded microprocessors, and PC networking components. Samsung (Korea)The diversified electronics giant is No. 1 in DRAM, SRAM, NAND flash, and display drivers; it also supplies microcontroller and logic devices. Texas Instruments (US)The top supplier of analog ICs, DSPs, and components for handsets. TI is also a big player in broadband, PC peripherals, and digital consumer, and has smaller efforts in microcontrollers, ASICs, and standard logic. TI also has a MEMS division called DLP which supplies digital TVs and projectors. Toshiba (Japan)The semiconductor unit of the electronics conglomerate, Toshiba is the top supplier of discretes and is No. 2 in NAND. It is also a major supplier of ASICs, optoelectronics, microcontrollers, analog ICs, standard logic, and embedded MPUs. STMicroelectronics (Switzerland)ST is the top supplier of analog ASSPs and ASICs, strong in wireless, PC peripherals, automotive, and consumer. ST is also a big supplier of discretes, flash, EPROM, EEPROM, standard logic, and microcontrollers. Renesas (Japan)The merged, spun-out semiconductor businesses of Hitachi and Mitsubishi, Renesas is the No. 1 supplier of microcontrollers and is also strong in ASICs, discretes, display drivers, and embedded MPUs. Infineon (Germany)The spin-out from Siemens is a major player in analog ASSPs (especially for communications and automotive), discretes, microcontrollers, and EEPROM. Infineon carved out its memory unit, Qimonda, in May 2006the company is a leading provider of DRAM and is has been building a presence in NAND as well. NXP (Netherlands)The former semiconductor division of Philips is a top supplier of analog ASSPs and ASICs, strong in wireless and consumer. NXP/Philips is also a major supplier of discretes, DSPs, microcontrollers, standard logic, and EEPROM. Hynix (Korea)Hynix is the No. 2 supplier of DRAM and is rapidly building a position in NAND, now No. 3 worldwide. In 2005, Hynix spun out its LCD driver, image sensor, MCU, and foundry businesses as a new company, Magnachip. NEC Electronics (Japan)Spunout from parent NEC, NEC Electronics is a major supplier of microcontrollers, ASICs, display drivers, SRAM, discretes, and embedded MPUs.

60

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Key Semiconductor Competitors Key Semiconductor Competitors


Semiconductor Market Top Competitors By Product
Analog
Analog SLICs Texas Instruments Analog Devices National Semi Maxim Linear Tech ON Semi Intersil STMicro Fairchild Intl. Rectifier Analog ASSPs STMicro Infineon Texas Instruments NXP/Philips Freescale Renesas Toshiba Matsushita Sanyo Rohm
DRAM Samsung Hynix Micron Infineon Elpida Nanya PowerChip ProMOS SRAM Samsung Intel NEC Spansion Cypress STMicro Renesas Hynix NOR Intel Spansion STMicro Samsung Renesas Sharp Silicon Storage Macronix Toshiba EEPROM Atmel STMicro Microchip Tech. Rohm NXP/Philips Infineon Renesas Catalyst Semi

Memory
Flash NAND Samsung Toshiba Hynix Sandisk Renesas STMicro Micron M-Systems IM Flash Tech.

Digital Processors
Microprocessors PC & Server Intel AMD Sun/TI Via Freescale IBM Embedded Freescale Intel IBM Micro Renesas Toshiba NEC AMD PMC-Sierra AMCC Broadcom Microcontrollers Renesas Freescale NEC Infineon Microchip Toshiba Matsushita Fujitsu STMicro Texas Instruments DSPs Texas Instruments Freescale Agere Analog Devices NXP/Philips

Mask ROM Macronix Sun Plus NEC Amic

EPROM Oki STMicro Asahi Kasei Cypress

Discretes and Optoelectronics


Discretes Toshiba STMicro Rohm Renesas Fairchild Semi Infineon Vishay Intl. Rectifier NXP/Philips Mitsubishi ON Semi NEC Fuji Sensors Robert Bosch Infineon Freescale Analog Devices Allegro Micro VTI Mitsubishi NXP/Philips Optek Technology Micronas

Digital Logic
Special Purpose Logic PC-Related Intel NVIDIA ATI SiS Via Broadcom Communications Qualcomm Infineon Texas Instruments Freescale Broadcom Intel Conexant Agere Marvell PLD Xilinx Altera Lattice Actel Cypress QuickLogic STMicro Storage Marvell Agere MediaTek STMicroelectronics LSI Logic Texas Instruments Renesas NEC PMC-Sierra GP Logic Texas Instruments Toshiba NXP/Philips Fairchild Semi Renesas ON Semi STMicro Consumer Sony Matsushita STMicroelectronics Sharp Broadcom Fujitsu Mediatek NXP/Philips Zoran Display Drivers Samsung NEC Renesas Novatek Himax Solomon Systech Sharp Matsushita

Optoelectronics Image Sensors Sony Micron Toshiba STMicro Sharp Avago Matsushita Sanyo Omnivision All Other Sensors Sharp Mitsubishi Nichia Chemical Rohm Avago Matsushita OSRAM Toyota Gosei Stanley Electric Sanyo Toshiba Q-Cells

ASICs IBM Fujitsu NEC Toshiba Sony LSI Logic Samsung

Source: Dataquest, IDC, Forward Concepts, Databeans, CIBC World Markets Corp.

61

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Key Semiconductor Competitors Key Semiconductor Competitors


Semiconductor Market Top Competitors By End Market
Computing
PCs CPU Intel AMD Chipsets Intel Via SiS NVIDIA ATI PC DRAM Samsung Hynix Micron Infineon Nanya Powerchip ProMOS Elpida Graphics ATI NVIDIA BIOS SST Macronix Spansion STMicro Sharp PC Clocks IDT/ICS Cypress Realtek Audio Creative Realtek Analog Devices IDT/Sigmatel Analog Modem Conexant Silicon Laboratories Agere Analog Devices I/O Controller Standard Micro Winbond/National ITE PC Ethernet Broadcom Intel Realtek Marvell Agere Vitesse PC Wireless LAN Intel Atheros Broadcom Texas Instruments Marvell Realtek Bluetooth CSR Broadcom NXP/Philips Texas Instruments Infineon STMicro RF Micro Power Mgmt Fairchild Texas Instruments STMicro Intl. Rectifier On Semi Toshiba Maxim Mitsubishi

PC Displays LCD Controllers Genesis MorningStar Realtek Novatek Pixelworks LCD Drivers Samsung Novatek NEC Himax Magnachip Seiko Epson Matsushita Sharp Toshiba

Hard Disk Drives Marvell Agere STMicro TI LSI Logic QLogic/Marvell Renesas Infineon IBM Micro

Optical Drives Controllers MediaTek Renesas NEC Matsushita Rohm Ricoh Toshiba Seiko-Epson Sanyo Optoelectronics Sharp Avago Stanley Electric Sony Nichia Chemical Toshiba OSRAM Mitsubishi NEC

Printers/MFPs STMicroelectronics Texas Instruments Avago/Marvell Freescale NEC Electronics PMC-Sierra Toshiba Seiko-Epson Kawasaki Micro NXP/Philips

Networking
Ethernet PC Controllers Broadcom Intel Realtek Marvell Agere Vitesse PHY & Switch ASSPs Broadcom Marvell Avago Realtek Vitesse Intel ASICs LSI Logic IBM Agere Avago TI Fujitsu Wireless LAN Intel Broadcom Atheros Marvell Texas Instruments Conexant Airgo NXP/Philips Bluetooth CSR Broadcom NXP/Philips Texas Instruments Infineon STMicro RF Micro HBA/SAN ASICs IBM LSI Logic Agere Mindspeed Broadcom Enterprise Storage RAID Processors/ Controllers Intel LSI Logic IBM AMCC Promise/Marvell Vitesse/Adaptec Broadcom Disk Array ICs PMC-Sierra IBM Vitesse LSI Logic

Telecom/Datacom
CO Line Cards Infineon Legerity STMicro IDT Freescale NEC Winbond Analog Modem Conexant Silicon Laboratories Agere Austria Microsystems Analog Devices DSL Conexant Texas Instruments Broadcom Infineon Ikanos Centillium STMicro Cable Modem Broadcom Texas Instruments Microtune Conexant PON PMC/Passave Teknovus BroadLight Freescale Centillium Conexant Mindspeed Maxim VoIP Texas Instruments Broadcom Freescale Mindspeed Centillium

Communications Infrastructure T/E Infineon PMC-Sierra Agere Maxim/Dallas Mindspeed Freescale Exar TDK SONET/SDH Agere PMC-Sierra Vitesse AMCC Infineon Broadcom Cortina/Intel TranSwitch ATM PMC-Sierra Agere Fujitsu NEC Zarlink Mindspeed TranSwitch Ethernet WAN AMCC/Quake Maxim Vitesse Infineon PMC-Sierra Vitesse TranSwitch AMCC Agere Comm Processor Freescale Intel AMCC Broadcom PMC-Sierra Renesas Toshiba IBM ASICs IBM Fujitsu LSI Logic Agere TI Network Processing NPUs CAMs AMCC Netlogic Intel IDT Agere Cypress Wintegra Renesas Mindspeed Hifn Vitesse Bay Micro Other Hifn Cavium Broadcom Safenet Vitesse Tarari Switching ASICs ASSPs IBM AMCC Fujitsu Vitesse LSI Logic Zarlink Agere Interconnect PMC-Sierra PLX Dune Tundra TranSwitch Intel

Source: IDC, Dataquest, Linley Group, CIBC World Markets Corp.

62

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Key Semiconductor Competitors Key Semiconductor Competitors


Semiconductor Market Top Competitors By End Market
Wireless
Handset ICs Digital Baseband Texas Instruments Qualcomm Freescale NXP/Philips Infineon Mediatek Agere Analog Devices Analog Baseband STMicro Texas Instruments Qualcomm Freescale Dialog Agere Analog Devices Rohm App. Processor Texas Instruments STMicro MtekVision Corelogic ATI Intel NVIDIA Renesas RF & PA Qualcomm Skyworks RF Micro Devices Infineon Renesas NXP/Philips STMicro Freescale Silicon Labs Memory Intel Samsung Spansion STMicro Renesas Sharp Toshiba Elpida Baseband ASIC Qualcomm IBM Agere Fujitsu Texas Instruments NEC NXP/Philips Analog Devices Infrastructure DSPs Texas Instruments Freescale Analog Devices Agere Power Amp Freescale NXP/Philips Infineon Eudyna STMicro

Digital Consumer
Set-Top Box STMicro Broadcom Conexant NXP/Philips ALPS LSI Logic Intel/Zarlink Fujitsu Digital TV Matsushita NXP/Philips Sharp Genesis ATI Trident Sony Pixelworks Toshiba Renesas ALPS Micronas Silicon Image Zoran Samsung DVD Mediatek LSI Logic Zoran SunPlus NEC ESS Toshiba TI Cheertek Sony NXP/Philips Digital Camera ASIC/ASSP Fujitsu Sony Renesas SunPlus TI Zoran Matsushita Fuji Micro Samsung Image Sensor Sony Matsushita Sharp Sanyo Omnivision MP3 Player SigmaTel PortalPlayer Actions NXP/Philips TI Telechips Broadcom Wolfson Sony Samsung ALi Videogame Sony NEC Sharp IBM Intel Nvidia LSI Logic ATI Marvell Memory Card Samsung Toshiba Hynix SanDisk Renesas STMicro Micron M-System Infineon

Automotive
Automotive Auto MCUs Freescale Renesas NEC Fujitsu Texas Instruments Infineon Intel Toshiba Samsung ASICs/ASSPs STMicro NXP/Philips Infineon Robert Bosch Freescale Texas Instruments NEC Toshiba Auto SLICs Micrel National Semi Sanken NewJRC Renesas Maxim Intersil Analog Devices SIPEX Linear Tech Auto Discretes Infineon Robert Bosch Rohm NXP/Philips STMicro Vishay OSRAM On Semi Intl. Rectifier Denso

Source: IDC, Dataquest, Linley Group, CIBC World Markets Corp.

63

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Section IV: Product Summary Section IV: Product Summary

This section discusses the major types of semiconductor products. For each product we discuss: Product Definitions Market Forecast Competitive Environment

64

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The Semiconductor Market The Semiconductor Market


The CIBC Semiconductor Revenue Forecast
We expect total semiconductor industry revenue to grow 9% in 2006, as the industry recovers from the mild supplydriven downturn experienced in 2005. Demand has moderated in key end markets, particularly in PC and handset, dampening the outlook for the next expansion. Overall, we expect a 2005-2010 5-year CAGR of 8%, growing from $227 billion in 2005 to $332 billion in 2010.
Revenue ($M)

$350,000 Memory Logic $300,000 Microcomponents Analog $250,000 Discretes, Sensors, & Optoelectronics

$200,000

$150,000

$100,000

$50,000

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: WSTS, CIBC World Markets Corp.

Product Forecast Highlights


Logic should slightly outgrow the market with a 10% CAGR. Special purpose logic for communications and automotive, FPGAs, and display drivers should lead growth. Microcomponents should grow at a 6% CAGR. MPUs should grow just 5% as prices decline for PC processors. DSPs should outperform driven by healthy wireless growth. Analog should outgrow the market at 10%, driven by strong demand from both application-specific analog (11%) and standard linear ICs (8%). Memory should remain a growth segment, but will continue to show the most volatility; up the most in boom years and down the most in downturns. Despite perpetual price pressure, discretes, sensors, and optoelectronics should grow in line with the market given their multi-market application.

The semiconductor industry totaled $227 billion in sales in 2005, a 7% increase over 2004. The industry experienced a mild cyclical downturn in 2005 even during a period of relatively strong and stable demand, characterized by moderate oversupply but low inventories across the sector. We expect semiconductor revenue to grow 9% in 2006 as the industry recovers from this modest oversupply and as inventories are replenished (though they are correcting in 2H06). We expect 2007 to 2008 to be expansion years, followed by the next downturn and recovery in 2009 and 2010. Overall, we estimate a 2005-2010 five-year CAGR of 8%. By segment, we expect the following: MOS Logic should grow at a 10% CAGR, with communications and automotive ASSPs, along with FPGAs and display drivers, growing fastest. Microcomponents should grow at a 6% CAGR. MPUs should grow just 5% as PC growth is tempered by new, more intense price pressure. DSPs should outperform with 14% growth, driven by healthy wireless demand. Analog should grow 10%, driven by strong demand for both ASSPs (11%) and standard linear ICs (8%). Memory should remain a growth segment with a 6% CAGR, but will continue to show the most volatility; up the most in boom years and down the most in downturns. Despite an always-difficult pricing environment, discretes, sensors, and optoelectronics should grow roughly in line with the market given their multi-market application.

65

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Analog Product Summary Analog Product Summary

The semiconductor market can be divided into five key product areas

Five Key Markets:

Analog Digital Processors Digital Logic Memory Discretes and Optoelectronics Discretes and Optoelectronics
We first examine the semiconductor industry by product group. The industry can be divided into five segments: Analog: Analog circuits are used to process real world signals, using electronic voltage patterns within the device to represent the original signal. Standard linear integrated circuits (SLICs) are standard analog components used in multiple applications, usually purchased off the shelf, often through distribution. Application specific standard products (ASSPs) are more specialized analog devices targeted for a specific application. Digital Processors: Also called microcomponents, these devices perform intensive compute processing and system control. Microcomponents includes microprocessors, microcontrollers and DSPs. Digital Logic: Logic devices perform digital processing, and include gate arrays and standard cells (ASICs), programmable logic devices (FPLDs, FPGAs), display drivers, general purpose logic and specialty purpose logic. Memory: Memory devices are used to store data either for short periods of time or permanently. Volatile memory loses stored information when the power is lost; key types are DRAM and SRAM. Non-volatile memory retains the information when the system is powered off; key types are flash, mask ROM, EPROM and EEPROM. Discretes, Optoelectronics and Sensors: This category includes all non-integrated circuit semiconductors. Discretes are devices that contain just a single transistor in a package. Optoelectronics include devices designed to emit and detect light. Sensors include those devices that measure physical, chemical or biological properties.

66

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Analog Analog

Devices designed to receive, process, and create analog real world signals

Analog ICs

Two Key Markets:


Analog SLICs Analog ASSPs

Analog ICs are devices used to process and convert real world signals. Whereas digital circuits perform processing functions by determining the absence or presence of an electrical charge (the 1s and 0s of the digital world), analog circuits are concerned with the actual voltages and currents in circuits. When electrical devices interact with the real world, analog circuits measure real world input in the form of voltage/current strength (the electrical charges in the device form an analogy of the real world inputhence the term analog). They can also convert the signal to a digital form for processing. Conversely, when the device creates real world output, analog circuits convert the digital instructions to analog voltages/currents to drive the creation of output. Analog circuits are divided into two broad categories: SLICs and ASSPs. SLICs (standard linear integrated circuits) are standard analog components that used in multiple applications. Although designed for a specific function within a system, a SLIC is not optimized for particular types of systems. Key types are amplifiers, interface circuits, voltage regulators, data converters and comparators. ASSPs (application specific standard products) are more specialized analog devices targeted for a specific application. These devices are often designed as multiple SLICs that are integrated and then modified to serve a specific function in a particular end market; but they can also be designed from the ground up. Examples include radio frequency (RF) ICs in cellular phones, interface ICs for networking, read channels in HDDs, tuners and demodulators in set top boxes, battery management in notebooks and power train control devices in automobiles.

67

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Analog SLICs Analog SLICs


Analog SLICs (Standard Linear Integrated Circuits) - Standard analog components used in multiple applications. While designed to perform a specific function, SLICs are not optimized for any particular type of system. Key Types: Amplifiers, interface drivers, voltage regulators, data converters (analog-to-digital, digital-to-analog, codec), comparators.

Market Forecast
Analog SLICs are expected to grow at a 5-year CAGR of 8%, rising from $11.7 billion in 2005 to $17.2 billion in 2010. Growth to be led by data converters.
$20,000 $18,000 $16,000 $14,000
Revenue ($M)

Key Competitors
Texas Instruments 15 % Others 32 % Analog Devices 13 %

Comparators Data Converters Voltage Regulators Interface Amplifiers

$12,000 $10,000 $8,000 $6,000 $4,000 $2,000 $0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Intl. Rectifier 2% Fairchild 2% STMicro 2% Intersil 3% ON Semi 3%

National Semi 11 % Maxim 9%

Linear Tech 7%

Source: Dataquest, CIBC World Markets Corp.

Source: WSTS, CIBC World Markets Corp.

Analog SLICs (standard linear integrated circuits) are general purpose analog components that can be used in multiple types of electronic devices. Although designed to perform a specific function within a system, SLICs are not optimized for particular applications. For example, a line driver (a type of interface SLIC) is designed to produce or receive a signal on a copper wire, but can be used in any application that requires this function (e.g. a telephone set, an analog modem, etc.). Similarly, a video ADC (analog-to-digital converter) is designed to convert analog video signals to digital, but can be used in any video application (e.g. digital camera, digital TV, set-top box, etc.). SLICs can be sub-divided into five types of devices: Amplifiers: ICs designed to amplify a signal, also can be configured to perform certain related operations. Interface: ICs that produce output or receive input, usually in the transmission of signals to and from systems. Voltage Regulators: Devices that control power flow by providing a stable flow of current at a particular voltage. Data Converters: Includes mixed signal (analog/digital) devices designed to convert a signal from analog to digital (an ADC) or from digital to analog (a DAC). A codec combines the two functions in a single device. Comparators: ICs used to compare two quantities, usually used in power management. These are often grouped with amplifiers, as it is a small market and is similar in nature to amplifiers.

68

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Market Forecast: The market for analog SLICs is expected to grow at an 8% CAGR over the next five years. While most electronics OEMs and semiconductor companies are focused on driving the performance of the digital portion of the system, most applications still require significant analog content to interface with the real world. In total, we expect the market to grow from $11.7 billion in 2005 to $17.2 billion in 2010. Growth should be led by data converters, with a CAGR of 10%, followed by voltage regulators at 8%. Interface circuits and amplifiers are expected to grow 7% and 6%, respectively. Comparators should grow 11%, but should remain around 2% of the market. Competitive Environment: Note that the SLIC market is one of the most consistent and profitable segments in the semiconductor industry. Barriers to entry in the standard linear market remain extremely high, due in part to the nature of analog SLIC device design. Analog engineering is more of a refined art rather than a school-taught competency, with most analog engineers honing their craft over years of real world design. Although digital devices are more dependent on scaling processing architectures and migrating to smaller geometries, SLICs are more dependent on a deep knowledge of the intricacies of the particular medium for which the device is being designed (e.g., sound waves, light waves, heat, etc.). The pool of analog design engineering talent is therefore more limited than for digital devices, and a multitude of SLIC vendors have been able to develop defensible franchises. Further, as SLICs are extremely diverse in application, the supplier chain is extremely fragmented. Only three vendors held above 10% market share in 2005. Texas Instruments held 15%, with more than 10% of each device category. Analog Devices held 13%, with a dominant 43% share of data converters and a strong amplifier position. National Semiconductor was third with 11%, No. 1 in voltage regulators, and No. 3 in interface ICs and in amplifiers. Maxim, the high-performance standard linear pure-play, held 9% and was No. 2 in interface ICs and No. 3 in converters. Rounding out the top 5 was Linear Technology, Maxims chief high-performance competitor, with 7% overall and the No. 3 spot in voltage regulators. Other major vendors include On Semi, Intersil, STMicroelectronics, Fairchild, and International Rectifier. We provide additional competitive detail in the charts below. Amplifiers and Comparators
Others 21 % Texas Instruments 18 %

Voltage Regulators
Others 21 % National Semi 16 %

NewJRC 4% Maxim 4% NEC 5% Avago 5% STMicro 5 % Linear Tech 6%

Analog Devices 17 %

Analog Devices 3% STMicro 3% Micrel 3% Fairchild 5% ON Semi 5%

Texas Instruments 16 %

National Semi 15 %

Linear Tech 10 % Intersil 5% Maxim 10 %

Converters
Others 12 % National Semi 3% Cirrus Logic 4% Asahi Kasei 4% Linear Tech 5% Wolfson Micro 6% Maxim 10 % Analog Devices 43 % Others 18 %

Interface Circuits

Texas Instruments 25 %

Linear Tech 5% Toshiba 6%

Philips 8%

Maxim 17 % Intl. Rectifier 9%

Texas Instruments 14 %

National Semi 12 %

Source: Dataquest, CIBC World Markets Corp.

69

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Analog ASSPs Analog ASSPs


Analog ASSPs (Application Specific Standard Products) - Specialized analog ICs targeted for a specific application. Often integrate multiple SLICs or even some digital circuitry. Often called mixed signal analog. Key Types: Computer (PC and storage), communications (wireline, wireless, networking), consumer, automotive, and industrial ASSPs.

Market Forecast

Key Competitors
Others 19 % STMicro 20 %

Analog ASSPs are expected to grow at an 11% CAGR, rising from $20.2 billion in 2005 to $33.3 billion in 2010. Growth to be led by communications Broadcom (particularly wireless) ASSPs with a 12% RF Micro 3 % Devices CAGR; consumer, automotive, and 3% industrial should all grow 10%. Skyworks
$40,000 Industrial Automotive Communications Computer & Storage Consumer

3% Renesas 4% Qualcomm 4% Freescale 6%

Infineon 14 %

$35,000

$30,000

$25,000
Revenue ($M)

$20,000

$15,000

NXP/Philips 12 %

Texas Instruments 12 %

$10,000

Source: Dataquest, Databeans, CIBC World Markets Corp.

$5,000

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: WSTS, CIBC World Markets Corp.

Analog ASSPs (application specific standard products) are analog components designed to service a particular application. These are sometimes designed as multiple SLICs that are integrated to serve a specific function in a particular end market, but can also be designed from the ground up. They often include some digital circuitry, and hence are sometimes referred to as mixed signal analog. Key examples of analog ASSPs include radio frequency (RF) ICs in cellular phones, physical layer interface devices for networking applications, read channels in hard disk drives, demodulators in consumer electronics, battery management ICs in notebook computers and engine and power train control devices in automobiles. Unlike analog SLICs, analog ASSPs are not usually sold through the distribution channel, and instead go through design process at OEMs that is similar to that used for digital ASSPs. In fact, many of the highest volume electronics applications are served by a chipset that includes one or more analog ASSPs paired with one or more digital ASSPs offered by the same vendor, or by two vendors on a single reference design. The analog and digital ASSPs are thus designed in, and subsequently purchased by the OEM, at the same time. As mentioned above, some analog ASSPs actually incorporate more mixed signal and digital functionality, often blurring the line between analog ASSP and special purpose logic. Further, the same vendors who design digital ASSPs are now increasingly the vendors who design analog ASSPs. While the SIA has for now decided to keep the line

70

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

between analog and digital ASSPs, some third-party research firms like Gartner Dataquest have done away with this distinction and group all application specific analog and digital devices as simply ASSPs. ASSPs are divided into five categories by application: Automotive: Devices specifically designed for use in automobile engine, safety, comfort, entertainment and power controls. Consumer: ASSPs used in audio/video equipment and home appliances, usually for processing, production or reception of audio and video input. Computer: ASSPs used in computing applications, including storage devices like hard disk and optical drives. Communications: ASSPs used in wireless, telecom, and datacom client and service provider devices and equipment. These are usually used in the production and reception of transmission signals, including optical, electrical and radio. Industrial and Other: ASSPs used in industrial equipment and other applications.

Market Forecast: The market for analog ASSPs is expected to grow at an 11% CAGR over the next five years, rising from $20.2 billion in 2005 to $33.3 billion in 2010. The largest and fastest growing market segment in analog ASSPs is communications; it is expected to post a 12% CAGR through 2010 to represent just over 50% of the market. Growth in communications is primarily being driven by RF, analog baseband and power management ICs for wireless handsets. Certain wireline networking technologies should also drive communication ASSP growth, such as Gigabit Ethernet, SONET, Fibre Channel, wireless LAN and Bluetooth. DSL and cable modems should see unit growth, though much of the analog content in these applications has moved into a digital system-on-a-chip (SoC). Consumer, automotive, and industrial should all grow at around a 10% CAGR through 2010. Consumer benefits from nice unit growth, but is subject to more price pressure as well as the trend to pull analog functions into the digital SoC, moderating the TAM a bit. Automotive and industrial, on the other hand, see little price pressure and even less integration risk, as analog processing remains the focus of many of these systems. Competitive Environment: The ASSP market is highly fragmented, not surprising given the diversity of applications served. STMicroelectronics has historically been the leading analog ASSP provider and held 20% share in 2005, with strong positions in hard disk drive, PC peripheral, wireless, wireline and automotive ASSPs. Infineon was second with 14%, strong in automotive, wireless and wireline ASSPs. Texas Instruments, which is also the largest provider of analog SLICs, was third, with 12%; TI has a strong portfolio of products for hard disk drives, PC peripherals, displays, wireless and wireline. NXP/Philips was fourth with 12%, a top provider of consumer ASSPs and also strong in wireless and automotive. Other suppliers include Freescale, Qualcomm, Renesas, Skyworks, RFMD, Broadcom, Matsushita, Sanyo, Toshiba, Rohm, MediaTek and Analog Devices. The table below displays some additional detail. Communications Key device types Analog basebands Handset power management RF transceivers Power amplifiers DSL front ends Key Competitors STMicroelectronics Qualcomm Skyworks RF Micro Devices Texas Instruments Infineon Renesas Freescale Broadcom Conexant Silicon Labs Computing Printer front ends and cartridge heads HDD read channels, preamps, and motor controllers ODD front ends PC interface ICs Texas Instruments STMicroelectronics MediaTek Renesas NEC Rohm Marvell/Avago NEC Electronics Freescale Ricoh Sanyo Consumer Demodulators Tuners Audio/video interface Digital camera front end DVD front end Sony Matsushita NXP/Philips STMicrolectronics Renesas Micronas Broadcom Conexant Texas Instruments STMicroelectronics NXP/Philips Infineon Renesas Robert Bosch Freescale Texas Instruments NEC Toshiba Auto/Industrial Engine control Automotive audio Climate control Motor control

Source: Dataquest, CIBC World Markets Corp.

71

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Digital Processors Digital Processors

Devices designed to perform intensive compute processing and system control

Digital Processors:

Three Key Markets:


Microprocessors Microcontrollers Digital Signal Processors

Microcomponents include all digital processors, including microprocessors, microcontrollers and DSPs. These devices are built around the famous 1s and 0s of the digital world and are designed to be programmable; i.e., they run software which is very customizable and can change the function of the device. Microprocessors (CPUs or MPUs) are the largest, most complex processors, and are generally found in computing and other compute-intensive applications and typically form the brains of an entire electrical system. Raw performance is generally the most important metric for microprocessors, though increasingly performance per watt is becoming important as well. Little integration is done in the microprocessor market, as they almost always sit within a larger system with other ICs to handle supporting functions. Microcontrollers (MCUs) are smaller stand-alone processors that perform dedicated or embedded computer function within a system and usually contain on-chip RAM, ROM, I/O logic and timers in order to minimize cost and to speed design and implementation cycles. Integration is most important, followed by power consumption and performance; though performance will usually vary by application given the diverse use of MCUs. Digital signal processors (DSPs) are specialized high-speed digital processors typically used to process real-time data. Performance and power consumption are of equal importance. Integration has also been critical in driving DSPs into select applications (like handsets), though many systems use DSPs as a standalone high-speed processor.

72

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Microprocessors Microprocessors
Microprocessors (MPUs) - Digital processors that execute instructions and perform system control functions as programmed with an assembly language. MPUs are optimized for general purpose data processing. Key Types: x86, PowerPC, ARM, MIPS. 16-, 32-, and 64-bit external bus.

Market Forecast
Microprocessors are expected to grow from $35.0 billion in 2005 to $43.6 billion in 2010, a 5% CAGR. The market is almost entirely PC-based and is tied to PC unit growth and the mix between desktop, notebooks, and servers.
$50,000 Microprocessors $45,000 $40,000 $35,000
Revenue ($M)

Key Competitors
IBM Freescale 1 % 3% Others 4%

AMD 11 %

$30,000 $25,000 $20,000 $15,000 $10,000 $5,000 $0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Intel 81 %
Source: Dataquest, CIBC World Markets Corp.

PC

Server

Embedded
Freescale Toshiba Intel NEC IBM AMD Renesas PMC-Sierra

Source: WSTS, IDC, CIBC World Markets Corp.

Intel Intel AMD Sun/TI Freescale (Apple) IBM Via AMD

A microprocessor (MPU) is a digital processor that executes external instructions and performs system control functions as programmed via software with an assembly language. The architecture is optimized for general purpose data processing and includes an instruction decoder, arithmetic logic unit, registers and additional logic to support operation as per the assembly language. An MPU can receive input commands, manipulate data, direct storage of data and initiate application commands to the outside world. The PC and server market represents approximately 90% of microprocessor revenues. Most PC microprocessors use the x86 CISC (Complex Instruction Set Computer) architecture, which uses microcode to execute very comprehensive instructions. Instructions may be variable in length and use all addressing modes, requiring complex circuitry to decode them. Both Intel and AMD, the major PC microprocessor suppliers, have added their own enhancements to the x86 architecture, both in the actual processor core and the surrounding logic. These enhancements optimize certain common features used in the PC, such as operating system tasks, multimedia functions, Internet usage, etc. Microprocessors used for PC peripheral, communications and digital consumer applicationsusually referred to as embedded processorsgenerally use either a generic x86 processor or one of many RISC (Reduced Instruction Set Computer) architectures. RISC cores reduce chip complexity by using simpler instructions and keeping the instruction size constant without microcode layer or associated overhead. There are several mainstream RISC architectures, most notably MIPS, PowerPC and ARM. These architectures usually require a license.

73

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Microprocessors are also classified according to bit width, which defines the size of the instructions sent over the external bus (some processors use a higher bit-width on the interior of the processor, only the external bus is used for classification). Most PC microprocessors today use a 32-bit external bus, but can support 64 bit extensions. AMD first introduced 64-bit extension in 2003 and Intel followed suit in 2004; virtually the entire market has transitioned at this point. In the server space, most low-end and mainstream servers use standard 32-bit processors with 64-bit extensions prevalent in the PC market, while high-end servers will use proprietary or licensable 64-bit processors (e.g. Sun Microsystems SPARC or IBMs PowerPC). Intel has also introduced a standard 64-bit processor for this market (branded Itanium), but has had trouble gaining share from proprietary systems. In the embedded market, processors use a mix of 8-, 16-, 32-, or 64-bit processors, depending on the application. The highest end processors would be used in communications infrastructure equipment like softswitches or firewalls, while the lowest end would be used in consumer electronics or industrial applications. Note that microprocessors can also contain more than one processor core on a single die. Many proprietary RISC server processors utilize a dual-core or multi-core design, which efficiently ramps performance for high-end computeintensive applications. Faced with thermal limitations in PC processor design, both Intel and AMD launched dual-core products in 2005, with Intel first to market in the desktop and AMD first to market in the server. Dual core processors are now starting ramp significantly for the desktop, notebook, and server markets in 2006. Market Forecast: Microprocessors are heavily tied to the PC market and thus are more dependent on demand for this one application than on broader semiconductor or electronics demand. Overall, we expect microprocessors to grow from $35.0 billion in 2005 to $43.6 billion in 2010, a 5% CAGR. This assumes a high single-digit growth rate for PCs annually, offset by pricing pressure that is now the new norm in the PC market following AMDs rise as a stronger No. 2 to Intel in recent years. Competitive Environment: The microprocessor market is one of the most concentrated markets in semiconductors. Intel has effectively maintained a near-monopoly on the PC and server microprocessor market, with just over 86% revenue share overall. Note that Intel has a somewhat lower unit share (79%), since its microprocessor ASPs are higher than its competitors. AMD held 12% revenue share in 2005 (and about 18% unit share) and competes with Intel in all market segments. IBM and Freescale also participated in the PC processor market with their PowerPCbased processor for Apple, though Apple began a migration to Intel in 2006. Other x86 microprocessor vendors include Via, which competes at the low end. The diagrams below depict revenue and unit share in the PC and server microprocessor markets for 2005. 2005 PC and Server Microprocessor Revenue Share
Others 2%

2005 PC and Server Microprocessor Unit Share


100% 90% 80% 70% 60% 50% 40%

AMD 12 %

Intel 86 %

30% Others 20% 10% 0% Total Units Desktop Mobile Server AMD Intel

Source: IDC, CIBC World Markets Corp.

The embedded microprocessor market is somewhat more fragmented, not surprising given the diversity of applications served and architectures required. Freescale is the largest player overall with 28% market share and the top position in the communications and other categories, and the No. 2 and 3 positions in the consumer and computing markets, respectively. Freescales PowerQUICC is the dominant processor choice for wireline equipment and its DragonBall and i.MX family have historically been well positioned in the PDAs and smartphone markets.

74

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Intel was the No. 2 supplier in 2005 with 25% share; Intel holds the top position in computing and the No. 2 positions in communications and other applications. Intels PC processors, find lots of use in non-PC applications; typically these will be the lower end or lower-voltage versions of its product portfolio. Intel also supplies a low-power ARM-based family for portables under the Xscale brand, though it sold the handset and PDA portion of XScale to Marvell 2006. IBM was the No. 3 supplier in 2005 with 11% share and holds the No. 2 position in computing. Renesas followed as No. 4 with 8% share and the top position in consumer. Toshiba followed with 5%, NEC with 4%, and AMD with 3%. Rounding out the top 10 are comm IC players PMCSierra, AMCC, and Broadcom, all of whom entered the market through acquisition.

2005 Embedded Processor Market Share


Broadcom 2% AMCC PMC-Sierra 3 % 3% Others 7%

AMD 3% NEC 4% Toshiba 5% Renesas 8%

Freescale 28 %

IBM Micro 11 %

Intel 25 %

2005 Computing Embedded MPU Market Share


Transmeta 2% PMC-Sierra 3% AMD 3% Toshiba 5% NEC 4% Others 9%

2005 Comm Embedded MPU Market Share


Others 13 % PMC-Sierra 3% NEC 4% Renesas 4% Toshiba 4% Broadcom 6% AMCC 7% Freescale 48 %

Intel 42 %

Freescale 10 %

IBM 20 %

Intel 10 %

2005 Consumer Embedded MPU Market Share


Toshiba 5% NEC 5% AMD 5% IBM 6%
Others 6%

2005 Other Embedded MPU Market Share


Others 12 % Fujitsu 3% IBM 5% Toshiba 7%

Renesas 32 %

Freescale 34 %

Intel 17 % Freescale 24 %

Renesas 13 %

Intel 25 %

Source: Dataquest, CIBC World Markets Corp.

75

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Microcontrollers Microcontrollers
Microcontrollers (MCUs) Stand-alone devices that perform dedicated or embedded computer functions within an overall system. MCUs contain single or multiple processing mini-architectures, as well as on-chip RAM, ROM, and I/O logic. Key Types: 4-, 8-, 16-, 32-bit (or greater). Automotive, consumer, computer, IC card, wireless communication, wired communication.

Market Forecast
Microcontrollers are expected to grow at a 6% CAGR, from $12.1 billion in 2005 to $16.3 billion in 2010. Growth to be led by 32-bit MCUs.
$18,000 32-Bit MCUs $16,000 $14,000 $12,000
Revenue ($M)

Key Competitors
Others 19 % Texas Instruments 4% STMicro 4% Fujitsu 4% Matsushita 5% Toshiba 5% Microchip 6% Infineon 7% NEC 11 % Freescale 14 % Renesas 23 %

16-Bit MCUs 8-Bit MCUs 4-Bit MCUs

$10,000 $8,000 $6,000 $4,000 $2,000 $0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: Dataquest, CIBC World Markets Corp.

Source: WSTS, CIBC World Markets Corp.

Microcontrollers are stand-alone computation devices that perform dedicated or embedded computer functions within a system. Microcontrollers incorporate simpler processing architectures than microprocessors and are often used to perform only a specific function within a system. They contain some form of RAM and ROM memory as well as additional I/O logic on-chip and thus can perform relatively simple processing tasks without the need for traditional support ICs found in more complex systems. This lowers the system cost and speeds design and implementation. Like microprocessors, microcontrollers are classified by external bit width. Four-bit devices are still in use, primarily to control the simplest electronic devices (e.g., a toaster oven) or for very simple embedded functions within a larger system (e.g., controlling the LCD screen inside a digital camera). Eight-bit devices are the largest single segment in revenue terms at 39% of total microcontroller revenue in 2005 and represented roughly 45% of all microcontroller units shipped. Sixteen-bit microcontrollers totaled 29% of revenue, and 32-bit MCUs have grown rapidly to 28% of total revenue. By end market, the largest single consumer of microcontrollers is the automotive market, which accounted for 33% of total microcontroller sales in 2005, as automotive electronics have typically been implemented with a combination of microcontrollers and analog ASSPs and SLICs. Consumer devices accounted for 13%; this includes the simple consumer appliances controlled by MCUs as mentioned above, as well as the embedded functions. Smart cards followed at 11% of total sales and data processing accounted for 10%. Wireline and wireless applications accounted for less than 5%, as these devices tend to use dedicated ASSPs and DSPs for their implementation.

76

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The exhibits below display microcontroller revenue share by bit width and application. 2005 Microcontroller Revenue Share By Bit Width
4-bit 3% 32-bit 28 % Other 29 % Automotive 33 %

2005 Microcontroller Revenue Share By Application

8-bit 39 % Wireless 1% Wireline 3% Data Processing 10 % Consumer 13 %

16-bit 29 %

IC Card 11 %

Source: WSTS, CIBC World Markets Corp.

Market Forecast: Microcontrollers are expected to see healthy growth given their broad exposure to the electronics landscape. Overall, we expect microcontrollers to grow from $12.1 billion in 2005 to $16.3 billion in 2010, a 6% CAGR. The increasing complexity of automotive and consumer systems should favor the higher bit width segments. 32-bit (or greater) devices are expected to lead growth, with a 9% CAGR through 2010. 8-bit and 16-bit devices are expected to grow 6% and 5% through 2010, respectively, while 4-bit is expected to trail the overall market with negative 10% CAGR. Competitive Environment: The microcontroller market is highly fragmented due to the diversity of applications served. The largest supplier in 2005 was Renesas, with 23% market share. Renesas is the No. 1 supplier of 16-bit MCUs, the No. 2 supplier of 4-bit and 8-bit MCUs, and the No. 3 supplier of 32-bit MCUs. Renesas is also the leading supplier to both communications and consumer applications and the No. 2 supplier to computing and automotive applications. Freescale was the No. 2 supplier in 2005 with 14% share; the No. 1 supplier of 8-bit and the No. 2 supplier 32-bit microcontrollers and strong in automotive applications. NEC was third with 11%, strong in all applications. Infineon, Microchip, and Toshiba each followed with 7%, 6%, and 5%, respectively. Other major suppliers include Matsushita, Fujitsu, STMicroelectronics, Texas Instruments, Samsung, Atmel, and NXP/Philips. The exhibits below display microcontroller market share, first by bit-width and then by application. 4-bit Microcontrollers
Others 15 % NEC 21 %

8-bit Microcontrollers
Freescale 15 % Others 25 %

Oki Electric 6%

Renesas 14 %
Seiko Epson 8%

NXP/Philips 4% Toshiba 5% Atmel 6%


Samsung 17 %

Toshiba 11 %

Renesas 22 %

Microchip 13 % STMicro 8% NEC 10 %

77

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

16-bit Microcontrollers
Sunplus NEC 2 % 3% Others 9%

32-bit Microcontrollers
Samsung 3% Others 4%

Fujitsu 4% Toshiba 4% Texas Instruments 5% Intel 5% Renesas 40 %

Toshiba 4% Infineon 4%

NEC 21 %

Texas Instruments 7% Fujitsu 7% Freescale 20 % Matsushita 13 %

Freescale 9% Infineon 19 %

Renesas 17 %

Source: Dataquest, CIBC World Markets Corp.

Computing Microcontrollers
Others 14 % Microchip Tech. 4% Philips 6% Infineon 21 %

Communications Microcontrollers

Others 23 %

Renesas 38 % Toshiba 4%

Freescale 6% Renesas 21 %

Samsung 4% Matsushita 6% Freescale 6%

NEC 9% Matsushita 9% STMicro 10 %

NEC 8%

Microchip Tech. 10 %

Consumer Microcontrollers

Automotive and Other Microcontrollers


Others 14 %

Others 28 %

Renesas 30 %

STMicro 3% Samsung 3% Intel 4%


Toshiba 5%

Freescale 22 %

Sony 4% Atmel 4% Toshiba 5% Matsushita 9% NEC 10 % Microchip Tech. 11 %

Fujitsu 6% Infineon 6%

Renesas 18 %

Texas Instruments 6%

NEC 12 %

Source: Dataquest, CIBC World Markets Corp.

78

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Digital Signal Processors Digital Signal Processors


Digital Signal Processors (DSPs) - Specialized processors designed for stand-alone use, most often used for real-time processing of digital signals, e.g. handsets, digital cameras, MP3 players, IP set-top boxes, etc. Key Types: Consumer, computer and peripheral, wireless communication, wired communication, automotive, multipurpose.

Market Forecast
DSPs are expected to grow at a 14% CAGR through 2010, rising from $7.6 billion in 2005 to $14.8 billion in 2010. Handset unit growth, the move to 3G and beyond, and increased usage in consumer electronics and automotive should drive revenue growth.
$16,000 DSPs $14,000

Key Competitors
NXP/Philips 6% Analog Devices 6% Others 7%

Agere 9%

$12,000

$10,000
Revenue ($M)

Freescale 14 %

Texas Instruments 58 %

$8,000

$6,000

$4,000

$2,000

Note: Includes DSPs that are customized for specific applications, but excludes digital logic circuits with embedded DSPs. As an example, because of the nature of their design the majority of TIs handset basebands are included as DSPs but Qualcomms are not. Source: Forward Concepts, Dataquest, CIBC World Markets Corp.
2002 2003 2004 2005 2006 2007 2008 2009 2010

$0

Source: WSTS, CIBC World Markets Corp.

Digital signal processors (DSPs) are specialized standalone digital processors that are optimized for performing intensive computations typically required in communications and multimedia applications. DSPs are almost always used in applications that require real-time processing of digital signals. Examples include handset basebands, digital camera image processors, IPTV decoders, and MP3 player audio de-compressors. By end market, 75% of all DSP revenue is derived from the wireless market, both from handsets and basestations. Six percent is derived from the consumer market, primarily for portables like digital cameras and MP3 players as well as high-end consumer devices like DVD recorders; 5% sell into wireline, mostly broadband modems; 4% into computing applications; and 3% is derived from the automotive market. The exhibit at right displays revenue share in DSPs by application in 2005.

2005 DSP Segmentation by Application


Automotive 3% Other 7%

Data Processing 4% Wireline 5% Consumer 6%

Wireless 75 %

Source: WSTS, CIBC World Markets Corp.

79

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Market Forecast: DSPs are at the core of next generation personal communications, audio and video devices. They are fast, easy to program, extensively libraried and inexpensive. As such, we expect them to continue to drive highperformance applications like cellular handsets and to benefit from new applications such as digital portable audio and video players, cameras and smart handheld devices. In some cases, unit growth will be bolstered by ASP growth notably in the handset market when 3G penetration begins to rise. Overall, we expect DSPs to grow from $7.6 billion in 2005 to $14.8 billion in 2010, a 14% CAGR. Competitive Environment: The DSP market is highly concentrated, with five players controlling over 90% of the market in 2005. TI held the No. 1 position in 2005 with 58% market share, with its leadership position in GSM, GPRS, and UMTS baseband chips and its strong ties to handset leader Nokia. Freescale was No. 2 with 14% share in 2005, much of it derived from its relationship with its former parent company, Motorola. Agere fell one spot to No. 3 as NEC discontinued Agere based 3G platforms in 2004; Agere still remains well positioned at Samsung for 2G and 2.5G/EDGE products. Agere is also an incumbent in the 2.5G wireless infrastructure DSP market. Rounding out the top five were Analog Devices and NXP/Philips, both with 6% share. Other vendors include Toshiba, NEC, Cirrus Logic, Sunplus, and Fujitsu. Note that most special purpose logic devices (discussed later) incorporate some DSP circuits. Most mainstream definitions of DSPs exclude these devices, though the lines of distinction are somewhat subjective. For instance, both Texas Instruments and Qualcomm make digital baseband chips for cellular handsets, but TIs devices are considered DSPs and Qualcomms are considered special purpose logic. The defining line usually rests on how the device is designed: in the case of digital basebands, TI tends to wrap logic around its standard programmable DSP core: thus, it is classified as a DSP. Qualcomm designs a baseband ASSP from the ground up and incorporates a DSP-engine; thus, its devices are usually classified as special purpose logic.

80

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Digital Logic Digital Logic

Devices designed to process signals in a the digital realm

Digital Logic:

Five Key Markets:


Special Purpose Logic Display Drivers General Purpose Logic ASICs FPGAs/PLDs

Digital logic includes all non-microcomponent logic circuits. These devices are more hard-coded than microcomponents and thus typically are designed from the ground up for a particular function. Some software programmability is always incorporated, but it is usually to adjust certain details about the devices core functioning (e.g., which features to turn on, how fast to run the device, etc.). Key types include special purpose logic (digital ASSPs designed for a specific application), display drivers, general purpose commodity logic, custom logic (gate array and standard cells, also called ASICs), and field programmable logic devices (PLDs and FPGAs).

81

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Special Purpose Logic Special Purpose Logic


Special Purpose Logic - Application specific digital ICs sold as standard products, designed to service a particular application. Includes ASSPs for PC, display, communication, consumer, and automotive applications. Key Types: PC core logic, GPUs, communication ICs, consumer ASSPs.

Market Forecast
We expect special purpose logic to grow at a 9% CAGR through 2010, from $37.5 billion in 2005 to $58.6 billion in 2010. Growth to be led by communications and automotive ASSPs.
$70,000 Other Automotive Communications Computer & Peripheral Consumer $60,000

Key Competitors
PC and Server
Intel NVIDIA ATI/AMD Via SiS Broadcom

Storage and Peripheral


Marvell Agere MediaTek STMicroelectronics LSI Logic Texas Instruments

$50,000

Revenue ($M)

$40,000

Communications
Qualcomm Freescale Broadcom NXP/Philips Infineon Conexant Texas Instruments Marvell Agere PMC-Sierra

Consumer
STMicroelectronics NXP/Philips Sharp Broadcom Fujitsu Mediatek Zoran SigmaTel PortalPlayer Genesis

$30,000

$20,000

$10,000

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: WSTS, CIBC World Markets Corp.

Specialty purpose logic devices are digital logic circuits sold as standard products, usually designed specifically for a particular application (though not for a specific socketthose are classified as custom ASICs). Designers typically incorporate some degree of system-level knowledge in the architecture, as these devices typically form the core of the equipment into which they are designed. As discussed in the section on analog ASSPs, many of the highest volume electronics applications are served by a chipset that includes one or more analog ASSPs paired with one or more digital logic devices or ASSPs offered by the same vendor, or by two vendors on a single reference design. The analog and digital ASSPs are thus designed in, and subsequently purchased by the OEM, at the same time. A great example of this would be a wireless LAN chipset, where most vendors offer a chipset that includes an analog RF chip along with a digital baseband/MAC device. In almost all cases, a design win is awarded for the entire chipset as opposed to individual components. The category is segmented by application; i.e., consumer, communications, computing and peripheral, automotive and multipurpose. In 2005, computing and peripheral logic represented roughly 42% of total market revenue. Communications and consumer logic represented 29% and 20%, respectively, and automotive and other logic totaled 9%. The automotive and other segments are understandably small, as these equipment types are often implemented with microcontrollers as opposed to hard wired special purpose logic devices.

82

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Market Forecast: The market for special purpose logic is expected to outpace the overall market as electronic devices use an increasing amount of specialized processing power. There is also some effect of integration, as it is common for digital ASSP vendors to integrate increasing amounts of general purpose logic, embedded memory, and mixed signal circuits into their devices over time. Overall, we expect the market to grow at a 9% rate through 2010, rising from $37.5 billion in 2005 to $58.6 billion in 2010. Communications and automotive should see the strongest CAGRs at (12% and 11% through 2010, respectively), followed by consumer and computing (at 9% and 7%). Competitive Environment: Given the broad application and customer base that use special purpose logic, the market is rather fragmented. The top suppliers for PC applications include chipset and graphics vendors Intel, NVIDIA, ATI/AMD, Via, SiS, and Broadcom. For storage and peripheral, top vendors include Marvell, Agere, MediaTek, STMicroelectronics, LSI Logic, and Texas Instruments. Top consumer electronics IC vendors include STMicroelectronics, NXP/Philips, Sharp, Broadcom, Fujitsu, Mediatek, Zoran, Sigmatel, PortalPlayer, and Genesis. Note that the many of the big, vertically integrated Japanese IC makers that supply the consumer electronics market are excluded since their products would be classified as custom ASICs; the same goes for most suppliers into the video game console and handheld markets. Top vendors for communications include wireless IC leaders Qualcomm, Freescale, NXP/Philips, Infineon, Mediatek, and Agerenote that the majority of TIs wireless business is not included here, as its basebands are generally classified as DSPs. On the wireline side, key ASSP vendors include Broadcom, Conexant, Texas Instruments, Marvell, CSR, PMC-Sierra, Agere, and Atheros. The table below provides some additional detail, listing some of the higher volume special purpose logic devices and their key vendors. Market Segment Computing & Peripheral Device PC core logic Graphics controllers PC display controllers HDD SoCs & hard disk controllers ODD controller Printer SoC Communications Cellular baseband Ethernet controllers and switch ICs Wireless LAN baseband/MAC Bluetooth baseband/MAC/SoC Modem CPE SoC Telecom/Datacom Consumer Set-top box back-end Digital TV back end DVD decoder/encoder Digital camera image processor MP3/PMP decoder
Source: IDC, Dataquest, CIBC World Markets

Key Vendors Intel, Via, SiS, NVIDIA, ATI/AMD, Broadcom NVIDIA, ATI/AMD, Matrox Genesis, MorningStar, Realtek, Novatek, Pixelworks Marvell, Agere, STMicroelectronics, LSI Logic, Infineon, Renesas MediaTek, Renesas, NEC, Matsushita, Rohm, Ricoh, Toshiba, SeikoEpson STMicroelectronics, Texas Instruments, Avago/Marvell, Freescale, NEC Electronics, PMC-Sierra Qualcomm, Freescale, NXP/Philips, Infineon, Mediatek, Agere, Analog Devices, NEC (TI excluded as its basebands are classified as DSPs) Broadcom, Marvell, Intel, Realtek, Avago, Standard Micro, Vitesse, Agere Intel, Broadcom, Atheros, Marvell, Texas Instruments, Conexant, Airgo CSR, Broadcom, NXP/Philips, Texas Instruments, Infineon, STMicroelectronics Conexant, Broadcom, Texas Instruments, Infineon, Agere, Ikanos, Centillium PMC-Sierra, Agere, Infineon, AMCC, Vitesse, Cortina/Intel, Mindspeed, TranSwitch STMicroelectronics, Broadcom, Conexant, NXP/Philips NXP/Philips, Sharp, Genesis, ATI/AMD, Trident, Pixelworks, Renesas, Silicon Image, Zoran Mediatek, LSI Logic, Zoran, SunPlus, NEC, ESS, Toshiba, TI, Cheertek Renesas, SunPlus, TI, Zoran SigmaTel, PortalPlayer, Actions, NXP/Philips, TI, Telechips, Broadcom

83

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Display Drivers Display Drivers


Display Drivers - Standard logic devices designed to control and drive flat panel displays such as LCD or plasma. Key Types: LCD driver, plasma display driver, source driver, gate driver.

Market Forecast
We expect display drivers to grow at a 10% CAGR through 2010, from $6.3 billion in 2005 to $10.1 billion in 2010. Flat panel TV, desktop and notebook monitors, wireless handsets, and portable gaming are the primary growth drivers.
$12,000 Display Drivers $10,000

Key Competitors
Others 26 % Samsung 18 %

Matsushita 5% Sharp 5% Solomon Systech 5%

NEC 13 %

$8,000
Revenue ($M)

$6,000

Renesas 12 % Himax 7% Novatek 10 %

$4,000

$2,000

Source: Dataquest, CIBC World Markets Corp.

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: WSTS, CIBC World Markets Corp.

Display drivers are specialized logic devices designed to control and drive flat panel displays such as LCD or plasma. Traditional CRT display technology uses a phosphorous coated screen to trap light flashed from a light source; flat panel displays, on the other hand, use semiconductor driver ICs to manipulate pixels on an LCD or plasma screen. There are basically two kinds of drivers: source and gate. Source drivers usually sit across the top of the panel and generate signals that penetrate individual columns of pixels. Three source drivers are needed for each row of pixels: red, green and blue. Gate drivers usually run along the side of the panel and manipulate individual rows of pixels. The gate drivers intersect the signals driven by the source drivers, turning individual pixels on or off. Market Forecast: We expect display drivers to be one of the strongest segments in the semiconductor market, with a CAGR of 10%. Revenue should grow from $6.3 billion in 2005 to $10.1 billion in 2010. In some cases, growth is being driven by a migration from CRTsuch as desktop monitors and LCD/plasma TV. In other cases, LCD screens are an integral part of a category that itself is growing nicely, such as notebooks, wireless handsets, and portable gaming. In all cases, price pressure is being partially offset by larger screens which require more drivers. Competitive Environment: Display drivers are quasi-commodity parts, though obviously driver performance is critical to winning designs. The top suppliers in 2005 include Samsung, with 18% market share, NEC, with 13%, Renesas, with 12%, and Novatek, with 10%. Other significant suppliers include Himax, Solomon Systech, Sharp, Matsushita, Magnachip, Seiko Epson, and Oki Electric (which also acquired TIs LCD driver business in 2005).

84

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

General Purpose Logic General Purpose Logic


General Purpose Logic - Standard commodity catalog logic parts designed for use in multiple applications in various end markets. Key Types: Simple gates, flip-flop circuits, switches, clocks, registers.

Market Forecast
We expect general purpose logic to grow at a 5% CAGR over the next few years, rising from $1.4 billion in 2005 to $1.8 billion in 2010. Growth should trail overall semi market growth as integrated ASSPs eliminate the need for separate logic circuits in many applications.
$2,000 $1,800 $1,600 $1,400
Revenue ($M)

Key Competitors
STMicro 5% ON Semi 7% Others 7%

Texas Instruments 30 %

Renesas 9%

General Purpose Logic

$1,200 $1,000 $800 $600 $400 $200 $0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Fairchild Semi 12 % NXP/Philips 13 % Toshiba 17 %

Source: Dataquest, CIBC World Markets Corp.

Source: WSTS, CIBC World Markets Corp.

General purpose logic includes standard logic parts designed for use in multiple applications in equipment in various end markets. Though designed to perform a specific function, theses ICs are not tailored for use in any one application. Parts in this category are usually very low ASP, very high volume devices; examples include simple gates, flip-flop circuits and registers. Many of these parts are commodities, though some are more proprietary. Market Forecast: The market for general purpose logic is heavily tied to the overall semiconductor cycle, as these devices sell into every end market and application in the electronics landscape. They are also heavily commoditized and pricing can compress severely in down-cycles and rise in up-cycles. We expect growth in this segment to trail that of the overall semiconductor sector, as ASSP vendors continue to integrate additional functionality into their devices that eliminate the need for external general purpose circuits in many applications. Overall, we expect a 5% CAGR through 2010, as the market grows from approximately $1.4 billion in 2005 to $1.8 billion in 2010. Competitive Environment: The general purpose logic category includes a wide variety of device types, but most are commodities, and thus a few dedicated players dominate the market. Most of these are fully fabbed, though a few niche players are fabless. The top standard logic producer in 2005 was Texas Instruments, with 30% market share. Toshiba was second with 17%, NXP/Philips was third with 13%, and Fairchild was fourth with 12%. Renesas and On Semiconductor followed with 9% and 7%, respectively. Other competitors include STMicroelectronics, National Semi, IDT, Rohm, and Pericom.

85

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

ASICs ASICs
ASICs (Application Specific Integrated Circuits) - Customized ICs designed for specific customers to be used in specific platforms. Key Types: Standard cell (cell-based IC), gate array.

Market Forecast
We expect the ASIC market to grow at an 10% CAGR, from $9.0 billion in 2005 to $14.7 billion in 2010. New game console cycles should drive growth; in most other applications the trend away from ASICs should continue. The shift from gate arrays to standard cells should continue as OEMs demand higher performance devices.
$15,000 Standard Cells Gate Arrays $12,000

Key Competitors
Rohm 6% Samsung 6% Others 8% IBM 26 %

LSI Logic 8%

Sony 8%

Fujitsu 14 % Toshiba 11 %

Revenue ($M)

$9,000

NEC 13 %

$6,000

Source: Dataquest, CIBC World Markets Corp.

$3,000

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: WSTS, CIBC World Markets Corp.

Moving away from standard logic devices and into the custom-designed realm, the ASIC (application specific integrated circuit) market includes all customized integrated circuits designed for specific customers to be used in a specific platform. The most prominent examples of ASICs come from the game console market, where most silicon is designed for a single platform, which, given the high volume it is likely to generate, justifies the cost of designing a custom chip. Note that despite being classified under the logic category, ASICs can include analog, digital, or memory circuits, as long as they are custom designed. Most ASICs, however, are pure logic devices. ASIC devices are either designed by the equipment manufacturer in-house or contracted out to an ASIC house and the device is then fabbed by the ASIC house or a foundry. The level of cooperation between OEM and ASIC house can vary from program to program. ASICs are divided into two broad categories: Gate Arrays: Circuits consisting of fixed and regular arrangement of transistor cells forming a matrix of logic gates of various standard densities. An OEM using a gate array uses a series of software tools and macro libraries in order to design the mask for the top layers of the device, which determines its eventual function in the customer system. Standard Cells: Also known as cell-based ICs, these circuits consist of a user-specified arrangement of predefined and fixed sub-circuits of any function (analog, logic, memory). These ICs are designed from the

86

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

ground up and use standard dimensions of components or gates to pack them together more uniformly. These are more costly than gate arrays but are more efficient. ASICs tend to be the first solutions to be used in emerging technologies, as most electronic devices require silicon processing, but few standard, off-the-shelf components exist during the introduction phase. These ASICs tend to be costly to design, but usually deliver the performance necessary to power the device given the custom design; and the costs can be spread over the life of the product if it takes off. Also, given the early stage of the technology life cycle, the OEM is likely earning high gross margins, justifying the cost of the ASIC. As the technology matures and the market grows, other OEMs will enter the market with competitive offerings that either improve upon existing platforms or offer the same functionality at lower cost, or both (usually both). These competitive offerings will sometimes use newer ASIC solutions. Other times, they will leverage standard products (ASSPs) that may have come to market a few years into the life cycle of the emerging technology by IC vendors who were attracted by the growth prospects of the market as it took off. Over time, more and more OEMs will migrate to ASSP solutions, and the ASICs will disappear (and the ASIC vendors will be on to the next new market). Although there will always be a need for ASICs, most OEMs will choose a standard product if it is available. This is the primary reason why the transition from ASICs to ASSPs is a constant theme in semiconductors (though it varies by the stage of maturity in the end market). There are two notable exceptions: the first is where a single platform is likely to do so much volume that it justifies having a custom part designed for it. A prime example of this is the game console ICs mentioned above. The second case is where silicon performance is such a differentiating factor for the OEM that, to retain its competitive advantage, OEMs will develop ASICs instead of leveraging outside IC designers. A good example is Cisco, which views the performance of its switches and routers to be so integral to its strategy that it employs 250 ASIC engineers. Market Forecast: Overall, we expect the market for ASICs to grow at an 10% rate through 2010, rising from $9.0 billion in 2005 to $14.7 billion in 2010. We would normally expect the ASIC CAGR to trail that of the specialty purpose logic segment, largely due to slow but steady cannibalization of the ASIC market by ASSPs (specialty purpose logic revenue has outperformed ASICs for the past five years). However, the timing of new game console and handheld introductions (Sony PSP in 2004, Xbox 360 in 2005, Nintendo Wii and Sony PS3 in 2006), and the ramps that follow them, are reinvigorating this market for the next few years. Within ASICs, we expect the trend toward standard cell-based ICs to continue as the market continues to favor higher-performance devices (with lower performance devices most susceptible to ASSPs or FPGAs). Gate arrays should decline to less than 5% of total ASIC revenue over time. Competitive Environment: Many OEMs use ASIC houses to help design and manufacture custom ICs for high performance equipment, while others maintain internal semiconductor design teams that develop custom silicon designs that are then sent to an ASIC manufacturer for gate construction and fabrication. This yields a vendor list that includes both OEMs (e.g., NEC, Matsushita, Sony) and IDMs (IBM, STMicroelectronics, LSI Logic, Agere). The largest ASIC house in 2005 was IBM, with 26% total share and a major presence in the game console market as well as in ASICs for wireline and storage applications. Fujitsu was second with 14%, and NEC was third with 13%. Toshiba followed with 11% and Sony and LSI Logic each held 8%. Samsung and Rohm followed with 6% share. Other significant ASIC suppliers include Renesas, Matsushita, Agere, and Texas Instruments.

87

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

FPGAs/PLDs FPGAs/PLDs
Programmable Logic Devices (PLDs) - Standard logic circuits that can be configured into higher level logic patterns by the customer during system integration. Key Types: Programmable Array Logic (PAL), Field Programmable Gate Array (FPGA), Complex Programmable Logic Device (CPLD).

Market Forecast
We expect FPGAs to grow at a 12% CAGR, rising from $3.2 billion in 2005 to $5.7 billion in 2010. While communications will remain the focus, the emergence of low-cost highdensity FPGAs should broaden the application base for FPGAs.
$7,000 FPGAs $6,000

Key Competitors
Actel 6% Lattice 7%
Others 3%

Xilinx 51 % Altera 33 %

$5,000

Revenue ($M)

$4,000

$3,000

$2,000

Source: Dataquest, CIBC World Markets Corp.


$1,000

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: WSTS, CIBC World Markets Corp.

Programmable logic devices (PLDs) are standard logic circuits consisting of one or more switch matrices that can be configured into higher level logic patterns by the customer during system integration. With a PLD, the OEM purchases a catalog part and uses the PLD vendors software to design and configure the top layers of the device to perform the desired function. The OEM then inputs the architecture into the device, which stores it and adopts it for permanent use (some PLDs can be programmed multiple times). PLDs therefore provide the customizability of an ASIC without the need to design and fab new devices for each platform. There are several key types of PLD devices: Programmable Array Logic (PAL): Simple PLD devices with a small number of gates and a set number of inputs and outputs. Often used for glue logic. Complex Programmable Logic Device (CPLD): High-density PLD containing macrocells that are interconnected through a central Global Routing Pool. This type of architecture provides high speed and predictable performance. Generally the preferred architecture for implementing high speed logic. Field Programmable Gate Array (FPGA): High density PLD containing small logic cells interconnected through a distributed array of programmable switches. This type of architecture produces statistically varying results in performance and functional capacity, but offers high register counts. Programmability is typically achieved using SRAM or one-time-programmable antifuses.

88

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

In terms of end market use, PLDs are closely tied to the communications market. Approximately 47% (on a revenue basis) wind up in communications systems. Industrial consumes 21%, and data processing, primarily storage and server applications, consume an additional 12%. Consumer applications are generally not large users of PLDs, though newer flat panel TVs are adopting them; and in 2005 consumer represented 10% of the PLD market. Military/aerospace and automotive represent the remaining share of the market at 8% and 2%, respectively. The exhibits below display the end market usage for PLDs as well as top PLD applications by end market. 2005 PLD Revenue Share By Application
Military/ Aerospace 8% Automotive 2%

Top PLD Applications, By End Market Data Processing Servers Wireline Routers Broadband infrastructure SONET add/drop multiplexers MSPPs LAN switches SAN switches CO switches PBXs VoIP gateways Consumer LCD TV Plasma TV Digital CRT TV DVD recorders DVRs Set-top boxes Digital cameras/ camcorders Industrial Manufacturing systems Instruments Medical equipment Security/ energy management

Consumer 10 %

Enterprise storage Printers

Data Processing 12 %

Comm 47 %

Wireless Infrastructure (baseband subsystem)

Industrial 21 %

Source: Dataquest, CIBC World Markets Corp.

Market Forecast: Programmable logic has become an important solution in advanced wireline and wireless communications, with this end market representing nearly 50% of PLD revenues. As such, the market was hit especially hard in 2001 and 2002 as demand for components in the wired telecom infrastructure market declined significantly. The market rebounded in 2003 and grew at a healthy rate in 2004, with sequential annual growth of 42% and 12%, respectively. The rebound is mostly attributed to a recovery in wireline applications and wireless basestation sales. The market cooled in 2005, growing 5%, as the industry worked through some modest inventory in the first half of the year. Looking ahead, PLD vendors remain focused on the communications market, but are also broadening their product offerings to include smaller, lower-cost PLDs to compete in consumer, industrial, data processing applications that have traditionally favored ASICs or ASSPs. While communications should continue to outweigh this opportunity, these new markets should add to the long-term SAM for PLDs. In total, we expect programmable logic revenues to grow at a 12% rate, rising from $3.2 billion in 2005 to $5.7 billion in 2010. Competitive Environment: The PLD market is highly concentrated and has undergone a fair amount of consolidation in the past few years as the complexity and size (in terms of number of transistors) has steadily increased. In 2005, the top three vendors held over 90% share. Xilinx was the No. 1 vendor in 2005 with 51% share. Altera followed at No. 2 with 33% share and Lattice held 7%. Other PLD vendors include Actel, Cypress, Quicklogic and Atmel. Note that the PLD businesses of Agere and AMD have each been consolidated into Lattice.

89

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Memory Memory

Devices designed to store electrical data in a system, either temporarily or permanently

Memory:

Four Key Markets:


Volatile Memory DRAM SRAM Non-Volatile Memory Flash Mask ROM, EPROM, Other Non-Volatile Memory

Memory devices are digital logic circuits designed to store data on either a temporary or permanent basis. Though they are digital in nature, the manufacturing processes used to build memories are often quite different from a bulk CMOS process, especially in the case of DRAM and NAND flash. In nearly all cases, they are optimized for cost and high volume production. Volatile memory includes those memory devices that lose their stored information when they lose power. The most important types are DRAM (dynamic random access memory) and SRAM (static random access memory). Non-volatile memory retains the stored information when the system is powered off. The most important types are flash memory, mask ROM, EPROM and EEPROM.

90

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

DRAM DRAM
DRAM (Dynamic Random Access Memory) - The most common type of volatile memory, used primarily as system memory in PCs and servers. Currently the highest volume, highest density, lowest cost memory type. Key Types: SDRAM (Synchronous DRAM), DDR (Double Data Rate), DDR II, DDR III, QDR (Quad Data Rate), RDRAM (Rambus DRAM).

Market Forecast
We expect DRAM revenue to grow at a 5% CAGR, from $25.6 billion in 2005 to $33.0 billion in 2010. The market should remain volatile, up the most in boom years and down the most in downturns; pricing has historically swung wildly based on supply-demand trends. DRAM should remain heavily PC-based.
$45,000 DRAM $40,000 $35,000 $30,000
Revenue ($M)

Key Competitors
ProMOS 3% PowerChip 4%
Others 4%

Nanya 6% Elpida 7% Samsung 32 %

Qimonda/ Infineon 13 %

$25,000 $20,000 $15,000 $10,000 $5,000 $0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Micron 15 %

Hynix 16 %

Source: Dataquest, CIBC World Markets Corp.

Source: WSTS, CIBC World Markets Corp.

DRAM (dynamic random access memory) is the most common type of volatile memory, which is digital memory that loses its stored information once the power is removed. The dynamic heading means that DRAM, unlike SRAM, must be continually refreshed by the system, slowing down memory access, but allowing for a low-cost design and easy scalability to large densities. Like most memory, bit words in DRAM can be written, stored and read in any sequence. DRAM is used in all electronics markets, but has an especially high concentration in computing applications, which also tend to use the fastest and highest density DRAM devices. In 2005, approximately 75% of DRAM revenue was derived from PCs and servers. Approximately 35% of these were shipped with new desktop PCs, 13% into mobile PCs, 16% into servers and workstations (including memory modules), and the remainder into graphics cards and PC peripherals. Industrial, consumer and communications applications each represented less than 10% of DRAM bit shipments, as these applications are generally better suited to low-density DRAM as well as SRAM. In terms of memory type, the most common type of DRAM is DDR (double data rate), which can send twice as many bits across the bus than standard SDRAM (synchronous DRAM), effectively doubling throughput. This DRAM first appeared in 2001 andalong with the follow-on DDR2represented approximately 85% of bits in 2005. Standard SDRAM still accounts for 14% of bits, mostly in value PCs and in PC peripheral, consumer and other non-PC applications. Other types such as EDO and Rambus DRAM have basically withered away as they have lost support from Intel on its chipset designs.

91

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The exhibits below display 2005 DRAM unit shipments by application and by interface. 2005 DRAM Revenue By Application
Other 2% Desktop PCs 29%
Unit Shipments (M)

2005 DRAM Bit Shipments By Interface


4,000 3,500 3,000 2,500 2,000 1,500 1,000

Industrial 10% Consumer 8%

Comm 6%

Other Computing 15%


Servers & Memory Modules 13%

Notebook PCs 13%


Graphics Cards 4%

500 FPM/EDO DRAM SDRAM DDR SDRAM DDR2 SDRAM RDRAM

Source: Dataquest, IDC, CIBC World Markets Corp.

In addition to being classified by interface type, DRAM is quoted by density, with the most common size in use today being 256 Mb. The 128 Mb and 64 Mb devices are being phased out of mainstream PCs and are moving to non-PC applications. In 2004, 512 Mb started to ramp meaningfully and is expected to have crossed over 256 Mb in 2006. Still widely used, 16 Mb is typically found in non-PC applications. The exhibits below display 2005 DRAM unit shipments by density, as well as historical and forecast unit shipment data showing the constant migration to larger densities over time. 2005 DRAM Unit Shipments By Density
4,000 3,500 3,000
Unit Shipments (M)

DRAM Unit Shipments By Density, 1997-2010


100% 90% 80% 70% 60% 50% 40%
64Mb 16Mb 128Mb 256Mb 512Mb 2+Gb 1Gb

2,500 2,000 1,500 1,000 500 4Mb 16Mb 64Mb 128Mb 256Mb 512Mb 1Gb 2Gb

30% 20% 10% 0%


97

4Mb & below

00

01

02

04

05

07

09 20

98

99

03

06

19

20

20

20

20

20

20

08

Source: Dataquest, CIBC World Markets Corp.

Note that most DRAM sold into PCs and servers is not sold in chip form, and instead is sold onto standard memory modules. A standard SIMM or DIMM module is a small PCB containing eight or sixteen DRAM chips (there are eight bits to a byte, so, for example, a 256 MB DRAM module would usually have eight 256 Mb DRAM chips or sixteen 128 Mb DRAM chips). A DRAM module is pictured on the next page. In the server market, memory must be more reliable and robust, and therefore generally includes an extra DRAM IC along with a timing device to perform register functions; these are known as registered DIMM (R-DIMM) modules. In 2006, Intel-based servers began migrating to fully buffered DIMMs (FB-DIMMs), which replace the register with a high-speed serial I/O interface IC known as an AMB (Advanced Memory Buffer) directly on the module for interface to the server chipset.

92

20

19

19

20

20

20

10

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PC DRAM Module

Source: LSI Logic, Denali Software, CIBC World Markets Corp.

Market Forecast: DRAM is a commodity product; as such, its market is subject to periods of steep price inflation during expansions as capacity tightens and precipitous price declines during downturns as vendors drop prices to levels that approach variable cost. In 2001, the DRAM market saw its worst year ever, and prices cratered as vendors continued to fill their fabs despite slower demand. The market recovered in 2003-04 as PC demand rebounded and as OEMs took advantage of the low cost of DRAM by aggressively ramping the number of megabytes per PC. The market worsened slightly in 2005 despite relatively healthy PC demand, as pricing softened due to moderate oversupply. Looking ahead, the DRAM market should remain volatile, up the most in boom years and down the most in downturns; pricing should be volatile, varying with supply trends. Overall, we expect the market to grow modestly from 2005 levels, rising from $25.6 billion in 2005 to $33.0 billion in 2010, a CAGR of 5%. Note that our forecast assumes a cyclical peak in 2008 at $39.9 billion. Competitive Environment: With the exception of Micron in the United States and Qimonda/Infineon in Europe, all major DRAM vendors are located in Asia. The top vendor in 2005 was Samsung, which held an impressive 32% share. Hynix and Micron followed with 16% and 15%, respectively. Qimonda, the memory subsidiary of Infineon, was next with 13%. Elpida, the combined DRAM units of NEC and Hitachi, was No. 5 with 7%. Rounding out the top vendors are three Taiwanese DRAM players: Nanya (6% share), PowerChip (4%), and ProMos (3%).

93

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

SRAM SRAM
SRAM (Static Random Access Memory) - A type of volatile memory that is configured in a flip-flop circuit, allowing cells to remain without being refreshed. A higher performance memory than DRAM, used mostly in communications applications and for high-speed data caching. Key Types: Synchronous, asynchronous, PSRAM, FIFO, multi-port, CAM.

Market Forecast
We expect the SRAM market to grow at a 5% CAGR through 2010, from $2.8 billion in 2005 to $3.5 billion in 2010. Rising bit demand in wireless is likely to be offset by declining ASPs. Unlike DRAM, SRAM is unlikely to approach 2000 peak levels.
$4,000 SRAM $3,500

Key Competitors
Others 21 % Samsung 24 %

Hynix 3% Renesas 5% STMicro 5% Cypress 8% Intel 12 %

$3,000

$2,500
Revenue ($M)

$2,000

$1,500

Spansion 10 %

NEC 10 %

$1,000

$500

Source: Dataquest, CIBC World Markets Corp.

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: WSTS, CIBC World Markets Corp.

SRAM (static random access memory) is a type of volatile memory, which, like DRAM, loses stored information when powered down. Unlike DRAM, however, SRAM stores data bits in a flip-flop circuit using a minimum of four transistors, as opposed to the single transistor/single capacitor design used in DRAM. This allows current to flow through one side or the other based on which of the two transistors is activated, allowing for faster access time. The structure of SRAM means that it does not need to be refreshed, making it suitable for high-performance communications applications. Older asynchronous SRAMs perform read and write operations sequentially; newer synchronous SRAM overlaps reads and writes. In terms of application, 64% of SRAM (by revenue) is used in communications, mainly wireless. Fifteen percent is used in computing; and 10% is used in industrial. The exhibit to the right displays SRAM revenue by application.

2005 SRAM Revenue By Application


Industrial 10 % Automotive 2% Consumer 5% Wireless Handsets 50 % Computing 15 % Other 4%

Other Comm 14 %

Source: Dataquest, CIBC World Markets Corp.

94

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

SRAM is classified by both density and speed. The most common density is 8 Mbs, used primarily in mainstream handsets. Sixteeen Mb, 32 Mb, and 64 MB SRAMs are used in high-end handsets and in high-performance communications applications. Lower-density SRAMs tend to be used in computing, industrial and consumer applications. SRAMS are available in a wide variety of speeds. High performance communications applications typically require the fastest access times, typically 10-19 ns (nanoseconds) and below. However, the majority of SRAM unit shipments are 45-70 ns and above, used in most other applications that do not require quick access times. A major trend that has emerged in the past three years in the rise of PSRAM (Pseudo SRAM); which is actually DRAM modified to behave like SRAM. PSRAM has built-in refresh and address control circuits to mimic SRAM, but leverages the low-cost and high-density benefits of DRAM. Since PSRAM is sold into the same sockets as SRAM (primarily wireless), we include revenue and market share data here rather than in the discussion of DRAM. The exhibit below displays SRAM unit shipments by density and speed. 2005 SRAM Unit Shipments By Density
200 180 160 140
Unit Shipments (M) Unit Shipments (M)

2005 SRAM Unit Shipments By Access Time


700 600 500 400 300 200 100

120 100 80 60 40 20 Kb Kb Kb M b M b 1M 2M 4M 64 25 6 51 2 8M 16 32 64 12 8 M b M b

0-9 ns 10-19 ns 20-44 ns 45-70 ns 71+ ns Pseudo

Source: Dataquest, CIBC World Markets Corp.

Market Forecast: The SRAM market is less commoditized than the DRAM market given the more differentiated product offerings required for communication applications. Demand for SRAM, however, was negatively impacted in 2001 and 2002 by a weak environment for communications equipment. This, combined with pricing pressure as bit supply continued to grow, drove a 42% drop in revenues in 2001 and a 33% drop in 2002. SRAM recovered in 20032004 during the industrys cyclical expansion, but saw a 9% decline in 2005 as the communications market dealt with excess inventory. Looking ahead, we see growth in SRAM bit demand from both wireline and wireless, but expect declining ASPs to partially offset this growth. Also, longer-term, we expect SRAM demand to moderate as DRAM competes more effectively in certain applications. Overall we expect the market to grow 5%, from $2.8 billion in 2005 to $3.5 billion in 2010, though as with DRAM we expect a cyclical peak in 2008, at $3.6 billion. Competitive Environment: SRAM sells in a variety of densities and speeds and thus is a more fragmented market than DRAM. It also includes several vendors of NOR flash who include SRAM in their multi-chip packages for wireless handsets. The top supplier in 2005 was Samsung, with 24% market share. Intel was second, with 12% share; followed by NEC and Spansion, each with 10%; Cypress followed, with 8% share. Other important suppliers include STMicroelectronics, Renesas, Hynix, Sharp, and Toshiba. Although by definition still SRAM, certain types of proprietary memory warrant special mention, as they generally incorporate richer design content and are thus less commoditized than the general market. These include FIFO and multiport SRAMs, non-volatile SRAMs, CAMs (sometimes called network search engines) and FRAMs. The leaders in this group include IDT, Cypress, NetLogic, Oki, Hitachi, and Renesas.

95

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Flash Flash
Flash Memory - The most common type of non-volatile memory, which retains its stored information even when the power supply is removed. Key Types: NOR, NAND.

Market Forecast
We expect the flash memory market to grow at a 7% CAGR, from $18.6 billion in 2005 to $25.9 billion in 2010. NOR flash should decline at a 3% CAGR; while NAND grows 13%. NAND is benefiting from the growth of digital consumer devices and is also encroaching on NOR in handsets.
$30,000 NAND Flash NOR Flash $25,000

Key Competitors
Sharp 2% Renesas 4% SanDisk 5% STMicro 6% Hynix 8% Others 8%

Samsung 34 %

$20,000
Revenue ($M)

Spansion 9% Intel 10 %

$15,000

Toshiba 13 %

$10,000

Source: Dataquest, CIBC World Markets Corp.

$5,000

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: WSTS, CIBC World Markets Corp.

Flash memory is the most common type of non-volatile memory, which can store data continuously in its cells even when a power supply is removed. Like DRAM and SRAM, flash can be written and rewritten, but when the system is powered down, the memory cells retain all the information stored. Flash memory is used primarily in devices that are frequently turned on and off, but must keep its stored information intact because the device lacks an external storage device (like a hard disk drive). Fitting this profile are a number of very high volume markets: notably wireless handsets and consumer portables, which use flash memory to store the system operating software as well as user data. Flash is also used to store boot-instructions in a variety of electronic devices, including PCs. A large portion of flash memory is also sold as external flash cards, to be used for external storage of user data, primarily for digital consumer devices and PDAs. These use a variety of standard interfaces in order to make them compatible with a wide set of electronics from multiple vendors. Another application is USB flash drives, which combine flash memory and a USB controller, to form a small external storage device for PCs. Two main types of flash memory exist. NOR memory (43% of total flash revenue in 2005) is the most reliable flash memory, supporting random access and memory card program execution, known as execute in place. Erasing and writing can take several seconds, but reading is fast. For all these reasons, NOR is primarily used for program and code storage. The largest application for NOR is the handset market, which accounted for approximately 61% of revenue in 2005. Consumer was another 12%, followed by communications at 7% and data processing and industrial, each with 5%. Automotive and flash cards represented the remaining 6%.

96

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

NAND flash (about 57% of total flash revenue in 2005) is cheaper and higher density than NOR, but is somewhat less reliable. It can be read, written and erased at rates considerably faster than NOR. NAND is primarily used for file storage and is most often the type used in removable memory modules for consumer electronics. In fact, in 2005, roughly 48% of NAND flash (by revenue) was sold into flash memory cards, with another 17% sold into flash drives. About 22% of NAND revenue was derived from memory sold directly into consumer device. Six percent of flash revenue was derived from handsets, as NAND has begun shipping alongside NOR in multimedia handsets. The remaining 7% was sold into industrial, data processing, and other applications. The exhibits below display 2005 NOR and NAND revenue by application. 2005 NOR Flash Revenue By Application
Flash Cards Automotive and Other 2% 4% Industrial 5% Data Processing 5% Comm 11%

2005 NAND Flash Revenue By Application


Data Industrial Processing 3% 3% Handsets 6% Other 1%

Other consumer 22% Handsets 61%

Flash Cards 48%

Consumer 12%

Flash Drives 17%

Source: Dataquest, CIBC World Markets Corp.

Flash is also classified by size. NOR flash comes in all sizes ranging from 256Kb to 1Gb, with no single size range dominating the category. The highest volume density in terms of units, 64 Mb is the dominant solution in mainstream handsets. One hundred and twenty-eight Mb is used primarily in high-end handsets and 256 Mb is used in 3G phones. The lower density units are used primarily in non-handset applications. On the NAND side, most NAND flash is sold in densities of 256Mb and above, with a constant push upward as data storage demand in consumer portables is far above what can be economically supplied with todays NAND chips and is continually rising as these devices become more complex. The exhibits below display 2005 flash unit shipments for NOR and NAND by density. 2005 NOR Flash Unit Shipments By Density
600 550 500 450
Unit Shipments (M)

2005 NAND Flash Unit Shipments By Density


440 400 360 320
Unit Shipments (M)

400 350 300 250 200 150 100 50 0


Kb b b b b Kb M b 1G b 2G b M b M b M b M b 1M 2M 4M 8M M b 25 6 51 2 51 2 16 64 32 12 8 25 6

280 240 200 160 120 80 40 0


32 16 64 12 8 25 6 51 2 8G b 16 G b b b M b M b M b M b M b 1G b 2G b 4M 8M 4G b M b

Source: Dataquest, CIBC World Markets Corp.

97

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Market Forecast: The flash memory market is largely driven by demand for memory in wireless handsets, which represents the largest single use of NOR flash, as well demand for digital consumer devices, which drives sales of NAND flash. In 2001, the industry downturn had an impact on flash pricing and unit demand, though the effect was far more muted in flash than it was in DRAM or even SRAM. Overall, the market declined 29% in 2001 and saw a slight bounce in 2002 as pricing remained weak. In 2003, the flash market rebounded, up 51% to $11.7 billion, as demand for 2.5G handsets and consumer device-related flash memory drove strong growth of units and average density. This trend continued into 2004 and 2005, with the market growing 33% and 19%, respectively. Looking ahead, we expect growth for the overall market, but with a divergence between NAND and NOR. NAND should benefit from continued strong demand in consumer electronics, both for external storage cards and embedded applications (like MP3 players). NAND should also begin to take share from NOR in handsets, especially multimediaenabled handsets with megapixel camera, video, and MP3 capability. This should erode the NOR opportunity, itself plagued by lingering overcapacity. In total, we expect NAND to grow at a 13% CAGR, rising from $10.6 billion in 2005 to $19.2 billion in 2010. NOR, on the other hand, should decline at a 3% CAGR, from $8.0 billion in 2005 to $6.7 billion in 2010. Overall, we expect flash to grow from $18.6 billion in 2005 to $25.9 billion in 2010, a CAGR of 7%. Note that embedded in our forecast is some incremental growth from the cannibalization of mask ROM, EPROM, EEPROM and other non-volatile memories, though the bulk of this transition has already taken place. Competitive Environment: The top vendors of flash memory in 2005 included Samsung, Toshiba, Intel, Spansion, Hynix, STMicroelectronics, SanDisk, Renesas, and Sharp. Samsung has seen surging demand for NAND flash in consumer applications; in total the company held 34% share in 2005. Toshiba was No. 2 with 13% share, also mostly from NAND. Intel and Spansion followed with 10% and 9%, respectivelyboth entirely NOR-based, mostly from handsets. Hynix followed at 8%, entirely from NAND flash. Other major flash vendors include STMicroelectronics in NOR and NAND, SanDisk in NAND, Renesas in NOR and NAND, and Sharp in NOR. By flash memory type, the top vendors of NOR flash memory, which represented 43% of the market in 2005, were Intel at 26% and Spansion at 23%. STMicroelectronics followed with 14%, and Samsung and Renesas each had 7%. Other NOR vendors include Sharp, SSTI, Macronix, and Toshiba. NAND flash memory leaders were Samsung at 51% revenue share and Toshiba at 19%. Hynix followed with 12%. Sandisk had 8%, and Renesas had 3%. Other major NAND flash suppliers include, STMicroelectronics, Micron, and MSystems. The most recent newcomers to the NAND market include Qimonda/Infineon, STMicroelectronics (in cooperation with Hynix), Micron, and Intel (the latter two recently formed a JV called IM Flash Memory Technologies to produce NAND). The exhibits below display 2005 market share data for the NOR and NAND markets separately. 2005 NOR Flash Market Share
Toshiba 3% Macronix 3% Silicon Storage 5% Others 7%

2005 NAND Flash Market Share


Micron STMicro 2 % 2% Renesas 3% M-Systems Others 2% 1%

Intel 26 %

Sandisk 8%

Sharp 6%

Renesas 7%
Spansion 23 %

Hynix 12 %

Samsung 51 %

Samsung 7% STMicro 14 %

Toshiba 19 %

Source: Dataquest, CIBC World Markets Corp.

98

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Other Non-Volatile Memory Other Non-Volatile Memory


Mask ROM (Read-Only Memory) - Read-only memory that is locked in a permanent position during the fabrication process. EPROM (Electronically Programmable Read-Only Memory) - ROM that is programmed by electrical means during system integration. EEPROM (Electronically Erasable Programmable Read-Only Memory) - EPROM that can be erased and reprogrammed electronically.

Market Forecast
We expect the other non-volatile memory market to decline at an 8% CAGR, from $1.6 billion in 2005 to $1.1 billion in 2010.
$1,800 EEPROM and Other Memory $1,600 $1,400 $1,200
Revenue ($M)

Key Competitors
Catalyst Semi 3% Renesas 3% Infineon 3% NXP/Philips 4% Rohm 5% Microchip 8% Macronix 17 % Others 8% STMicro 18 %

EPROM Mask ROM

$1,000 $800 $600 $400 $200 $0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Oki 15 %

Atmel 15 %

Source: WSTS, CIBC World Markets Corp. Source: Dataquest, CIBC World Markets Corp.

Although flash has grown to dominate the non-volatile memory market, older technologies such as mask ROM, EPROM and EEPROM have remained for applications that require permanent memory storage, but dont necessarily need the easy re-writability of flash. Although similar in basic design, these devices differ slightly in composition and function: Mask ROM (mask read only memory): Mask ROM is read-only memory that is locked in a permanent position during the fabrication process through the use of a custom mask. Mask ROM is typically used for boot instructions, and is most commonly found in consumer appliances and printers. EPROM (electronically programmable read only memory): EPROMs are similar to mask ROMs except that they are programmed by electrical means rather than a mask. This includes OTP (one time programmable) and rewriteable devices that can be reprogrammed using ultra-violet light. EPROM is most commonly found in consumer electronics, game consoles, industrial applications, and some PC peripherals. EEPROM (electronically erasable programmable read only memory): EEPROMs are similar to re-programmable EPROMs, but can be erased and reprogrammed using electrical means. EEPROM is found in a wide variety of consumer, computing, automotive, and industrial devices.

In 2005, 19% of the other non-volatile memory category was mask ROM and another 24% was EPROM. The remaining 57% was other memoryprimarily EEPROM.

99

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Market Forecast: The market for legacy non-volatile memory has been in a steady decline as flash memory has penetrated the market. Looking to 2010, we expect the market to decline at an 8% rate, falling from $1.6 billion in 2005 to $1.1 billion in 2010. We expect EPROM to decline the fastest through 2010 at an 11% rate; Mask ROM and EEPROM should decline at 8% and 7% rates, respectively. Competitive Environment: The market for mask ROM is dominated by Macronix with a 90% share in 2005. SunPlus held a 7% share, followed by NEC with 2% and Amic with 1%. In EPROM, Oki increased its lead in the market in 2005, growing its share to 66%. STMicroelectronics remains a strong No. 2 with 19%, followed by Asahi Kasei with 9%. Other vendors include Cypress, Atmel, and Macronix. In EEPROM, the market leaders as of 2005 were Atmel and STMicroelectronics, at 26% and 24% share, respectively. Microchip was third with 13%, followed by Rohm with 8%. NXP/Philips, Infineon, Renesas, and Catalyst Semi each held about 6% share. The exhibits below display market share data for mask ROM, EPROM and EEPROM. 2005 Other Non-Volatile Memory Revenue Share 2005 Mask ROM Market Share
Sun Plus 7%
NEC 2% Amic 1%

Mask ROM 19%

EEPROM and Other Memory 57%

EPROM 24%
Macronix 90 %

2005 EPROM Market Share


Cypress 3% Others 4%

2005 EEPROM Market Share


Catalyst Semi 6% Renesas 6%
Infineon 6%
Others 4%

Asahi Kasei 9%

Atmel 26 %

STMicro 19 %

NXP/Philips 6%

Rohm 8%
Oki 66 %

STMicro 24 %
Microchip Tech. 13 %

Source: Dataquest, CIBC World Markets Corp.

100

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Discretes and Optoelectronics Discretes and Optoelectronics

Semiconductor devices that consist of a single transistor in a package

Discretes

Three Key Markets:


Discretes Sensors and Actuators Optoelectronics

This section deals with all types of discretes, which are semiconductor devices consisting of a single transistor in a package (i.e., not an integrated circuit). These devices sell in extremely high volume (hundreds of billions per year) at extremely low ASPs (sometimes as low as a penny each). They are used in all types of electrical systems primarily as enabling deviceshelping to move signals around the board or alternatively providing the bridge between the electronic system and the outside world. These devices are generally application agnostic, and are almost always sold through distribution. We discuss the three key types below: Discretes: Discretes are semiconductor devices consisting of a single transistor in a package. They are often used to direct signals around a board and to control power flow in a system. Sensors and Actuators: Sensors are specialized discrete semiconductor devices whose electrical properties can be translated into measurements of external stimuli. These include things like temperature, pressure, displacement, velocity, acceleration, stress, strain, or any other physical, chemical or biological property. Actuators are the opposite of sensors; actuators create mechanical motion by converting various electrical signals to rotating or linear mechanical energy. Optoelectronics: Optoelectronics are semiconductor devices designed to produce and receive light waves. This includes displays, lamps, couplers and other opto-sensing and emitting semiconductor devices.

101

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Discretes Discretes
Discretes - Discretes are semiconductor devices consisting of a single transistor in a package. They are used on the electrical board as key enabling devices, and are generally application agnostic. Key Types: Diodes, small signal transistors, switching transistors, power transistors, rectifiers (power diodes), thyristors.

Market Forecast
We expect the market for discretes to grow at a 6% CAGR, from $15.2 billion in 2005 to $20.3 billion in 2010. Growth for these multi-market devices should slightly trail the broader industry as ASSPs seek to integrate discretes in some high volume applications.
$25,000 Other Discretes Thyristors Rectifiers Power Transistors Small Signal Transistors Diodes

Key Competitors
Toshiba 9% Others 25 % STMicro 8%

Rohm 7% Fuji 3% NEC 3% ON Semi 4% Mitsubishi 5% NXP/Philips 5% Vishay 6%

$20,000

Renesas 7% Fairchild Semi 6% Infineon 6%

Revenue ($M)

$15,000

$10,000

Intl. Rectifier 5%

$5,000

Source: Dataquest, CIBC World Markets Corp.

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: WSTS, CIBC World Markets Corp.

Discretes are semiconductor devices consisting of a single transistor in a package. These devices sell in extremely high volume (hundreds of billions per year) at extremely low ASPs. They are used in all types of electrical systems primarily as enabling deviceshelping to move signals around the board as well as to control power flowand are generally application agnostic. There are several major categories of discrete devices, including: Diodes: Devices that act as a one-way valve. Key types include small signal diodes, zener diodes, transient protection diodes and RF/microwave diodes. Small Signal and Switching Transistors: Transistors with power dissipation of less than 1W. Power Transistors: Transistors with power dissipation of more than 1W. Key types include RF and microwave power transistors, bipolar general purpose transistors and insulated gate bipolar transistors (IGBTs) and modules. Rectifiers (Power Diodes): Devices used to convert AC (alternating current) to DC (direct current). These devices are classified by the amount of current supported. Thyristors: Devices that can turn on when activated by a gate signal. Other Discretes: These include varactor tuning diodes, selenium rectifiers and other polycrystalline devices.

102

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

On a revenue basis, more than half of discrete sales are power transistors, at 51% of total in 2005. Rectifiers, diodes and small signal transistors are 15%, 14%, and 13%, respectively. Note that on a unit basis, diodes and small signal transistors total 74% of discrete shipments, as these devices generally have very low ASPs. In terms of application, discretes go into every electronic device produced. In 2005, the revenue split was 25% consumer, 19% industrial, 18% data processing, 16% communications, 15% automotive, and 7% aerospace. The exhibits below display discrete semiconductor revenue share by product type and application. 2005 Discrete Revenue By Product Type
Thyristors Small Signal 5 % Transistors 13 % Other Discretes 2%

2005 Discrete Revenue By Application


Aerospace 7% Automotive 15 % Consumer 25 %

Diodes 14 %

Power Transistors 51 %

Comm 16 % Industrial 19 %

Rectifiers 15 %

Data Processing 18 %

Source: WSTS, Dataquest, CIBC World Markets Corp.

Market Forecast: Discretes sell into every electronic application and are therefore closely tied to the semiconductor cycle and to demand for electronics. Many of these products are commodities and thus are subject to rapid changes in pricing as fab capacity rises and falls with the semiconductor cycle. Countering this effect, however, is the fact that discretes and sensors are heavily used in industrial and automotive applications, which tend to have more stable unit demand than consumer and IT-focused products. Still, historically, discretes have shown more unit demand and ASP volatility than ICs, and the projected growth rates are lower than for other classes of devices. Overall, we expect the market to grow from $15.2 billion in 2005 to $20.3 billion in 2010, a CAGR of 6%. Note this CAGR is slightly below our semiconductor industry CAGR, as we have seen analog and digital ASSPs seek to integrate out discretes in a number of higher volume applications across end markets. Competitive Environment: The market for discretes is large and fragmented. The largest vendors in 2005 were Toshiba and STMicroelectronics, with 9% and 8% share, respectively. Rohm and Renesas were next with 7% each. Other important vendors include Fairchild, Infineon, Vishay, International Rectifier, NXP/Philips, Mitsubishi, ON Semi, NEC and Fuji. Most of the other large Japanese semiconductor manufacturers compete here as well.

103

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Sensors and Actuators Sensors and Actuators


Sensors and Actuators- Sensors are discrete semiconductor devices that translate real world input into electrical signals. Actuators perform the opposite function, taking electrical signals and translating them into some form of real world output. Key Types: Temperature sensors, pressure sensors, acceleration and yaw rate sensors, magnetic field sensors, LEDs, motors, solenoids, speakers.

Market Forecast
We expect the market for sensors and actuators to grow at a 12% CAGR, rising from $4.5 billion in 2005 to $7.9 billion in 2010.
$9,000 Actuators $8,000 $7,000 $6,000
Revenue ($M)

Key Competitors
Others 20 % Robert Bosch 17 %

Sensors

Micronas 4% Optek Technology 4% NXP/Philips 4%

Infineon 12 %

$5,000 $4,000 $3,000 $2,000 $1,000 $0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Mitsubishi 5%
VTI 5% Allegro MicroSystem 9%

Freescale 11 % Analog Devices 9%

Note: Forecast includes actuators beginning in 2003. Source: WSTS, CIBC World Markets Corp.

Note: Market share data is for sensors, and excludes actuators. Source: Dataquest, CIBC World Markets Corp.

Sensors are specialized discrete semiconductor devices whose electrical properties can be translated into measurements of external stimuli. These include things like temperature, pressure, displacement, velocity, stress, strain, acceleration, or virtually any other physical, chemical or biological property. The category also includes actuators, which translate signals into movement, acting as the opposite of a sensor (the mechanisms that move the head assembly on a disk drive or an arm of a robot are two types of actuator). In 2005, actuators represented 59% of total sensor and actuator revenue. Magnetic field sensors represented 17% of total, accelerate & yaw sensors were 12%, pressure sensors were 10% and temperature sensors were 1%. Note that image sensors are not included in this category and are instead grouped with optoelectronics. Market Forecast: The market for sensors and actuators is expected to grow at a 12% CAGR over the next five years, from $4.5 billion in 2005 to $7.9 billion in 2010. The growing digitization of devices, especially in automobiles, as well as the move to silicon-based sensors from other materials, is expected to drive this growth. Competitive Environment: The market for sensors is smaller and thus less fragmented than for general discretes, but many niche players have been able to build defensible positions. The largest vendor in 2005 was Robert Bosch with 17%, followed by Infineon with 12%. Freescale held 11% and Analog Devices and Allegro each held 9%. Other suppliers include VTI Technologies, Mitsubishi, NXP/Philips, Optek Technology, and Micronas. Note that this market share data does not include a breakout for actuators (data for this segment was not available).

104

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Optoelectronics Optoelectronics
Optoelectronics - Semiconductor devices designed to produce and receive light waves, including displays, lamps, couplers, image sensors, and other opto-sensing and emitting devices. Key Types: Displays, lamps, couplers, laser devices, image sensors (CCDs, CPDs, SSPs, CMOS), infrared devices.

Market Forecast
We expect the market for optoelectronics to grow at a 10% CAGR through 2010, rising from $14.9 billion in 2005 to $24.0 billion in 2010. After several years of rapid growth, we expect image sensors to see a more stable CAGR of 7% through 2010.
$30,000 All Other Optoelectronics Image Sensors $25,000

Key Competitors
Sharp 12 %

Sony 11 % Others 43 % Nichia Chemical 7% Toshiba 6% Avago 6% Matsushita 5%

$20,000
Revenue ($M)

$15,000

Stanley Electric SANYO 3% 3%

OSRAM 4%

$10,000

Source: Dataquest, CIBC World Markets Corp.


$5,000

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: WSTS, CIBC World Markets Corp.

Optoelectronics are semiconductor devices designed to produce and receive light waves. This includes displays, lamps, couplers and other opto-sensing and emitting semiconductor devices (though it excludes liquid crystal devices and displays, incandescent lamps and displays and other non-semiconductor optoelectronics components). The best-known type of optoelectronic device is the LED (light emitting diode), a diode that emits light when charged and is usually housed in a small bulb and used for readouts (see diagram). About 43% of optoelectronics revenue is actually derived from image sensors, including CCDs and CMOS image sensors. Seventeen percent is derived from lamps, 7% from laser pickup devices, and 7% from couplers. The remaining 26% includes infrared devices, displays, laser transmitters and other optoelectronics. Like discretes, optoelectronic semiconductors are generally application agnostic, selling into every type of electrical device. They are more heavily weighted toward the industrial and communications markets, which represented 33% and 27% of sales in 2005, respectively. Other markets include 20% sold into consumer, 15% into data processing and the remaining 5% into automotive and other.

Light Emitting Diode

Source: I.Weiss, CIBC World Markets Corp.

105

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The exhibits below display optoelectronics revenue by device type and application. 2005 Optoelectronics Revenue By Product Type
Other Opto 16 % Laser Transmitter 1% Displays 3% Infrared 6% Couplers 7% Laser Pickup 7% Lamps 17 % CCD & Image Sensors 43 % Consumer 20 %

2005 Optoelectronics Revenue By Application


Automotive 4% Military/ Aerospace 1%

Data Processing 15 %

Industrial 33 %

Comm 27 %

Source: WSTS, Dataquest, CIBC World Markets Corp.

Market Forecast: The market for optoelectronics is expected to outperform the overall semiconductor industry through 2010, growing at a 10% rate. After several years of rapid growth driven by digital cameras and camera phones, we expect image sensors to see a more stable CAGR of 7% through 2010, as pricing pressure partially offsets continued unit growth. Overall, revenue is expected to grow from $14.9 billion in 2005 to $24.0 billion in 2010. Competitive Environment: The market for optoelectronics is large and highly fragmented. The top producer in 2005 was Sharp with a 12% market share, followed by Sony with 11%. Nichia Chemical, Toshiba, Avago, and Matsushita each held 5% or more. Other significant vendors include OSRAM, SANYO, Stanley Electric, Rohm, Omnivision, Toyota Gosei, and STMicroelectronics.

106

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Section V: End Market Summary Section V: End Market Summary

This section discusses the major semiconductor end markets. For each end market we discuss: Key technologies & end market fundamentals End market unit forecast Semiconductor revenue forecast Semiconductor block diagram Competitive environment Semiconductor trends

107

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

End Market Summary End Market Summary

The semiconductor sector serves six key end markets

Six Key End Markets:

Computing Networking Telecom/ Datacom

Wireless Consumer Consumer Automotive

We now move from a discussion of semiconductor devices to the electronic systems that use them. We discuss several key equipment types grouped into six end market categories, and for each type of equipment, we discuss the end market unit forecast, our semiconductor revenue forecast, the competitive environment for both equipment and ICs, and market and technology trends. Note that in general, we limit the equipment types we discuss to those that are implemented primarily using application specific custom and/or standard ICs. There are a multitude of electronic devices that are implemented using mostly generic componentse.g., analog SLICs, microcontrollers, general purpose logic, discretes, etc.which, while they may ship in significant volume, do not present themselves as opportunities for specialization. We break down the key semiconductor end markets into the following categories: Computing: Equipment used in personal computing and data processing, including PCs and servers, PC displays, hard disk drives, optical storage, and printers/multi-function peripherals. Networking: Client and infrastructure equipment for local area, wireless, personal area and storage networks. Telecom/Datacom: Service provider and enterprise equipment used to implement voice and data networks. Wireless: Cellular handsets and wireless infrastructure equipment. WiMAX is also discussed. Consumer: Digital consumer devices, such as set-top boxes, DVD players, MP3 players, game consoles, etc. Automotive: Electronic systems and subsystems used in automobiles.

108

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Block Diagram Example: A Basic System Block Diagram Example: A Basic System
In the sections that follow, we will use block diagrams to identify key semiconductor opportunities. A typical electrical system block diagram looks like this:
Provides temporary or permanent storage of data used by the system processor Processes the signal at the highest level, implements system software, performs overall system control functions Controls power flow into the system as well as battery management in portable devices

Performs timing functions to enable system performance Senses real world input

Memory Memory Clock

Microprocessor/ Microcontoller/ Application Processor

Power Supply
Aids in system functioning by controlling signal flow, power delivery, etc.

Real World Input

Sensor

Analog Processor

Digital Processor

Digital Processor

Analog Processor

Real World Output

Source: CIBC World Markets Receives real-world input, performs analog processing, converts it to digital Performs digital processing on the incoming signal Performs digital processing on the outgoing signal Converts the digital signal to analog and produces real-world output

The diagram above, which we showed earlier in Section III when we introduced the different types of semiconductor devices, displays a generic electronic system. In the sections that follow, we show application specific electronic systems using the same format. We will use black to denote analog devices and white to denote digital devices. Double borders denote microcomponents, while a checkered pattern denotes memory. Note that in most cases, we will not show the multitude of enabling devices (grey) that populate the board.

109

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Computing Computing

Five Key Markets:


Desktop and Notebook PCs and Servers PC Displays - Flat Panel and CRT Hard Disk Drives Optical Storage Printers and Multi-Function Peripherals

Computing remains the largest single application for semiconductors and includes equipment such as desktop PCs, notebooks, servers, displays, hard disk drives, optical storage drives, printers and other peripherals. Although these devices are pervasive in the United States, Western Europe and Japan, emerging markets are major growth opportunities. In discussing the end markets for computing semiconductors, we focus on five key product areas: PCs and Servers: Includes all semiconductor devicesincluding memorythat are used on desktop and notebook motherboards and in industry-standard servers. Excludes semiconductors sold into hard disk and other mass storage devices, displays, printers and other peripherals, as well as proprietary servers. PC Displays: Includes all ICs sold into PC displays. While we include CRT monitors in our unit forecast, most traditional CRT displays are implemented using standard components such as microcontrollers, analog SLICs, discretes and general purpose logic. Flat panel displays, on the other hand, generally use ASSP solutions. Consequently, our discussion of this segment is heavily focused on flat-panel displays and their ICs. Hard Disk Drives: All hard disk drives used in PCs and servers, enterprise storage, and consumer applications. Optical Storage: Includes PC-based CD, CD-R/W, DVD, DVD-R/W and combo drives; excludes consumer DVD. Printers and Multi-Function Peripherals: All laser and inkjet based printers. Includes multi-function peripherals, which include printer, scanner, and fax functions.

110

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PCs and Servers PCs and Servers


PCs remain the largest single consumer of semiconductors and a huge portion of technology spending. After stalling during the post-Y2K tech downturn, growth has resumed, mostly in notebooks, servers, and emerging markets.
PC Unit Shipments and YoY Growth
250
PC Unit Shipments Year-over-year Growth

40% 35% 30% 25%

200

Unit Shipments (M)

20% 150 15% 10% 100 5% 0% 50 -5% -10% 00


4 19 85 19 86 19 87 19 88 19 89 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 19 8

-15%

Source: IDC, CIBC World Markets

PCs remain at the center of the technology universe and are a huge portion of IT and home technology spending. The market has matured a great deal over the past decade, with the commoditization of PCs forcing consolidation in the industry and average selling prices declining rapidly as system differentiation diminishes. Growth has slowed in the developed markets as corporate PC penetration has peaked, and consumer PC adoption has decelerated. Still, emerging markets present growth opportunities for PC makers and there is still a healthy replacement market in the developed regions. After the post-Y2K tech downturn drove anemic PC sales in 2001 and 2002, unit growth recovered nicely from 2003 to 2005 (averaging 14% per year), led by strong notebook growth. For 2006, PC and server shipments are expected to approximate 223 million units, up about 11%. While 11% is still a healthy number, it has taken the supply chain (and investors) some time to deal with the decline in growth from the former mid-teens rate.

111

Year-over-year-Growth

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PCs and Servers PCs and Servers


Desktops are the most mature market, seeing only incremental growth. Notebooks have gained significant ground against desktops as mobility trends have accelerated and LCD prices have come down; servers have also been strong.
PC & Server Unit Share
100%

Desktop Shipments
160 35% 30% 25% 20%
Unit Shipments (M)

140

120

100 15% 80 10% 60 5% 40 Desktops 20 Year-over-year Growth -5% -10% 1998 1999 2000 2001 2002 2003 2004 2005 0%

00

Notebook Shipments
70 Notebooks 60 Year-over-year Growth 35% 40% 30%
Year-over-year-Growth

Server
Unit Shipments (M)

50 25% 40 20% 30 15% 20 10% 10

Notebook
80%

5%

60%

00 1998 1999 2000 2001 2002 2003 2004 2005

0%

Server Shipments
7.0 35% Servers 6.0 Year-over-year Growth 30%

40%
5.0
Unit Shipments (M)

25%
Year-over-year-Growth

20%

Desktop

4.0

20%

3.0

15%

2.0

10%

0% 1998

1.0

5%

1999

2000

2001

2002

2003

2004

2005
0.0 1998 1999 2000 2001 2002 2003 2004 2005 0%

Source (all graphs): IDC, CIBC World Markets

By platform, desktops remain the largest market, still 65% of total units shipped in 2005. Desktops are also the most mature market, having grown just at a 4% CAGR over the past five years vs. the total market at 8%. In general, growth in desktops is being driven by emerging markets, as PC penetration in the developed world is pretty stagnant at present. Notebooks have grown at a much faster rate20% over the past five yearsand now represent 31% of shipments. This has resulted primarily from desktop replacement in both corporate and consumer PC channels. Notebooks have also increased the total market for PCs by offering the advantages of mobility, opening up new applications for PCs, primarily with thin-and-light notebook offerings. x86 servers have also outperformed, growing 12% over the past five yearsthough still small at 3% of unit shipments. Server OEMs have done an excellent job of packing more performance and functionality and improving ease-of-use into low-end and mid-range servers over the past few years, opening the market to more small and medium businesses, which have historically depended on either custom-built systems or simple small-business software running on a normal desktop PC. PC servers are also cannibalizing the mid- and high-end proprietary server market, offering cost and configurability advantages over proprietary systems from OEMs like IBM, Sun, and HP.

112

Year-over-year-Growth

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PCs and Servers PCs and Servers


PC shipments are shifting from the mature markets of North America, Western Europe, and Japan to developing economies in Asia, Eastern Europe, and Latin America. The distribution channel is growing in importance, serving local consumers in these fast-growing emerging markets. By customer type, PCs are split roughly 36% consumer and about 64% corporate and small business.
PC Unit Shipment Share By Region
100%

2005 PC Unit Shipments By Customer Type


Consumer 36%

ROW
80%

Asia Japan

60%

Europe
40%

20%

North America

Commercial 64%

0% 1998

1999

2000

2001

2002

2003

2004

2005

Source: IDC, CIBC World Markets

Source: Dataquest, CIBC World Markets

Looking at the market by geography, PCs have shown a sustained shift away from mature markets and toward developing economies, especially in Asia. North America remains the largest market for PCs, at 31% of total shipments in 2005. This market has grown at a modest CAGR of 4% over the past five years (note that the CAGR for North America was impacted heavily by the post-Y2K tech bubble; it has actually recovered nicely in the past couple of years, growing roughly 10% from 2003-2005, driven by healthy demand for notebook PCs). Growth in Western Europe23% of PC shipmentshas been slightly better at 8% over the past five years. Japan was the weakest region, with a flat CAGR over the past five years and declining to just 7% of total shipments. Asia and the other emerging markets, on the other hand, have seen runaway PC growth. Unit shipments have grown at a 15% CAGR over the past five years in Asia and in the rest of world segment. As a percentage of total shipments, Asia now accounts for 20% of total, up from 15% five years ago. Rest of world accounts for 19%, up from 14% five years ago. In terms of customer base, 36% of PC unit shipments in 2005 were sold into the consumer channel, with 64% sold into the commercial channel. These percentages have been relatively stable for the past few years, only slightly favoring the consumer PC market, due to two offsetting factors: 1) the consumer PC market in the U.S. and Europe has grown faster than the mature corporate sector, favoring consumer PCs and 2) Asia/Pacific has been the strongest area of growth, primarily for large and small enterprises, favoring commercial PCs.

113

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PCs and Servers PCs and Servers


Desktop and Notebook PC Unit Forecast
Total PCs growing 11%, from 201 million units in 2005 to 336 million units in 2010. Desktops growing 5%, from 136 million units in 2005 to 177 million units in 2010. Desktop shipments in developed markets should remain relatively flat through 2010, with growth mainly in Asia and ROW. Notebooks growing 19%, from 65 million units in 2005 to 159 million units in 2010. Notebooks should grow in all regions, though less in Japan given high penetration rates today.
400,000
Notebooks

Desktop and Notebook PC Suppliers


Dell 18 % Others 39 %

HP 16 %

Sony 2% Gateway 2%

Apple 2%

NEC 3%

Toshiba 3%

Fujitsu/F-S 4%

Acer 5%

Lenovo 6%

350,000
Desktops

300,000

250,000

200,000

150,000

100,000

50,000

Dell HP Lenovo Fujitsu/F-S NEC Gateway Acer Apple


2002 2003 2004 2005 2006 2007 2008 2009 2010

Desktops

Notebooks
Dell HP Toshiba Acer Lenovo Fujitsu/F-S Sony NEC

Source: IDC, CIBC World Markets Corp.

Note: The PC business of IBM was acquired by Lenovo in 2005. Source: IDC, CIBC World Markets Corp.

PC Unit Forecast: The past few years have seen a solid recovery in PC unit shipments following a challenging 200102. 2001 was a difficult year, as the industry posted its first decline in unit shipments since the introduction of the IBM PC in the early 1980s. The downturn in the U.S. and European economies prompted many corporations to pull in IT budgets, and falling consumer confidence and rising unemployment hurt sales of consumer PCs. Worse still, the falling unit number prompted a PC price war, exacerbated by a shift in strategy at the major PC OEMs to focus on services (which effectively lowered the margins they would need to charge on sales of PCs). This caused PC OEM revenue to fall far more dramatically than unit sales. 2002 was a stabilization year in unit terms, with worldwide shipments growing 3%. Substantially all of the growth came from the Asia/Pacific market with its strong white box channel. The market started to reaccelerate in 2003 with shipments growing 12% for the year, highlighted by a massive 28% surge in notebooks. In 2004, desktops and notebooks each grew double digits and the market as a whole grew 15% on strong economic activity globally. In 2005, desktop growth moderated to high single digits, though notebook experienced robust growth, up 33%; in total, the market grew 16% and reached 201 million units. Going forward, we expect PC unit growth to settle in at an 11% CAGR, growing from 201 million units in 2005 to 336 million units in 2010. We expect the trend away from desktops to continue, with a lagging 5% CAGR through 2010; overall we expect desktop PCs to grow from 136 million units in 2005 to 177 million units in 2010. We expect mobile, which is becoming more of a focus at every major PC and PC component supplier worldwide, to outperform with a

114

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

19% CAGR, rising from 65 million units in 2005 to 159 million units in 2010. Much of this growth will be driven by emerging markets, which should finally begin to see meaningful desktop replacement toward the back half of the forecast period. Within each platform, we expect the shift to lower price points to continue, though not as sharply as in the past few years. We believe most PC vendors have come to terms with the lack of elasticity in the market; i.e., their price cuts can be successful in winning market share but have been less than effective in growing unit demand for the industry as a whole. We also believe the market is sufficiently consolidated, with top OEMs in control of pricing. The exhibits below display our unit shipment forecast for desktops and notebooks. Desktop PC Unit Shipments, 2002-2010E
200,000 180,000 160,000 140,000
Desktops

Notebook PC Unit Shipments, 2002-2010E


180,000
Notebooks

160,000 140,000 120,000

120,000

100,000
100,000

80,000
80,000

60,000
60,000 40,000 20,000 -

40,000 20,000 2002 2003 2004 2005 2006 2007 2008 2009 2010 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: Dataquest, CIBC World Markets Corp.

PC Suppliers: The PC market is served by a two large double-digit share holders, Dell and HP, along with a number of tier-2 OEMs as well as a few notebook specialists, who together compose 61% of the PC market. A huge base of small white box or clone PC makers and system integrators, most of which operate in emerging markets, supply the remaining 39%. In 2005, Dell was No. 1 with 18% unit share and the top position in both desktops and notebooks; HP followed with 16% and was also No. 2 in both desktop and notebook. Lenovo was No. 3 with 6% share; note that Lenovo purchased IBMs PC business in 2005. Prior to the acquisition, Lenovo was the largest PC supplier in China and held a 2% share of the worldwide market. Acer, particularly strong in the European notebook market, was next with 5%, followed by Fujitsu/Fujitsu-Siemens with 4%. Notebook specialists Toshiba and NEC followed with 3% overall unit share each. Apple, Gateway, and Sony rounded out the top 10 with 2% unit share each.

115

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PCs and Servers PCs and Servers


Server Unit Forecast
Total servers growing at a 12% CAGR, from 6.5 million units in 2005 to 11.5 million units in 2010. x86 servers already represent 92% of the market in 2005 and are forecast to approach 94% of shipments by 2010, growing at a 12% CAGR. Itanium is expected to grow to 1.6% of total shipments by 2010. RISC should peak in 2007 at 613 million units, as it is challenged by x86 and Itanium; overall, RISC growth should be flat through 2010.
14,000
Itanium x86 RISC CISC

Server Suppliers
Industry Standard Servers (x86)
Others 19 % HP 31 %

Acer Sun 1 % 1% NEC 2%

Fujitsu/F-S 4%

IBM 16 % Dell 26 %

12,000

10,000

Low- and MidRange Proprietary


Sun IBM HP Apple Fujitsu/F-S

High-End Proprietary
IBM HP Sun Fujitsu/F-S NCR

8,000

6,000

4,000

2,000

2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, CIBC World Markets Corp.

Note: The semiconductors used in proprietary servers are not discussed in this document given their custom nature; however, we provide unit forecast and vendor data for context. Source: IDC, Gartner, CIBC World Markets Corp.

Server Unit Forecast: The server market has undergone a lot of change over the past decade, as the market has shifted away from proprietary systems designed from the silicon up by server OEMs. The market is now composed of a mix of proprietary RISC/CISC systems for high performance applications and industry standard (x86-based) servers used for mainstream IT infrastructure. As of 2005, industry standard servers had grown to 92% of unit shipments, though they were still less than half of server industry revenue due to the dominance of RISC/CISC at the high end. Like the PC market, server shipments were sluggish in 2001 and 2002 but grew double digit from 2003 to 2005, reaching 6.5 million units. We expect the market to reach 11.5 million units by 2010, a 12% CAGR. x86 servers represented 92% of the market in 2005 and are forecast to approach 94% of shipments by 2010, growing at a 12% CAGR. RISC should peak in 2007 at 613 thousand units, as it is challenged by x86 and Itanium; overall, RISC growth should be flat through 2010. Itanium, the high-end 64-bit processor architecture pioneered by Intel and HP, is expected to grow to 1.7% of total shipments by 2010 from a small base today. Server Suppliers: The x86 server market is served by three key players: HP with 31% unit share, Dell with 26%, and IBM with 16%. The rest of the market remains fragmented, with Fujitsu/Fujitsu-Siemens, NEC, Sun and Acer each holding less than 5% and representing less than 10% of the market in aggregate. The proprietary server market is served by Sun, IBM, and HP for low-end and mid-range and IBM, HP, Sun, and a number of small high-performance specialists at the high end.

116

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PCs and Servers PCs and Servers


Memory used to run programs One half of the PC core logic (also called the chip set), the north bridge interfaces between CPU and memory and graphics Generates timing frequencies to synchronize IC operations The other half of the core logic, the south bridge provides the PCI/PCI-E interface for internal peripherals, USB interface for external peripherals, and ATA/SATA interface for disk drives; it also integrates audio processing Manages PC power systems Stores boot instructions Controls keyboard, mouse, serial port, printer

DRAM DRAM Clock BIOS


ATA

Power Management

Hard Disk Drive Optical Disk Drive Floppy Disk Drive Super I/O

Microprocessor
Cache
High-performance 32-bit or 64bit CPU, runs the operating system and software

North Bridge

South Bridge
PCI, PCI AC 97, Express HD Audio

ATA/SATA I/O Controller

USB

PCI-Express x16/AGP Bus PCI/PCI-Express Bus DRAM Graphics Processor


DAC TMDS

Modem Codec Ethernet Controller Line Driver

BT/WLAN BB/MAC BT/WLAN RF/IF/PA Audio Codec

Graphics processor, memory, and interface ICs

Analog Monitor

Digital Monitor

LAN

Phone Line
Modem line driver and codec

Antenna

Speakers, Microphone

USB Devices

Keyboard, Mouse, Serial Port

Source: CIBC World Markets Corp.

Ethernet interface controller, including the physical layer interface (PHY) and protocol processor (MAC)

Bluetooth or wireless LAN radio generation (RF/IF/PA) and protocol processing (BB/MAC)

Audio DAC interfaces with speakers and microphone

Note: The diagram above shows a typical full-featured desktop PC. Notebook PCs contain some additional semiconductor content, including battery management and PC card controllers. Servers look slightly different in that they can have multiple processors, more memory, larger and less integrated chipsets and less multimedia content. The nature of computing requires a variety of powerful devices with diverse functionality to handle a wide set of data types. The easiest way to think of a PC is to draw an analogy to the human body. The CPU (central processing unit), a microprocessor, performs all high-level processing, essentially the high-level brain functions of critical thinking and problem solving. RAM (random access memory) acts as short-term memory, used by the CPU as a workspace to process its instructions. BIOS, another form of memory, stores boot instructions (similar to autonomic nerves), while a clock provides synchronization functions, keeping all systems in check. The chipset (includes north bridge and south bridge) acts as the base of the spinal cord, relaying instructions to and from the CPU and connecting it to the rest of the PC over the system bus (usually PCI or PCI-Express), which functions as the spinal column. The hard disk, floppy disk and CD/DVD drives provide long-term storage, which is retrieved by the CPU and moved into RAM when needed, similar to a human beings long-term memory. Together, these systems function as a cohesive central nervous system, powerful enough to perform complex computations and versatile enough to handle a wide variety of inputs. Outside these core systems, various peripheral devices perform the specific input and output processing to connect the PC to the outside world. The I/O controller interfaces with simple input devices such as the keyboard, mouse,

117

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

parallel and serial ports (used by older, pre-USB printers and other peripherals). Graphics processor and interface chips produce the visual image displayed on the monitor (allowing the computer to draw pictures for display), while audio controllers and codecs produce sound (allowing the computer to literally speak and be spoken to). Ethernet, modem and wireless LAN/Bluetooth ICs connect the PC with other data sources (analogous to a person placing a phone call) and USB/Firewire ports interface with additional peripherals (e.g., scanners, MP3 players, digital cameras). PC and Server Semiconductor Device Summary Processing Microprocessor: The central processing unit (CPU) in a PC is a high-performance 32- or 64-bit microprocessor. It performs all major processing of software code and communicates with all other devices in the system. It uses a modified CISC (Complex Instruction Set Code) architecture to process the lengthy sets of instruction used in typical computer software. Aside from this compute core, the CPU usually includes on-chip memory called a cache as well as additional logic extensions, which are embedded instructions designed to optimize the execution of certain types of critical functions (i.e., multimedia, security). Microprocessor performance is measured in clock speed, usually in Gigahertz (GHz), which measures the number of instructions per second (in billions) the processor can handle. Examples of PC processors are Intels Core 2 Duo and AMDs Athlon 64 X2. Note that desktop and notebook PCs usually have a single CPU, while servers can have 2 or more.

Chipset, BIOS, Clock and I/O Chipset (North Bridge and South Bridge): The chipset supports the CPU by performing some related processing functions and by connecting the CPU out to the rest of the system. The north bridge connects the CPU out to main memory and also to the graphics processor (many include a graphics processor on-chip). The south bridge is the gateway to other PC interfaces (besides graphics) and provides access to the hard disk drives and optical drives via the ATA/SATA, as well as I/O, networking, multimedia, and other peripherals via the PCI or PCIExpress bus. Examples of chipsets include Intels 965, Vias K8 series, and NVIDIAs nForce family. USB Controller and USB Host (integrated into south bridge): The USB host sits in front of the USB port and receives input from and sends output to a peripheral device using USB (Universal Serial Bus). USB has been widely adopted for its ease and speed: devices connect seamlessly and sometimes require no drivers. USB 2.0 can run at speeds up to 480 Mbps. IEEE 1394/Firewire Controller and Host (not shown): An alternative standard to USB is IEEE 1394/Firewire. Chipset leader Intel has not pushed this standard; thus most PCs that support Firewire use discrete controllers. BIOS: BIOS stores boot instructions read by the microprocessor whenever the computer starts up. BIOS is usually implemented with low density NOR flash memory. Clock: Motherboard clocks provide timing and synchronization functions, keeping all systems in the PC working together. Examples include clock families from IDT, Cypress, and Realtek. I/O Controller: The I/O controller interfaces the south bridge with the PCs primary simple input and output devices, including the keyboard, mouse, floppy drive, printer and serial devices. PC Card Controller (not shown, only used in notebooks): This controller is a specialty I/O device that provides the interface with PCMCIA cards. Controllers are only found in notebook PCs. Note that notebooks that support PCI-Express cards do not need a discrete controller, since this is integrated into the south bridge.

Memory DRAM (Dynamic Random Access Memory): PCs use DRAM to store program data in use by the CPU at any given time. DRAM is blank when the computer boots up and goes blank again when the computer is shut down. DRAM is implemented on modules that can be inserted and removed from the PC motherboard. DRAM is a commodity and is measured in MBs, segmented by interface (SDRAM, DDR, DDR-II, etc.). Memory Register or AMB I/O (not shown, only used in servers): Servers typically use higher performance memory modules than PCs. These are known as registered DIMM modules, and they include register ICs (similar to a timing IC) directly on the module. Newer fully-buffered DIMMs include a specialty I/O IC known as an AMB I/O IC directly on the module, which manages the interface to the server chipset, which would include AMB I/O logic on chipset. NAND Cache (not shown, only beginning to appear in 2007): PC OEMs are beginning to add a memory cache comprised of NAND flash to store frequently used operating system commands and other files.

118

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Graphics and Audio Graphic Processor (GPU): The graphics processor is a digital processor optimized to produce 2D and 3D graphics images. In desktops, GPUs are found on separate add-in cards (these cards usually contain DRAM specifically used for graphics); in notebooks they are down on the motherboard. GPUs connect up to the north bridge using PCI-Express x16, though older cards still use the specialized AGP graphics bus. Note that many systems eliminate the GPU by integrating graphics functionality into the chipset. Examples of GPUs include NVIDIAs GeForce 7 and ATI/AMDs Radeon family. Sound Processor (integrated into south bridge, can also be a discrete add-in board [not shown]): The sound processor is a digital processor that produces stereo and surround sound audio processing. Sound processors were originally implemented with add-in boards, and a minority of high-performance PCs implement separate PCI or PCI-Express sound cards with discrete audio processors. Most PCs, however, utilize the sound processor integrated into the south bridge, with discrete controllers reserved for the highest-end multimedia desktops. Audio Codec: The audio codec is a specialized digital-to-analog/analog-to-digital converter and amplifier that performs the conversion of audio signals from the digital sound processor to an analog format for use with speakers/headphones (output) and microphones (input).

Networking Note: Networking ICs, including devices found in PCs and servers, are discussed in further detail later on in the networking section. Analog Modem Codec and Line Driver: The modem codec encodes digital signals from the PC into analog signals for transmission over the phone line and vice versa. The line driver is the analog chip that interfaces with the RJ-11 phone line. In many PCs, a soft modem is used, eliminating the need for the discrete codec; a discrete line driver, however, remains. Ethernet Controller: The Ethernet controller is a single-chip integrated device that interfaces between the PC and the Ethernet local area network. It integrates both the MAC and PHY. The Ethernet MAC encodes data from the PC into Ethernet frames and affixes overhead and vice versa. The PHY (physical layer) interfaces with the RJ45 Ethernet cable and converts the serial Ethernet stream into parallel digital form for processing and vice versa. Bluetooth/Wireless LAN Baseband/MAC and RF/IF/Power Amp: Many PCs include wireless networking functionality, especially notebooks. Wireless LAN is the most common for PCs, but some will also include Bluetooth. In most implementations, a baseband/MAC chip encodes data from the PC into 802.11 or Bluetooth format and interfaces with the radio. The RF/IF and power amp performs the radio transmission and reception and interfaces with the antenna. Note that we have displayed the chipset as a 2-chip solution; in reality chipsets can be less integrated and will split up some of the functions (separate power amps, multiple radios, specialized MACs, etc.); others will include the solution on a single chip. Also note that the diagram shows a system with either a Bluetooth or wireless LAN solution; in reality a system can have one or the other or both. UWB/Wireless USB chipset (not shown, only beginning to appear in 2007): In 2007, PCs will begin to include UWB solutions to implement wireless USB. Solutions will likely look very similar to wireless LAN or Bluetooth implementations, with a two chip solution comprising a baseband/MAC and RF/IF. Unlike wireless LAN or Bluetooth, however, UWB is targeted for short-range cable replacement as opposed to a full networking solution.

Power Management Note: Our block diagram does not show specific power management components; these are used throughout the board and on add-in boards as well as within and around the PC power supply. Power Management Power management content in a PC includes both analog ICs and discretes. Key types include power transistors, power diodes, AC/DC regulator ICs, PFC pre-regulator ICs, linear regulator ICs, battery charging and management ICs, hot-swap controllers and MOSFET driver ICs.

119

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PCs and Servers PCs and Servers


PC and Server Semiconductor Forecast
PC and server ICs growing 7%, from $59.5 billion in 2005 to $84.3 billion in 2010. Microprocessors to grow 7% as unit growth is tempered by pricing pressure. DRAM to grow 5% with a peak in 2009. NAND cache should enter PCs in 2007 and grow quickly to $1.6 billion in 2010. Chipset, I/O, graphics and audio to face integration and price pressure. Communications growing at 12%, driven by wireless networking ICs. Power management growing 10%, boosted by notebook penetration.
$90,000 $80,000 $70,000
Semiconductor Revenue ($M)

2005 2010E 05-10 Market Size Market Size CAGR Microprocessor Main Memory NAND Cache Chipset and I/O Core Logic All Other I/O Graphics and Audio Discrete GPU Audio Communications Ethernet Controller Analog Modem Wireless LAN Bluetooth UWB Power Management $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ 30,748 14,491 6,702 5,501 1,201 3,542 3,246 296 1,402 622 286 422 72 2,643 $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ 42,865 18,623 1,613 9,652 8,492 1,160 4,916 4,496 420 2,474 737 226 930 171 411 4,180 7% 5% N/A 8% 9% -1% 7% 7% 7% 12% 3% -5% 17% 19% N/A 10%

Power Management Communications Graphics and Audio Chipset and I/O NAND Cache Main Memory Microprocessor

$60,000 $50,000 $40,000 $30,000 $20,000 $10,000 $0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Desktops Notebooks Servers

$ $ $

32,487 20,202 6,840

$ $ $

32,150 40,752 11,421

0% 15% 11%

Total PC & Server Semiconductors


Data is revenue in millions

59,529

84,324

7%

Source: IDC, Dataquest, CIBC World Markets Corp.

Note: Our entire discussion of PC and server ICs includes chips used in desktop PCs, notebooks, and x86 servers. It excludes chips used in proprietary RISC/CISC servers, as these are typically custom solutions that are difficult to cost within the server BOM. It also excludes LCD drivers used in notebook PCs, as these are included in our PC display forecast. Also note that our forecast includes all PC and serverrelated ICs used in full systems, add-in boards, and the retail after market. PC and Server Semiconductor Forecast: We estimate the computing semiconductor marketwhich includes desktop, notebook and x86 server semiconductorstotaled $59.5 billion in 2005, up 5% year over year. The market benefited from healthy PC units and a rational pricing environment for microprocessors, chipsets and graphics, partially offset by a difficult DRAM market, which declined 10% year over year. Excluding DRAM, the computing semiconductor market would have been up 11%. Looking to 2010, we expect the combination of PC unit growth, a recovery in DRAM pricing, and the addition of new features to drive a 7% CAGR and 2010 TAM of $84.3 billion. Tempering the growth rate somewhat is a more intense pricing environment in the PC microprocessor segment - significant since it represents about half the market revenue. Drilling down further, we segment the market by subsystem, consistent with the IC segments we used in the block diagram on page 117. The largest device segment in 2005 was microprocessors, at $30.7 billion, roughly 52% of industry revenues by our estimate. Though the core processing device of the computing industry, microprocessor

120

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

revenue growth should trail units as competition from Intel and AMD combined with moderating PC demand pressures pricing. We expect microprocessor revenue to grow at a 5-year CAGR of 7% to $42.9 billion in 2010, or 51% of total IC computing revenue. Chipsets and I/O totaled $6.7 billion in 2005; we expect this segment to grow 8% to $9.7 billion in 2010. The PC core logic (north bridge and south bridge) is the largest piece at 82% of total in 2005. I/O controllers represent another 5%, and the BIOS, PC clocks, PC card controllers, Firewire controllers, and security devices represent the balance. Main memory totaled $14.5 billion in 2005, down 10% YoY on weak DRAM prices. Main memory is expected to remain a volatile market, peaking at $19.1 billion in 2009 and then declining to $18.6 billion in 2010, a 5% CAGR over the forecast period. Note that in our forecast we include memory registers and AMB I/O products, which are expected to grow from just over $60 million in 2005 to $175 million in 2010 as fully buffered DIMMs gain traction. NAND flash is beginning to make its way into PCs in the form of a cache, and we project 10% of PCs will ship with NAND in 2007. Though penetration in the out years is hard to project, we forecast $1.6 billion in revenue in 2010. The graphics and audio subsystem, at $3.5 billion in 2005, is expected to grow 7% to $4.9 billion in 2010. Revenue growth should trail PC units as discrete GPUs continue to be eroded by integrated graphics, though the trend has been noticeably decelerating. On the audio side, discrete audio controllers should grow 5%, again affected by integrated chipsets; audio codecs should grow 9% through 2010 as price erosion partially offsets unit growth. Communications is the fastest growing segment in the PC semiconductor market with a CAGR of 12%, growing from $1.4 billion in 2005 to $2.5 billion in 2010. Ethernet controllers should grow only modestly (3% CAGR) as penetration remains high and ASPs continue to decline. Analog modems should decline (-5% CAGR), as OEMs design analog modems out of PCs and as pricing gradually declines. Wireless LAN should experience solid growth (17% CAGR), benefiting from increased PC penetration and robust notebook growth. Bluetooth should also outgrow the market (19% CAGR) on rising PC penetration, but should still represent only a small portion of total. Ultra wide band (UWB), still in its infancy today, should exceed $400 million in 2010 by our estimate. Finally, power management is expected to grow at a 10% CAGR, from $2.6 billion in 2005 to $4.2 billion in 2010. The faster growth in notebooks and servers, each of which include more power management content than desktops, should offset price erosion in this market. The exhibits below depict our forecasts for the various PC and server semiconductor sub-systems. On the following page, we display individual device forecasts by platform (desktop, notebook and server). PC & Server Semiconductor Subsystem Forecasts Microprocessor
50,000 45,000 40,000
Semiconductor Revenue ($M) Semiconductor Revenue ($M)

Memory
25,000 NAND Cache Main Memory (DRAM)

Chipset, BIOS, Clock and I/O


12,000 Other Device I/O PCMCIA Controller Firewire/IEEE 1394 PC Clock BIOS Memory I/O Controllers Core Logic

Microprocessor

10,000 20,000
Semiconductor Revenue ($M)

35,000 30,000 25,000 20,000 15,000 10,000 5,000 2002 2003 2004 2005 2006 2007 2008 2009 2010

8,000

15,000

6,000

10,000

4,000

5,000 2,000

2002 2003 2004 2005 2006 2007 2008 2009 2010

2002 2003 2004 2005 2006 2007 2008 2009 2010

Graphics and Audio


6,000 Audio Codec Audio Controller 5,000 Discrete Graphics Processor 2,500 3,000

Communications
4,500 UWB Bluetooth Wireless LAN Analog Modem Ethernet Controller
Semiconductor Revenue ($M)

Power Management
4,000 3,500 3,000 2,500 2,000 1,500 1,000 Power management

Semiconductor Revenue ($M)

Semiconductor Revenue ($M)

4,000

2,000

3,000

1,500

2,000

1,000

1,000

500 500

2002 2003 2004 2005 2006 2007 2008 2009 2010

2002 2003 2004 2005 2006 2007 2008 2009 2010

2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, Dataquest, CIBC World Markets Corp.

121

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

2005 PC Semiconductor Revenue By Platform


40,000

Desktop PC Semiconductor Forecast


Microprocessor NAND Cache Graphics and Audio Power Management Main Memory Chipset and I/O Communications

Servers 16%

35,000

30,000
Semiconductor Revenue ($M)

25,000

20,000

Desktops 53% Notebooks 31%

15,000

10,000

5,000

2002 2003 2004 2005 2006 2007 2008 2009 2010

Notebook PC Semiconductor Forecast


45,000 Power Management 40,000 35,000
Semiconductor Revenue ($M)

x86 Server Semiconductor Forecast


12,000 Power Management Communications Graphics and Audio Chipset and I/O NAND Cache Main Memory Microprocessor

Communications Graphics and Audio Chipset and I/O NAND Cache Main Memory Microprocessor
Semiconductor Revenue ($M)

10,000

30,000 25,000 20,000 15,000 10,000

8,000

6,000

4,000

2,000 5,000 2002 2003 2004 2005 2006 2007 2008 2009 2010 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: Dataquest, CIBC World Markets Corp.

122

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PCs and Servers PCs and Servers


PC Semiconductor Market Key Competitors
Microprocessor Revenue
AMD 12 % Others 2%
100% 90% 80% 70% 60% 50% 40%

CPU Unit Share

Intel 86 %

30%
Others

20% 10% 0%

AMD Intel

Total Units

Desktop

Mobile

Server

Chipset Revenue
Broadcom Others ATI/AMD 1 % 1% 6% NVIDIA 6% SiS 6%
100% 90% 80% 70% 60% 50%

Chipset Unit Share

Via 9%

40% 30% 20% 10%

Intel 71 %

Others ATI/AMD NVIDIA SiS Via Intel

0% Total Units Desktop Notebook Server

Source: IDC, CIBC World Markets Corp.

The PC semiconductor competitive environment is divided firmly along the lines of devices and subsystems. Although many vendors play in several devices within the PC, the standards, design competencies and competitive dynamics in each subsystem are sufficiently different to warrant separate discussion of each subsystem. In the exhibit above, we display 2005 revenue market share data for microprocessors and chipsets, as well as unit data for desktop, mobile, and server. On the next page, we display discrete graphics controllers, DRAM, BIOS, and power management. The page following displays PC communications market share, plus PC clocks, audio controllers, and I/O controllers. Microprocessors: The PC CPU market is among the most concentrated in the semiconductor industry. In 2005, Intel held roughly 86% revenue share and 79% unit share, down roughly a few percentage points from 2004. AMD gained a few percentage points and held 12% revenue share and 18% unit share. AMD is well positioned in consumer desktop, with a strong presence at both the low end and also for the very high end for gamer systems. AMDs presence in mobile and server also improved in 2005 as it ramped the Opteron for servers and the Turion 64 for thin & light notebooks; OEMs serving the corporate desktop market are also beginning to offer AMD based systems. Other microprocessor vendors share 2% revenue share and 3% unit share. These include Via and Transmeta, as well as IBM and Freescale, which supplied CPUs for Apple in 2005, before the transition to Intel began in 2006. Chipsets: Intel dominates the PC chipset market, with 71% revenue share in 2005. The balance is split between Taiwanese chipset makers Via and SiS and graphics leaders NVIDIA and ATI (acquired by AMD in 2006). For servers, Intel and AMD each supply chipsets for their processors, and Broadcom and NVIDIA also support AMD.

123

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PCs and Servers PCs and Servers


PC Semiconductor Market Key Competitors
Graphics
Matrox 3%

DRAM
Elpida ProMOS 4 % 4% Powerchip 5% Nanya 7% NVIDIA 50 % Others 1% Samsung 28 %

Others 2%

ATI/AMD 45 %

Qimonda/ Infineon 15 %

Micron 17 %

Hynix 19 %

PC BIOS (NOR Flash)


Others 13 % SST 28 %

Power Management
Fairchild 13 % Others 25 % Texas Instruments 9% STMicro 7% Intl. Rectifier 7%

Toshiba 4% Samsung 6% Winbond 6%

Linear Tech 2% Lite-On 3% Vishay 3% NXP/Philips 4% Fuji Electric 4% Mitsubishi 5%

Sharp 8% STMicro 10 %

Macronix 13 % Spansion 12 %

On Semi 7% Maxim 5% Toshiba 7%

Source: Dataquest, Mercury, CIBC World Markets Corp.

Graphics: The graphics market has consolidated and ATI (acquired by AMD). Excluding integrated total share in 2005, gaining share at the high end manufacturing issues. ATI fell to No. 2 at 45% of

in the past few years into the hands of two key players: NVIDIA graphics chipsets, Nvidia jumped to the No. 1 position with 50% of the desktop market as ATI experienced product delays and the market.

Memory: The top vendor of DRAM for PCs in 2005 was Samsung, which held 28% share in 2005. Hynix followed with 19%, Micron with 17% and Infineon (now Qimonda) with 15% share. Other vendors include Nanya, Powerchip, ProMOS, and Elipida. Note that this data differs slightly from the numbers for DRAM as a whole, which includes nonPC DRAM (generally lower density and lower speed). Note that in the server market, memory registers for R-DIMMs are supplied by IDT/ICS, TI, and Hitachi. AMB devices for FB-DIMMs are supplied by IDT, Intel, NEC, and Inphi. BIOS: BIOS is generally implemented with low density NOR flash. The top vendors for PCs in 2005 include SST, Macronix, Spansion, STMicro, Sharp, Winbond, and Samsung. Again, share differs from the broader NOR market. Power Management: The PC power management market, which includes vendors of discretes and analog components, is highly fragmented. The top vendors of PC power components in 2005 were Fairchild and Texas Instruments, with 13% and 9% share, respectively. STMicroelectronics, International Rectifier, On Semi, and Toshiba followed, each with 7% share. Other significant vendors include Maxim, Mitsubishi, Fuji Electric, NXP/Philips, Vishay, LiteOn Semiconductor and Linear Tech.

124

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PCs and Servers PCs and Servers


PC Semiconductor Market Key Competitors
PC Ethernet
Others 3% Marvell 13 %
Agere 22 %

PC Analog Modem
Analog Devices 3% Others 4%

PC Wireless LAN
Others 5%
Broadcom 11 %

Realtek 14 %

Broadcom 41 %

Conexant 41 %

Atheros 13 %

Intel 29 %

Silicon Laboratories 30 %

Intel 72 %

Audio Controller & Codec


Others 9% IDT/Sigmatel 7%
Cypress 15 %

PC Clocks
Realtek 7%

Others 3%

I/O Controllers
Standard Micro Winbond/National Semi ITE

Analog Devices 12 %

Creative 45 %

Bluetooth
CSR Broadcom NXP/Philips Texas Instruments Infineon

Realtek 27 %

IDT/ICS 75 %

Source: IDC, Dataquest, CIBC World Markets Corp.

Ethernet: PC Ethernet controllers are supplied by Broadcom, Intel, Realtek, and Marvell. Broadcom and Intel held 41% and 29% of the market, respectively, with Realtek and Marvell sharing the balance. Note that Marvell supplies its own controller to the market and also derives revenue from a partnership with Intel (Intels MAC, Marvells PHY). Modems: The top supplier of PC analog modems in 2005 was Conexant, with 41% market share. Silicon Labs, with its silicon DAA (Digital Access Arrangement), held the No. 2 spot with 30% share, followed by Agere at 22%. Wireless LAN: The largest supplier of 802.11 chipsets for PCs in 2005 was Intel, which bundles WLAN with the Centrino platform. Atheros and Broadcom were also significant players; others include Conexant, TI, and Marvell. Audio: Creative dominates the PC audio add-in board market, and therefore has a 45% share of PC audio IC revenue. The remaining 55% is comprised of codec vendors, Realtek being No. 1 with more than 50% share of audio codecs and therefore 27% of the total PC audio market. ADI followed with 12% and IDT/Sigmatel held 7%. PC Clocks: The PC clock market is dominated by IDT/ICS (IDT purchased ICS in 2005), which held roughly 75% market share in 2005. Cypress was No. 2 with 15% share, followed by Realtek at 7%. I/O Controllers: Standard Microsystems, Winbond/National Semi (Winbond acquired National Semis I/O controller business in 2005), and ITE supply I/O controllers. Bluetooth: The top Bluetooth suppliers in 2005 were CSR, Broadcom, NXP/Philips, TI, and Infineon.

125

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The exhibit below displays the competitive landscape for PC semiconductors. PC Semiconductor Competitive Landscape
Graphics Controller Audio Power Management

PC Ethernet PC Wireless LAN

Microprocessor

BIOS (PC NOR)

Analog Modem

NAND (Cache)

I/O Controller USB/Firewire

Intel AMD/ATI Samsung Hynix Semiconductor Micron Technology Infineon NVIDIA Via Technologies Silicon Integrated Solutions Realtek IDT/ICS Cypress Semiconductor Standard Micro SST STMicroelectronics Creative Technology Texas Instruments Broadcom Conexant Agere Systems Marvell Technology CSR Fairchild Semiconductor
Source: IDC, Dataquest, Mercury Research, CIBC World Markets Corp.

Bluetooth

PC DRAM AMB

Chipset Clock

= >15% market share

= 5%-15% market share

< 5% market share

126

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PCs and Servers PCs and Servers


PC Semiconductor Market Trends
Microprocessors are moving to dual and multi-core. Intel and AMD now compete head-to-head in all major segments of the microprocessor market; Intel-only shops are disappearing. Integrated graphics chipsets have gained momentum as graphics quality improved; share gains are decelerating, however, and in fact may reverse temporarily with the launch of Windows Vista. PCI-Express has arrived; so has SATA. The memory bus is continuing the move to DDR2; DDR3 is on the horizon and FB-DIMM is penetrating servers. Gigabit Ethernet pervades the corporate desktop and is ramping in the consumer PC. Wireless LAN is now standard in mainstream notebooks; Bluetooth penetration has been sluggish but UWB should be pervasive. USB 2.0 is pervasive; IEEE 1394/Firewire is still around but giving way without the support of Intel. Power consumption is becoming a major issue for all component suppliers into PCs and servers. NAND will begin to penetrate PCs in 2007.

Microprocessors are beginning to move to dual and multi-core. Historically, Intel and AMD have achieved greater performance with their microprocessors by focusing on driving Moores Law in order to ramp up clock speeds for their devices, as well as by boosting the cache size and increasing the pipeline. In the past few years, however, they have been challenged by spiraling heat dissipation figures as they ramp clock speed, and have had to look for new ways to drive performance. Both have opted to increase the number of compute cores on their CPUs, allowing parallel processing to take the place of raw speed. As mentioned above, both Intel and AMD have been aggressively moving to dual-core. Intel first introduced dual core for the desktop in 1H05, and followed with a server model in 2H05. AMD did the reverse; dual core server chips launched before desktops. Intel has disclosed aggressive targets for dual-core penetration: more than 75% for desktop, 90% for mobile, and 85% for server exiting 2006; these figures move to 90% for desktop and mobile and 100% for server exiting 2007. Intel also launched quad-core chips in 2H06.

Intel and AMD now compete head-to-head in all major segments of the microprocessor market. Historically, Intel and AMD competed aggressively in the desktop segment of the market as well as in the desktop replacement notebook market. Intel, however, dominated server and thin & light notebooks. In the past few years, AMD addressed both of these segments effectively, and has taken significant share in each. AMD now has design wins at all major players, including Dell for server and desktop, which previously sole-sourced all

127

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

processors from Intel. This game-changing event has produced intense price pressure for the segment as a whole, and likely will weigh on the CAGR for PC microprocessor revenue for the next several years. Integrated graphics chipsets have gained momentum as graphics quality improved. One of the most talked about trends in PC semiconductors is the shift to integrated graphics chipsets, which eliminate the need for a discrete GPU. Graphics quality using these solutions was historically seen as poor and they were originally reserved for low-end and corporate PCs. This changed as Intel and other chipset vendors ramped the performance of their graphics cores, and more than half of all PCs use these solutions today. Notebooks were particularly aggressive in adopting integrated, as eliminating the GPU also increased battery life. Looking ahead, we expect integrated graphics penetration to increase as the graphics quality of integrated solutions improves further still, though the rate of change is slowing. There is also some debate about what effect Windows Vista will have on the graphics market, as in the early going the new OS will tax integrated solutions, perhaps reversing the trend towards integrated, at least temporarily. The memory bus is continuing the move to DDR2. After a big transition from SDRAM to DDR memory, the market began yet another memory transition with DDR2. As in previous transitions, DDR2 first penetrated the performance desktop segment and over time has moved into mainstream desktops, servers, and notebooks. Note that DDR3 is coming to market now, and should follow a similar trajectory. An additional transition is taking place on the server front. Most x86 servers today use registered DIMM (RDIMM) which include register ICs (similar to timing ICs) directly on the module. Newer fully-buffered DIMMs include a specialty I/O IC known as an AMB I/O IC directly on the module, which manages the interface to the server chipset, which would include AMB I/O logic on chipset. PCI-Express has arrived; so has SATA. In 2004, PCs began a major architectural transition. The PCI bus, nearly ten years old, was replaced by the faster and more flexible PCI-Express. Both buses are supported by chipsets currently shipping to the market, but eventually PCI will be phased out completely in favor of PCIExpress. This transition will require completely new peripherals. Note that the graphics interface has also begun moving from the specialty AGP bus to PCI-Express graphics. Another interface to change in recent years was the disk drive interface. PCs have historically used the Parallel ATA (IDE) standard to communicate with disk drives. ATA/IDE uses a hierarchical system and wide parallel ribbon cables, which are often cumbersome inside the PC case. In 2003 an update to this bus was introduced, known as serial ATA (SATA), which uses a serial peer-to-peer hierarchy and thinner cables and also increases throughput. To enable an orderly transition, most chipsets support both legacy ATA and SATA. The market should accelerate when SATA II is introduced, as this standard ups throughput by a much greater amount than SATA I did. Gigabit Ethernet pervades the corporate desktop and is moving into the consumer PC. Now in the market for about five years, Gigabit Ethernet has successfully overtaken 10/100 Fast Ethernet in the corporate environment. Consumer PCs are also adopting Gigabit Ethernet, as are notebooks. Wireless LAN is now standard in mainstream notebooks; Bluetooth penetration has been sluggish but UWB should be pervasive. Nearly all notebooks now integrate wireless LAN networking. This trend began with the standardization of 802.11b and accelerated with Intels Centrino platform, which included wireless LAN with a chipset and processor. Bluetooth is gaining traction in both desktops and notebooks, though without the support of Intel this will be a slower, demand-driven trend. Ultra wide band (UWB) is the next major wireless technology targeted for PCs; a number of IC vendors are bringing chipsets to market using the WiMedia standard for wireless USB, with high volume projected for 2007. USB 2.0 is pervasive; IEEE 1394/Firewire is still around but giving way without the support of Intel. For peripheral interface, USB has proven to be a huge success, with nearly all PC peripherals standardizing around USB in the late 1990s. USB 2.0 provided much higher throughput (480 Mbps vs. 12 Mbps for USB 1.0) and has been included in Intels chipsets for some time; most peripherals have moved this way as well. IEEE 1394/FireWire has also gained some popularity, but lacks the support of Intel and thus is implemented with a separate host today. Still, it remains a popular feature, especially in consumer PCs. Power consumption is becoming a major issue for all component suppliers. This is true not only for notebooks (with the focus on battery life), but also for desktops and servers. Note that the trend to dual-core was in part driven by power consumption issues. NAND will begin to penetrate PCs in 2007. A major architectural enhancement for PCs is coming in 2007: the addition of a NAND flash memory cache. This cache will reduce I/O latency by allowing the system to write to flash memory instead of the hard disk drive, particularly for operating system or frequently used files. A major advantage would be quicker boot-up, as the system could store critical boot instructions and a portion of the OS on the NAND cache as opposed to the HDD. It also provides power savings for notebooks, as the system could access the hard drive less frequently.

128

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PC Displays PC Displays
The PC display market, long an afterthought, has become important again as flat-panel displays (FPD) monitors have reached mainstream price points. FPD adoption has surged in recent years, with desktop penetration reaching 72% in 2005. Notebook penetration has also surged as mobility trends have accelerated and lower FPD prices allowed the notebook premium vs. desktops to compress.
FPD Desktop Penetration
100% 100% 80%

Notebook Penetration
80%

CRT

Desktops

60%

60%

40%

40%

20%

20%

FPD
0% 2000

Notebook

0% 2000

2001

2002

2003

2004

2005

2001

2002

2003

2004

2005

Source: DisplaySearch, IDC, CIBC World Markets

Source: IDC, CIBC World Markets Corp.

Up until a few years ago, the PC monitor was an afterthought for PC OEMs. Although differentiated high-end monitors have represented a consistent percentage of the market (for workstations, etc.), PC OEMs have historically focused on supplying the lowest cost monitor possible in an effort to drive down the total system cost of a PC. This paradigm changed dramatically as LCD technology has matured and become more cost effective. Flat-panel LCD monitors have been available for years, but have been primarily reserved for the notebook market, as low yields and constrained LCD glass capacity had kept prices highenough to warrant a significant cost differential between otherwise comparably-equipped desktop and notebook PCs. With the increase in LCD capacity and an influx of competition, the price of large-area LCD displays has dropped dramatically, spurring demand for both notebooks and desktop LCD monitors. This has caused a new twist in the PC paradigm; the display is once again becoming an important part of the PC buying decision, and dozens of OEMs, primarily in Asia, are moving to meet the demand. In terms of units shipped, FPDs have been adopted more rapidly than most analysts had originally anticipated as LCD prices continue to decline, spurring demand. Flat panel display penetration in desktops, which exceeded 70% in 2005, is still increasing. Supply seems to be growing nicely and should be able to support continued strong notebook demand, further penetration of FPDs in the desktop market and the eventual volume deployment of LCD TVs.

129

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PC Displays PC Displays
PC Display Unit Forecast
Total PC displays growing at a 9% CAGR, from 230 million units in 2005 to 361 million units in 2010. Desktop CRT monitors declining 28%, from 47 million units in 2005 to 9 million units in 2010. Desktop LCD monitors growing 10% from 118 million units in 2005 to 193 million units in 2010. Mobile LCD monitors growing 19%, from 65 million units in 2005 to 159 million units in 2010 (1:1 with notebook sales).
400,000
Notebook LCD Desktop LCD Desktop CRT

LCD Monitor Suppliers


Others 32 % Dell 21 %

Samsung 11 %
Sony 2% NEC 2% ViewSonic 4% Lenovo 4% Acer 4%

HP 10 %
Philips 4%

LG 6%

CRT Monitor Suppliers


Others 20 % Samsung 21 %
Gateway+ eMachines 2% Acer 2% ViewSonic 3% AOC/Envision 4%

350,000

300,000

Unit Shipments (000s)

250,000

200,000

150,000

100,000

LG 15 %

50,000

Lenovo 7%
2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, DisplaySearch, Dataquest, CIBC World Markets

Philips 8%

HP 9%

Dell 9%

Source: IDC, CIBC World Markets Corp.

PC Display Unit Forecast: The market for PC displays includes desktop CRT and flat panel displays and the LCD screens attached to notebooks. For desktops, total monitor shipments actually exceed desktop PC unit shipments by about 20%, due to 1) an upgrade cycle for LCD monitors, as some users upgrade their monitor while keeping the same PC, and 2) the rising use of dual-display systems. Notebook displays, on the other hand, ship 1:1 with notebooks. In 2005, we estimate total PC monitor shipments were 230 million units. Looking ahead, we expect displays to grow 9% through 2010, rising to 361 million units. The big story in displays is the migration to flat panel, which has accelerated as falling LCD prices have spurred demand. In 2005, desktop flat panel penetration increased to 72% of desktop monitor shipments. As penetration rises, we expect desktop LCD monitors to grow from 118 million units in 2005 to 193 million units in 2010, a 10% CAGR. Notebooks are also growing quickly, and approximated 28% of total PC monitor shipments in 2005. Notebook displays should mirror notebook shipment growth, from 65 million units in 2005 to 159 million units in 2010, a 19% CAGR. CRTs are expected to decline 28% through 2010 to just 4% of desktop monitor shipments. PC Display Suppliers: The top supplier of desktop LCD displays in 2005 was Dell, with 21% share, followed by Samsung at 11% and HP at 10%. Other LCD display suppliers included LG, NEC-Mitsubishi, Philips, Acer, Lenovo, ViewSonic, NEC, and Sony. In CRT, Samsung and LG led with 21% and 15% share, respectively. Dell and HP followed, each with 9%. Other CRT suppliers include Philips, Lenovo, AOC/Envision, ViewSonic, Acer, and Gateway. The top notebook suppliers in 2005 were Dell, HP, Toshiba, Acer, Lenovo, Fujitsu, Sony, NEC, ASUS, and Apple.

130

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PC Displays PC Displays

Digital monitor interface receiver

Processes the video feed and scales it up to an image that can be displayed on the screen

Transmit and receive the digital image signal between the monitor printed circuit board and the glass pane

LCD drivers generate the image on the LCD panel

Monitor PCB

LCD Glass Panel Source Source Source Source Drivers Drivers Drivers Drivers

Digital Input Analog Input

TMDS/DVI Receiver Video ADC

Scaling Engine
Microcontroller

LVDS Transmitter

LVDS Receiver Timing Controller

Gate Drivers Gate Drivers Gate Drivers LCD Screen

DRAM
Retimes the signal received from the monitor PCB before sending it to the screen

Analog monitor interface receiver (analog-to-digital converter)

Graphics memory used in scaling

Notebook Display Desktop Display

Source: CIBC World Markets Corp.

Note: The diagram above and the discussion below focus on flat panel (LCD) displays, which represent a much larger opportunity for semiconductor vendors than the CRT market. Note that desktop and notebook LCD displays differ somewhat; in desktop monitors, the image is received from an external source and then scaled; in a notebook the digital image is sent directly to the glass panel via LVDS. Semiconductor devices in flat panel displays are found in two places: on the monitor board and in the monitor panel. The monitor board receives the signal from the PC and scales the image and then transmits it to the glass panel digitally over LVDS (Low Voltage Differential Signaling). Devices in the glass panel/CRT receive the signal, retime it and produce the image. Notebook displays employ a simpler design, since the monitor is directly attached to the source of the image. No scaler is needed in a notebook display; instead the digital image is sent directly to the glass panel via LVDS. We discuss below the key devices found in desktop and notebook flat-panel displays. Monitor Board (desktop monitor only) Video ADC: This analog-to-digital converter receives analog video input from the PC and converts it to a digital stream to be received by the scaler. This device is usually integrated into the scaling engine.

131

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

TMDS/DVI Receiver: This device receives digital video signals from the PC. Most monitors with a digital receiver also have a legacy analog receiver. Like the analog receiver, this device is increasingly being integrated into the scaling engine, though not at quite the same rate as ADCs. Scaling Engine: The scaler processes the video feed and scales it up to an image that can be displayed on the screen. DRAM: Most scalers require external memory to help process the video image. LVDS Transmitter: The LVDS transmitter sends the scaled image from the monitor board to the LCD panel in a digital format. Note that other types of transmission methods may be used; LVDS is a popular standard in notebook LCD screens and in desktop flat panels. Like the receiver chips, this device is increasingly being integrated into the scaling engine.

Notebook Base (notebook only) LVDS Transmitter: The LVDS transmitter sends the image from the graphics processor to the LCD panel in a digital format. This can be integrated into the PC graphics processor or chipset.

LCD Panel LVDS Receiver: The LVDS receiver receives the scaled image from the monitor board or notebook base. Timing Controller: This device retimes the signal received from the monitor board before sending it to the LCD row and column drivers. It is typically implemented as a custom ASIC. LCD Row/Column Drivers: LCD row (gate) and column (source) drivers are placed around the LCD panel, producing the viewable image. Source drivers usually sit across the top of the panel and generate signals that penetrate individual columns of pixels. Three source drivers are needed for each row of pixels: red, green and blue. Gate drivers usually run along the side of the panel and manipulate individual rows of pixels. The gate drivers intersect the signals driven by the source drivers, turning individual pixels on or off.

132

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PC Displays PC Displays
PC Display Semiconductor Forecast
PC display semiconductors are expected to grow at a 6% CAGR, from $4.0 billion in 2005 to $5.4 billion in 2010. LCD drivers growing 7%; they should eventually approach 80% of IC revenue. LCD controllers ICs should grow at a 2% rate due to tough price competition. Microcontrollers growing 1%; CRT MCU revenues declining 29% offset by 9% growth in desktop LCD.
$6,000 LCD Drivers Other ASSPs (Timing, Interface) $5,000 LCD Scaler/Controller Microcontrollers
Semiconductor Revenue ($M)

PC Display Semiconductor Competitors


LCD Scaler/Controller
Others 13 % Pixelworks 4% Novatek 7%

Genesis 36 %

Realtek 9%

MorningStar 31 %

LCD Row and Column Drivers


Renesas 3% Toshiba 3% Sharp 3% Matsushita 3% Seiko Epson 5% Others 8%

$4,000

Samsung 24 %

$3,000

$2,000

$1,000

Magnachip 6% Novatek 19 %

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, DisplaySearch, Dataquest, CIBC World Markets

Himax 9% NEC 17 %

Source: Dataquest, CIBC World Markets Corp.

PC Display Semiconductor Forecast: We expect PC display semiconductors to grow at a 6% rate through 2010, growing from $4.0 billion in 2005 to $5.4 billion in 2010. LCD drivers should grow at a 7% CAGR driven by desktop LCD and notebook penetration. The advance to larger size screens should also boost growth and help offset price pressure. We expect LCD drivers to rise from 76% of total PC display IC revenue in 2005 to 80% in 2010. LCD controllers, 8% of the total market in 2005, should grow at a 2% CAGR through 2010; intense price competition in this market is almost completely offsetting unit growth in desktop LCD monitors (most of the IC vendors have shifted their attention to the LCD TV market in the process). Microcontrollers, 7% of IC revenue in 2005, should grow just 1%; 9% growth in desktop LCD microcontrollers is offset by CRT microcontroller revenues declining at a 29% CAGR. Other ICs (mostly timing controller ASICs) should grow at a 4% rate through 2010. PC Display Semiconductor Competitors: The market for LCD scaling engines is dominated by Genesis Microchip and Taiwan-based MorningStar. In 2005, Genesis held 36% revenue share, while MorningStar held 31%. Other vendors include Realtek with 9% share, Novatek with 7%, and Pixelworks with 4%. The top suppliers of drivers for large-panel LCD panels in 2005 include Samsung, with 24% market share, Novatek with 19% and NEC with 17%. Himax held 9% share, Magnachip (the spinout from Hynix) held 6%, and Seiko Epson held 5%. Other vendors include Matsushita, Sharp, Toshiba, and Renesas. CRT monitor microcontroller suppliers include NXP/Philips, Weltrend, Samsung Semiconductor, Myson and Novatek.

133

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PC Displays PC Displays
PC Display Semiconductor Market Trends
Integration has become critical for low cost FPD monitor designs: scalers now integrate digital ADC & DVI interface receivers, microcontrollers, and memory. Digital interface attach rates are increasing. Many OEMs are adopting smart panel designs, integrating all devices onto a single PCB in the monitor panel. Desktop FPD monitors and notebooks screen sizes are increasing, requiring additional LCD driver ICs. LCD source driver channel density is on the rise as panel makers look to reduce costs.

Integration has become critical for low cost FPD designs. IC vendors now integrate more functionality into the scaler chip, including digital and analog input receivers, microcontrollers, memory, and the LVDS transmitter. Digital interface attach rates are increasing. These are still found primarily in higher-end and larger size monitors, both in CRT and LCD. Many OEMs are adopting smart panel designs, integrating all devices onto a single PCB in the monitor panel. An emerging trend is that of smart panels, which includes monitor inputs, scalers, timing devices and LCD drivers on a single circuit board in the monitor panel. This eliminates the need for a second PCB and reduces the number of connectors (and associated transceiver ICs) and even simplifies the manufacturing process. Desktop FPD monitor and notebook screen sizes are increasing, requiring additional LCD drivers. The mainstream desktop LCD monitor is moving from 15-inch to 17-inch, and notebooks are increasingly adopting larger screens sizes (in the case of notebooks, wider screens are becoming more mainstream as well). LCD source driver channel density is on the rise as panel makers look to reduce costs. Greater channel density reduces the number of drivers needed for a given resolution. For example, moving from 480 channels (mainstream today) to 642 channels (used in newer designs) reduces the number of source drivers for a typical 1280 x 1024 monitor from nine to six (recall that three are needed per pixel: one for each RGB color). Note that gate drivers face less pressure since they are lower cost to begin with, and only one driver is needed per pixel.

134

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Hard Disk Drives Hard Disk Drives


Hard disk drives (HDDs) are still closely tied to the PC market, but drives have dramatically outgrown the 1:1 ratio with PCs as enterprise storage networks, consumer devices, and personal storage are providing new opportunities for mass storage. In enterprise, terabytes are growing at a 50%+ CAGR. Networked systems are enabling the separation of processing and storage, allowing for easy, inexpensive capacity additions as corporate data needs grow. In consumer, DVRs and video game consoles are proving to be an effective usage model for HDDs as these markets require capacity points that cannot be served adequately by NAND flash. Personal storage is a fast-growing application, as increased familiarity with digital content is driving demand for solutions to manage that content.

Hard disk drives (HDDs) are the primary data storage devices used in data processing applications. The drives, usually integrated within larger host devices, use a set of rigid disks to magnetically store bits of data in extremely large quantities at very high speeds. All PCs ship with at least one hard disk drive, which stores the computers operating system and software applications and the users files. They are fast, cheap, carry very large capacities, and conform to industry-standard form factors. They have therefore become an integral part of PCs and servers. Though still closely tied to the PC market, hard disk drives have dramatically outgrown the 1:1 ratio with PCs. Enterprise storage, consumer devices, and personal storage are providing new opportunities for mass storage. On the enterprise side, IT managers can supplement their individual PC storage with enterprise storage systems to store and manage common-use applications and files. In these systems, arrays of drives are incorporated into special arrays or networks designed specifically for data storage traffic, which can be accessed by servers on the data network. This dramatically increases storage capacity and performance and also allows the IT manager the option to manage processing and storage independently. Terabytes of enterprise storage continue to grow in excess of 50% per year. On the consumer side, the rise of digital content has created new opportunities for HDDs in consumer electronics, though flash has presented an alternative to HDDs in certain applications (MP3 players, for example). DVRs and game consoles are proving to be effective usage models for HDDs as these markets require capacity points that cannot be served adequately by flash; other smaller opportunities exist as well. A related driver is personal storage, as consumer familiarity with digital content drives demand for storage solutions to manage and protect that content.

135

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Hard Disk Drives Hard Disk Drives


Hard disk drives are self-contained devices that use rigid disks coated with a magnetic material to store bits of data. They are available in standard form factors using standard interfaces, differentiated by storage capacity, speed, power consumption, and reliability. Desktop drives are designed for the highest capacity at the lowest cost. Mobile drives are smaller and consume less power. Enterprise drives focus on speed, performance and durability.

Hard disk drives are self-contained devices that use one or more rigid disks coated with a magnetic material to store bits of data. They are available in standard form factors using standard interfaces, but are differentiated by storage capacity, speed, power consumption and reliability. A hard disk drive has three basic subassemblies: A set of platterswhich are flat, round, rigid disksused as the storage media. Each platter is made up of a substrate that is coated with a special media material designed to store data in the form of magnetic patterns. The platters can be double-sided and are mounted on a spindle, which is attached to a spindle motor that spins the platters at high speed. A mechanical device known as an actuator, which sits in the corner of the drive and controls a set of arms that reach out over the surface of the individual platters. Mounted on the end of each arm is a slider and a specialized electromagnetic read/write device called a head. The read/write heads, which are in essence tiny electromagnets that perform the conversion from electrical signals to magnetic signals, are the interface between the magnetic physical media on which data are stored and the other electronic components on the hard disk, recording data to the disk and reading from it. Hard disk drives need one head for each platter surface (i.e., two are needed for a double-sided platter).

136

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

A printed circuit board, which contains all the logic needed to run the drive system. Devices on the board include a preamp, which amplifies signals between the head and the drive logic; a read channel, which performs read and write functions; and a hard disk controller and microprocessor, which control the system and implement the interface protocol. The disk drive also has some memory and motor control ICs for the spindle motor and the actuator.

Disk Drive Types While the technologies used are very similar across platforms, hard disk drives fall into three basic categories: Desktop. Desktop-class hard disk drives are used in desktop PCs and low-end workstations and servers. They have standard physical dimensions: 4 wide by 5.75 deep by 1 high (though some drives are less than 1 high) and are generally referred to as 3.5-inch drives. Most desktop drives use the ATA/SATA interface and have one or two platters. Mobile. Mobile-class hard drives are used in notebook PCs and small form factor desktops. They are smaller than desktop drives; the standard size has dimensions of 2.75 wide, 3.95 deep and 0.37 high and is generally referred to as a 2.5-inch drive. Some are even smaller and are often referred to as microdrives; these drives are found in consumer devices like portable media players. Most mobile drives use the ATA/SATA interface and have one or two platters; they typically consume less power than other drives. Enterprise. Enterprise-class hard drives are used in workstations, servers, networked storage systems and high-end desktop PCs. Most have similar physical dimensions to desktop drives, though some high-capacity drives can be greater than 1 in height while others adopt the 2.5-inch mobile form factor. Most enterprise drives use the SCSI interface, though drives designed for network storage often use Fibre Channel. Newer drives are implementing SAS (serial-attached-SCSI), the follow-on to SCSI. Most enterprise drives have more than one platter.

Note that the vast majority of hard disk drives are internal drives, meaning that they are designed to be implemented inside a PC or enterprise system or other host device. These generally conform to the specifications indicated above. External drives are sold as standalone units, which are fully enclosed and incorporate additional interfaces such as USB, FireWire, or Ethernet. A growing application for hard disk drives is the consumer electronics market. The larger HDD vendors have all introduced modified versions of their desktop and mobile drives specifically for this market. These drives can vary from standard HDDs in form factor, power consumption, speed, and cost, but generally fall into the same drive classes described above. Drive Storage Capacity Storage capacity in a hard disk drive is measured in Gigabytescolloquially known as a billion bytes but actually 1,073,741,824 bytes (computer specifications usually use binary numbers; therefore, a Kilobyte is 1024 bytes, a Megabyte is 1024 Kilobytes and a Gigabyte is 1024 Megabytes, which gets us to 1,073,741,824 bytes). Note that a byte is eight bits, which are the 1s and 0s of the digital world. Desktop class drives generally have the highest capacity points over the greatest range, catering to corporate PCs, consumer PCs, and the retail market. Most DVRs and personal storage solutions use desktop class drives as well. Mobile drive capacity points are substantially lower and have a tighter range; these are primarily used in notebook PCs but are also found in portable media players, game consoles, and other consumer electronics. Enterprise drives typically trail desktop drives in capacity, and also have a tighter range. A drives capacity is a function of 1) the amount of bytes that can be stored on each platter (this in turn is determined by the size of the platter and the platters areal density) and 2) the number of platters in the drive. In general, drive vendors seek to maximize the amount of bytes that can be stored on each platter (the first factor) and to minimize the number of platters in the drive (the second factor). In the end, however, the final platter and density implementation will depend on a number of tradeoffs that include engineering constraints, cost dynamics and marketing requirements. Drive Speed The speed of a hard disk drive is usually measured on two fronts: transfer speedwhich refers to the rate at which data is transferred from the disk to the host system and vice versa; and positioning speedthe rate at which the drive system can hop around the disk in search of the desired data. While many components within the drive affect these metrics, one of the most significant is the strength of the spindleso much so that spindle speed is usually quoted right after drive capacity when describing a drive and is also the most commonly used specification used to classify

137

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

different drives by performance. Spindle speed describes the rate at which the platters in the disk drive turn, expressed in revolutions per minute (RPM). The faster a platter spins, the faster the transfer rate when reading or writing as well as the positioning when searching for the correct track and sector. Spindle speed differs for different classes of drives. In the desktop market, most drives spin at 7,200 RPM, though some higher-end desktop drives spin at 10,000 RPM. Mobile drives spin at speeds ranging from 5,400 RPM to 7,200 RPM (the notebook market requires drives of varying power consumption to meet the different platform requirements). Enterprise drives also vary in speed: from 7,200 RPM at the low end to 15,000 RPM at the very high end, with the most common being 10,000 RPM. Interface Hard disk drives use one of several standard interfaces to send data between the drive and the host system. Up until the last few years, the most common interface was IDE (Integrated Development Environment), also known as ATA (Advanced Technology Attachment), which was used in most desktop and mobile drives. ATA drives are configured as master or slave using jumper pins on the drive itself, allowing for two drives to connect into each ATA interface controller on the host. They use wide 80-pin ribbon cables (18 in length) and connect directly to PC motherboards (most PC chipsets include one or more ATA controllers embedded in the south bridge). The data transfer rate for ATA increased over time to 133 Mbps. The evolution of the IDE interface is known as Serial ATA (SATA), which has quickly displaced ATA in most PCs. SATA changes the physical architecture from master-slave to point-to-point, so each drive is connected to its own interface. SATA also changes the cable interface to a narrow four-wire cable and extends the length to one meter. Transfer rate was increased to 150 Mbps in the first implementation, with 300 Mbps coming to market now and a roadmap planned to 600 Mbps. SATA is also able to use lower voltage power supplies. Parallel ATA vs. Serial ATA Cables

Source: The Computer Language Co. Inc., CIBC World Markets Corp.

While ATA/SATA has quickly become the most popular PC disk drive interface due to its lower cost, SCSI (pronounced scuzzy) remains an important interface in the enterprise market as well as in high-end desktops. The SCSI interface uses a daisy chain architecture allowing for up to 15 peripherals to be connected to a single interface controller (each peripheral has a second port to connect to the next device). SCSI has also consistently maintained a transfer rate advantage over ATA, with most SCSI drives in use today carrying a transfer rate of 320 MBps. Further, SCSI was the first interface to support RAID (Redundant Array of Inexpensive Disks) configurations. For all of these reasons, SCSI dominates the enterprise hard disk drive market, though the advent of Serial ATA may change the dynamics somewhat. Note that as with ATA, SCSI drive cables connect to a SCSI controller on the PC or server motherboard; unlike ATA, however, chipset vendors have not integrated this functionality into the PC chipset. Servers and SCSIenabled desktops therefore require a separate controller chip or add-in board to support SCSI drives. The follow-on to SCSIknown as Serial Attached SCSI (SAS)is currently in progress. SAS expands the number of supported devices to 128, uses smaller cable connections (like Serial ATA) and provide universal interconnect compatibility with SATA while retaining the reliability, performance and manageability advantages of SCSI. About 25% of high end drives incorporate support for Fibre Channel, which is a high-speed transport technology used in storage area networks. Fibre Channel serializes SCSI commands into Fibre Channel frames, which can be sent at high speed (1.0625 Gbps, 2.125 Gbps, 4.25 Gbps, or even 10.625 Gbps) either point-to-point or through a switched network of Fibre Channel switches. Most Fibre Channel networks, along with their associated arrays of hard disk drives, are set up independent of the network of servers used for data processing, allowing any server to access data stored on any hard disk drive without a direct connection. See the discussion of enterprise storage in the networking section of this report for a more in-depth discussion of Fibre Channel.

138

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Hard Disk Drives Hard Disk Drives


HDD Unit Forecast
We expect the HDD market to grow at a 13% CAGR, from 381 million drives in 2005 to 706 million drives in 2010. Mobile drives should grow >20%, driven by desktop replacement trends and consumer devices. Desktop class drives should grow 9% despite the shift to notebook; personal storage and consumer electronics devices will be the primary growth drivers. Enterprise class drives should grow modestly as desktop class SATA drives start to penetrate storage systems.
800,000
Enterprise Class HDD

HDD Suppliers
Fujitsu 6% Toshiba 9% Others 1%

Seagate 28 %

Samsung 9%

Maxtor 14 % Western Digital 17 %

Hitachi 15 %

700,000
Mobile Class HDD Desktop Class HDD

600,000

Source: IDC, CIBC World Markets Corp.

Unit Shipments (000s)

500,000

400,000

Desktop

Mobile

Enterprise

300,000

200,000

100,000

2002 2003 2004 2005 2006 2007 2008 2009 2010

Seagate Seagate Hitachi GST Western Digital Fujitsu Fujitsu Maxtor Hitachi GST Toshiba Samsung Maxtor Seagate Hitachi GST Western Digital Samsung Others Western Digital

Source: IDC, CIBC World Markets Corp.

Hard Disk Drive Unit Forecast: The market for hard disk drives is significantly larger in unit terms than that of PCs. In addition to the 1:1 unit shipments going into PCs, the market includes drives sold into mass storage arrays to support servers and data centers, SAN and NAS systems (networked storage systems), personal storage solutions, and emerging applications in the consumer, automotive and industrial markets. There is also a significant retail market for add-on HDDs, both internal and external. In total, we estimate the market for hard disk drives totaled 381 million units in 2005, 83% larger than the PC market. The hard disk drive market is also growing faster than the PC market. On the enterprise side, the number of multidrive enterprise storage systems continues to increase, as does the number of drives per storage system. This is being enabled by more mature large enterprise storage solutions, the industry transition to SATA for the low and midend (which enables cheaper storage systems for enterprise and SMB), as well as the emerging market for personal storage systems. On the consumer electronics side, digital media devices are incorporating hard disk drives, most notably DVRs, game consoles, and portable media players. Overall, we expect the hard disk drive market to grow from 381 million units in 2005 to 706 million units in 2010, a CAGR of 13%. By drive class, we estimate that 63% of drives shipped in 2005 were desktop class, 30% were mobile classincluding small form factorand 7% were enterprise class. Looking ahead, we expect mobile devices to lead growth with a 21% CAGR and to represent 43% of total units in 2010, driven by desktop replacement trends and growth in consumer applications. Desktop drives should still grow at a 9% CAGR despite the shift to notebook; personal storage

139

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

and digital consumer devices (mostly DVRs) should be healthy growth drivers. Our forecast for enterprise drives is a more modest 5% CAGR, as lower-end enterprise storage systems are increasingly being populated with desktop (or even mobile) class drives leveraging SATA (which we regard as desktop class) instead of SCSI or SAS. Looking at the end application for HDDs, computing represented the largest opportunity in 2005 at 84% of total shipments, with consumer applications representing the balance. Within computing, desktop PCs represented 62% of shipments, followed by notebook at 23%, enterprise at 10% and personal storage at 5%. Looking forward, drives that sell into notebooks and desktops should grow 20% and 5% respectively, in line with PC unit growth, and should represent over 60% of the market in 2010. Drives that sell into enterprise storage (including all drive classes) should grow at a 14% CAGR to roughly 10% of the market. Personal storage should grow the fastest at a 35% CAGR, to represent over 10% of the market in 2010. In total, we expect the computing storage market to grow at 12% CAGR, from 319 million units in 2005 to 564 million units in 2010, to represent 80% of total HDDs in 2010. Note that outside of small-form factor notebooks and ultra-mobile PCs, we do not expect to see flash eat into the computing market in any major way during the forecast period. The consumer market represented the remaining 16% of HDD units in 2005 and is expected to grow at an 18% CAGR, to roughly 20% of HDD units in 2010. The largest segment in 2005 was MP3/portable media players at 47% of consumer HDDs (8% of total HDDs), though this market took a big hit in 2006 due to Apples decision to replace the HDD-based iPod Mini with the flash-based iPod Nano (the video iPod remains HDD-based). We expect the HDD MP3 market to remain under constant pressure from flash, though the largest players should remain HDD-based. The other segments of consumer HDDs require capacity points that cannot be served adequately by flash, and therefore see more consistent growth. DVRs, which represented 35% of consumer HDD shipments in 2005 (6% of total units), should grow at a 19% CAGR as these devices gain adoption in cable, satellite, IPTV, and DVR/DVD-R implementations. Game consoles and handhelds should grow at an even faster 40% CAGR, as the new game consoles have a high attach rate of HDDs. Automotive should grow at a 32% CAGR to roughly 13% of total consumer HDDs during the forecast period. Handset HDDs will face pressure from flash until higher capacity microdrives hit the market in volume in 2008; handset HDDs should grow to 11% of units in 2010. Digital cameras and camcorders (mostly camcorders) will probably also see nice growth but represent less than 5% of the market in 2010. The exhibits below display our computing and consumer hard disk drive unit forecast. Computing HDD Unit Forecast
600,000 Personal storage Enterprise storage 500,000 Notebook PCs Desktop PCs 400,000
Unit Shipments (000s) Unit Shipments (000s)

Consumer HDD Unit Forecast


160,000 Digital audio and PMP Digital cameras and camcorders Automotive infotainment Game consoles and handhelds PDAs/Handsets Set-top box, DVR, DTV

140,000

120,000

100,000

300,000

80,000

60,000

200,000 40,000 100,000 20,000

2002 2003 2004 2005 2006 2007 2008 2009 2010

2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, CIBC World Markets Corp.

We can also segment the market by drive size. In 2005, 69% of HDDs shipped were 3.5-inch drives, 22% were 2.5inch, and 9% were 1.8-inch and below. 3.5-inch drives dominate desktop PCs, enterprise storage systems, personal storage solutions, and DVRs. 2.5-inch drives dominate notebook PCs and automotive applications, and should come to dominate game consoles as the new generation of consoles rolls out. 2.5-inch is becoming an important form factor in the enterprise, and is also becoming popular in small form factor desktops portable personal storage solutions. 1.8-inch and below dominates portable media players and the small (but growing) market in digital cameras and camcorders, and over time should take a nice chunk of the notebook PC market. In terms of growth rates, 3.5-inch drives should grow at an 8% CAGR, as strength in personal storage (38% CAGR) and DVRs (13% CAGR) is partially offset by share loss to smaller 2.5-inch drives, which are penetrating desktops and server markets. 2.5-inch should grow at an impressive 25% rate, driven by strong notebook growth (18% CAGR) and

140

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

increasing penetration of 2.5-inch in desktop (63% CAGR), enterprise storage (78% CAGR), and DVRs and gaming consoles (84% CAGR). 2.5-inch drives are also seeing good growth in automotive and personal storage markets, though this will likely represent less than 10% of total 2.5-inch drives in 2010. Though flash will continue to pressure HDDs presence in MP3 players, 1.8-inch and below drives are expected to grow at a 17% rate, as these drives move beyond the portable media player market and into thin & light notebooks. The exhibits below display our HDD unit shipment forecast by application for each size segment. 3.5 HDD Unit Forecast
450,000 Game consoles and handhelds 400,000 350,000 300,000 250,000 200,000 150,000 100,000 50,000 50,000 2002 2003 2004 2005 2006 2007 2008 2009 2010 2002 2003 2004 2005 2006 2007 2008 2009 2010 10,000 Set-top box, DVR, DTV Personal storage Enterprise storage Desktop PCs
Unit Shipments (000s) Unit Shipments (000s)

2.5 HDD Unit Forecast


300,000 Other Consumer Automotive infotainment STB, DVR, DTV, Gaming Personal storage Enterprise storage Desktop PCs Notebook PCs

1.8 and Below HDD Forecast


80,000 Automotive infotainment PDAs/Handsets Digital cameras and camcorders Digital audio and PMP Personal storage Notebook PCs

70,000

250,000

60,000
Unit Shipments (000s)

200,000

50,000

150,000

40,000

30,000

100,000 20,000

2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, CIBC World Markets Corp.

Hard Disk Drive Suppliers: The largest supplier of hard disk drives in 2005 was Seagate with 28% unit share. Even before it acquired Maxtor, Seagate was the top supplier of desktop drives with 33% share in 2005 and controlled just over half the enterprise market. Seagate also has a growing presence in mobile drives with 12% share. Aggregating Seagate and Maxtors share in 2005 would give the combined entity 43% overall unit share, though there has been some attrition since the closing of the deal in 2006. Western Digital was No. 2 in 2005 with 17% total unit share; Western Digital is No. 2 in desktop drives with 26% share and has a small presence in enterprise and mobile. Hitachi GST, which acquired IBMs hard disk drive business three years ago, was No. 3 overall with 15% unit share; Hitachi was the top supplier of mobile drives with 32% share and has a smaller position in enterprise and desktop. Maxtor as a standalone company followed with 14% share; Maxtor was No. 3 in desktop drives with 21% share and had a small presence in enterprise. Other suppliers include Samsung in desktop and mobile, Toshiba in mobile, and Fujitsu in mobile and enterprise drives. Each held less than 10% total unit share in 2005. The exhibits below display 2005 unit share data for desktop, mobile, and enterprise hard disk drives. 2005 Desktop HDD Unit Share
Hitachi 8% Others 1%

2005 Mobile HDD Unit Share


Samsung 7%
Western Digital 4%

2005 Enterprise HDD Unit Share


Maxtor 14 % WD 3%

Samsung 11 %

Seagate 33 %

Seagate 12 %

Hitachi 32 %

12

Hitachi 13 %

Seagate 51 %

Maxtor 21 % Toshiba 23 % Western Digital 26 % Fujitsu 23 %

Fujitsu 20 %

Source: IDC, CIBC World Markets Corp.

141

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Hard Disk Drives Hard Disk Drives


Performs interface processing and sends signals between the host system and the read channel Memory used to temporarily store data being transferred to and from the host system

Amplifies the analog signal for reading and writing to the disk

Performs the data encoding and conversions needed to read and write data to and from the disk

Preamp

Read Channel

Hard Disk Controller


Microcontroller

Buffer (DRAM) Cache (SRAM) Boot (Flash)

Host System

Head

Hard Disk Platter


Source: IDC, CIBC World Markets Corp.

Motor Control

Servo ASIC

SRAM used in processing by the embedded microcontroller

A driver which controls current flow to spindle and voice coil motors to spin the drive

A DSP that interprets cylinder numbers and controls the the timing of servo gate assertions, controlling movement of the hard disk

Boot memory is used to store initial instructions required to start up the drive
SoC Solution Integrates HDC, read channel, servo ASIC, and memory

Note: the diagram above shows a hard disk drive implemented with discrete ICs for each separate function of the system. In practice, most HDDs are implemented with an SoC solution which combines read channel, HDC, servo ASIC, and memory in a single IC. Semiconductors in hard disk drives perform two basic functions: control of the movement of the hard disk spindle and actuator, and read/write of data to and from the hard disk. Hard Disk Read/Write Preamp: The preamp amplifies the analog signal from the read/write heads on the platter and sends the data to the read channel. Read Channel: The read channel performs the data encoding and conversions needed to read and write the data to and from the hard disk. Hard Disk Controller (HDC): This controller is the interface processor that sends instructions from the host system to and from the servo ASIC and the read channel. It also determines when to raise and lower read, write and servo gates and is responsible for defect management and error correction. Hard disk controllers usually include a microcontroller to run the entire HDD system.

142

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Buffer Memory (DRAM): Buffer memory temporarily stores the data being transferred to and from the host system. Cache Memory (SRAM): Cache memory is used to store processor instruction code once the drive is up and running. Code/Boot Memory (Flash): Boot memory is used to store initial instructions required to start the drive. System-On-a-Chip (SoC) (not pictured in block diagram above): As mentioned above, most HDDs incorporate an SoC solution that includes most of the read/write functions on a single IC. Most SoCs today incorporate the read channel, HDC, servo ASIC and code and cache memory. Some SoCs also integrate buffer memory, removing the need for a separate DRAM module. The exhibits below display block diagrams without and with SoCs.

Hard Disk Movement Control Motor Control: The motor control IC controls the amount of current to the spindle motors and the voice coil motor (used to move the head stack), controlling drive commutation and speed. Motor controllers sold as ASSP solutions are specifically designed for hard disk drives. Servo ASIC: The servo ASIC is a DSP that interprets cylinder numbers and controls timing of servo gate assertions and events during the assertion periods. This digital device is most often integrated into the hard disk controller or SoC. Hard Disk Drive Without SoC Integration
Buffer (DRAM) Cache (SRAM) Boot (Flash)

Hard Disk Drive With SoC Integration

Preamp

Read Channel

Hard Disk Controller


Microcontroller

Host System

Preamp

Read Channel

DRAM SRAM Flash

Host System

Hard Disk

Motor Control

Servo ASIC

System-ona-chip Hard Disk Motor Control


Servo

MCU

Source: IDC, CIBC World Markets Corp.

143

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Hard Disk Drives Hard Disk Drives


Hard Disk Drive Semiconductor Forecast
HDD semiconductors are expected to grow at a 7% CAGR, rising from $3.4 billion in 2005 to $4.8 billion in 2010. SoCs are expected to rise to 68% of ASIC/ASSP revenue by 2010, absorbing discrete HDC/servo ASICs and read channel ICs, which are expected to decline steadily over the forecast period. Pre-amps and motor controllers should see stable growth of 6% through 2010.
$5,000 System-on-a-chip $4,500 $4,000 $3,500 $3,000 $2,500 $2,000 $1,500 $1,000 $500 $0 2002 2003 2004 2005 2006 2007 2008 2009 2010 Read channel HDC/servo logic Preamplifier Motor Controller

HDD Semiconductor Competitors


Renesas 4% QLogic/Marvell 4% Infineon 3% IBM 1%

LSI Logic 6%

Marvell 32 %

TI 17 %

STMicro 17 %

Agere 17 %

Semiconductor Revenue ($M)

Source: IDC, CIBC World Markets Corp.

SoC

Read Hard Disk Preamps Controllers Channels TI STMicro Agere Marvell

Motor Control STMicro TI Renesas Marvell Agere

Source: IDC, CIBC World Markets Corp.

Marvell LSI Logic Marvell Agere Qlogic/Marvell Agere STMicro Renesas IBM Infineon Agere Infineon TI STMicro IBM

Hard Disk Drive Semiconductor Forecast: The hard drive IC market has seen nice growth on the back of strong unit growth in the HDD market. It has also undergone some dramatic changes because of the increasing adoption of SoC solutions, which now dominate desktop and mobile class drives. Looking ahead, we expect the market to continue to benefit from unit growth, somewhat offset by the benefits of integration and normal component price pressure. Overall, we expect the market to grow from $3.4 billion in 2005 to $4.8 billion in 2010, a 7% CAGR. In terms of SoC adoption, we estimate that in 2005, 85% of drives used an SoC solution, up from 71% in 2004 and 57% in 2003. The 2005 figure included 97% of desktop drives and 80% of mobile drives, including small form factor. By 2010, we expect desktop, mobile, and consumer SoC adoption will approach 100%. Enterprise SoC adoption should begin to happen in 2007-08, reaching 45% by 2010. The exhibit on the following page right displays SoC adoption trends for hard disk drives. In terms of our device forecast, SoCs are, not surprisingly, expected to grow the fastest, at a 14% CAGR through 2010. In the process, SoCs should rise from 50% of total HDD IC content to 68% in 2010. Discrete HDC/Servo and read channel devices are expected to decline rapidly as these devices get integrated into the SoCwe expect negative CAGRs of 20% and 21%, respectively. Preamp and motor control ICs, which remain outside the SoC, are expected to each grow at CAGRs of 6% apiece, representing 12% and 15% of the total HDD IC market in 2010, respectively. The exhibits on the next page display SoC adoption trends as well as the percentage shift in content within the HDD.

144

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

SoC Adoption By Drive Class


120% Total SoC Adoption Desktop Class 100% Mobile Class Enterprise Class 80% 90% 80% 70%
Semiconductor Revenue

HDD Semiconductor Content Shift


100% System-on-a-chip Read channel HDC/servo logic Preamplifier Motor Controller

60% 50% 40% 30% 20%

60%

40%

20% 10% 0% 2002 2003 2004 2005 2006 2007 2008 2009 2010 0% 2002

2003

2004

2005

2006

2007

2008

2009

2010

Source: IDC, CIBC World Markets Corp.

Hard Disk Drive Semiconductor Competitors: The hard disk drive IC market is dominated by a handful of diversified analog and DSP vendors with competencies in mixed signal design and HDD system IP. As drive capacity and performance have risen and the end market consolidated, the playing field has been limited to those with a strong IP base, integration competencies and the desire to invest in the design of complex and integrated solutions. The largest vendor of hard disk drive ICs in 2005 was Marvell, with 32% share; Marvell is strong in SoCs and read channels. Note that Marvell acquired QLogics HDD semiconductor business in late 2005. Agere followed with 17% share and has a strong presence across device types (note that LSI announced in 2006 its intention to acquire Agere). STMicroelectronics followed with 17%; ST participates in the SoC market at Seagate in partnership with Agere and also is strong in preamps and motor controllers. Texas Instruments followed with 17%; TI is the leader in preamps and is No. 2 in motor controllers. Other players like LSI Logic, Qlogic, Renesas, Infineon, and IBM have been marginalized for the most part, though some niches exist for their products. The clearest competitive trend has been the advancement toward SoCs, which replace discrete read channels and hard disk controllers along with their suppliers, unless they can come up with an SoC solution. Thus, those competitors without a clear road map to an SoC have pared back drive business over the past few years: the market has witnessed the exits of Texas Instruments (from read channels), Cirrus Logic (read channels, HDCs), NXP/Philips (preamps, motor control ICs) and most recently Qlogic with the sale of its motor controller business to Marvell. Still others (Infineon and STMicroelectronics) have seen their market share decline as they got integrated out. We describe below the competitive environment for hard disk drives ICs by device, and follow with detailed exhibits. System-on-a-chip (SoC): Marvell is the clear leader in HDD SoCs, holding 53% share in 2005. Marvell supplied Western Digital and Samsung for the desktop and Toshiba, Fujitsu, Samsung and Western Digital for mobile. Note that Marvell gained Maxtor as a new customer for desktop drives in 2006, though the business went away after Seagate acquired Maxtor and shut down those platforms. Agere was No. 2 with 26% share, supplying Seagate for desktop and mobile and Maxtor for desktop (that business was transitioning to Marvell until it was shut down). STMicroelectronics, with 15% share, supplied SoCs to Seagate via a partnership with Agere. Infineon, with 6% share, entered the SoC market in 2004 supplying desktop SoCs for Hitachi GST. Hard Disk Controllers (HDC): The top supplier of discrete HDCs in 2005 was LSI Logic, with 37% share. Other HDC suppliers include Qlogic/Marvell (27%), Renesas (9%), Agere (8%), TI (7%), STMicro (5%), Marvell (4%) and IBM (3%). Note that most of these are hard disk controllers for enterprise drives, which havent been affected by SoCs. Read Channels: Marvell is the top supplier of read channels, with 67% share in 2005, and supplies virtually all enterprise read channels. Agere followed with 20%, mostly from mobile with Hitachi GST (though this business was lost in 2006). IBM followed with 11%; most of that volume is consumed by Hitachi GST in the old IBM platforms. Preamps: TI is the top supplier of preamps, with 71% share in 2005. STMicroelectronics followed with 18% and Agere held 13%, though its share recovered nicely in 2006. Marvell is also a minor supplier of preamps. Motor Control ICs: Motor control ICs are specialized high performance ICs that require both HDD platform knowledge and more generalized expertise in analog motor control. They are dominated by three players: STMicroelectronics (42% share), Texas Instruments (41% share), and Renesas (14%). Marvell and Agere also participate, but are small.

145

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

2005 Hard Disk Drive SoC Market Share


Infineon 6% STMicro 15 %
Agere Marvell

Key Hard Disk Drive SoC Wins


Desktop Western Digital Samsung Maxtor* Mobile Toshiba Fujitsu Western Digital Samsung Seagate** Seagate**

Maxtor* Seagate** Seagate** Hitachi GST

STMicroelectronics

Marvell 53 %

Infineon

Agere 26 %

* Seagate acquired Maxtor in 2006. Agere supplied all of Maxtors SoCs until 2006; Marvell began supplying Maxtor in 2006 but lost the business after the acquisition by Seagate. ** Agere and STMicroelectronics share the Seagate business, which is developed primarily with read channel technology from Agere and control IP from STMicroelectronics. Note that not all platforms indicated in the table above use SoCs. For example, some drives use discrete read channel and hard disk controller solutions.

2005 Hard Disk Controller Market Share


Marvell 4% STMicro 5% TI 7% LSI Logic 37 % Agere 8% IBM 3%

2005 Read Channel Market Share


Infineon 2%

IBM 11 %

Agere 20 %

Renesas 9%

Marvell 67 %
QLogic/Marvell 27 %

2005 Motor Control IC Market Share


Marvell 2% Agere 1%

2005 Preamp Market Share


Marvell 2%

Renesas 14 %

Agere 13 %

STMicro 42 %

STMicro 18 %

TI 41 %

Texas Instruments 71 %

Source: IDC, CIBC World Markets Corp.

146

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The exhibits below display market shares for hard disk drive semiconductors by drive class. 2005 HDD Semiconductor Share By Class 2005 Desktop HDD Semiconductor Market Share
Renesas 4%
IBM 1%

Enterprise 20 %

Infineon 5%

Marvell 25 %

TI 15 %

Desktop 54 % Mobile 26 % STMicro 23 % Agere 25 %

2005 Mobile HDD Semiconductor Market Share


Renesas 2% Agere 9% IBM 2%

2005 Enterprise HDD Semiconductor Market Share


Renesas IBM 3 % Infineon 1 % STMicro 1% 4%

Agere 7%

LSI Logic 28 %

STMicro 13 %

TI 15 % Marvell 54 %

TI 21 %

QLogic/Marvell 21 %

Marvell 21 %

Source: IDC, CIBC World Markets Corp.

147

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Hard Disk Drives Hard Disk Drives


Hard Disk Drive Semiconductor Market Trends
The transition to SoC continues; adoption reached 85% in 2005 and should reach 97% by 2010. Enterprise SoCs should begin to appear in 2007. SATA now dominates the desktop and notebook and is enabling new low-cost enterprise systems. The transition from SCSI to SAS in the enterprise has begun. Consumer HDDs continue to experience strong growth despite the challenge from flash. Hybrid Hard Disk Drives are coming to market and will enable intelligent caching with Windows Vista.

SoC adoption reached 85% in 2005 and should reach 97% by 2010. SoC solutions pervade HDD designs; by 2005 all major desktop and mobile drive vendors had adopted SoCs for all or some of their volume. SoC penetration already reached 85% in 2005, and should receive another boost when enterprise drives begin implementing SoCs in 2007. Note that Marvell bought QLogics enterprise HDC business with this goal in mind. SATA now dominates the desktop and notebook and is enabling new low-cost enterprise systems. Though it had a slow start when launched in 2003, SATA has now taken over the mainstream desktop and notebook PC market. We are also seeing SATA drives penetrate the enterprise market in new low-cost systems. The transition from SCSI to SAS has begun. For the enterprise market, HDD OEMs have begun to migrate their SCSI product lines to SAS. We expect 2007-08 to be big adoption years for SAS. Consumer HDDs continue to experience strong growth despite the challenges from flash. The consumer HDD market continues to grow double digit annually despite the challenge from NAND flash. Some applications, most notably MP3 players, have already been conceded to flash, but applications which require higher capacities (most notably DVRs and game consoles) remain captive to HDDs. Note that HDD IC vendors continue to design low-cost, low-power solutions, and ramp capacity specifically for consumer applications. Hybrid HDDs will enable intelligent caching with Windows Vista. New caching features in Vista can be enabled either with a flash module (Intels Robson) in the PC system or flash embedded in a hybrid HDD (HHD).

148

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Optical Drives Optical Drives


Optical drives use direct access disks written and read using lasers. They offer much higher capacity than magneto-optic (floppy) disks and have become ubiquitous in computing over the past few years. CD and DVD are the mainstream technologies today; Blu-Ray and HD-DVD, two blue-laser DVD formats, are being promoted as the next-generation successor to DVD.
Read Only CD DVD Blu-Ray HD-DVD CD-ROM DVD-ROM BD-ROM HD-DVD-ROM Write Once CD-R DVD-R DVD+R BD-R HD-DVD-R Re-write CD-RW DVD-RW DVD+RW BD-RE HD-DVD-RW Capacity, Single Layer 650 MB 4.7 GB 25 GB 15 GB 8.5 GB 50 GB 30 GB Capacity, Dual Layer

The PC optical drive market includes CD- and DVD-based read and record data storage drives. These storage devices use direct access disks read and written using a laser. The disks are removable, and the drives use lasers to read from and write to them, similar to the audio compact discs from which the technology evolved (but using a different format). On the back end, the drives use the standard PC ATA or SATA interface to transfer data to and from the host system. Sold in standard form factors at consumer-oriented prices, they are bundled with PC systems and have become ubiquitous in computing over the past few years. Optical disks offer much higher capacity points than magneto-optic (floppy) disks. A standard 120 mm CD-ROM holds 650 MB of data, though some are slightly larger and hold 700 MB. Standard transfer speed is 150 KBps; CD-ROM speed is quoted as a multiple of this number (e.g., a 24X CD-ROM operates at 3600 KBps). DVD-ROMs are more dense than CD-ROMs and can store 4.7 GB of data on a single-layer disc or 8.5 GB on a dual-layer disc. Standard 1X transfer speed is 9X the speed of a CD-ROM drive, or 1.35 MBps. Originally sold as read-only devices, most PC drives include the capability to write (R) or re-write (RW) to the disks. Two blue-laser DVD formats are currently being promoted to succeed DVD: Blu-Ray and HD-DVD. Blu-Ray can store 25 GB of data on a single-layer disc or 50 GB on a dual-layer disc; transfer speed is 36 Mbps. Blu-Ray is supported by such hardware vendors as Sony, Samsung, Panasonic, Philips, Pioneer, Hitachi, Dell, HP, and LG. HD-DVD can store 15 GB of data on a single-layer disc or 30 GB on a dual-layer disc; transfer speed is also 36 Mbps. HD-DVD is supported by such hardware vendors as Toshiba, NEC, and Sanyo as well as Intel and Microsoft.

149

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Optical Drives Optical Drives


Optical Drive Unit Forecast
We expect the optical drive market to grow from 301 million units in 2005 to 456 million units in 2010, a 9% CAGR. DVD-Recorder drives should grow 24% through 2010, from 41% of drive shipments in 2005 to 80% in 2010. DVD-ROM drives should grow at a 2% rate; CD-ROM, CD-RW, and Combo drives should decline steadily. Blue-laser formats wont be significant in the PC market for several years.
500,000 450,000 400,000 350,000
Unit Shipments (000s)

Optical Drive Suppliers


Sony 2% Pioneer 4% NEC 5% Panasonic 8% Others 6% Hitachi-LG 26 %

Philips-BenQ 10 % ToshibaSamsung 22 %

HD-DVD & Blu-Ray DVD-Recorder Combo CD-RW DVD-ROM CD-ROM

Lite-On 17 %
Source: IDC, CIBC World Markets Corp.

300,000 250,000 200,000 150,000 100,000 50,000 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, CIBC World Markets Corp.

Optical Drive Unit Forecast: Optical drives have become ubiquitous in the PC market and in fact exceed PCs in terms of unit shipments due to a significant retail drive market and the growth of dual-drive desktops in the consumer market. We estimate the market totaled 301 million units in 2005. By technology, optical disk drives were broken out 41% DVD-Recorder, 20% combo (CD-R/W and DVD-ROM), 16% DVD-ROM, 14% CD-ROM, and 9% CD-R/W. Looking ahead, we expect the total market to grow at a 9% CAGR, reaching 456 million units in 2010. DVD-recorder should see the strongest growtha 24% CAGR, rising to 80% of drives in 2010. DVD-ROM should grow at a 2% CAGR to 11% of the market in 2010. CD-RW and CD-ROM drives should decline at a more than 45% CAGR and combo drives should decline at a CAGR of more than 25%, together representing less than 5% of the market in total. We do not expect blue-laser formats to be significant in PCs over the next few years (they will be much more important in the consumer electronics market for high-definition DVDs); we forecast a slow ramp to about 26 million units in 2010. Optical Drive Suppliers: The ODD market has seen some unusual consolidation trends: three of the top four suppliers are actually JVs between two separate companies, with each partner company based in another country. The top supplier in 2005 was Hitachi-LG with 26% unit share, followed by Toshiba-Samsung with 22% and Lite-On IT with 17%. Philips-BenQ and Panasonic followed with 10% and 8%, respectively. Other suppliers include NEC, Pioneer, and Sony.

150

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Optical Drives Optical Drives


This ASSP demultiplexes the analog signal and converts it to digital This ASSP performs digital processing on the signal, preparing it for decode by the host system

CD/DVD Drive Bay

RF Amp Laser Driver

Servo/Motor Front End Drivers Analog DemodASSP ulator


Source: IDC, CIBC World Markets Corp.

Servo Control Read Channel DSP

Front End Digital ASSP


Front End Decode/ (Encode*)

MPU/ MCU
ATA Controller

Host System

SRAM

DRAM

ROM

Amplifies the signal coming from the DVD read laser The digital ASSP uses SRAM, DRAM, and flash memory for processing and boot code storage

Controls the DVD read (and write) laser

Optical drive semiconductors perform only front end functions; this includes reading the CD or DVD and performing the analog to digital conversion and preliminary decoding, as well as drive bay and laser control. Back end functions, which include decoding the signal and processing it into data signals, are done by the host PC. Laser Driver and RF Amp: This optoelectronic driver controls the CD/DVD read and write laser. The optoelectronic amplifier boosts the signal coming off the laser. Front End Analog ASSP: This device includes both the servo/motor driver and the demodulator/analog front end. The servo driver controls the movement of the drive spindle. The demodulator receives the signal from the RF amp and converts it to a digital format to be processed. Front End Digital ASSP: This device includes the servo control IC, read channel DSP, front end decoder/encoder, the ATA/SATA controller and a microcontroller or microprocessor. The servo DSP controls the servo driver and is also responsible for error correction control. The read channel DSP and front end decoder perform the first layer of digital processing on the digital signal and prepare it for back end video/audio processing; in a recorder drive, it also encodes the stream for writing it to the CD/DVD. The ATA/SATA controller provides the interface to the host system using the ATA/SATA protocol. DRAM/SRAM/ROM: The digital ASSP and embedded microcontroller/microprocessor use several external memories for cache and boot instructions.

151

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Optical Drives Optical Drives


Optical Drive Optical Drive Semiconductor Forecast Semiconductor Competitors
Optical drive semiconductors are growing at an estimated 6% CAGR, from $2.9 billion in 2005 to $4.0 billion in 2010. The optical drive chipset, which includes the analog front end and digital SoC, should grow in line with the market from $1.9 billion in 2005 to $2.6 billion in 2010. Optoelectronics should grow 5%, from $981 million in 2005 to $1.2 billion in 2010. MCUs/DSPs, used in higher-end drives, should grow from $82 million in 2005 to $165 million in 2010, a 15% CAGR.
$4,500 Optoelectronics $4,000 $3,500
Semiconductor Revenue ($M)

Optical Drive Chipsets


MediaTek Renesas NEC Matsushita Rohm Ricoh Toshiba Seiko-Epson Sanyo Intersil Marvell

ODD Optoelectronics
Sharp Avago Stanley Electric Sony Nichia Chemical Toshiba OSRAM Mitsubishi NEC Vishay Rohm

Analog Front End & Digital SoC Microcontroller/DSP

Source: CIBC World Markets Corp.

$3,000 $2,500 $2,000 $1,500 $1,000 $500 $0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: Dataquest, IDC, CIBC World Markets Corp.

Optical Drive Semiconductor Forecast: The market for optical drive semiconductors includes two device segments: drive logic (an analog front end and digital SoC, sometimes augmented by a discrete DSP or microcontroller) and optoelectronics. In total, we estimate the market was $2.9 billion in 2005, and is growing at a 6% rate to $4.0 billion in 2010. We estimate 64% of IC revenue was ASIC/ASSP solutions in 2005, 3% was MCUs/DSPs and 34% was optoelectronics. The mix should stay mostly consistent, though MCUs and DSPs should rise slightly as a percentage as more complex DVD-recorder drives ramp up; these sometimes require a discrete microcomponent. Optical Drive Semiconductor Competitors: The optical drive market is served by both merchant chipset and custom ASIC vendors, with merchant vendors supplying most lower end CD-ROM and CD-RW drives and about half of all mainstream DVD-ROM and combo drives. ASIC suppliers serve the higher end DVD-R/RW drives, though over time these are likely to migrate to merchant as well. Blue laser drives are likely to remain captive for some time. The largest supplier of optical disk drive chipsets is Mediatek, which supplies merchant solutions for all segments of the market but dominates the low-end and mainstream. Other suppliers include Renesas, NEC, Matsushita, Rohm, Ricoh, Toshiba, Seiko-Epson, and Sanyomost of these are ASIC suppliers. High performance analog player Intersil also has a solid position in laser drivers. Note that HDD IC leader Marvell recently entered the market. The top optoelectronic suppliers for disk drives are Sharp, Avago, Stanley Electric, Sony, Nichia Chemical, Toshiba, OSRAM, Mitsubishi, NEC, Vishay, and Rohm.

152

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Optical Drives Optical Drives


Optical Drive Semiconductor Market Trends
Integration has reduced chip count for mainstream optical drives to one digital IC and one analog IC; single-chip solutions dominate the low end. The shift to DVD-Recorder drives is changing the competitive landscape in favor of high end suppliers. The standards battle continues for the successor to DVD, though consumer electronics concerns will dominate.

Integration has reduced chip count for mainstream optical drives to one digital IC and one analog IC; single-chip solutions dominate the low end. Our block diagram on page 151 reflects the integration that has already taken place. Note that vendors now offer single-chip solutions for the lower end that incorporate both digital and analog functions (leaving only optoelectronics and memory separate). Mainstream and high end drives still use the two-chip architecture, but IC vendors will be under pressure to integrate over time. Note that Marvells entrance strategy is grounded in its single-chip SoC, which it is offering for the high end x16 DVD recorders. The shift to DVD-Recorder drives is changing the competitive landscape in favor of high end suppliers. Like any major technology transition, the shift to DVD-Recorders is favoring suppliers positioned in these drives, primarily the captive vendors like NEC and Matsushita, which lead the market for DVD-Recorder ICs. The standards battle continues for the successor to DVD, though consumer electronics concerns will dominate. Both Blu-Ray and HD-DVD have solid hardware backing at this point, with no clear winner evident yet. We expect consumer electronics concerns will dominate the debate; the industry is primarily looking to highdefinition video content to drive interest in the new formats; PCs will be somewhat of an afterthought and will therefore follow the CE market. Note that the battle should have minimal impact on the merchant optical drive semiconductor vendors for now, as next-generation models will likely be implemented with only captive solutions.

153

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Printers and MFPs Printers and MFPs


The printer market has seen continued innovation despite its maturity. The razor business model continues to produce affordable printers for consumer and enterprise. Multi-function peripherals, which include printing, fax, scanner, and copier functions, have enjoyed success in the home and SMB, and now represent over 40% of total shipments. Photo printers have grown even more quickly, and in 2005 were about half of total shipments.
Printers vs. MFPs
100% 100%

Printer Type
Color Laser Mono Laser
80%

MFP
80%

Photo Inkjet
60% 60%

40%

40%

Standard Inkjet
20%

Printer

20%

Impact
0% 2000 2001 2002 2003 2004 2005 0% 2000 2001 2002 2003 2004 2005

Source: DisplaySearch, IDC, CIBC World Markets

Source: IDC, CIBC World Markets Corp.

The printer market has seen continued innovation despite its maturity. OEMs have focused on driving steady replacement by packing additional functionality as well as improving print quality and lowering cost. The razor business model continues to produce affordable printers for consumer and enterpriseOEMs continue to subsidize the initial cost of printers and rely on high-margin replacement toner and ink for profitability. This model drives a huge focus on unit share which drives the installed base and therefore future purchases of ink cartridges. Innovation has happened along two main vectors: features and imaging. On the features front, OEMs are incorporating printing, fax, scanner, and copier capabilities into all-in-one devices, known as multi-function peripherals, or MFPs. In 2005 over 40% of total shipments were MFPs. On the imaging side, advancements in digital photography are driving increased demand for photo printing solutions, and more than half of all printers sold in 2005 were photo capable. OEMs even offer small form factor printers designed to print photos only. Note that the MFP and photo printer trends are not mutually exclusivemany MFPs are photo-capable. Over time, we would expect MFP penetration to continue to rise at a steady pace to well over half of total shipments. Photo printers (and MFPs) should rise even more rapidly until they represent nearly all consumer shipments. The market for standard inkjet printers is continually declining.

154

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Printers and MFPs Printers and MFPs


Printer and MFP Unit Forecast
We expect printers and MFPs to grow at at a 3% CAGR, from 120 million units in 2005 to 138 million units in 2010. MFPs should grow at 10% CAGR, from 44% of units in 2005 to 60% in 2010. Inkjet printers should decline at a 9% CAGR, giving way to MFPs. Mono and color page printers should grow at a 5% CAGR; new lower-cost color laser printers should drive most of this growth.
160,000
Multi-Function Peripherals Mono and Color Page Printers Inkjet Printers

Printer and MFP Suppliers


Brother Dell 4 % 6% Samsung 3% Xerox 1%

Lexmark 11 %

HP 42 %

Canon 16 %

140,000

Epson 17 %

120,000

Unit Shipments (000s)

100,000

Printers
HP Canon Epson Lexmark Dell Samsung Brother OKI

MFPs
HP Epson Lexmark Canon Brother Dell Samsung Xerox

80,000

60,000

40,000

20,000

2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, CIBC World Markets Corp.

Source: Dataquest, CIBC World Markets Corp.

Printer and MFP Unit Forecast: Though clearly tied to the PC market, printers represent a separate purchase for enterprise and consumer PC users and do not track PC shipments all that closely. In recent years, printer shipments have bounced between 57% and 66% of PC shipments. MFPs and photo printers have generated some additional demand, driving several years at the high end of the range. This effect appears to be waning, and in 2005 the printer ratio dropped to 59%, with 56% forecast for 2006 and down from there on. In total, we expect printer shipments to trail PC growth, growing at a 3% CAGR, from 120 million units in 2005 to 138 million units in 2010. The fastest growth segment is still expected to be multi-function peripherals, already at 44% of unit shipments in 2005. MFPs are expected to grow from 53 million units in 2005 to 83 million units in 2010, a 10% CAGR, and should represent 60% of the market in 2010. Mono and color page printers should follow at a 5% CAGR, with particular strength in the color laser segment as aggressive pricing by OEMs spurs increased adoption in the enterprise. Inkjet printers will give way to MFPs over time, and should decline at a 9% rate, representing just 20% of units in 2010. Printer and MFP Suppliers: HP dominated the printer market in 2005 with 42% unit share; HP is the market leader in both standalone printers and MFPs. Epson followed with 17% and held the No. 2 position in MFPs and No. 3 position in printers. Canon followed with 16% share and held No. 2 position in printers and No. 3 position in MFPs. Lexmark, which plays primarily in the enterprise market, was No. 4 with 11% share. Other major suppliers include Dell (6% share), Brother (4%), Samsung (3%), Xerox (1%), followed by Oki, Lenovo, Sharp, and Konika Minolta.

155

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Printers and MFPs Printers and MFPs


A digital-to-analog converter converts printing instructions to analog, and the printer head driver interfaces with the cartridge; a corresponding IC on the consumable drives the print cartridge Generates the image on the LCD screen A custom ASIC or SoC controls the overall printer/MFP system A high-performance 32-bit or 64-bit CPU performs the intense computations that drive the printing function

LCD Screen

LCD Drivers

DRAM Microprocessor Flash

The MPU uses DRAM and Flash This block can include either a Bluetooth or wireless LAN chipset

Cartridge Print Head Head Motor Head Positioning Motor Paper Loading Motor
A motor driver controls the movement of the printing mechanism

Printer Head Driver

D/A Converter Motor Driver

LCD Controller

Bluetooth/ WLAN Block Interface Controller

Antenna USB Cable


Interfaces with the USB, FireWire, or serial cable to connect up to the PC

System Control ASIC/SoC

Scan Area Scan Motor

CCD Image Sensor

A/D Converter Motor Driver

MFP Block Image Processor Fax Codec


MCU Data Pump

Line Driver

RJ-11 Phone Line

An image sensor captures the image from the scan/copy area

An analog-to-digital converter converts the analog image signal to digital for processing

A motor driver controls the movement of the scanning mechanism

Controls the scanning subsystem; also performs digital processing on the scanned image

Implements the fax transmission/reception scheme

Interfaces with the telephone line, driving the analog signal

Source: STMicroelectronics, NEC, TDK, CIBC World Markets Corp.

Printing Block System Control/ASIC/SoC: A custom ASIC or SoC controls the primary printing function. In mainstream inkjet or mono laser printers, the ASIC/SoC usually integrates a microprocessor to perform the intense computations involved in printing. Microprocessor: In high-end color laser printers, a discrete high-performance 32-bit or 64-bit CPU performs the intense computations that drive the printing function. DRAM and Flash: The embedded or discrete microprocessor uses DRAM and flash. D/A Converter (DAC): A DAC converts digital printing instructions to analog and interfaces with the print head. Printer Head Driver: The printer head driver interfaces with the print cartridge. These are typically custom to individual OEMs, as a means to insure only OEM-branded print cartridges can be used. Motor Driver: The motor driver controls the movement of the printing mechanism.

MFP Block Scanner/Copier CCD Image Sensor: The image sensor receives the image from the scan/copy area.

156

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

A/D Converter: The ADC converts the image from analog to digital for processing, printing, and storage. Motor Driver: The motor driver controls the movement of the scanning mechanism. Image Processor: An image processor controls the scanning subsystem and performs image processing.

MFP Block Fax Fax Codec: The Fax AFE/codec performs the analog-to-digital conversion to generate the fax signal. Similar to analog modem codecs, they are specialized mixed-signal processors and typically include a data pump and embedded microcontroller. Line driver: This analog SLIC interfaces with the RJ-11 copper twisted pair (phone line), performing transmission and reception functions.

Interface and Display USB/1394 Controller: A USB/Firewire controller manages the interface to the PC. Bluetooth Baseband and RF/IF (not pictured in diagram) Bluetooth-enabled printers can interface wirelessly with a PC via a personal area network. They can also use Bluetooth to connect directly to a digital camera to print out photos images. In a Bluetooth chipset, the baseband performs the digital processing and protocol encoding and the RF/IF performs the radio transmission/reception. Wireless LAN Baseband/MAC and RF/IF/PA (not pictured in diagram): Wireless LAN-enabled printers can interface wirelessly with a PC or camera via an 802.11 wireless LAN in order to transfer images. In an 802.11 chipset, the baseband/MAC performs digital processing and protocol encoding and the RF/IF/PA performs the radio transmission/reception and power amplification. Display Drivers: Display drivers produce the image on the LCD screen.

Print Cartridge/Consumable Print Head: A specialized print head IC sits on each individual print cartridge and controls the functioning of the cartridge. Like the head driver on the printer, these are typically custom to individual OEMs, as a means to insure only OEM-branded print cartridges can be used.

157

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Printers and MFPs Printers and MFPs


Printer and MFP Semiconductor Forecast
We expect the market to grow at a 5% CAGR, from $1.9 billion in 2005 to $2.4 billion in 2010. ICs should slightly outpace units as mono and color laser printers outpace inkjet, driving higher ASIC/SoC ASPs and greater microprocessor content. Much of the growth generated from ICs for scanning and fax in MFPs is now behind us; integration and ASP pressure should now mitigate the effects of stillrising MFP penetration.
$3,000 Fax Codec and Line Driver Scanning/Copying Subsytem Embedded Microprocessor Printer SoC/ASIC and Head Driver

Printer and MFP IC Competitors


STMicroelectronics Texas Instruments Avago/Marvell Freescale NEC Electronics PMC-Sierra Toshiba Seiko-Epson Kawasaki Micro NXP/Philips
Source: CIBC World Markets Corp.

$2,500

Semiconductor Revenue ($M)

$2,000

$1,500

$1,000

$500

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: Dataquest, CIBC World Markets Corp.

Printer Semiconductor Forecast: We expect the market for printer and MFP semiconductors to grow at a 5% CAGR, from $1.9 billion in 2005 to $2.4 billion in 2010. Semiconductor revenue should slightly outpace units due to the rising penetration of laser vs. inkjet, both in the printer category and in MFPs. These laser printers use more robust SoCs, and in the case of high-end or color laser, also include significant discrete microprocessor content. Note that the rapid growth of MFPs has generated significant growth on the IC side, as these units require image processing and analog ICs to enable scanning and copier functions as well as the codecs and line drivers ICs for fax capability. In 2005, we estimate these two IC opportunities combined yielded more than $300 million in market revenue. Much of the IC growth generated by this trend is now behind us, however, as OEMs shift focus to lowering the cost of these features, primarily through integration. We expect the revenue associated with these two IC subsegments to grow only modestly through 2010 and to be modestly dilutive to the overall IC CAGR. Printer Semiconductor Market Share: The printer IC market is dominated by a number of large, diversified semiconductor vendors with design competencies in SoC/ASIC, microprocessing, and analog design. Major vendors in this space include STMicroelectronics, Texas Instruments, Avago/Marvell (Marvell acquired Avagos printer business in 2006), Freescale, NEC, PMC-Sierra, Toshiba, Seiko-Epson, Kawasaki Microelectronics, and NXP/Philips. A second set of companies provides fax ICs into MFPs; these include Conexant, Silicon Labs, Agere, Austria Microsystems, and Analog Devices.

158

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Printers and MFPs Printers and MFPs


Printer and MFP Semiconductor Market Trends
The market remains heavily ASIC-based as OEMs continue to value control over the design to keep out standardized ink cartridges. The rise of multi-function peripherals has opened up opportunities for imaging and communication ICs. All but the highest-end printers are moving to SoC solutions. 802.11, Bluetooth, and UWB are all candidates for wireless connectivity within printers.

The market remains heavily ASIC-based as OEMs continue to value control over the design to keep out standardized ink cartridges. To preserve the economics of the razor business model, OEMs need to preserve the proprietary nature of their designs, especially with regards to the print cartridge. They have therefore kept their printer designs predominantly ASIC-based. The rise of multi-function peripherals has opened up opportunities for imaging and communication ICs. MFPs can include printing, scanning, copying, and fax functions, and therefore require specialized ICs to enable these functions. Note that while the core printing functions in MFPs are enabled by ASICs, as mentioned above, imaging and communications functions has allowed merchant suppliers to penetrate the printer market. All but the highest-end printers are moving to SoC solutions. Integration is a major focus for OEMs, especially considering they usually lose money on printer hardware. The market has seen many discrete ICs move into the core ASIC/SoC over the past few years, in particular the microprocessor. At this point, only highend color laser printers utilize a discrete MPU, and while this market is growing now with the rising penetration of color laser, it saw a major drop just a few years ago as inkjet and mono laser printers moved to SoC. 802.11, Bluetooth, and UWB are all candidates for wireless connectivity within printers. Though not included in our IC forecast, we expect to see wireless networking technology move into printers over time. Penetration thus far has been slow, but could jump nicely when UWB is available in volume.

159

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Networking Networking
We now move from computing applications, which allow users to manipulate data stored locally within a single PC or server as well as to interact with peripherals, to networking applications, which enable the transfer of data between these devices. This section deals primarily with equipment used to connect PCs, servers, and storage in a single physical location, e.g., a small business, a single corporate branch office, or a home network environment. Equipment used to send data over longer distances (e.g., modems) typically utilize public telecom and cable infrastructure or corporate wide area networks and are discussed later in the section on telecom and datacom. Voice-over-IP, which utilizes corporate local area networks but sends traffic over service provider networks, is also discussed later.

We now move from computing applications, which include equipment that allows users to manipulate data stored locally within a single PC or server as well as to interact with directly attached peripherals, to networking applications, which enable the transfer of data between PCs and servers. This section deals primarily with equipment used to connect PCs, servers and storage in a single physical location e.g., a small business, a single corporate branch office, or a home network environment. Given the close ties to computing, networking equipment is usually designed with the PC in mind. Metrics like speed, ease of install and configurability and low maintenance cost generally take precedence over factors such as ensuring quality of service or multiple classes of service support (a situation that is reversed in telecom networks). Equipment used to send data over longer distances (e.g., modems) typically utilizes public telecom and cable infrastructure or corporate wide area networks and is discussed separately (in the section on telecom and datacom). This equipment, though datacentric in its operation, is designed to leverage the existing voice or video infrastructure to send data rather than act as a standalone network, though some standalone wide area data networks do exist. The economics are also dramatically different, as telecom and datacom equipment is tied to service provider capital expenditures, while networking equipment is tied more heavily to IT spending.

160

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Networking Networking
The OSI Reference Model The OSI reference model provides a framework for the development of networking technologies and standards. The model divides networking functions into seven layers, all of which can be designed independent of each other. The upper layers (4-7) deal with user applications, and function independently of the type of physical network used to carry information. The lower layers (1-3) are dedicated to issues relating to the actual transport of information around the network.
Source: Computer Desktop Encyclopedia, The Computer Language Co.

Networking technologies are designed to share data between computing devices. These devices connect using a wide variety of terminal equipment and operate using disparate communications protocols, transmission media and data transfer rates. Given the disparate technologies, the challenge of internetworking becomes how to connect these networks such that data can be reliably transmitted among them. In order to meet the challenges of internetworking, network architects began to segment communications tasks into separate and self-contained layers of functionality. Each layer is designed to perform a certain task or range of tasks and is independent of the layers above and below it. This process was formalized in 1984 by the International Organization for Standardization (ISO) with the publication of a seven-layer reference model. The reference model governs how information from a user application in one computer is transmitted across a network or inter-network to a user application in another computer. In this way, it provides a framework for the coordination of standards development, allowing existing and evolving standards to be set within a common framework. The OSI model has therefore become the primary architectural model for internetworking communications. The layers of the OSI reference model can be grouped into two categories: upper layers (4-7) and lower layers (1-3). The upper layers deal with actual user software applications (like Microsoft Outlook or Internet Explorer), and function independently of the type of physical network used to carry information. The upper layers are typically implemented in software running on the networked device. The lower layers are dedicated to data transport issues and can be implemented in either software or hardware.

161

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

In order to understand the function of each layer of the model, we can look at the process of sending an E-mail from a PC. A computer is a layer 7 device, meaning that it can operate from the highest layer (which is where applications are run) to the lowest layer (the PC can put signals directly onto a network cable plugged into a port). At the application layer (7), Outlook (or another E-mail client) deals with the composition of the E-mail, including formatting, attachments, and security features. At the presentation layer (6), the E-mail is converted into a standard format compatible with all E-mail programs. At the session layer (5), a channel is opened between the PC and the network for message transmission. At the transport layer (4), the operating system passes the E-mail to the computers networking subsystem. At the network layer (3), the system makes sure it has a clear path to the rest of the data network. At the link layer (2), a network address for the E-mail is attached to the message. Finally, at the physical layer (1), the PC sends out the actual signal over the network cable. In receiving an E-mail, the process is reversed, though the functions performed are the same.

The OSI Reference Model: Layer 1-7 Summary The Physical Layer (1) This layer defines the electrical, mechanical and functional specifications for transmitting a data signal over a network link. The physical layer is also responsible for establishing, maintaining and terminating the physical signal between network devices. Physical layer protocols typically define parameters such as voltage level, shape or template of transmission signal, data transfer rates, maximum transmission distances and physical connectors. The Data Link Layer (2) The data link layer controls the transfer of data across a physical network link. As such, data link layer protocols define the characteristics of network links including network topology, physical addressing, frame formats, frame sequencing, error notification and flow control. The IEEE (Institute of Electrical and Electronics Engineers) further subdivides the link layer into two sub-layers: logical link control (LLC) and media access control (MAC). The LLC sublayer manages communications between devices separated by a physical link, while the MAC sub-layer manages access to the physical network medium. The Network Layer (3) The network layer is responsible for establishing, maintaining and clearing a network-wide connection between two transport layer entities. The network layer principally provides routing and other functions that enable data to be transmitted over disparate physical links. The network layer provides both connection-oriented and connectionless service to layers above it. Most network layer protocols are routing protocols. Note that network layer routing functions are based on logical addressing rather than physical addressing. The Transport Layer (4) This layer functions as the interface between the application-oriented upper layers and their underlying networkdependent layers, providing transparent yet reliable inter-network data transport services to the higher layers. Effectively, the transport layer hides the detailed operation of the underlying network from the upper layers. Transport layer functions include flow control, multiplexing, virtual circuit management and error correction. Flow control manages the rate of data transmission between two entities such that the transmitting devices do not send data at a rate greater than what the receiving entity can process. Multiplexing allows several communications channels to be carried over the same physical link without any data loss. Virtual circuit management involves the creation, maintenance and termination of virtual circuits (a virtual circuit is a lower speed channel carried over a higher speed physical link). Finally, error checking involves mechanisms for the identification and correction of transmission errors. The Session Layer (5) This layer is responsible for establishing, managing and terminating a communication channel between presentation layer entities through a series of communications requests and responses. In addition, the session layer determines whether communication will be half-duplex (two-way transmission, but only one direction at a time) or full-duplex (simultaneous two-way transmission), maintains synchronization between communicating entities, and performs exception reporting. The Presentation Layer (6) This layer is concerned with the syntax of data during transfer, i.e., it ensures that data is formatted using a common interface for user applications. To achieve network interoperability, a number of transfer syntaxes have been defined. The presentation layer negotiates and selects the appropriate transfer syntax for each communication so that data transmitted may be understood by both systems. If the transfer syntax differs from the application syntax, the presentation layer performs the necessary conversion. Some common transfer syntaxes include the ASCII and EBCDIC for text and data, motion picture experts group (MPEG) and QuickTime for video and graphics interchange format (GIF), joint photographic experts group (JPEG) and tagged image file format (TIFF) for graphics images. The presentation layer also performs data compression/decompression and encryption/decryption if required.

162

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The Application Layer (7) This layer interacts with and provides services directly to user software applications. It deals directly with the communications component within an application (e.g., the Inbox in Microsoft Outlook) and provides the interface to the services of the network. Functionality provided by this layer typically includes identification of communication partners, determination of network resource availability, synchronization, file transfer, message interchange services such as E-mail, agreement of privacy (encryption) standards and identification of constraints on data syntax.

Networking Equipment, Components, and the OSI Model Though it all seems technical and theoretical, it is important to understand the layers of the OSI model in order to identify the intended functionality of the various types of networking equipment. For instance, a layer 1 device would deal only with actual physical signaling. An example of a layer 1 device would be a repeater (a device used to boost a signal over a length of cable), which simply receives a signal (at the physical layer [1]) and spits it back out. Layer 2 deals with addressing. A layer 2 example would be a switch, which receives traffic (at the physical layer [1]), checks the address of the traffic (at the data link layer [2]), and sends the traffic out again (at the physical layer [1]) over the proper channel to reach the recipient. Layer 3 deals with routing. A layer 3 example would be a router, which receives traffic (at the physical layer [1]), checks the address of the traffic (at the data link layer [2]), determines the best route for the data to take to its destination (at the network layer [3]) and then re-addresses and sends out the traffic (at the data link [2] and physical [1] layers). We wont carry the example all the way up to layer 7, but it is clear why we called a networked PC a layer 7 device. It can do all the functions of a switch, router, or other more complicated devices because of its robust and flexible operating system. In fact, layer 4-7 network devices are often designed as specialized servers designed to run networking software. We can also carry the analysis forward to the component world (though we are skipping ahead a bit). An Ethernet PHY is a physical layer (1) device, as it deals with reception and transmission of the signal over copper wire or optical fiber. An Ethernet PHY would be found in all types of Ethernet equipment, from a repeater (a layer 1 device) up to a load balancing device or a networked PC (a layer 4-7 device). An Ethernet MAC is a data link layer (2) device, as it applies and removes an address to a data packet. An Ethernet MAC would be found in all types of Ethernet equipment that operate at layer 2 and above, such as an Ethernet switch or NIC card all the way up to the load balancer. A network search engine is a network layer (3) device, as it looks up addresses in a route table. A network search engine would be found in Ethernet equipment that operates at layer 3 or above, from routers up to our load balancer. As mentioned above, layer 4-7 devices are generally implemented in software running on a specialized networking OS. For high-performance equipment, OEMs often use a combination of microprocessors, specialized ASICs, ASSPs, and FPGAs to perform these tasks.

163

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Networking Networking
Communications Flow in the OSI Reference Model For two PCs to communicate, data must travel down the layer stack to the physical layer so it can be passed between nodes in the network. At each node, the packet is processed and forwarded; how far up the layer stack it is processed depends on the function served by the node.
Application Presentation Session Transport Network Data Link Physical PC LAN Switch Router LAN Switch PC

2
Source: CIBC World Markets Corp.

We now look at the OSI reference model in action, and track how it governs communication between networked devices. For data to be transferred from a user application program (e.g., Outlook or Internet Explorer) on one computer (or other connected computing device) to a user application program on another computer, it must pass through each of the OSI layers on both the transmitting and receiving computers. This is because user applications operate at layer 7 of the OSI model, while the port used to transmit the data operates at layer 1. To get the data from layer 7 to layer 1 on the transmitting device, control is passed from one layer to the next, starting at the application layer, down through the middle layers until it reaches the physical layer. The data then travels as layer 1 traffic over the physical network link to the receiving device and back up the layer hierarchy. If more than one physical link separates communicating devices, for instance if the devices are on different floors in an office building, the information contained within the transmitted signal will be processed by each intermediate node sitting on the network between those two devices. These nodes can be hubs, switches, routers, servers, or anything else connected to the network. How far up the OSI layer stack the information will be processed depends on the function served by the node. For instance, LAN switches tend to process information at layer 2, meaning that the information will travel up the layer stack from layer 1 to layer 2, will be processed by the switch at layer 2 and then will go back down to layer 1 within the switch before it is sent on its way. Routers process data at layer 3; the path is similar except that the information will travel as high as layer 3 before being sent back down. More complex devices operate at layers 4-7, as they need to look at the type of data being transmitted in order to perform their functions.

164

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Networking Networking
Encapsulation in the OSI Reference Model Each layer of the OSI model can communicate directly with peer devices at the same layer. In order to communicate with layers above and below, control information in the form of headers and trailers is used. Higher layer headers and trailers are treated as payload data for lower layers. This technique is known as encapsulation.
Device 1
Application Presentation Session Transport Network Data Link Physical L4H L4T = = Layer 4 header Layer 4 trailer L3H L3T = =
L2H L3H L4H Data Data Data Data Data L4H L3H L2H Message Segment Packet or Datagram Frame or Cell

Information Unit

Device 2
Application Presentation Session Transport Network Data Link Physical L2H L2T = = Layer 2 header Layer 2 trailer

Layer 3 header Layer 3 trailer

Source: CIBC World Markets Corp.

Each layer in the OSI model can communicate with only three other places in the network: the layer directly above it within the device, the layer directly below it in the device and its peer layer in other networked devices. The latter process, communication with peer devices, is straightforward: at any given layer, networked devices use the same protocols and are designed to interact together. In order to communicate with other layers, however, control information in the form of headers and trailers prepended or appended to the payload data stream is used. Each layer uses the information in the header/trailer below it in order to process the received payload data. As information travels up the stack, each OSI layer removes the header attached by its corresponding peer layer and forwards the remaining information to the next higher layer. On the way down, each layer will add its own headers and trailers; these will be treated as payload data by lower layers. This is known as encapsulation. Some common terminology used in referring to the information units passed between OSI layers is defined below. Message: An information unit whose source and destination layers lie above the network layer. Segment: An information unit, including header/trailer, whose source and destination layers are transport layer entities. Packet or Datagram: An information unit, including header/trailer, whose source and destination layers are network layer entities. Datagrams are used in connectionless services.

165

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Frame or Cell: An information unit, including header/trailer, whose source and destination layers are data link layer entities. A frame is of variable length while a cell has a fixed length (like in an ATM network).

Encapsulation Demonstration To demonstrate the principles of encapsulation, we will again follow an E-mail as it makes its way through the PC, but will now complete the process by sending it over the network and opening the E-mail on the recipients PC. An E-mail application at layer 7 on the transmitting PC generates the E-mail and tells the system to send it over the network (thats Microsoft Outlook telling Windows it has an E-mail to send). The PC initiates a session (Windows accesses the communications software on the Ethernet NIC/LOM connection) and puts headers and trailers on it and sends it to the lower layers. As it travels down the network stack (within the NIC card at this point), each layer adds headers and trailers to the outside of the previous layer until a fully stacked frame or cell is created. The frame or cell is then placed on the network by the physical layer device (the PHY device on the NIC card). The frame or cell, encapsulated in some physical layer protocol (e.g., Ethernet), is then read by the physical layer and data link layers of all other devices to determine if the frame or cell was intended for it. In the case of switches and routers, the devices will forward the frames out toward other network devices (the distinction being that switches will dump the frame out the port closest to the destination, whereas a router will actually find the best route for the frame and forward it on). Once it finally reaches its destination, the cell will travel up the layer stack. Each layer (starting with those in the NIC card, then the communications software, then Windows) will strip off the headers and trailers from the layer below and pass the data up until it reaches the application (Outlook again or some other E-mail client).

166

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Networking Networking

Four Key Markets:


The most pervasive local area networking protocol, Ethernet is used to connect PCs and other networked devices in a single physical location.

Ethernet (LAN)

Wireless LANs connect users in a single location using radio waves. They are usually used in conjunction with a LAN or broadband Internet connection to extend the wired network wirelessly.

Wireless LAN (802.11)

The most common personal area networking protocol, Bluetooth connects two devices in close proximity wirelessly. Bluetooth is particularly prevalent in the handset market.

Bluetooth

Enterprise storage systems, including direct attached storage (DAS), storage area networks (SAN), and network attached storage (NAS).

Storage

This section deals with equipment used to implement pure data network technologies, i.e., those networks that connect multiple computers, peripherals, or storage devices together in a single physical location. Networking equipment is usually designed around the needs of corporate IT networks, though SMB/home/consumer networking has gained significant traction in the past few years as well. Key networking technologies include: Ethernet: The primary wired data networking technology for computers and servers, Ethernet is a fast, scalable, cost-effective technology used to connect PCs in a corporate or home network. Wireless LAN (802.11): Wireless LAN technology mimics a wired LAN but replaces the Ethernet cable with radio wavesallowing for the networking of PCs at distances up to 300 feet without wires. It is usually used in conjunction with a LAN or broadband Internet connection to extend the wired network. Wireless LAN technology uses unlicensed spectrum, primarily 2.4 GHz and 5 GHz. Bluetooth: A personal area networking technology, Bluetooth is a fast, cheap, low power wireless protocol used to connect two devices in close proximity. Bluetooth is particularly prevalent in the handset market. Storage: Enterprise storage, while primarily implemented via direct-attached hard drives today, is increasingly moving to Fibre Channel-based storage area network and network attached storage systems, mimicking an Ethernet network and providing increased efficiency, manageability, scalability and upgradeability.

167

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Ethernet Ethernet
Ethernet is the most popular protocol used in local area networks. PCs, servers, and printers equipped with Ethernet ports connect to switches that direct traffic around the network. Data packets bound for locations outside the LAN are addressed and forwarded by routers.

LAN Switch
Network Printer with Ethernet Port
Fast/Gigabit Ethernet

Hub
Fast/Gigabit Ethernet

PCs with NICs or LANon-Motherboard (LOM) Server PCs with NICs or LOM
Gigabit Ethernet

Router or Layer 3 Switch


Source: Cisco Systems, CIBC World Markets Corp.

Connection to the WAN

Ethernet is the most pervasive protocol used in local area networks. Ethernet is a packet-based broadcast technology, meaning that it sends addressed information out over the network rather than establish a dedicated line between two devices (we discuss packet switched and circuit switched networks in more detail in the datacom/telecom section). A PC that has data to send broadcasts the data, along with the recipients address, over the network. All devices attached to that network link read the address; if the packet is bound for that device it extracts the payload; otherwise that packet is ignored. In practice, most Ethernet networks use a switched architecture, meaning that switches are used to forward traffic between clients on the network; this cuts down on redundant traffic and frees up bandwidth on the network. To connect to an Ethernet network, client devices require a Network Interface Card (NIC), which provides the link to the network. Most PCs today ship with LAN-on-motherboard (LOM), which eliminates the need for a discrete NIC card. Each client connects via Category 5 cable up to an Ethernet switch (fiber can also be used, though this is less common). As the client sends packets out, the switch reads the address and determines which of its ports the recipient sits on and broadcasts the packet out on that port, where it is received and opened by the recipient client. Switches in the LAN are generally connected to one another via uplink ports (an uplink port is the same as a standard Ethernet port but is sometimes a higher speed port) to form a LAN backbone. This LAN backbone allows communication between clients throughout the LAN and also interfaces with a router to send and receive data to and from the Internet (or to other corporate LANs over a WAN).

168

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Ethernet Ethernet
Types of Ethernet Equipment
NIC (Network Interface Card) - Add-in card installed in PCs, servers, and other Ethernet clients to connect the device to the network. Most client devices have migrated to LAN-on-motherboard solutions. Hub - Simple LAN device that connects multiple clients and forwards data traffic without much protocol processing. Bridge - Simple LAN interconnect device used to connect two LANs. Layer 2 Switch - The basic LAN building block, layer 2 switches connect multiple LAN segments and provide packet forwarding between input and output ports through a switch fabric in order to reduce network congestion. Router - Intelligent packet forwarding device that operates at layer 3 of the OSI stack. Routers are placed strategically within a network and communicate with each other to determine the optimal route through the network for packet forwarding. Routers also perform protocol translation. Layer 3 Switch - Adds routing capability to the basic Ethernet switch, eliminating the need for packets to be routed centrally. Layer 4-7 Switch - Performs packet forwarding and routing based on deeper data packet information (as per layer 4 of the OSI model), enabling multiple classes of service (CoS) and better quality of service (QoS). It can also do load balancing and intelligent provisioning.

Types of Ethernet Equipment Network Interface Card (NIC): Network interface cards are add-in cards that provide network terminals with an interface to the LAN. NICs operate at the lowest two layers of the OSI reference model. NICs are responsible for segmenting and framing the user data stream (data link layer functions) and for signal shaping and conditioning (physical layer functions) required to transmit data over the LAN medium. IEEE compliant NICs ship with a predefined, hard-coded MAC address that uniquely identifies the terminal device from all others connected to the LAN. LAN-on-Motherboard (LOM): Most PCs now ship with a LAN-on-Motherboard, which essentially captures the entire functionality of a NIC card on a single chip down on the PC motherboard. Hub: Hubs reflect the trend in LAN network topologies away from bus architectures toward star architectures. Hubs connect multiple terminal devices, each via a dedicated cable, and provide a connection to other hubs, switches or routers. Hubs operate at the data link layer and the physical layer and simply forward data packets based on the destination MAC address. Hubs do not perform significant packet processing and do not make routing decisions. Hubs provide congestion control in LANs. Unlike a bus topology, in which all devices on a LAN segment are physically connected, the star topology connects terminal devices to the hub only. This way, when a terminal device has data to transmit, it does not prevent other devices from transmitting simultaneously. In addition, unlike the bus topology in

169

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

which every packet is examined by each device on the network, the hub forwards the packet only to the intended recipient. Hubs also serve as a point of data aggregation in the LAN. Bridge: Bridges are simple LAN interconnection devices that connect two bus-based LANs running the same protocol. Like hubs, bridges operate at the lowest two layers of the OSI reference model (the data link layer and the physical layer) and forward traffic from one LAN segment to another based on the MAC address of the data packet. They do not perform any significant processing of data packets, nor do they make any routing (layer three) decisions. They simply extend the physical size of a LAN. Switch: Switches serve to interconnect multiple LAN segments and/or terminal devices. Like hubs and bridges, switches operate at the data link layer and physical layer of the OSI reference model and provide packet forwarding based on destination MAC addresses. Switches differ from bridges and hubs by offering packet forwarding between multiple input ports and multiple output ports through a central switch fabric. They therefore prevent network congestion on LAN segments that do not include either the originating or receiving device. Furthermore, switches allow multiple simultaneous communications across the fabric on a point-to-point basis. Switches may support simultaneous communications over a single, high speed physical link through time division multiplexing. Router: Routers are intelligent packet forwarding devices that operate at the three network dependent layers (physical, data link and network layers) of the OSI reference model and serve two basic functions: determining optimal routes or paths through the network and packet forwarding. Routers typically serve to interconnect LAN segments as well as to interconnect LAN segments to the WAN. As such, routers typically support multiple data networking protocols, such as T1/E1, T3/E3, Ethernet, Token-Ring, FDDI, Frame Relay, ATM and SONET/SDH. Routers, especially high speed routers, perform the packet routing in hardware in order to provide the greatest possible performance. Unlike switches that view the network on a link-to-link basis, routers view the network on a global basis and have knowledge of neighboring routers. Routers dynamically maintain network topology data, including neighboring routers, route distance, route delay and route congestion levels, and update this information as network conditions change. In addition, routers are self-learning devices. When introduced into a network, a router exchanges information as to network topology with neighboring routers. In this process, the newly introduced router learns the topology of the network from its neighbors, while simultaneously allowing its neighbors to update their view of the network to reflect the existence of the new router. When making packet forwarding decisions, routers depend on this accumulated network intelligence. Routing decisions may be based on such factors as destination address, payload type, packet priority level (contained in the IP header), least-cost route, route delay and route congestion levels. Layer 3 Switch: As discussed, switching typically takes place at layer 2 of the OSI reference model. Layer 3 switching is the switching of packets through a network based on decisions made by edge routers and carried out by switches. The network is divided into sub-segments, with each sub-segment given a unique tag or label. When a packet reaches an edge router, the router identifies the destination address, selects the appropriate output port from its routing table and prepends a tag or label to the packet header. The edge router then forwards the packet to a layer 3 switch and the packet passes between layer 3 switches that forward the packet based solely on the prepended tag or label. By making intelligent routing decisions at the edge of the network and performing only switching in the network core, the number of network routers may be reduced, thereby limiting the number of router peers with which each router must interact. Layer 4-7 Switch: Layer 4-7 switching operates at the transport layer (layer 4) and above of the OSI reference model. Layer 4-7 switching goes beyond IP to transport layer protocols such as TCP or UDP to make packet forwarding decisions. Layer 4-7 switches are able to look deep into data packets to extract information regarding the packet content and the application protocols exchanging the data packet. Using this information, together with details about network congestion, delay and loss, the layer 4-7 switch makes packet forwarding decisions. With the ability to differentiate classes of traffic, layer 4-7 switches have the ability to prioritize those flows with more strict requirements, thereby meeting quality of service demands. They also perform load balancing functions, keeping congestion down to manageable levels.

170

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Ethernet Ethernet
Strong corporate profits and a focus on productivity drove strong IT spending through the 1990s, pushing corporate PC LAN penetration to over 90%. In 2001, switch port faltered as IT budgets compressed with the economic downturn. The market stabilized in 2002, and has grown nicely from 2003 to 2005 on stronger economic activity. Continued growth and healthy upgrades of corporate LANs should combine with rising consumer usage to drive growth going forward.
Switch Port Shipments
250,000 LAN ATM Shared FDDI 200,000 Token Ring Shared Ethernet Switched Ethernet
Ports in Thousands

PC LAN Penetration
100% 90% 80% 70% 60% 50%

LAN penetration in business, government, and education LAN penetration in home and small office

150,000

100,000

40% 30%

50,000

20% 10%

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

0% 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Source: DellOro, CIBC World Markets Corp.

Source: DellOro, CIBC World Markets Corp.

After being introduced in the early 1990s, Ethernet equipment enjoyed a steady ramp as corporate IT departments began building LANs to connect their PCs. From under 10 million switch/hub ports shipped in 1993, the market grew to more than 160 million port shipments in 2000 (larger than the PC market, actually). The economic boom of the 1990s coupled with the growth in productivity and the increasing use of e-commerce and the Internet quickly drove corporate PC LAN penetration above 90%. With Ethernet ubiquity in the enterprise basically achieved by 2000, the economic downturn in 2001 and the compressed IT budgets that accompanied it caused the industrys first decline in port shipments. The market then stabilized in 2002 and experienced healthy growth from 2003 to 2005 as business spending rebounded with stronger economic activity. On the consumer and SMB, or small to medium-sized business, side, LAN usage has grown steadily but more slowly. Lower-cost switch and router equipment and the increasing penetration of broadband are accelerators in this market, though we doubt LAN usage will ever approach similar levels to the corporate environment. Looking ahead, growth in the Ethernet market will be dependent on continued growth in the enterprise as well as upgrades of corporate PCs (layer 3 switching, power-over-Ethernet, etc.), boosted by continued penetration of Ethernet in the home and small business environment as well as the consumer space.

171

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Ethernet Ethernet
Ethernet Port Forecast
Total Ethernet ports are expected to grow at a 10% rate, from 549 million ports in 2005 to 866 million ports in 2010. Ethernet connectivity is ubiquitous in PCs and penetration has plateued. Switch ports are still growing from both new LAN buildouts and healthy upgrades; networking OEMs continue to pack better performance and more advanced security and manageability to keep these going. Networking CPE should remain >15% of the market; these include printers, game consoles, set-tops, modems and other PC peripheral and consumer devices.
1,000,000 900,000 800,000 700,000
Unit/Port Shipments (000s)

Ethernet Equipment Suppliers


Ethernet Switches
Foundry Alcatel 1% 1% Enterasys 1% Extreme 2% Allied Telesyn 3% NetGear 3% D-Link 4%

Others 11 %

Nortel 4% 3Com 8% HP 8%

Cisco 53 %

Infrastructure Networking CPE Switching NIC/LOM

600,000 500,000 400,000 300,000 200,000 100,000 2003 2004 2005 2006 2007 2008 2009 2010

PCs
Dell HP Lenovo Acer Fujitsu/F-S Toshiba

Servers
HP Dell IBM Fujitsu/F-S NEC Sun

Source: DellOro, IDC, Dataquest, CIBC World Markets

Source: DellOro, IDC, CIBC World Markets Corp.

Ethernet Port Forecast: Key unit opportunities in the Ethernet market include PCs, switches, networking CPE, and infrastructure equipment. PC ports are usually implemented either with discrete NICs (network interface cards) or LOM (LAN-on-motherboard) solutions. Switches include the traditional LAN infrastructure, from the smallest SOHO switches to the most advanced layer 4-7 switches. Networking CPE includes devices such as broadband modems, settop boxes, game consoles and PC peripherals that incorporate Ethernet connectivity. Infrastructure includes a variety of high-end computing, communications, and storage hardware found in large IT and telecom networks. PC NIC and LOM ports are expected to grow roughly in line with PCs, growing from 221 million ports in 2005 to 363 million ports in 2010, a 10% CAGR. Nearly all PCs ship with Ethernet capability today; servers use a mix of NIC and LOM solutions and often carry more than one Ethernet port for additional bandwidth as well as for redundancy. The second major opportunity is Ethernet switches, which include simple layer 2 switches (simpler devices that switch and forward packets), layer 3 switches (add routing functionality) and layer 4 and above switches (switches with higher layer intelligence, for load balancing and other management capabilities). Looking ahead, switch ports are expected to grow 8%, from 224 million ports in 2005 to 324 million ports in 2010. A shift toward higher layer switches should continue; in 2010, estimated switch ports are split 51% layer 2 and 49% layer 3 and above. Note that although we forecast them separately, the line between layer 2 and 3 switches is blurring.

172

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

The third unit opportunity is networked CPE devices that incorporate Ethernet functionality (boadband modems, printers, game consoles, set-top boxes, VoIP phones and other PC peripheral and consumer devices). We expect this segment to grow from 103 million ports in 2005 to 158 million ports in 2010, a 9% CAGR. The fourth unit opportunity is infrastructure equipment, which includes blade servers, storage networking devices (iSCSI), broadband infrastructure (DSLAMs, CMTS, etc.), and ATCA equipment. We expect this segment to grow from 1.2 million ports in 2005 to 21.4 million ports in 2010, a 66% CAGR. In terms of speed/protocol, the majority of Ethernet traffic today uses 100 Mbps Fast Ethernet links, though the market is ramping quickly to Gigabit Ethernet. In 2005, we estimate the FastE/GigE split was 64%/36%. By 2010, the split should be 18%/80% with the balance shifting to 10 GigE (for infrastructure and high end switching market). By device type, 35% of NIC/LOM ports, 76% of switch ports, and 97% of networking CPE ports were at FastE in 2005. GigE totaled 65% of NIC ports, 24% of switch ports, 3% of networking CPE ports, and 98% of infrastructure ports in 2005. By 2010, GigE should represent 90%-100% of NIC/LOM and switch ports and 68% of infrastructure ports (the rest being 10 GigE). Note that Fast Ethernet should continue to dominate networking CPE ports. The exhibits below display the mix of switch ports by functionality, as well as by speed for switches, NIC/LOM and other networking CPE and infrastructure applications. Ethernet Switch Ports By Layer Served
100% 90% 80% 70% 60% 50% 40% 30% 20% Layer 3 Switches 10% 0% 2003 Layer 2 Switches
10% 0% 2003 100% 90% 80% 70% 60% 50% 40% 30% 20% 10 Gigabit Ethernet Gigabit Ethernet Fast Ethernet

Ethernet Switch Ports By Speed

2004

2005

2006

2007

2008

2009

2010

2004

2005

2006

2007

2008

2009

2010

NIC/LOM Ports by Speed


100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2003 10 Gigabit Ethernet Gigabit Ethernet Fast Ethernet

Networking CPE and Infrastructure Ports by Speed


100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2003 10 Gigabit Ethernet Gigabit Ethernet Fast Ethernet

2004

2005

2006

2007

2008

2009

2010

2004

2005

2006

2007

2008

2009

2010

Source: IDC, CIBC World Markets Corp.

Ethernet Equipment Suppliers: Cisco was the dominant supplier of Ethernet switches in 2005 with 53% port share. HP and 3Com followed with 8% share each. All other suppliers of Ethernet equipment each held less than 5% share; these include Nortel, D-Link, NetGear, Allied Telesyn, Extreme, Enterasys, Foundry, and Alcatel. PC and server vendors (for NIC/LOM solutions) include Dell, HP, Lenovo, Acer, Fujitsu/F-S, Toshiba, NEC, and Sun. Vendors of networked CPE devices include manufacturers of modems, printers, game consoles, set-top boxes, and a variety of other PC peripherals and consumer electronics devices.

173

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Ethernet Ethernet
The microprocessor and core logic control overall switch functioning A switch fabric chipset switches data packets between ports Usually found in layer 3 switches, aids in the routing process by quickly looking up network addresses

Microprocessor
Memory used by the processor

Core Logic

Switch Fabric Chipset


MAC MAC MAC MAC

Look-up Engine/CAM (Layer 3 Switch) SRAM

Ethernet Switch

SRAM

Flash

SRAM memory is used by the switch fabric and the look up engine Performs Ethernet protocol processing, removing and applying Ethernet overhead

Ethernet PHY RJ-45 Ethernet Cable

Ethernet PHY RJ-45 Ethernet Cable

Ethernet PHY RJ-45 Ethernet Cable

Ethernet PHY RJ-45 Ethernet Cable

Interfaces with the Ethernet cable, performing transmission and reception function and converting the signal between an analog serial stream and a digital parallel stream, and vice versa

NIC/LOM

RJ-45 Ethernet Cable


Interfaces between the host system and the Ethernet cable, performing physical layer (a/d conversion) and MAC layer (protocol processing) functions

Ethernet PHY MAC SRAM


PCI

An embedded PCI or PCIExpress controller provides the interface with the host system

Host System
SRAM memory is used by the MAC

Source: IDC, CIBC World Markets Corp.

Ethernet Port (NIC and Switch) Ethernet PHY: Sometimes called a transceiver, the PHY interfaces with the RJ-45 Ethernet cable (or optical fiber) and converts the serial analog stream to a digital parallel stream and vice versa. In a switch, these are implemented as multi-port (quad, octal, etc.) PHYs. In a NIC/LOM, they are part of the Ethernet controller. Ethernet MAC, SRAM: The MAC assigns and removes Ethernet headers and overhead. In a switch, these are usually integrated into the switch fabric chipset. In a NIC/LOM, they are part of the Ethernet controller. MACs typically use a small quantity of SRAM memory.

Switching Block (Switch Only) Switch Fabric: The switch fabric receives packets from the various Ethernet ports and switches them between ports or sends them to the uplink port (if the destination is a port on another switch or the WAN or Internet). Lookup Engine/CAM: Usually found in layer 3 switches, a lookup engine aids in routing by quickly finding addresses. A specialty SRAM known as a CAM (content addressable memory) is used to store the route table. Microprocessor, Core Logic, SRAM, Flash: This blockwhich contains a microprocessor, one or more coprocessors, core logic, FPGAs, SRAM and flashcontrols the overall system and performs high-end functions.

174

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Ethernet Ethernet
Ethernet Semiconductor Forecast
We expect the market to grow at a 6% CAGR, from $2.1 billion in 2005 to $2.8 billion in 2010. We expect Gigabit Ethernet to grow from $1.3 billion in 2005 to $2.1 billion in 2010, an 11% CAGR. 10/100 Mbps Ethernet is expected to decline at a 29% CAGR, from $867 million in 2005 to $157 million in 2010. 10 Gigabit Ethernet should be used in high-end switches and servers; we expect it to grow to over $550 million in 2010.
$3,000 10 Gigabit Ethernet $2,500 Gigabit Ethernet Fast Ethernet
Semiconductor Revenue ($M)

Ethernet Semiconductor Competitors


Vitesse 1% Standard Micro 2% IBM Micro 4% Others 5%

Agere 5% Avago 5% LSI 6% Realtek 7%

Broadcom 34 %

Intel 12 %

Marvell 18 %

Source: IDC, CIBC World Markets Corp.

$2,000

$1,500

PC LAN PHY & Switch Controllers ASSPs (PHY + MAC)


Broadcom Intel Realtek Marvell Agere Vitesse Broadcom Marvell Avago Realtek Vitesse Intel Agere

Switch ASICs
LSI Logic IBM Agere Avago TI Fujitsu

$1,000

$500

$0 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, CIBC World Markets Corp.

Note: Our Ethernet semiconductor forecast includes all silicon used in PC LAN connections as well as in other client applications. On the infrastructure side, we include all Ethernet-specific devices, including PHYs, MACs, and switch fabric ICs. Our forecast excludes network processing, CAM, and control plane devices that are used in high-end switches and switch/routers. These devices are included in our communications infrastructure semiconductor forecast, found in the telecom/datacom section. Ethernet Semiconductor Forecast: The market for Ethernet ICs is expected to grow at a 6% CAGR, rising from $2.1 billion in 2005 to $2.8 billion in 2010. Revenue growth should trail units due to aggressive ASP declines, though this will be offset partially by 1) the mix shift toward Gigabit, 2) faster growth rates for higher ASP infrastructure ports and 3) growth of layer 3+ switch solutions. Growth is expected to be led by Gigabit Ethernet, expected to rise from $1.3 billion in 2005 to $2.1 billion in 2010, an 11% CAGR. 10 Gigabit Ethernet, initially being used in LAN backbones and high-end servers, is expected to grow from modest revenue in 2005 to $554 million in 2010. 100 Mbps Fast Ethernet is expected to decline at a 29% rate, from $867 million in 2005 to $157 million in 2010. By device, about 29% of Ethernet IC revenue in 2005 was derived from PC clients (NIC/LOM solutions). 65% of revenue was derived from switch applications, split roughly 75/25 in favor of switch ASICs/ASSPs vs. PHY chips. 6% was derived from networking CPE applications and 1% from infrastructure. We expect this mix to change just slightly

175

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

over the forecast period. Switching is expected to grow at a 4% rate through 2010, followed by NIC/LOM at 3%; both segments should see aggressive ASP declines, though this should be offset modestly by the transition to GigE and growth of layer 3 switching. Networking CPE devices should see the strongest growth at an 8% CAGR as penetration of Ethernet technology in these devices increases, outweighing price declines that will accelerate as Ethernet hits the consumer electronics market. Infrastructure Ethernet ICs should grow from a small base to 7% of total in 2010. The exhibits below display Ethernet IC revenue mix by speed and device. Ethernet IC Revenue Mix By Speed
100% 100%

Ethernet IC Revenue Mix By Device

80%

80%

60%

60%

40%

40%

10 Gigabit Ethernet 20% Gigabit Ethernet Fast Ethernet 0% 2003 0% 2003 20% Infrastructure Networking CPE Switch/Infrastructure ASSP/ASIC Switch/Infrastructure PHY NIC/LOM 2004 2005 2006 2007 2008 2009 2010

2004

2005

2006

2007

2008

2009

2010

Source: IDC, CIBC World Markets Corp.

Ethernet Semiconductor Competitors: The Ethernet semiconductor market has grown more competitive as LANs have become ubiquitous in corporate PCs. In 2005, the top vendors were Broadcom with 34% share, Marvell with 18% and Intel with 12%. Note that Broadcom and Marvell each took share due to their strong GigE portfolios in both NIC/LOM and switches. Following the top three were Realtek and LSI, with 7% and 6% share, respectively, followed by Avago and Agere each with 5%. Other major vendors included IBM, Standard Micro, and Vitesse. In terms of devices, the PC controller space is served by Intel, Broadcom, Marvell, and Realtek. Note that Intel and Marvell have a joint solution for GigE controllers; Marvell integrates Intels MAC with its PHY, which Intel resells; Marvell and Intel also offer their own controllers to the market independent of the partnership products. On the switch side, Broadcom and Marvell dominate the mainstream and performance GigE PHY and switch ASSP markets, while Vitesse has carved out a nice niche at the low end for SMB applications. Realtek and Intel lead in Fast Ethernet, though Broadcom and Marvell also have a presence there. Agilent and Agere also offer ASSP solutions, though they have failed to gain significant traction. A great percentage of switches still use custom ASICsthe best example is Cisco (over 50% of the market in port shipments), which uses mostly custom ASICs. The leading switch ASIC vendors (to Cisco and others) include LSI Logic, IBM, Agere, Agilent, Texas Instruments, and Fujitsu.

176

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Ethernet Ethernet
Ethernet Semiconductor Market Trends
Gigabit Ethernet has moved from the corporate desktop to the mainstream consumer PC; the Gigabit premium is now below $3 for a PC controller and heading to negligible levels. Chipset vendors may begin pulling the MAC into the PC chipset, leaving only a PHY external. High end NICs are coming to market incorporating RDMA, iSCSI, TOE engines, and network management capability. 10 Gigabit merchant ICs are starting to hit the market. Gigabit Ethernet switching continues to ramp aggressively from top to bottom. While ASICs continue to play a significant role in switch designs, all but the largest switch vendors are moving to ASSP solutions. GigE switch designs have consolidated around Broadcom and Marvell, though Vitesse is now established at the low end. Switches based on integrated switch/MAC/PHY SoCs are beginning to ramp for the low end. Multi-port (4-, 8-port) PHYs are enabling high-density (24-, 48-, 72-port) Gigabit switches.

Gigabit Ethernet has moved from the corporate desktop to the mainstream consumer PC. 2005 was a big year for Gigabit Ethernet in PC clients, with most PCs shipping moving to Gigabit as the price premium came down. In total, we estimate that nearly 65% of NIC/LOM ports shipped with Gigabit in 2005 (vs. 35% in 2004). We expect this trend continued in 2006 and expect nearly 85% of PCs to ship with Gigabit Ethernet. By 2007, this percentage should increase to over 95% according to our forecast. Chipset vendors may begin pulling the MAC into the PC chipset. This would allow a PC OEM to implement the GigE port with just an external PHY, lowering the total bill of materials. We would expect this architecture to start as a low end solution, with most higher-end PC platforms opting for the traditional MAC+PHY controller. High end NICs are coming to market incorporating RDMA, iSCSI, TOE engines, and network management capability. With Gigabit Ethernet now fully penetrated in the server market (and 10 Gigabit ramping in some cases), NIC/LOM suppliers are looking to add increased functionality to enable lower latency, more efficient data throughput, and greater management features. Broadcoms C-NIC is a prime example; it incorporates RDMA technology and a TOE engine for lower latency and better performance, iSCSI for storage, and network management features to give IT managers better control over their server client devices. 10 Gigabit merchant ICs are starting to hit the market. A number of credible start-ups are now shipping merchant PMD and PHY devices for 10 Gigabit over fiber, while others have focused on the controller (most of

177

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

these are shipping full NIC card solutions to enable the ecosystem). Others are pursuing 10 Gigabit copper PHYs, while a third group attacks switching. We would expect these products to see meaningful design-in traction in 2007 with more volume in 2008 and beyond. Gigabit Ethernet switching continues to ramp aggressively from top to bottom. Originally a high-end technology with a hefty premium, Gigabit Ethernet is now showing up across the entire switch market, from high end modular switches to 8-port switches for SMB. In 2005, at least eight switch vendors shipped more than one million ports of Gigabit Ethernet. Total ports increased 70% year-over-year and topped 52 million, or about 24% of total switch port shipments and 51% of switch revenue. With the Gigabit premium continuing to come down, we expect the market to accelerate toward our 2010 estimate of 90% port shipment penetration. While ASICs continue to play a significant role in switch designs, all but the largest switch vendors are moving to ASSP solutions. ASIC suppliers continue to dominate the landscape for Ethernet switch fabric ICs. The best example is Cisco (over 50% of the market in port shipments), which jointly designs its fabric ICs with a multitude of suppliers (Cisco itself has more than 250 ASIC engineers). As the market moves to Gigabit, we expect only the largest switch vendors to retain the same level of ASIC investment; most should move to ASSP solutions. To meet this demand, the major ASSP vendors have been introducing more advanced switch fabric chipsets, with integrated MACs, layer 3-7 functionality (for routing support) and built-in support for uplink ports. Next-generation designs are even incorporating network search engines for more robust routing. Note that the maturity level of these ASSP solutions is also incubating the ODM market, with vendors like Accton, Delta, Alpha, BATM, Runtop, Cameo and others. GigE switch designs have consolidated around Broadcom and Marvell, though Vitesse is now established at the low end. While designs using Gigabit Ethernet ASSPs are still in the early part of their product life cycles, design wins to date point to a consolidation around two vendors: Broadcom and Marvell. These vendors have mature PHY and switch fabric IC solutions and strong customer relationships. Vitesse has made some significant progress at the low end, specifically with unmanaged layer 2 switches and CPE devices, and is also becoming a significant supplier to this market. Other players such as Agere, which acquired both a switching ASSP start-up and a PHY start-up to add to its ASIC capability, and SwitchCore, a start-up based in Sweden, have been less successful at penetrating this market. Switches based on integrated switch/MAC/PHY SoCs are beginning to ramp for the low end. Already shipping are 8-port switches based on single SoCs incorporating switch, MAC, and PHY functionality, and 24-port designs should ramp soon. Much of Vitesses success on the low end is due to its aggressive roadmap with singlechip switches. Note that these switches are targeted at the low end; we do not expect the mainstream market to migrate to integrated SoCs for some time. Multi-port Gigabit PHYs are enabling the high-density Gigabit switch market. The earliest Gigabit Ethernet deployments on the switch side were uplink ports, requiring only one or two per switch. In order to enable Gigabit throughout the LAN, higher density switches (those with 24 or 48 ports) were required, which are enabled by higher density multi-port PHYs. Both Broadcom and Marvell are shipping 4- and 8-port PHY devices and others are ramping multi-port devices now as well.

178

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Wireless LAN Wireless LAN


Wireless LANs use the 802.11 protocol to send data traffic over unlicensed 2.4 GHz and 5 GHz radio waves, mimicking Ethernet but with a wireless connection. Wireless LAN has seen runaway success in home networking, specifically for distributing a broadband Internet connection wirelessly around the home. The technology got a big boost from Intels Centrino program, which has driven notebook penetration to north of 80%.
Wireless LAN in a Home Application
Wireless NIC
1 802.1

Fast Ethernet

DSL/Cable Connection

Wireless Access Point


Source: CIBC World Markets Corp.

DSL/Cable Modem

Wireless LANs connect multiple PCs (or other connected devices) in a single location and provide the ability to share files, hardware and resources in a fashion similar to traditional wireline Ethernet LANs. The distinction, of course, is that no physical wiring is needed to connect any single client PC to its neighbor (or the server, or printers, or PDAs etc.). Instead, wireless LAN encodes data using the 802.11 protocol and broadcasts it over unlicensed spectrum. Although most networking technologies to date have been geared to corporate networks, wireless LAN offers the first true connectivity solution for the home. In fact, it is the home networking market that has really driven wireless LAN sales to date, boosted by Intels Centrino program, which incentivized OEMs to include WLAN in notebooks, eventually driving penetration notebook WLAN penetration north of 80%. In the home setting, wireless LANs act as the primary method of sharing a broadband Internet connection. Access points designed for the home interface directly with a cable or DSL modem (using an Ethernet connection) and distribute the bandwidth to wireless-enabled PCs. Many industry participants believe that wireless LAN will eventually evolve into a home media distribution system, used to send video, voice and data around the home wirelessly. Under this topology, modems, PCs, set-top boxes, game consoles, and VoIP-enabled phones would all be networked together wirelessly, with the ability to share content and an Internet connection between them. Voice service could easily be distributed wirelessly to WLAN-enabled VoIP phones. Video could be distributed from wireless-enabled set-tops and digital TVs. Game consoles could access the Internet (this is already common) as well as the content stored in the PC. Consumer portables and data-enabled cellular handsets could easily hook up to the network and access a variety of these devices.

179

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Wireless LAN Wireless LAN


In the corporate environment, 802.11 can be used to extend the LAN without wires, allowing for greater mobility among users. We have seen slow but steady uptake as security, bandwidth, handoff, and channelization concerns have been addressed.
Wireless LAN in a Corporate Application
802.1 1

Wireless Access Point


Fast/Gigabit Ethernet

Wireless NIC

Fast/Gigabit Ethernet

LAN Switch
Server PCs with NICs or LOM
Gigabit Ethernet

Router or Layer 3 Switch


Source: Cisco, CIBC World Markets Corp.

Connection to the WAN

In the corporate setting, a WLAN is meant to augment the connection from user to user currently accomplished with NICs and Ethernet cable. It is not designed to replace the traditional LAN switch and backbone architecture, as it requires this equipment to switch and route data and to provide the interface to access and edge routers and out to the WAN. Rather, 802.11 extends the LAN without wires. In a typical corporate setting, wireless-enabled PCs access specialized transmit/receive access points distributed at regular geographic intervals throughout the building or campus. Access points act both as transmit/receive bases and as a connection between the wireless and wired LAN. Wireless LAN technology presents even greater utility when used in a campus setting. Wireless LANs provide true mobility throughout the campus and can be used to provide intranet and Internet access to notebook computers, PDAs and other handheld devices across the campus with no need to connect or configure equipment. They are especially popular with campuses that have inherently mobile users, such as universities and airports. In terms of adoption, enterprise wireless LAN has been slow to gain steam, owing in large part to IT concerns over issues like security, bandwidth, handoff, and channelization. The major enterprise networking OEMs, however, have made a lot of progress addressing these issues. Tougher security standards have been implemented directly into the 802.11 standard, and computing and networking software have additional safeguards as well. 802.11n offers more bandwidth and channelization. Switch management software and hardware solutions are now coming to market to support dynamic roaming between access points. Power-over-Ethernet is enabling easier access point deployments. Some challenges still remain, but we are seeing steady progress in corporate WLAN.

180

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Wireless LAN Wireless LAN


Wireless LAN Protocols 802.11b - The first wireless LAN standard to take off, 802.11b runs at 11 Mbps and operates in the 2.4 GHz band. 802.11a 802.11a runs at 54 Mbps and operates in the 5 GHz band, and offers more channels than 802.11b. It was never a success as a standalone protocol, but is still included in multi-band (a/b, a/g) solutions. 802.11g - 802.11g adds additional modulation schemes to 802.11b, increasing speeds to 54 Mbps while remaining in the 2.4 GHz band. Backward compatible with 802.11b. 802.11n A next generation standard, expected to boost throughput to 108 Mbps or higher. 802.11n uses MIMO technology in both the 2.4 and 5 GHz bands.

Although most wireless LAN technologies use the same topology, they can vary widely in terms of standard, each with its own frequency band, data transfer rate and modulation scheme. A major hurdle holding up volume wireless LAN deployments was the lack of a clear standard for the equipment makers to center around. Until about five years ago, two key standards, IEEE 802.11 and HomeRF, were competing for OEM adoption. However, aggressive actions by OEMs and chip manufacturers, together with the formation of the Wireless Ethernet Compatibility Alliance (WECA), helped the industry settle on 802.11 as the dominant standard. Within 802.11, the first sub-standard to take off was 802.11b, an 11 Mbps technology that operates in the 2.4 GHz band (the same band used by other consumer devices such as cordless telephones, microwave ovens, and Bluetooth devices). This band is free from government regulation, allowing users to set up wireless LANs without license or approval. The FCC has defined three spectrum ranges, known as ISM bands, for use by unlicensed continuous operation up to 1 Watt: industrial (902-928 MHz), scientific (2.4-2.4835 GHz) and medical (5.15-5.35 GHz and 5.725-5.825 GHz). 802.11b was the first wireless LAN standard, operating in the 2.4 GHz ISM band, though it was quickly replaced with the faster 802.11g. 802.11g uses the same frequencyand thus remains backward compatible with 802.11b productsbut supports data rates of up to 54 Mbps. Another sub-standard, 802.11a, offers data transfer rates up to 54 Mbps using the less crowded 5 GHz band. 802.11a received a lot of attention but never much deployment as a standalone technology. The primary advantage of 802.11a, which offers similar data transfer rates as 802.11g, is its higher channel density that provides better

181

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

average user throughput in congested networks. However, because of its 5 GHz frequency, 802.11a suffers from narrower range and greater power consumption than 802.11g. It is also not compatible with the large installed base of 802.11b and 802.11g. As a result, 802.11a deployment has been included as a second band in multi-band implementations (a/b, a/g solutions), but as mentioned above, very little 802.11a-only equipment ever shipped. The newest WLAN standard is 802.11n (not yet finalized), which uses both the 2.4 GHz and 5 GHz bands and leverages MIMO (multiple input multiple output) radio technology to boost bandwidth to 108 Mbps or higher. 802.11n is expected to be an effective solution for both the home and enterprise. The first products, which are being sold as pre-n or draft-n solutions, came to market in 2006. The standard should be ratified some time in late 2007 or early 2008. The exhibit below displays WLAN standards and working groups.

IEEE Standards 802.11a The leading standard at 5 GHz, 802.11a operates at speeds up to 54 Mbps. The standard employs a more advanced modulation scheme, called orthogonal frequency division multiplex (OFDM), than 802.11b that allows data to be encoded on multiple radio frequencies concurrently. Products based on 802.11a currently deliver a much shorter range than 802.11b/g because of the higher frequency band, though there is less interference as this band is less crowded. 802.11b The first standard to gain wideband adoption, 802.11b operates in the 2.4 GHz band at speeds up to 11 Mbps. The standard uses one of several Direct Sequence Spread Spectrum (DSSS) modulation schemes, including Complementary Code Keying (CCK). 802.11b products typically have a range of 200-300 feet. 802.11d 802.11d enables mobile client access between additional countries. 802.11e The 802.11e standard adds a MAC enhancement for quality of service (QoS) features and multimedia support that are critical for voice, video and audio delivery over wireless LAN. This standard is seen as a critical enhancement toward wide-scale usage in the networked home. 802.11f enables multi-vendor access point-to-access-point communication for DSSS-based equipment. 802.11g operates at 2.4 GHz, but uses the OFDM modulation scheme for speeds above 11 Mbps to boost data rates. 802.11gbased equipment runs at data rates of up to 54 Mbps, making it competitive with 802.11a. A key advantage of 802.11g is that it is backward compatible with 802.11b; thus, 802.11b NICs will operate with 802.11g access points, though only at 11 Mbps and vice versa. Enhancements incorporated in 802.11h automate frequency band conformance and tuning, employing dynamix frequency selection (DFS) and transmit power control (TPC). The technology enables 5 GHz compatibility in Europe and is also included in HiperLAN/2. 802.11i adds security features to improve on WEP and supports 802.1x authentication, Temporal Key Integrity Protocol (TKIP) and advanced encryption standard (AES). The next-generation 802.11 standard should support speeds of 108 Mbps or higher, and theoretically can evolve to 540 Mbps over time. 802.11n builds upon previous 802.11 standards by adding MIMO (multiple-input multiple-output). MIMO uses multiple transmitter and receiver antennas to allow for increased data throughput and range. Though pre-standard products are currently shipping, the finalized 802.11n standard is not due for final approval until 2007 or 2008. A working group that is extending the 802.11 standard to automobiles. As part of a dedicated short range communications (DSRC) system, 802.11p operates in the 5.9GHz band and deals with roaming from cell to cell in a fast-moving vehicle.

802.11f 802.11g

802.11h

802.11i 802.11n

802.11p

Source: IEEE, WECA, TechWeb Encyclopedia, company reports and CIBC World Markets Corp

182

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Wireless LAN Wireless LAN


WLAN Unit Forecast
Wireless LAN units are expected to grow at a 23% CAGR, from 165 million units in 2005 to 464 million in 2010. Mobile, consumer, and PC peripheral devices to lead growth with a 39% CAGR. PC connections should grow 23% as notebook growth continues and desktop WLAN penetration increases; after-market PC adapters should decline 16%. Access points should grow 17%, with the client/AP ratio increasing from 5.4:1 in 2005 to 7.1:1 in 2010.
500,000 450,000 400,000 350,000
Unit Shipments (000s)

WLAN Equipment Suppliers


WLAN NICs and Access Points
SMC 2%
Symbol 1% 3Com 1%

Others 10 %

Cisco 3% Buffalo 7% Belkin 7%

Linksys 31 %

NetGear 16 %

D-Link 21 %

Access points & gateways Printers/Peripherals Consumer Electronics & Other Mobile Devices After-market NIC/adapter Desktop PC embedded Notebook PC embedded

Notebook PCs
Others 19 % Dell 17 %

300,000 250,000 200,000 150,000 100,000 50,000 2002 2003 2004 2005 2006 2007 2008 2009 2010

ASUS 2% Apple 3% NEC 4% Sony 4% Fujitsu/F-S 6% Toshiba 11 % Acer 10 %

HP 16 %

Source: IDC, CIBC World Markets Corp.

Lenovo 7%

Source: DellOro, IDC, CIBC World Markets Corp.

Wireless LAN Unit Forecast: We forecast a breakout in wireless LAN unit over the next few years as the market moves beyond PCs into the realm of consumer devices and wireless handsets. Notebooks, access points, and residential gateways are still growing at fairly solid clips, but should be surpassed over the long term by these new connected devices. Consumer electronics, PC peripheral, PDA, and wireless handset OEMs are all beginning to integrate wireless LAN into their devices and plan for an increased number of models using the technology over time. Declining chipset price points and improved power consumption should bolster this trend. Overall, wireless LAN units are expected to grow at a 23% rate, from 165 million units in 2005 to 464 million units in 2010. Note that we forecast faster growth for wireless LAN clients than for access point devices. To illustrate, we expect the client to access point ratio to increase from 5.4:1 in 2005 to 7.1:1 in 2010. This reflects the growing number of client devices over and above the PC that will leverage the existing access points and gateways currently in use in the home and enterprise. There should still be healthy replacement in the home driven by newer standards (802.11n), plus continued growth in the enterprise market, but clients are still growing at a faster clip. In total, we expect wireless LAN clients to grow at a 24% CAGR, from 139 million units in 2005 to 406 million units in 2010. Access points and gateways are growing at a 17% CAGR, from 26 million units in 2005 to 57 million in 2010. In terms of application, we expect PCs and PC adapters to grow at a healthy rate, but to decline as a percentage of the market over time, from 58% of total wireless LAN clients in 2005 to 40% in 2010. Notebooks remain the bulk of these units, and should benefit from strong unit growth plus some marginal penetration growthin total we forecast a

183

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

22% CAGR through 2010, with notebook unit shipments topping 150 million. Desktops should grow at a more than 40% CAGR but should still represent less than 10% of total PC wireless LAN units in 2010. Though they will get a boost from 802.11n in 2007-2008, we expect after-market NICs and adapters to decline at a 16% CAGR, to less than 15 million units in 2010. Consumer, mobile, and PC peripheral applications should become more meaningful as 802.11 is incorporated into such devices as game consoles, wireless handsets, digital cameras, MP3 players, set-top boxes, printers, MFPs, and PDAs. We expect these devices to grow at a 39% CAGR through 2010, growing from 26% of client connections in 2005 to 48% in 2010. Note that more than half of 2010 units are forecast to be derived from the handset and converged mobile device market with its huge unit TAM, though our forecast assumes less than 10% penetration in handsets by 2010. On the access point side, we forecast a 17% CAGR as mentioned above, with the market growing from 26 million units in 2005 to 57 million in 2010. About 23% of all access points in 2005 were standard access points, 70% were residential gateways, and 7% were enterprise access points. We expect a gradual shift away from standard access points over time in favor of more advanced residential gateways in the home and enterprise access points in the corporate market. The exhibits below display unit share for NICs and access points by application. Wireless LAN NIC Unit Share By Application
100% 100%

Access Point Unit Share By Application

80%

80%

60%

60%

40% Printers/MFP Consumer Electronics Mobile Devices Desktop PC (embedded) Notebook PC (embedded) Aftermarket NIC 2003 2004 2005 2006 2007 2008 2009 2010

40%

20%

20%

Enterprise access points Residential gateways Standard access points

0% 2002

0% 2002

2003

2004

2005

2006

2007

2008

2009

2010

Source: IDC, CIBC World Markets Corp.

Wireless LAN Equipment Suppliers: The top supplier of dedicated wireless LAN equipment (NICs, access points, and gateways) in 2005 was Linksys (a subsidiary of Cisco) with 31% unit share. D-Link was next with 21% share, followed by Netgear with 16%. Belkin and Buffalo each held 7% share and Cisco (excluding Linksys) followed with 4% share. Note that Cisco specializes in enterprise access points, where it held a 73% share in 2005. Other major vendors include SMC, 3Com, and Symbol. As embedded notebook wireless LAN represents a huge chunk of all wireless LAN shipments, it is important to note the top suppliers of notebook PCs. In 2005, the top notebook suppliers were Dell with 17% share, HP with 16%, Toshiba with 11%, and Acer with 10%. Lenovo and Fujitsu followed with 7% and 6%, respectively. Other major suppliers include Sony, NEC, Apple, and ASUS; each held less than 5% of the market in 2005.

184

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Wireless LAN Wireless LAN


Controls radio wave generation/reception and converts the analog radio signal to a digital stream (and vice versa) Sits in front of the antenna and boosts signal strength Baseband - Performs the first level of processing, implementing the modulation scheme (encoding technique) MAC - Performs protocol processing, applying and removing 802.11 overhead SRAM and Flash used by the 802.11 MAC

Enterprise Access Point Only

DRAM Microprocessor/ Microcontroller Ethernet Controller RJ-45 Ethernet Cable

Antenna

Power Amp

RF/IF

Baseband/ MAC

SRAM Flash

The microprocessor controls overall access point/gateway functioning and manages network data flow and subsystems; DRAM is used by the MPU

An Ethernet controller (in an access point) or switch (in a gateway) provides the interface to the wired LAN

DRAM

Microprocessor/ Microcontroller

Ethernet Switch
Ethernet Switch MAC MAC PHY PHY

Ethernet Cable Ethernet Cable Analog Phone Lines

DSL/Cable Connection

Modem Line Driver/Tuner

DSL/Cable Modem SoC

VoIP Codec DSP Codec

SLIC SLIC

Modem SoC - implements the DSL or DOCSIS protocol Modem Front End or Tuner interfaces with the wire or coax Source: CIBC World Markets Corp.

Residential Gateway Only


VoIP DSP decodes VOIP calls Codecs convert VoIP calls to analog SLICs interface with analog telephones

Wireless LAN NICs and access points blend devices from the Ethernet and wireless handset worlds. The wireless LAN signal is received and transmitted through the RF block, which includes a power amp, RF and IF transceiver. The signal is processed by a baseband, which implements the modulation scheme, and then an 802.11 MAC removes or applies 802.11 overhead (the baseband and MAC are almost always integrated into a single digital chip). Data packets are then relayed to the host system (in a NIC/client) or to a processor (in an access point or gateway). Note that the diagram above (and discussion below) shows the core 802.11 chipset (PA, RF/IF, and baseband/MAC), as well as the additional silicon that would be required to enable an enterprise access point or a residential gateway. Though we thought it useful to show them in the block diagram, our forecast and market share discussion of wireless LAN semiconductors excludes these ICs and focuses only on wireless LAN-specific ICs, i.e., the 802.11 chipset.

RF (NIC, Access Point, and Residential Gateway) Power Amp: The power amp sits in front of the antenna and boosts signal strength. RF/IF Converter (RF): The RF front end controls the radio wave generation on the transmit side and receives the radio wave signal on the receive side, translating between low-frequency analog signals used by the system (specifically by the IF transceiver) to high-frequency RF signals used in radio communications. In newer

185

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

implementations, the RF converter and IF are integrated in a single direct conversion transceiver, as shown in the block diagram on the preceding page. I/Q Modem (IF): The IF performs all quadrature modulation of I and Q baseband signals, converting the analog radio signal to a digital stream to be processed by the baseband and vice versa. In most implementations, the IF is integrated with the RF in a single direct conversion transceiver, as shown in the block diagram above.

Baseband/MAC (NIC, Access Point, and Residential Gateway) Baseband Processor: The baseband processor performs the first level of digital processing on the signal, implementing the modulation scheme (encoding technique). The baseband and MAC are usually integrated in a digital protocol IC, as shown in the block diagram. MAC: As in wired Ethernet, the MAC applies and removes overhead and implements the 802.11 protocol. Note that some wireless LAN chipsets implement additional security, QoS, and multimedia functions on top of the base 802.11 standard; the logic that facilitates these features is integrated into the MAC. The MAC is usually integrated with the baseband in a single ASSP, as shown in the block diagram. SRAM and Flash: These memories are used by the 802.11 MAC.

Control Block and Ethernet Interface (Enterprise Access Point Only) Processor: An embedded microprocessor performs control functions in the access point along with some data path functions. As enterprise access points are often centrally controlled, the processor will often perform advanced access control, security, and management functions not found in consumer wireless LAN. DRAM: DRAM is the primary memory used by the embedded processor. Ethernet Controller: The Ethernet controller provides the interface between the wired network (through the Ethernet cable) and the access point system. As in an Ethernet NIC application, the MAC handles the Ethernet protocol and the PHY interfaces with the cable and performs signal conversion.

Gateway Block (Residential Gateway Only) Processor: An embedded microprocessor performs control functions in the access point along with some limited data path functions. In most residential gateways they are integrated into the baseband/MAC. DRAM: DRAM is the primary memory used by the embedded processor. Ethernet Switch: Though the basic wireless access point needs only one Ethernet port to interface with the PC or wired network, most access points are actually gateway devices which usually include multiple Ethernet ports. These are usually implemented with a single-chip Ethernet switch, which includes Ethernet PHY, MAC, and switch functions. Most gateways have 4-5 ports of 10/100 Ethernet, though this can vary. DSL Line Driver (used in DSL gateways): A DSL line driver interfaces with RJ-11 telephone line. DSL Data Pump/SoC (used in DSL gateways): A DSL modem SoC implements the DSL protocol; it has an integrated DSP data pump to implement the DSL protocol and an analog front end to interface with a line driver. Cable Tuner (used in cable modem gateways): A tuner interfaces with the coax and tunes the cable RF frequency for transmission and reception. DSL Data Pump/SoC (used in cable modem gateways): A cable modem SoC implements modem functions, and has an integrated front end processing (QPSK/QAM modulation for transmission and QAM demodulation for reception) and a DOCSIS MAC. VoIP DSP: A VoIP packet processor DSP performs the TDM-to-IP conversion on IP-based phone calls. Integrated codecs perform the analog-to-digital and digital-to-analog conversion. Line Driver/SLIC: This SLIC is a line driver that interfaces with the telephone line, sending and receiving the analog signals to and from analog telephones.

186

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Wireless LAN Wireless LAN


WLAN Semiconductor Forecast
We expect wireless LAN semiconductors to grow 15% through 2010, from $1.1 billion in 2005 to $2.2 billion in 2010. 802.11b revenue peaked in 2002; 802.11g likely peaked in 2006; 802.11n should ramp meaningfully through the end of the forecast period. Chip count has been reduced from 5 or 6 in 2001 to 2 chips (an RF/IF chip and a baseband/MAC) + a power amp. Consistent with units, client chipsets are growing faster than access points.
$2,500 802.11n (includes pre-N) Dual Band (802.11a + 802.11b/g) 802.11g $2,000 802.11a 802.11b
Semiconductor Revenue ($M)

WLAN Semiconductor Competitors


Airgo 3% Conexant 10 % Others 5% Intel 23 %

Texas Instruments 10 %

Marvell 11 %

Broadcom 22 % Atheros 16 %
Source: IDC, CIBC World Markets Corp.

$1,500

NIC and AP/Gateway Broadcom Atheros TI Marvell Conexant Airgo Ralink Realtek

Embedded PC Chipset Intel Atheros Broadcom Marvell

Consumer & Wireless Marvell TI Atheros NXP/Philips Conexant Broadcom STMicro

$1,000

$500

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, CIBC World Markets Corp.

Wireless LAN Semiconductor Forecast: Although a focus on mobility and the growth of broadband have brought wireless LAN to the forefront of the technology sector, it is its low cost of implementation that has driven the technology to high volume, with nearly 165 million units shipped in 2005. Almost the entire cost of a wireless LAN client implementation is in the chipset, and that semiconductor BOM has rapidly dropped from nearly $30 in 2001 to under $7 in 2005 (lower for the biggest OEMs). This has been enabled by integration, as the chip count for a wireless LAN chipset has dropped from five or six in 2000 to just two or even one (plus a power amp) today. Aggressive price declines should continue to partially offset unit growth, though we still expect a 15% CAGR for wireless LAN ICs through 2010. Overall, we expect the market to grow from $1.1 billion in 2005 to $2.2 billion in 2010. Wireless LAN chipsets are generally segmented by protocol. In 2005, 802.11b represented 5% of total chipset revenue, down from 18% in 2004 and 50% in 2003. 802.11g took most of that share, reaching 75% of total in 2005, up from 70% in 2004 and 40% in 2003. 802.11a/b/g chipsets are still a relatively small piece of the market at 16% in 2005, owing to higher chipset costs with little incremental benefit over b/g. 802.11n (including pre-N), which held 3% share in 2005, is starting to gain significant traction in 2006. Over the next few years, we expect the market to shift rapidly to 802.11n. By 2010, we expect 78% of total revenue will be from 802.11n devices. 802.11g revenue should peak in 2006 at 70% of the market and should still be a big factor for a few more years in the consumer market; eventually it should give way to 802.11n once chipset costs

187

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

come down; we forecast 802.11g to decline to about 20% of the market by 2010. 802.11a/b/g should peak in 2007 at 20% of the market and decline to just 2% of total in 2010. Consistent with our unit forecast, we expect WLAN client silicon to outgrow access point silicon. Client silicon should rise from 84% of the semiconductor TAM in 2005 to 87% in 2010, yielding a 16% revenue CAGR. Access point silicon should drop from 16% in 2005 to 13% in 2010, yielding a 10% CAGR. The exhibits below display the wireless LAN semiconductor revenue splits by client vs. access point and by device. Wireless LAN Revenue Share by Protocol
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2002 802.11n (includes pre-N) Dual Band (802.11a + 802.11b/g) 802.11g 802.11a 802.11b 2003 2004 2005 2006 2007 2008 2009 2010

Wireless LAN Revenue Share; Client vs. AP


100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2002 AP Semiconductor Client-side Semiconductor

2003

2004

2005

2006

2007

2008

2009

2010

Source: IDC, CIBC World Markets Corp.

Wireless LAN Semiconductor Competitors: The rapid growth weve seen in wireless LAN has been made possible by the large number of vendors on the supply side offering competitive solutions. In 2005, at least six vendors derived more than $100 million in revenue from this market. Two vendors combined for 55% share in 2005interestingly, neither had more than 5% share just three years earlier. The largest supplier in 2005 was Intel with 23% revenue share. Note that Intel sells its chipset only into embedded PC applications as part of the Centrino platform, and does not participate in access points, gateways, or consumer/handset wireless LAN. Broadcom followed with 22% share; note that Broadcom was only a minor player in 802.11b, and really made its mark with the transition to 802.11g. Even in 2005, the vast majority of Broadcoms sales were 802.11g, where the company held 38% of the market. Rounding out the top three was pure-play Atheros, which held 16% share in 2005. Like Broadcom, Atheros is strong in 802.11g with 17% share and is well positioned in dual band as well. Note that Atheros has been the most aggressive vendor in moving to single-chip solutions, which has allowed it to stay competitive even vs. suppliers with more diversified businesses and deeper pockets. Marvell held 11% of the market in 2005, up roughly 7 percentage points vs. 2004, owing primarily to the ramp of Sonys PSP and Microsofts Xbox 360, where Marvell is the sole supplier. Marvell has made embedded platforms a focus for the company, and in addition to its gaming wins (it also has the PS3), the company has excellent design win coverage in emerging wireless LAN-enabled handsets as well as in printers and digital cameras. Texas Instruments and Conexant each held 10% share in 2005. TI has moved away from PC applications and is instead focusing on handsets and residential gateways. Conexant has had the most interesting history in 802.11the company acquired GlobespanVirata, which had itself bought the wireless LAN business of Intersil, which was at the time the dominant player in the market with more than 60% share. Conexants WLAN business struggled initially as Centrino cratered the pricing environment for WLAN in PC applications; Conexant was also late to market with a 802.11g solution, which muted growth prospects in 2005. Note that the market for 802.11n was just starting to ramp in 2006. Broadcom, Atheros, and Marvell are early leaders in the market. Airgo, which pioneered MIMO technology for wireless LAN and for some time has been shipping to top tier OEMs, is also actively migrating its product portfolio to be compatible with competing solutions in the pre-n battle.

188

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Wireless LAN Wireless LAN


Wireless LAN Semiconductor Market Trends
The market has shifted hard to 802.11g; 802.11a/b/g has had a minor impact. 802.11n pre-standard products have started to ramp significantly in 2006, disrupting the market again; n should come to dominate the market over the next few years. Intel has been successful with its Centrino campaign, and has driven notebook penetration to upwards of 80%. Cost has become hugely important; vendors that cannot compete have pared back their investments. Most vendors have moved to all standard CMOS 2-chip solutions; single-chip solutions target handset and consumer. Most chipset vendors are looking to the consumer and handset markets for the next volume application to implement wireless LAN.

The market has shifted hard to 802.11g; 802.11a/b/g has had a minor impact. 802.11g is now the dominant standard for wireless LAN in both the installed base and current shipments. It was especially popular in the consumer PC market, where users embraced the faster speeds and backwards compatibility. Multi-mode solutions, on the other hand, dont boost the throughput speed and instead offer more channels, a feature more attractive to high end enterprise users. In terms of market share, the largest beneficiaries of the shift to 801.11g have been Broadcom and Atheros, which combined for nearly 60% of 802.11g share in 2005 and 38% overall. The biggest losers were Conexant, which was behind Broadcom and Atheros by several months (Conexants share losses are also due to share gains in notebooks by Intel), and Agere, which had no 802.11g product. 802.11n pre-standard products have started to ramp significantly in 2006, disrupting the market again; n should come to dominate the market over the next few years. The IEEE is currently finalizing the standard for 802.11n, which leverages MIMO technology in both the 2.4 GHz and 5 GHz bands and could boost the throughput to 108 Mbps and eventually to speeds approaching 540 Mbps. This should support ASPs, as these new chipsets will be more expensive. It will also likely cause some share shifts, as some vendors have chosen to hit the market with pre-standard solutions before the final standard is ratified (Marvell, Atheros, Broadcom, and Airgo are already shipping pre-N or MIMO product). Eventually, market share, at least in PC applications, should shift back as Intel and others develop solutions after the standard is finalized.

189

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Intel has been successful with its Centrino campaign, and has driven notebook penetration to upwards of 80%. Most client implementations of wireless LAN in the early years were PC cards that were plugged into laptops. Some PC makers then attempted to integrate wireless LAN directly into the notebook as a differentiating feature, though it was for a minority of notebook models, primarily at the high end. In 2003, Intel was set to launch a new mobile processor (the Pentium M), and decided to bundle wireless LAN in with the platform, which it branded Centrino. To use the Centrino name and marketing dollars (Intel allocated $300 million to the Centrino campaign), PC makers had to buy not just the Pentium M but also the core logic and wireless LAN chipset from Intel. This had the effect of shifting market share away from incumbent wireless LAN suppliers to Intel, but had the much more important effect of increasing wireless LAN penetration in notebooks. We now estimate that upwards of 80% of notebooks now ship with embedded wireless LAN. Cost has become hugely important; vendors who cannot compete have pared back their investments. When wireless LAN began gaining traction in 2000, only three major vendors were shipping in volume: Intersil (which later sold the unit to Globespan, which was bought by Conexant in early 2004), Agere, and NXP/Philips (which at the time just did RF chips). In 2001 a number of mature companies such as Texas Instruments, Cirrus Logic, RF Micro Devices, and National Semiconductor entered the market (most through acquisition) along with then-start-up Atheros. In 2002 Marvell and Broadcom entered and Intel joined the market in 2003 with the Centrino platform. These entrants turned the competitive landscape on its head, driving chipset costs lower and levels of integration and performance higher. Today, only three vendors serve all segments of the wireless LAN market: Broadcom, Atheros, and Marvell. Intel has limited its exposure from the start to embedded notebook designs. Conexant, Texas Instruments, NXP/Philips, RF Micro Devices, and STMicroelectronics no longer target PC applications, and instead focus on other embedded handset, gateway, or consumer clients. Agere, Cirrus Logic, National Semiconductor, and a host of others (including start-ups) have exited entirely.

Most vendors have moved to all standard CMOS 2-chip solutions; single-chip solutions target handset and consumer. When wireless LAN began to ramp to volume in 2000, most solutions used as many as five chips, using a mix of CMOS (for the digital chips), SiGe (for radio components), and GaAs (for the power amp). Today, most offerings have migrated to a 2-chip solution; baseband/MAC and RF, all in CMOS; though some power amps (especially for 5 GHz) are external and are still done in GaAs. Most vendors also offer single-chip solutions, mostly targeted at the handset and consumer markets where power consumption and cost are even more critical. Atheros has been more aggressive with single-chip and uses it throughout its portfolio. Most chipset vendors are looking to the handset and consumer markets for the next volume application to implement wireless LAN. Chipset vendors are working hard to enable the consumer electronics and handset wireless LAN connection discussed in our unit forecast. Low-cost, low-power, small-form factor wireless LAN chipsets are the key here, and vendors have been devoting more resources toward addressing the needs of this demanding application. Already, a number of big designs have gone to volume, mostly for gaming (PS3 and Wii embedded; Xbox and Xbox 360 as an add-on). Printers and consumer portables are also contributing, though handsets are viewed as the next real big wave.

190

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Bluetooth Bluetooth
Bluetooth is a personal area networking protocol, used to send data between devices in close proximity at rates of up to 3 Mbps. Bluetooth has found its primary volume application in the handset market, where it functions as a reliable protocol to connect the handset to other devices, including wireless headsets, PCs, and consumer devices. PC peripherals and consumer portables are also using Bluetooth as a cable replacement for connection to the PC, or to connect to one another without the PC.

Bluetooth is the leading personal area networking technology, a short-range, low-power, high-bandwidth wireless transmission protocol aimed at connecting wireless handsets, PC peripherals and consumer devices. Bluetooth is standards-based, universal and cost effective, offering relatively high bandwidth (up to 3 Mbps) over a short distance (30-300 feet, depending on the device class) in an unregulated frequency band (2.4 GHz), with extremely low power consumption and low implementation cost ($2-$4). Bluetooth has found its primary volume application in the wireless market, where it functions as a reliable protocol to connect the handset to other devices, including wireless headsets, automotive stereo systems, PCs and consumer devices. Smart handheld devices, PC peripherals, and consumer portables are also using Bluetooth to perform data transfers without wires; and in many cases the entire PC can be removed from the equation (e.g., a digital camera can print straight to a Bluetooth printer, a PDA can transfer phone numbers to a cell phone, etc.).

191

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Bluetooth Bluetooth
Bluetooth Unit Forecast
We expect Bluetooth units to grow at a 33% CAGR, from 308 million units in 2005 to 1.3 billion units in 2010. Bluetooth in wireless handsets is expected to grow from 209 million units in 2005 to 764 million units in 2010; this represents a 58% penetration rate. Other significant applications include headsets, PCs, peripherals, consumer electronics, and PDAs. Bluetooth will be implemented almost entirely in silicon, with little additional content added by OEMs.
1,400,000
Add-ons, access points, and other Gaming Consumer Electronics (ex. Gaming) PCs & Peripherals Headsets Mobile Handsets
Unit Shipments (000s)

Handset Suppliers
Others 17 %

Sagem 2% BenQ/Siemens 5%

Nokia 33 %

Sony Ericsson 6% LG 7% Samsung 13 % Motorola 18 %

PC Suppliers
Dell 18 % Others 39 %

1,200,000

1,000,000

800,000

600,000

HP 16 %

400,000

200,000

2002 2003 2004 2005 2006 2007 2008 2009 2010

Sony 2% Gateway 2%

Apple 2%

NEC 3%

Toshiba 3%

Fujitsu/F-S 4%

Acer 5%

Lenovo 6%

Source: IDC, CIBC World Markets Corp.

Source: IDC, CIBC World Markets Corp.

Bluetooth Unit Forecast: We estimate that Bluetooth unit shipments totaled 308 million in 2005. Of these, nearly 80% sold into cellular handsets and wireless headsets. The remainder found homes in PCs, PC peripherals, consumer devices, and add-ons and access points. We expect the market to ramp quickly over the next few years, growing at a 33% CAGR to 1.3 billion units in 2010. The primary driver should be increasing penetration in wireless handsets and the accompanying headsets; these should represent 69% of the 2010 total. Gaming and consumer devices, which represented less than 5% of the market in 2005, should also be important growth drivers and should each represent 8% of total units in 2010. PCs and peripherals should remain 8%-10% of the market through 2010. Bluetooth Suppliers: Bluetooth is almost always implemented as an embedded solution within a larger system, and therefore Bluetooth equipment is a somewhat misleading term. As handsets represented 68% of Bluetooth unit shipments in 2005, it is important to note the leading handset vendors. In 2005, the top handset suppliers were Nokia with 33% unit share, Motorola with 18%, Samsung with 13%, LG with 7% and Sony Ericsson followed with 6%. The top suppliers of Bluetooth headsets are these same handset OEMs, all of which sell branded headsets to go with their phones (though they are interoperable with any Bluetooth-enabled handset). Third party headsets are also available from vendors like Plantronics, Logitech, and dozens of other smaller players. PCs are also a large end market for Bluetooth; top PC makers in 2005 were Dell, HP, Lenovo, Acer, Fujitsu/FujitsuSiemens, Toshiba, NEC, and Apple. On the gaming side, new consoles from Sony and Nintendo integrate Bluetooth.

192

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Bluetooth Bluetooth

The baseband uses flash memory

Antenna Flash

Power Amp

Radio (RF)

Bluetooth Baseband

Host System

Sits in front of the antenna and boosts signal strength Controls radio wave generation and reception, converts the analog radio signal to a digital stream and vice versa

Performs protocol processing, implementing the modulation scheme (encoding technique)

Source: CIBC World Markets Corp.

Bluetooth modules have similar configurations to those of wireless LAN client implementations. A power amp interfaces with the antenna and relays signals to a radio (RF) chip. The radio chip processes the RF signal and passes it to the Bluetooth baseband, which implements the Bluetooth modulation scheme (the encoding technique) and extracts the data packet. On the transmit side, the process is reversed: the baseband encapsulates the data using the Bluetooth modulation scheme and sends it through the RF and power amp to the antenna. Note that most Bluetooth solutions combine all functions on a single chip, sometimes leaving out the power amp but still integrating the RF and baseband. Other implementations use just a Bluetooth RF, integrating baseband functionality with a host processor (e.g., Qualcomm integrates the Bluetooth baseband in its handset baseband IC). Power Amp: The power amp sits in front of the antenna and boosts signal strength. Radio (RF): The RF controls radio wave generation on the transmit side and receives the radio wave on the receive side. Baseband Processor: The baseband processor implements the Bluetooth modulation scheme, extracting the packet from the Bluetooth frame on the receive side and encapsulating in on the transmit side. Flash: Flash memory is the primary memory used by the baseband.

193

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Bluetooth Bluetooth
Bluetooth Semiconductor Bluetooth Semiconductor Forecast Competitors
We expect the market to grow at a 17% CAGR, from $1.1 billion in 2005 to $2.3 billion in 2010. Revenue growth should trail units as ASP declines help drive units shipments into the mainstream. Revenue growth could start to exceed unit growth in 2010 and beyond as converged Bluetooth-UWB solutions ramp to volume; these will carry higher ASPs than standalone Bluetooth.
$2,500 Bluetooth Chipset $2,000

RF Micro STMicro 2 % Others 2% Infineon 2 % 4%

Texas Instruments 7%

NXP/Philips 13 %

CSR 46 %

Broadcom 23 %
Source: Dataquest, CIBC World Markets Corp.

Semiconductor Revenue ($M)

$1,500

$1,000

$500

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, CIBC World Markets Corp.

Bluetooth Semiconductor Forecast: We expect a strong implementation cycle for Bluetooth as falling chipset ASPs drive wireless handset and PC peripheral OEMs to increase the usage of Bluetooth in their product portfolios. Overall, we expect the market to grow from $1.1 billion in revenue in 2005 to $2.3 billion in 2010, a 17% CAGR. Bluetooth Semiconductor Competitors: The emerging Bluetooth competitive landscape has consolidated somewhat as handset, headset, and PC designs have ramped, with a few suppliers winning out among a wide variety of competitive solutions first available to the market. The largest supplier in 2005 was Bluetooth pure-play Cambridge Silicon Radio with 46% share; CSR is the largest supplier of Bluetooth ICs to handsets, with Nokia and Samsung as key customers. CSR also dominates the headset market. Broadcom held 23%, gaining 6 points of share from 2004, primarily due to key wins at Motorola and Samsung for handsets, a partnership with Qualcomm for Bluetooth RF chips and exclusive deals with Palm for PDAs and Apple for PCs. Broadcom intends to more aggressively target headsets in 2007. NXP/Philips was the No. 3 supplier in 2005 with 13% share, with a mix of both handset and headset chipsets. Other vendors in the space include Texas Instruments, Infineon, STMicroelectronics, and RFMD. Note that several vendors have exited the market over the past few years, including Agere, Conexant, Microtune, and Freescale. Marvell is the only significant new entrant to the market in the recent past, with products set to ramp in 2007.

194

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Bluetooth Bluetooth
Bluetooth Semiconductor Market Trends
Cellular handsets and headsets are proving to be the largest market for Bluetooth; voice quality and power consumption are therefore key factors in winning designs. Single-chip Bluetooth solutions have taken over the market. Some implementations will integrate the Bluetooth baseband into the host system. Bluetooth is starting to penetrate consumer and gaming devices. 3.0 Mbps Enhanced Data Rate (EDR) products are ramping quickly following standardization in 2004. Bluetooth-wireless LAN combo solutions are coming to market targeting next-generation handsets. Bluetooth and UWB will adopt the same front end later in the decade.

Cellular handsets and headsets are proving to be the largest market for Bluetooth. The center of the Bluetooth universe is clearly the handset market, with all major OEMs quickly proliferating the technology throughout their high end and mainstream models. Penetration was likely in the 45%-50% range in 2006, marching into the 50s through 2010. The headsets associated with the cellular market are also growing nicely, with shipments estimated to approach 50 million units in 2006 and to top 100 million units in 2010. Voice quality and power consumption are key factors in winning designs. The major Bluetooth IC vendors recognize that handsets will continue to dominate the Bluetooth market, and organize their Bluetooth design teams around the handset market. Factors such as voice quality, power consumption, and cost top the list of design goals in order to meet the needs of handset OEMs. Single-chip Bluetooth solutions have taken over. Although the first Bluetooth chipsets used separate baseband and radio chips, most vendors have moved to single-chip implementations. Note that single-chip designs have been critical to driving down the cost of the chipset bill-of-materials to the $2-$4 range; this has been critical to driving Bluetooth into the mainstream. Some implementations will integrate the Bluetooth baseband into the host system. Although we expect single-chip devices to dominate over the next five years, we expect many wireless handset IC vendors will seek to incorporate Bluetooth baseband processing into their devices, eliminating the need for a discrete baseband IC (or

195

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

the baseband logic within a single chip). A key example is Qualcomm, which has incorporated Bluetooth baseband functionality into some of its handset chipsets. Note that most of these designs require an external RF IC and several chip vendors address this opportunity with seamless, standards-based solutions. Bluetooth is starting to penetrate consumer and gaming devices. Given the buzz generated by Bluetooth in the handset market, it is no surprise that the technology is being adopted by consumer and gaming devices. In some cases the goal is to interact directly with the handset (e.g., a picture phone can print to a Bluetooth photo printer, or a PDA can sync contacts with a handset), while in other cases the range and bandwidth requirements are suitable for Bluetooth (Sonys PS3 and Nintendos Wii each use Bluetooth to talk wirelessly to their controllers). We estimate consumer and gaming will represent more than 15% of the total units in 2010 (from less than 5% in 2005). 3.0 Mbps Enhanced Data Rate (EDR) products are ramping quickly following standardization in 2004. The Bluetooth 2.0+ EDR specification alters the signal encoding system used by Bluetooth to allow it to transmit at up to 3.0 Mbps and extends the range to 100 meters. EDR products are backward compatible with conventional Bluetooth 1.1 products. Note that EDR products ramped quickly in 2005 to represent more than 10% of Bluetooth chipset shipments; EDR should represent the vast majority of shipments by 2008. Bluetooth-wireless LAN combo solutions are coming to market targeting next-generation handsets. Many comm IC players view wireless LAN as the next major connectivity technology to penetrate the handset. In order to ease adoption, combo solutions are now coming to market that integrate Bluetooth and wireless LAN in a single chipset or reference design, or in some cases on a single chip. The drivers for wireless LAN in a handset are not quite as clear as they were for Bluetooth (headsets are an easy concept given that the primary function of a handset is voice telephony), so it will be some time before the adoption curve becomes clear for wireless LAN or combo solutions in handsets. Bluetooth and UWB will adopt the same front end later in the decade. The Bluetooth SIG (special interest group) announced in 2006 that the next generation of Bluetooth would adopt a common front end interface with ultra wide band (UWB) technology. This paves the way for much higher data rates for Bluetooth, and also for combo solutions to handle Bluetooth and UWB in a single chipset or SoC. We expect Bluetooth products based on the UWB front end to come to market in 2008 and reach 15% of chipset units in 2010.

196

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Storage Storage
The buildout of high-performance computing infrastructure, adoption of higher-bandwidth LAN and WAN connections, new applications offering richer content, and increased focus on data security and redundancy have driven continued strong demand for high-capacity storage in the enterprise. Terabyte shipments continue to grow at more than 50% per year. The type of storage systems being implemented is also changing. Storage embedded within PCs and servers offers insufficient capacity, manageability, and scalability for the modern enterprise. OEMs have responded with more complex direct-attached systems (RAID) as well as networked topologies like SAN and NAS. Over time, DAS should replace most PC- and server-attached storage, while SAN and NAS continue to penetrate IT systems where greater scalability, manageability, and performance are required.
The three preceding networking technologies we have discussed in this section are all used in the transference of data between computing devices. We now turn from data networking to enterprise storage, which is used behind the scenes in the enterprise network for the storage, retrieval, backup, and management of corporate software and data. Over the past few years, storage equipment makers have attempted to bring the benefits of networking to enterprise data storage. The legacy model, which stores applications and data files on disk drives within individual servers, is being replaced by newer networked storage systems. By treating storage as a separate resource from processing, enterprise IT managers are realizing greater efficiency, reliability, scalability and upgradeability in their infrastructure. There are a wide variety of storage topologies and technologies employed in enterprise IT systems today. Most systems use a direct attached storage (DAS) architecture, which connects servers directly to drives and drive arrays. Though they do not offer the flexibility, scalability, manageability, and performance of networked storage, DAS systems are robust and easy to set up and maintain, and therefore remain an important part of the IT infrastructure. High-performance and high-reliability storage systems use storage area networks (SAN), predominantly based on the Fibre Channel protocol. These mimic the LAN topology, using specialized storage switches and adapter cards. SANs are the topology of choice for the highest-end storage applications. Some simpler IT systems use network-attached storage (NAS), which connect drive appliances directly to the Ethernet network. This brings the benefits of networked storage while leveraging the LAN (no new SAN network is needed).

197

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Storage Storage
Direct Attached Storage Direct attached storage systems are primarily implemented with RAID controllers, which are directly connected to disk array systems or internal drives. The cheapest, simplest, and easiest to implement, direct attached is the most widely used storage topology.
Internal RAID with Internal Drives
RAID

Ethernet

Internal RAID Controller


RAID

LAN Switch

Internal RAID with Disk Array Storage Arrays External RAID

Server
Ethernet
RAID

External RAID Controller

Server
Source: CIBC World Markets Corp.

Tape Library

Direct attached storage systems are the cheapest, simplest and easiest to implement and are currently the most widely used storage topology. In a DAS system, servers are connected via standard hard disk drive cables either to enterprise-class hard disk drives within the server or to storage arrays that contain multiple drives. The interaction between the server and the disks is controlled by RAID (redundant array of independent disks), a set of protocols for data striping, duplication, and drive management. A RAID controller, which sits between the server and the individual disks, implements the RAID software stack to enable these features. RAID controllers can be either internal or external. In an internal configuration, the RAID controller is housed inside the server and connects to drives within the server or to simple disk arrays like JBODs (just a bunch of disks). In an external configuration, the RAID controller is housed inside the disk array itself, and the server connects to the disk array over a standard hard disk drive cable. As mentioned above, DAS systems are cheap, simple and easy to implement. The multiple flavors of the RAID standard allow for varying degrees of redundancy and performance enhancement. However, they have limitations in terms of reliability, scalability and manageability that make DAS systems difficult to support as capacity grows larger. SANs have made major headway against large DAS systems over the past few years, especially for the highest-end storage applications.

198

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Storage Storage
Storage Area Networking (SAN) In a SAN environment, separate Fibre Channel networks switch storage traffic between disk arrays and servers. Storage is managed as a single resource, and tasks such as backup and disk maintenance can often be performed without any processing by host servers or PCs.
Tape Library
Fibre Channel Ethernet

LAN Switch
FC

Fibre Channel Switch


Fibre Channel

FC

Fibre Channel

Fibre Channel FC HBA


Ethernet

Storage Array

Server

Source: CIBC World Markets Corp.

Storage area networks (SANs) are implemented with disk arrays sitting on a high-speed storage-focused network separate from the data network. In this way, they separate processing functions from storage, and are therefore well-suited to large corporate networks. SANs allow IT managers to treat all storage as a single resource and also allow tasks such as backup and disk maintenance to be performed without any processing by host computers. In a SAN, servers do not connect directly to drives or arrays; they instead connect to a storage-specific network designed to carry storage traffic using the Fibre Channel protocol. These storage networks serialize SCSI traffic into Fibre Channel frames for high speed transport. Fibre Channel is somewhat similar to Ethernet, but is optimized for mass transfers of data as opposed to IP packet routing. The protocol uses optical fiber interconnects and runs at 1.0, 2.0 Gbps, or 4.0 Gbps. To connect to the SAN, each server is outfitted with a Fibre Channel host bus adapter (HBA), which connects to a network of specialized Fibre Channel switches. These switches provide the connection between servers and storage arrays, which also have to be specialized to run over Fibre Channel. Tape libraries can also be connected to the SAN, allowing for backup to take place without the use of servers on the data network. Note that as SANs have evolved, OEMs have begun building more intelligent appliances to help manage the network such as SAN routers which interconnect between islands of storage in disparate locations. As these are used in only the highest-end implementations, we do not picture them in our network diagram.

199

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Storage Storage
Network Attached Storage (NAS) In a NAS, storage appliances are connected directly to the Ethernet local area data network. Instead of a fullblown operating system, NAS systems use a slim microkernel OS specialized for handling only file reads and writes. NAS systems operate like mini-SANs in that they are self contained storage servers.
Tape Library
Ethernet
N A S

NAS Storage Appliance


Gigabit Ethernet

LAN Switch
Ethernet

Server
Source: CIBC World Markets Corp.

In a NAS environment, storage appliances are connected directly to the Ethernet LAN. NAS appliances operate like mini-SANs in that they are self contained storage servers. NAS systems usually contain a number of hard disks, often arranged into logical, redundant storage containers or RAID arrays. Instead of a full-blown operating system, NAS systems use a slim microkernel OS specialized for handling only file reads and writes. Storage capacity is added by connecting NAS appliances to the Ethernet network as simply as if a new PC or server were being added. Note that NAS technology has begun penetrating the consumer market in recent years, sparked by the rise of digital content and home networks. Home NAS appliances can do more than simple USB external HDDs, as they allow for direct access by more than one PC or connected device on the network, and can also connect to other drive arrays for automatic backup. Note that Gigabit Ethernet will be an important enabler for this market, as 10/100-based devices in the market today are dramatically outpaced by USB 2.0 at 480 Mbps or Firewire 800 in terms of raw data transfer.

200

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Storage Storage
Types of Storage Equipment
Enterprise Hard Disk Drive - Enterprise hard disk drives are the highestperformance drives on the market. They are usually housed either inside servers, in storage arrays, or inside NAS appliances, and use the SCSI or SAS protocol. Enterprise drives are included in our forecast for HDDs.
RAID

RAID Controller A client card used to connect servers to storage arrays; they can be housed either in the server (internal) or in the array (external). Storage Array - Storage appliance with multiple hard disk drives used in enterprise storage applications. In DAS systems, they either incorporate RAID controllers or connect to internal RAID controllers inside servers. In SAN systems, they connect to Fibre Channel switches.

FC

Fibre Channel HBA - A client card used to connect servers to the Fibre Channel network. Analogous to a NIC in an Ethernet network. Fibre Channel Switch - The basic SAN building block, Fibre Channel switches connect and switch storage traffic between servers and storage arrays. NAS Appliance - NAS appliances integrate storage and LAN connectivity. They include multiple HDDs and a slim OS to manage data transfers. Tape Library Tape library appliances are optimized for backup functions in an enterprise IT infrastructure system. They are implemented independently of the storage topology.

FC

N A S

Types of Storage Equipment Enterprise Hard Disk Drive: Similar to HDDs used in PCs, enterprise hard disk drives are the highest-performance drives on the market. They are usually housed either inside servers, storage arrays, or NAS appliances. Hard disk drives used in storage systems are not included in our forecast, and are instead discussed in the section on HDDs. RAID Controller: A client card used to connect servers directly to storage arrays. They can be housed either in the server (internal) or in the array (external), and use standard hard disk drive interfaces like SCSI, ATA, SATA, or SAS. Storage Array: Storage appliance with multiple hard disk drives used in enterprise storage applications. In DAS systems, they either incorporate RAID controllers or connect to internal RAID controllers inside servers. In SAN systems, they include Fibre Channel connectivity to connect up to SAN switches. Fibre Channel HBA and Switch: An HBA is a client card used to connect servers to the Fibre Channel network. A Fibre Channel switch is the basic SAN building block used to connect and switch storage traffic between servers and storage arrays. NAS Appliance: A mass storage array optimized for NAS systems, NAS appliances integrate storage with Ethernet LAN connectivity and can be added directly to the Ethernet network to provide enterprise storage. Tape Library: Storage appliance optimized for backup functions in an enterprise IT infrastructure system.

201

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Storage Storage
Storage Equipment Unit/Port Forecast
Total enterprise storage units/ports should grow at a 23% CAGR, from 14.0 million in 2005 to 38.7 million in 2010. RAID controllers should grow at a 16% CAGR, with external RAID outpacing internal RAID. Fibre Channel HBAs and switch ports should each grow at CAGRs of 30% to reach 55% of total ports in 2010. iSCSI targets should grow from a small base to 6% of total port shipments in 2010.
40,000
iSCSI SAN Targets

Key Storage Equipment Suppliers


Disk Storage Systems (TBs)
100% Others 90% 80% 70% Hitachi Sun Network App. 70% 90% 80% 100%

RAID Controllers
Highpoint AMCC/3Ware

Promise

Adaptec

Dell
60% 60%

EMC
50% 40% 30% 20% 10% 0% Disk Storage Systems (TB) 50%

LSI Logic IBM


40% 30% 20% 10% 0% RAID Controllers

HP

HP

Fibre Channel Switches


100% 90% 80% Others QLogic

Fibre Channel HBAs


100% Others 90% LSI Logic 80% 70%

35,000

Fibre Channel Infrastructure Fibre Channel HBAs (ports)

30,000
Unit/Port Shipments (000s)

RAID Controllers

Cisco

25,000

70% 60% 50%

20,000

McData

60% 50% 40% 30% 20% 10% 0%

Emulex

15,000
40%

10,000

30% 20% 10%

Brocade

5,000

QLogic

2003 2004 2005 2006 2007 2008 2009 2010


0% FC Switch Ports

Source: IDC, Dataquest, CIBC World Markets Corp.

HBAs (ports)

Source: Dataquest, DellOro, IDC, CIBC World Markets Corp.

Storage Equipment Unit/Port Forecast: The market for enterprise storage is expected to represent a growing percentage of IT budgets over the next five years as IT managers shift from adding network functionality to adding storage infrastructure, optimizing network efficiency and reducing IT maintenance costs. While sufficient bandwidth exists in both service provider wireline networks and corporate networks to support the transport of data and Internet traffic, the buildout of storage capacity to store that data in corporate networks and public hosting companies has been a lot slower relative to demand growth. Consequently, we expect units/ports used in enterprise storage systems to grow at a 23% rate through 2010, rising from 14.0 million units/ports in 2005 to 38.7 million ports in 2010. Within the storage pie, we expect the mainstream RAID controller market to grow at a more moderate but still-strong pace of 16%, rising from 4.8 million ports in 2005 to 10.1 million ports in 2010. External RAID controllers, growing at a 21% CAGR, should outgrow internal controllers at 14%. Over on the Fibre Channel side, we expect host bus adapter ports to grow at a 30% CAGR, from 2.1 million ports in 2005 to 8.0 million ports in 2010. This is derived from an 18% growth rate in discrete HBA units and an increase in the average port/HBA from 1.23 in 2005 to 1.51 in 2010, boosted by a 54% growth rate of blade server HBA ports, growing from just 328,000 ports in 2005 to 2.9 million in 2010. Fibre Channel switch ports are expected to grow 30%, from 3.6 million ports in 2005 to 13.1 million ports in 2010. Fixed port switches and chassis switches should grow at similar rates (23% for fixed; 26% for chassis). Blade server-

202

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

embedded switches, however, should see a big ramp, growing from just over 300,000 ports in 2005 to 27% of all switch ports in 2010. iSCSI SAN targets, still a developing market, are expected to lead growth in storage with a nearly 70% CAGR, growing from a very small base of 175,000 ports in 2005 to 2.4 million ports in 2010. iSCSI is IP-based and therefore does not require switches beyond the Ethernet-based data network; they also dont require discrete HBAs, though the client server or PC must either have an iSCSI enabled-NIC (like Broadcoms C-NIC) or be running iSCSI on the host processor. Storage Equipment Suppliers: The enterprise storage market includes vendors of disk storage systems, RAID controllers, Fibre Channel switches, and HBAs. Though there is some overlap, for the most part they are discrete markets, each with their own competitive dynamics. Note, however, that the vendors of disk storage systems typically resell components (as well as software) from the other segments within larger system solutions for their customers; these components are sometimes re-branded. The largest vendor of disk storage systems in 2005 (in terms of terabytes, which is a more meaningful figure than units) was HP with 31% share. IBM was second with 17%, followed by EMC with 12% and Dell with 11%. Other major systems suppliers include Network Appliance, Sun Microsystems (including StorageTek), and Hitachi. The largest vendor of RAID controllers in 2005 was HP with a 32% share. LSI Logic and Adaptec followed with 28% and 25% share, respectively. Other vendors include Promise Technology, AMCC/3Ware, and HighPoint. Over on the Fibre Channel side, QLogic and Emulex were the top suppliers of HBAs in 2005, with 44% and 37% port share, respectively. LSI Logic held a small 6% share. In Fibre Channel infrastructure, Brocade was the top supplier with 49% overall share, stemming from its 66% share of fixed-port switches and 33% of chassis-based switches. McData was No. 2 with 25% overall share, with 16% of the fixed-port switch market but nearly 44% share of the chassis-based switch market, where it is the No. 1 supplier. Note that in 2006 Brocade announced that it would acquire McData, which would give the combined company 74% share overall before any attrition. Cisco was the No. 3 player in 2005 with 14% share, almost entirely from chassis solutions. QLogic was next with 12%; most of its share was derived from fixed-port solutions.

203

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Storage Storage
A microprocessor runs the RAID software stack This ASIC provides the interface to the individual disks, as well as to the server in an external RAID system

RAID Controller
(Internal and External)
Host System

Microprocessor
PCI-X/PCI-E Controller

Hard Disk Controller/ Expander

Hard Disk Drives (Internal RAID) Disk Enclosure (External RAID)

SRAM

Flash

SRAM

Flash

Both the RAID controller and microprocessor use SRAM and flash memory Implements the Fibre Channel protocol, providing the disk arrays interface to the SAN

Performs control and monitoring

FC-RAID Sub-system Fibre Channel Controller

Disk Enclosure

Enclosure Management

Hard Disk Drive Hard Disk Drive

FC-RAID Storage Array

Optical Fiber

GBIC

SerDes

Interfaces with the GBIC (optical module), performing transmission and reception and converting the signal between an analog serial stream and a digital parallel stream

External RAID Controller

Port Bypass Controller Or Loop Controller SRAM Flash

Source: IDC, CIBC World Markets Corp.

Opens and closes nodes and distributes data to and from the systems drives; SRAM and flash memory are used by the PBC/LC

Hard Disk Drive

RAID systems enable critical storage functions such as data redundancy, striping (for performance), and disk management. Their architecture and implementation vary widely, from low end systems that control just a few disks within a single server, targeted for small/medium business, all the way to large switched disk arrays managing thousands of drives used in large corporate enterprises. At the core of a RAID system is the RAID controller, which performs the compute-intensive calculations needed to implement the RAID stack at high speed. A RAID controller is usually implemented as a separate add-in board. In internal RAID systems, the RAID controller is used directly inside the server, and is attached directly to the drives, which are also located within the server. In an external RAID system, the RAID controller is implemented within an array of disks as a subsystem to perform RAID processing. Larger disk arrays will usually have a number of subsystems, including a Fibre Channel subsystem to interface with a Fibre Channel SAN and a disk enclosure to manage the communication with individual disks.

RAID Controller Internal (used in servers) IOP/Microprocessor: A microprocessor runs the RAID protocol stack, enabling the striping, redundancy, and disk management features of RAID. The microprocessor in an internal RAID implementation is sometimes call an

204

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

I/O processor (IOP). These processors are usually optimized for high-speed RAID processing, and usually include the interface to the host controller, which is usually PCI-X or PCI-Express based. Hard Disk Controller: The hard disk controller provides the interface to the individual drives. ROC (RAID-on-Chip, not shown in above diagram): A ROC chip includes both the IOP and hard disk controller. ROC solutions can be used either on discrete RAID controller cards, but are mainly targeted for ROM (RAID-onmotherboard) implementations. SRAM and Flash: SRAM and flash are used by the microprocessor as well as by the hard disk controller ASIC.

RAID Controller - External (used within disk arrays) Microprocessor: A microprocessor runs the RAID protocol stack, enabling the striping, redundancy, and disk management features of RAID. In higher end systems, multiple high-performance processors are sometimes used. Hard Disk Controller: The hard disk controller performs the SCSI/ATA processing to communicate with the drives within the enclosure. In lower end PATA/SATA/SAS RAID systems that do not have separate interface cards, this controller also interfaces with the server on the front end. In higher end systems (especially Fibre Channel), this is not needed since a separate controller card performs this interface. SRAM and Flash: SRAM and flash are used by the microprocessor as well as by the hard disk controller ASIC.

FC-RAID Storage Array SerDes: The SerDes interfaces with the optical module, performing transmission and reception and converting the signal between an analog serial stream and a digital parallel stream. Fibre Channel Controller: A Fibre Channel controller ASIC implements the Fibre Channel protocol, providing the interface to the SAN. Port Bypass Controller (PBC): The primary logic device in the disk enclosure, a PBC opens and closes nodes between drives, distributing data to and from the systems drives. Loop Controller: More complex Fibre Channel disk arrays use loop controllers for a switched architecture instead of a PBC. Enclosure Management: In high-end systems, enclosure management chips monitor and control the drives in the storage array. SRAM and Flash: SRAM and flash are used by the port bypass controller or loop controller. Expander (used in lower-end SAS/SATA-based arrays, not shown in the diagram): An expander IC, used in serial arrays like SAS or SATA (instead of a PBC or loop controller), increases the number of drives that can be included in the array. Note that SAS expanders can support both SAS and SATA drives; therefore most IC designers are focusing on SAS expanders.

205

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Storage Storage
Control plane Comm processor - controls overall system functioning SRAM, flash - memory used by the MIPS processor

Communications Processor SRAM Flash


FC Port Controller

FC Port Controller FC Port Controller

Switch Fabric Chipset


FC Port Controller

SRAM

SRAM memory is used by the switch fabric and the FC port controllers

Fibre Channel Switch

SerDes

SerDes

SerDes

SerDes

This ASIC (or series of ASICs) switches storage packets between ports, and includes individual controllers for each port

GBIC Optical Fiber


Interfaces with the GBIC (optical module), performing transmission and reception functions and converting the signal between an analog serial stream and a digital parallel stream, and vice versa

GBIC Optical Fiber

GBIC Optical Fiber

GBIC Optical Fiber

This ASIC performs Fibre Channel protocol processing, removing and applying Fibre Channel overhead; it usually includes an integrated processor and system interface

Host Bus Adapter

Optical Fiber

Fibre Channel HBA Controller ASIC GBIC SerDes


MPU PCI-X/PCI-E Controller

Host System

SRAM

Flash
SRAM and flash memory are used by the controller ASIC

Source: IDC, CIBC World Markets Corp.

Fibre Channel Switch SerDes: The SerDes converts the serial analog stream from the GBIC to a digital parallel stream and vice versa. Switch Fabric Chipset: The switch fabric chipset is a set of ASICs that switch storage traffic between switch ports. The chipset usually has integrated port controllers that interface with the SerDes on the front of the switch. Communications Processor, SRAM and Flash: A processor, usually MIPS or PowerPC based, performs control plane functions. The processor will usually use SRAM and flash, while some additional SRAM is also used by the switch fabric chipset in switching and port control.

Host Bus Adapter (HBA) SerDes: The SerDes converts the serial analog stream from the GBIC to a digital parallel stream and vice versa. Fibre Channel Controller ASIC: The controller ASIC performs protocol processing on the Fibre Channel stream, applying and removing Fibre Channel overhead to storage packets. The controller usually incorporates a processor, though the HBA can have a discrete MPU as well. It also usually includes a host interface. SRAM and Flash: SRAM and flash are used by the controller ASIC and the embedded microprocessor.

206

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Storage Storage
Storage Semiconductor Forecast
Storage semiconductors growing at 7%, with the market growing from $1.1 billion in 2005 to $1.5 billion in 2010. IC revenue growth should lag port shipments as the market matures and more integrated solutions are introduced, driving down ASPs. RAID controller ICs growing 3% to 41% of the total storage IC market in 2010 Disk interconnect ICs growing 5%. HBA ICs growing 7%, Fibre Channel infrastructure ICs growing 14%.
$1,800 $1,600 $1,400
Semiconductor Revenue ($M)

Storage Semiconductor Competitors


Marvell 4% Silicon Image 4% Vitesse 5% Freescale 7% AMCC 2% Others 5%

LSI Logic 29 %

PMCSierra/Avago 14 % Intel 14 %

iSCSI SAN Targets Fibre Channel Infrastructure Fibre Channel HBAs Disk Arrays RAID Controllers

IBM 16 %

$1,200 $1,000 $800 $600 $400 $200 $0 2003 2004 2005 2006 2007 2008 2009 2010

HBA and SAN ASICs


IBM LSI Logic Agere Mindspeed Broadcom

RAID Processors/ Controllers


Intel LSI Logic AMCC Promise/Marvell Vitesse/Adaptec Broadcom

Disk Array ICs


PMC-Sierra IBM Vitesse LSI Logic

Source: IDC, Dataquest, CIBC World Markets Corp. Source: Dataquest, IDC, CIBC World Markets Corp.

Storage Semiconductor Forecast: Storage ICs should benefit from ramping port and unit growth of enterprise storage systems, but semiconductor revenue should significantly lag port shipments as the technology matures and system suppliers introduce more cost efficient, higher density solutions. To a certain extent, the industry is being forced to enable lower cost storage, especially in the Fibre Channel arena, as a number of lower-cost competitive technologies both from within the storage world (SAS and SATA) and from the networking world (Ethernet/iSCSI and Infiniband) threaten Fibre Channels hold on the enterprise. Overall, we expect a CAGR of 7%, with IC revenues growing from $1.1 billion in 2005 to $1.5 billion in 2010. Digging into the forecast, RAID controller ICs, which include both ASICs and comm processors used in internal controllers and disk arrays, totaled just over $540 million in revenue in 2005, representing about 49% of total semiconductor market revenue. About 63% of this was derived from Fibre Channel RAID (all from external RAID), with another 28% from SCSI (both internal and external). The remainder was split between legacy ATA and the emerging SATA and SAS. Looking ahead, we expect the RAID controller IC market to grow the slowest among device categories at a 3% CAGR, from $542 million in 2005 to $619 million in 2010. Normal ASP erosion should be augmented by the shift to lower-ASP SAS and SATA controllers. Disk interconnect ICs, which are used both to connect the front and back ends of the storage array and for disk management, totaled $238 million in 2005 according to our estimate, about 22% of the total storage semiconductor revenue. We expect this segment to grow at a 5% CAGR to $307 million through 2010. Much of this growth should

207

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

be driven by SAS expanders used in the back end of SAS-based storage arrays; Fibre Channel ICs (controller, PBC, loop switch, and enclosure management) should decline slightly. On the Fibre Channel network side, switch ICs totaled about $195 million in 2005 and HBA controllers were another $119 million, about 28% of market revenue in total. Infrastructure ICs should grow at a 14% CAGR through 2010. Fibre Channel HBA ICs should grow at a slower 7% CAGR, partially due to the increasing mix of dual-port HBAs (where the ASIC costs less than two discrete single-port ASICs). Finally, we expect iSCSI ICs to grow from a small base in 2005 to $60 million in 2010 as that technology gets off the ground. Note that there is likely an opportunity for iSCSI silicon only in disk arrays, as the protocol leverages the Ethernet data network for transport. The server connection will have to change to support iSCSI as well, but this will likely be done as an enhancement to the Ethernet NIC (e.g., Broadcoms C-NIC) rather than a discrete iSCSI HBA or controller IC. Storage Semiconductor Competitors: The enterprise storage market is served primarily by custom ASICs, with limited penetration to date by merchant suppliers. In many cases, particularly in the Fibre Channel market, the system vendor is heavily involved in the design of the ASIC (using the ASIC partner as little more than a foundry in some cases)in fact, we would say this is true to a greater extent than in any other market in semiconductors outside of maybe game consoles. Theoretically, then, we could list QLogic, Emulex, Brocade, McData, and Cisco as semiconductor participants, but for the purposes of this report we will focus on the actual ASIC supplier. The top two chip suppliers for the enterprise storage market in 2005 were LSI Logic and IBM, with 29% and 16% share, respectively. Both of these vendors have fairly broad market coverage, supplying ASICs for Fibre Channel switches and HBAs as well as RAID controllers and disk arrays. LSI also has done a nice job winning designs for nextgeneration SAS controllers and expanders. Note that in addition to third party customers, IBM serves its internal enterprise storage business, while LSI serves its Engenio storage systems business. Following the two big ASIC suppliers was Intel, which held 14% share in 2005. Intel is the major supplier of RAID controllers, which are typically implemented as embedded ARM processors, and also sells embedded processors into disk arrays and switches. PMC-Sierra followed, also with 14%; PMC is a key supplier of disk interconnect ICs for disk arrays, and further acquired the Fibre Channel controller business of Avago in 2006 (note that PMCs 11% market share includes Avagos business). Freescale rounded out the top five with 7% share, derived primarily from embedded processors sold into disk arrays and switches. Among the smaller vendors, Vitesse held 5%, with a large but eroding presence in disk interconnect and an emerging SAS business. Silicon Image and Marvell participated with SATA bridge products, while AMCC serves its internal 3ware SATA RAID controller unit. The following table lists storage semiconductor competitors and the markets in which they participate. Disk Array Interface and Interconnect ICs Vitesse LSI Logic PMC/Avago* QLogic Broadcom Fibre Channel HBA ASICs (partners for QLogic and Emulex) LSI Logic IBM Agere Fibre Channel Switch ASICs (partners for Brocade, McData*, Cisco, and Engenio) IBM LSI Logic Agere

RAID Controllers and Processors

Communications Processors

Intel LSI Logic IBM AMCC Promise/Marvell Vitesse/Adaptec Broadcom

Intel Freescale IBM Toshiba AMCC

*PMC acquired Avagos Fibre Channel Controller business in 2006. Brocade announced in 2006 its intention to acquire McData.

208

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Storage Storage
Storage Semiconductor Market Trends
SATA RAID controllers have made some inroads at the low end; the more meaningful transition will be from SCSI to SAS. ROC solutions seek to integrate the IOP and hard disk controller, and are enabling RAID on the motherboard. SANs are increasingly the architecture of choice for high-end enterprises, and continue to take share from direct-attached. Storage arrays are getting more intelligent, especially as Fibre Channel-based arrays increase in penetration. The transition to 4 Gbps Fibre Channel is well under way. Standard products have not gained much traction outside of a few pockets in RAID controllers and disk arrays. OEMs are looking to merge IP with SCSI and Fibre Channel.

SATA RAID controllers have made some inroads at the low end; the more meaningful transition will be from SCSI to SAS. The vast majority of RAID controllersover 70% in 2005utilize the SCSI protocol. This is because SCSI drives have been the only drives with enough performance for enterprise requirements. In the past few years, however, drive and storage OEMs have come to market with systems using the Serial-ATA (SATA) protocol (while it has already replaced ATA in performance PCs, SATA has made some inroads at the low end as a low-cost alternative with good-enough features and performance for small and medium business). The big transition, however, will be from SCSI to SAS (Serial Attached SCSI). SAS drives and RAID systems began to hit the market in volume in 2005, and should begin to replace SCSI both in servers and within disk arrays. Note that because it is a serial protocol, SAS also enables lower-cost systems that can bypass Fibre Channel altogether, though these will likely target the same small and medium business customers where SATA has had success.

ROC solutions seek to integrate the IOP and hard disk controller, and are enabling RAID on the motherboard. ROC (RAID-on-chip) solutions integrate the disk interface, embedded microprocessor (IOP), and PCI-X/PCI-E functionality on a single chip. These designs are now in the market from a number of vendors, especially for newer SATA and SAS controllers. They primarily target the internal RAID market, and are specifically intended to enable RAID-on-motherboard (ROM). ROM takes ROC one step further, by incorporating RAID functionality directly on the motherboard. This reduces bill-of-material costs for the server and also allows

209

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

for smaller form factor server designs. We believe these solutions will be used primarily at the low end of the server market. Note that the major server chipset vendors (Intel and Broadcom) are both pushing RAID-onmotherboard solutions. SANs are increasingly the architecture of choice for high-end enterprises, and continue to take share from direct-attached. The most obvious trend in storage has been the growth of Fibre Channel-based storage area networks. These are more robust than direct attached solutions and thus have been gaining traction in the high-end enterprise. While this segment is not fully penetrated just yet, the major HBA and switch vendors are beginning to expand into the small-andmedium business segment as well in order to increase the served market opportunity, though lower-cost SATA and SAS based systems could prove to be more effective solutions for this segment. Storage arrays are getting more intelligent, especially as Fibre Channel-based arrays increase in penetration. Most storage arrays connect via direct attached topologies and thus support only SCSI, ATA, SATA or SAS. As SANs have gained traction, however, more and more arrays sit within Fibre Channel networks and thus need to support the more robust features of the SAN. This is creating additional opportunities for IC providers, which are serving this market with port bypass controllers, loop controllers, and enclosure management ICs. The transition to 4 Gbps Fibre Channel is well under way. SAN technology was originally based on Fibre Channel running at 1.0 Gbps. As adoption has grown, the market quickly moved to 2.0 Gbps; the vast majority of systems sold in 2005 used 2.0 Gbps links. 4.0 Gbps Fibre Channel was next to roll out, with HBAs, switches, and disk arrays now shipping in high volume from a multitude of vendors. The transition to 4.0 Gbps will likely take 23 years, and the competitive landscape should shift to vendors of 4.0 Gbps along the way. Standard products have not gained much traction outside of a few pockets in RAID controllers and disk arrays. Most storage equipment designs use a combination of ASICs, microprocessors and FPGAs, sometimes employing ASSPs for specific functions. The projected unit/port growth of the market has prompted a handful of comm IC companies to plow additional R&D dollars into storage ASSPs and we are seeing positions build out steadily right now. Most are starting either with solutions for disk arrays or RAID controllers, as market share in Fibre Channel HBAs and switches remains concentrated in the hands of players with in-house ASIC capability. To date, the market has remained heavily ASIC-based, due primarily to the fact that the major system vendors view silicon and hardware design as key competitive advantages. OEMs are looking to merge IP with SCSI and Fibre Channel. OEMs are now developing system based on iSCSI, FCIP and other protocols to merge the advantages of Fibre Channel SAN systems and IP. The technology challenges are substantial, however, as Ethernet/IP networks approach most traffic with a best case effort in mind and as a result dont meet the predictable performance requirements of storage. Note that there is likely only an opportunity for iSCSI silicon in disk arrays, as the protocol leverages the Ethernet data network for transport. The server connection will have to change to support iSCSI as well, but this will likely be done as an enhancement to the Ethernet NIC (e.g., Broadcoms C-NIC) rather than a discrete iSCSI HBA or controller IC.

210

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Telecom/Datacom Telecom/Datacom
We now move from networking to wide area communications technologies. The WAN blends legacy equipment initially designed for voice telecom with newer technologies designed for data. Although the transmission media used is similar, the technologies and equipment that sit behind it are different: voice networks are generally circuit switched while data networks are generally packet switched. The wide area network can be thought of in three segments: long-haul, used for city-to-city or countryto-country transport, metro, used for intra-city transport, and access, used to connect individual enterprise and residential customers to the network. In each case, the traditional voice infrastructure has been overlaid with data networking equipment used to switch, route, and transport data alongside voice.

We now move from networking equipment used in discrete locations to the telecom and datacom realm, the wide area networks (WANs) that move voice and data traffic over longer distances. These networks are a lot older than the data-centric networks we have discussed so far; they are also a lot more challenging to design, build, and maintain. An extra layer of complexity is added with the blending of legacy technology originally designed specifically for voice telecommunications with newer technologies designed to send data traffic over the public telecom network. Voice Networks Wide area networks began as strictly voice networks, designed to offer local and long distance telephone service. As such, they emphasize the requirements of high-reliability voice telecommunications. In these networks, quality-ofservice is a key feature, as tolerance for latency, information loss and variations in speed of delivery are extremely low among telephone users. Further, information must be delivered strictly in the order it is received and it must be able to flow in both directions simultaneously (known as full-duplex). On the other hand, bandwidth requirements are minimal (a telephone line uses just 64 Kbps of bandwidth, less when compressed) and usage patterns are fairly stable and predictable. Thus, while voice networks have historically been expensive to install and run, they can be operated at close to full capacity (around 80%-90%), reducing the need for extensive headroom in the network. Voice networks have historically been circuit-switched; the transmission is enabled by establishing a dedicated communication channel between devices on a network. Such channels are established on demand and as available

211

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

and provide continuous and exclusive access between the terminal devices until the connection is terminated. In other words, upon initiation of a telephone call, a dedicated channel through the transmission network and switching matrices is established and a prescribed level of bandwidth is guaranteed for the duration of the connection. Data Networks With the growth of Internet data technologies such as E-mail, World Wide Web, instant-messaging and E-commerce, service providers have made significant investments to layer data network technologies on top of their voice infrastructure. Though the expertise and expense required to run data over voice-centric networks are significant relative to pure data networks, providers saw this as a better option than building entirely new data-only networks. We therefore find a public data network with a blend of voice and data equipment. In general, data traffic travels over the same pathways that voice traffic does in the WAN, though the equipment that handles data traffic at each point-of-presence (POP) may be different. Data networks do not require the same quality-of-service that voice networks do, and in fact data traffic does not necessarily need to be delivered in the same order it was sent. On the other hand, bandwidth requirements are large and extremely variable (often referred to as bursty), requiring lots of headroom in the network. In general, data traffic is better suited to packet switched networks, which involve encapsulating data in fixed- or variable-length packets and sending them across a shared network. Each packet contains headers encoded with a destination address and is switched at each node in the network over the most appropriate and available physical circuit. It bounces around in the network this way until it reaches the destination address. That is the big difference between circuit-switched and packet-switched networks. Packets travel through the network independent of each other, and in fact, packets addressed to the same destination may travel different physical paths, depending on the availability of network segments and the level of congestion across each network segment. As the route through the network is constantly changing, different packets may meet with varying levels of delay during transmission through the network. As a result, packet switching does not guarantee a prescribed level of bandwidth and packets are transmitted on a best efforts basis only; however, they are generally more efficient, as equipment is shared across a greater number of transmissions and as less dead air is transmitted. Network Segments, Speeds and Protocols The wide area network is usually thought of in three segments: long-haul, used for city-to-city or country-to-country transport; metro, used for intra-city transport; and access, used to connect individual enterprise and residential customers to the network. The protocols used to traverse these networks vary somewhat, though there is a lot of overlap. In each case, the traditional voice infrastructure has been overlaid with data networking equipment used to switch, route and transport data such as E-mail, Web, VPN and other data traffic. In the case of the access network, traditional cable video networks have been overlaid with data networking equipment in a similar fashion. In general, the closer to the core (long-haul portion) of the network, the faster the transmission speed. This is because fewer fiber connections are laid between cities than within them and thus the level of aggregationwhich translates into higher bandwidth requirements and faster speedsis greater for networks that span longer distances. Long haul networks typically employ transmission speeds of 2.5 Gbps (OC-48) or 10 Gbps (OC-192) and use the SONET protocol for transport over optical fiber. Metro networks also use SONET for transport, usually at speeds that range from 155 Mbps (OC-3) to 2.5 Gbps (OC-48). Some portions of the metro network use slower T3 (54 Mbps) or even T1 (1.5 Mbps) lines running on copper wire or optical fiber. Carriers also use these lines to connect to wireless basestations to reach cellular customers, and for large enterprise customers who want a direct connection to the WAN. Running on top of these SONET and T1 transport networks might be other protocols such as ATM, which is used for multi-service switching of data, voice, and video. The access network is a bit different, as it casts a much wider net and is used to reach individual voice and data customers, and therefore must interface with customer-owned equipment (which usually cant be controlled by the service provider). Speeds and protocols include 64 Kbps analog voice (over copper wire), 56 Kbps analog modems (over the same copper wire), 384 Kbps-6 Mbps DSL (again over the same copper wire but using separate frequencies), 1-10 Mbps cable modems (over coaxial cable), or newer 100-1000 Mbps PON/FTTH (over fiber). Depending on the source, destination and access technology, information sent over the WAN may employ a number of protocols in order to reach its destination. For example, an E-mail sent from our CIBC offices to a friend on the West Coast will first exit our corporate LAN via an access router, travel over a T1/T3 line in Verizons (or another LECs) network, be aggregated into an OC-3 or OC-12 SONET signal for transport to a larger network node where it will be further aggregated (multiplexed) into a higher-speed OC-48 or OC-192 SONET signal for transport over MCIs (or some other IXCs) long-distance network. Once it reaches the West Coast, the SONET signal will be disaggregated (demultiplexed) into a lower-speed SONET signal in AT&T/SBCs (or another LECs) network, and switched on an ATM network into a DSL signal for transport to the recipient.

212

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Telecom/Datacom Telecom/Datacom
The Voice Network
Digital Cross Connect (DXC)
Metropolitan T1/T3 / E1/E3 Network Metropolitan SONET/DWDM Network Long-Haul SONET/DWDM Network

Enterprise Voice Customers

PBX

Telco Switch

Home Voice Customers

Digital Loop Carrier (DLC)

SONET Add/Drop Multiplexer (ADM)

Access
Source: Cisco Systems, CIBC World Markets Corp.

Metro

Long-Haul

The voice network was originally designed to provide circuit switched communication between telephone users connected to a worldwide public network. It is designed for connection-oriented service, in which a user desiring service initiates a call by dialing the intended number, establishes a connection once the recipient picks up and then begins transmitting and receiving until one party hangs up, breaking the connection. As discussed above, this is a circuit-switched network; when the caller inputs the number, the service provider determines where on the network the recipient is located and allots the appropriate amount of bandwidth (usually 64 Kbps) over each segment of the call path for the duration of the call. This ensures a consistent, uninterrupted connection. Individual consumer telephones are connected to the public switched telephone network (PSTN) via copper twisted pair wire run from telephone poles or underground conduits directly into the home or office. These POTS (plain old telephone service) lines connect up to a line card within a telco switch housed in a carriers central office (CO) via a dedicated line. Sometimes, multiple telephone lines will be aggregated before transport to the central office. In residential networks, a digital loop carrier somewhere in the neighborhood aggregates many lines and forwards them to the CO. In a corporate environment, an enterprise PBX (private branch exchange) handles calls between extensions within the corporate system and forwards external calls to the CO. These voice aggregation devices will usually contain POTS line cards on the front end and a T1 network uplink on the back end, which will connect directly into the service providers network.

213

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Once the call reaches the central office, the telco switch (also called a class-5 switch or a CO switch) aggregates and switches calls. Calls headed to a local-area recipient will be connected at this juncture. Calls bound for destinations not connected to that central office are sent into the metro area network (MAN), which uses T1/T3 lines or low-speed SONET for transport. Digital cross connects located on the MAN aggregate and switch this traffic, and calls headed to recipients in the same city will be switched and sent to the recipients central office for call termination. These cross connects also interface with the long haul network, which uses higher-speed SONET transport equipment. The SONET ring will connect the call to other MANs around the country to enable long distance service. Next-Generation Voice Network Technologies (Includes Transport) Note that our network diagram shows a simplified representation of the traditional public voice network. A number of new technologies are being implemented to make the network faster and more efficient. Note that many of these are motivated by the desire to make voice networks more flexible when handling data and voice traffic, and thus are not pure voice or transport technologies. Some examples include: Multi-Service Provisioning Platforms (MSPPs): MSPPs enable the evolution of the optical network to support increasingly data- and voice-centric traffic streams. They combine SONET add/drop multiplexer, digital cross connect, switching, and even routing functions in a single device, and are used throughout the carrier metro and edge network. Optical Cross Connects (OCCs): Optical cross connects switch optical signals without performing the optical-electrical-optical (OEO) conversion that is performed in SONET add/drop multiplexers or digital cross connects. All switching is done in the optical realm. These switches tend to be used in the core of the network, where little intelligence is required to move signals around the network. VoIP for Transport: Voice-over-IP (VoIP) brings the benefits of packet-switched networking to traditional voice telephony. With VoIP technology, calls are digitized and translated into packets, and then are addressed like data packets and sent out over the data network. Most VoIP implementations to date have been behind the scenes within service provider networks for transport, especially for international calls. Service providers simply convert standard TDM-based voice calls to IP once the call reaches their network. They then transport the packets over the data network and switch them back to TDM somewhere near the destination for call termination. This leverages the efficiency of the data network and is transparent to the customer. Residential VoIP: Carriers are looking to offer lower-cost residential voice service by implementing VoIP at the customer premise, leveraging the installed base of analog phones and copper wire but moving call transport over to lower cost IP networks. Under this model, traditional analog phones hook into a specialized VoIP media gateway device (also called a VoIP terminal adapter or CPE device) that is connected to the customers broadband modem (which we will discuss in the data network section on the next page). This gateway translates voice into IP and sends it over the broadband connection. Back in the carriers network, a softswitch in the carrier network enables call features. As in the pure-IP enterprise implementation, calls placed over the IP network bound for a recipient not on the network must be terminated by the service provider; if the recipient is on a TDM network, the call must be translated back to TDM using a media gateway in the carriers network. Cable Voice: Cable MSOs are rapidly moving toward triple-play services, offering voice service in addition to the video and data they provide today. The long-term topology for this functionality is VoIP, with a terminal adapter or voice-enabled cable modem at the customer premise packetizing voice traffic and sending it out over the DOCSIS network. Many MSOs are rolling out this service today. In the interim, however, many MSOs have rolled out a bridge solution using circuit-switched voice; under this topology, a Network Interface Device (NID) is installed at the customer premise, which feeds traditional telephony signals over the coax network. The MSO then strips the voice signal out at the head end, and uses traditional class-5 switches and digital cross connects to enable voice service. Sometimes branded digital voice, this service is rapidly giving way to full cable VoIP.

214

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Telecom/Datacom Telecom/Datacom
The Data Network
WAN/LAN Switch

Web Hosting Companies Router Enterprise Data Customers Analog Modem


Servers

Router
2

Long-Haul SONET/DWDM Network

Metropolitan SONET/DWDM Network

T1/T3

Core Routers

Access Routers RAC WAN Switches

WAN Switches

DSL Modem DSLAM Cable Modem PON ONT CMTS PON OLT SONET Add/Drop Multiplexer (ADM)

Home Data Customers

Access
Source: Cisco Systems, CIBC World Markets Corp.

Metro

Long-Haul

The public data network uses the same physical network as voice does, but overlays data equipment both at the customer and at service provider points-of-presence to enable data capability. Conceptually, data equipment sits behind voice equipment at every point in the network, and uses the voice telecom equipment discussed in the preceding section for transport between network nodes. In the data network, the POTS and fiber infrastructure run by the RBOCs and the coaxial cable infrastructure run by the MSOs are used by consumers to connect to the public data network. Consumers can use analog modems or DSL to connect over the POTS infrastructure, cable modems to connect over the coaxial cable infrastructure, or PON ONTs or ONUs (passive optical networking optical network terminals or units) to connect over new FTTH deployments by telcos. These lines connect into remote access concentrator, DSL access multiplexer (DSLAMs), cable modem termination system (CMTS), or PON OLT (optical line terminal) data traffic aggregators located at the central office or cable headend. On the enterprise side, corporate customers will usually bypass the access network, connecting directly to the public network over a T1/T3 line or metro Ethernet. The more sophisticated data security, reliability, and bandwidth requirements of corporate enterprises warrant these always-on connections. Note that these T1/T3 lines can carry both voice calls and data traffic and are usually sold as a package by telco providers.

215

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Once in the public data network, packets are routed and switched by access routers and carrier-class WAN switches located at network POPs. These switches will typically use ATM, though newer designs use IP or MPLS (Multi-protocol Label Switching). The transport protocol here would normally be T1/T3 or low-speed SONET, as switching protocols like ATM, IP, and MPLS operate at layer 2 and therefore need to be encapsulated into a layer 1 technology like T/E, SONET, or Ethernet WAN. If the destination address is in a distant part of the WAN, packets are forwarded to the long haul network, where core routers and larger WAN switches perform similar functions. In the long haul network, high-speed SONET is used for transport. Note that packets that are bound for World Wide Web addresses are forwarded to web hosting companies or corporate web servers, which connect the public data network via enterpriseclass switches and routers the same way enterprise LANs are. Data networks are packet-switched, meaning that rather than establishing a dedicated pathway between sender and recipient, packets are addressed and sent out onto the network, where they are routed and switched until they reach their destinations. In a pure data network like an Ethernet LAN, both the switching and transport technologies are packet-switched. This makes for a very efficient network, since resources are only allocated when they are needed (i.e., the network only lights up when there is data to send). In WAN networks, however, voice-centric technologies like T/E and SONET are used for transport over long distances, and these are circuit-switched by design. This means that data packets sent over the WAN will be switched with packet-based technologies at network points of presence, but will actually be transported over a circuit-switched network. This is even clearer when we consider access technologies, which by definition use dedicated lines between the user and the service provider. Next-Generation Data Network Technologies Note that our network diagram shows a simplified representation of the traditional public data network. A number of new technologies are being implemented to make the network faster and more efficient, but did not make it into our diagram or the discussion above. Some examples include: Ethernet Access: Service providers now offer Ethernet-based access lines to enterprise customers to replace their T1/T3 lines. MSPPs: Discussed in the voice network discussion above, multi-service provisioning platforms are designed to better handle data traffic over the voice network. Video Servers and IPTV Set-Top Boxes: Next-generation IPTV services will require video servers to be installed in the metro network to deliver video content. IP set-top boxes are required at the customer premise to turn the data packets into video signals. IMS (IP Multimedia Subsystem): IMS is an attempt at creating a telephony standard infrastructure for IP services that is expected to evolve to a replacement to circuit switched voice. The goal of IMS is to provide today's telephony services along with all the services, current and future, that can be found on the Internet. IMS uses Internet Protocols (IP and SIP) to accomplish this goal. Service providers would like to use IMS infrastructure to provide ubiquitous access to services via wireless, wireless, and cable.

216

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Telecom/Datacom Telecom/Datacom
Types of WAN Equipment - Voice Networks
Digital Loop Carrier (DLC) - An aggregation device used in telecom networks to combine multiple telephone lines and send them over a single line to the central office. DLCs sit between the central office and customers. Private Branch Exchange (PBX) - An in-house telephone switching system used in enterprise networks to interconnect telephone extensions to each other and to the public switched telephone network. Telco Switch - Also called a class 5 switch, or a CO switch, these are the primary carrier access devices used in legacy voice networks. They are large-scale network devices containing thousands of line cards used to connect directly to voice customers. Digital Cross Connect (DXC/DCC/DCS) - Network device used by telecom carriers and large enterprises to switch and multiplex low-speed signals onto high-speed lines and vice versa. Typically used to aggregate several T1/T3 lines into higher speed optical SONET lines. SONET Add/Drop Multiplexer (ADM) Optical device used in fiberbased networks to enable new SONET signals to come in and existing signals to go out. ADMs are placed in intermediate positions on the SONET ring and serve as on/off ramps to other rings. Note that SONET ADMs are increasingly being replaced by Multi-Service Provisioning Platforms (MSPPs), which add digital cross connect, switching, and routing functions.

Types of Voice Networking Equipment Digital Loop Carrier (DLC): An aggregation device used in residential telecom access networks to combine multiple telephone lines and send them to the central office. DLCs sit between the central office and customers. Private Branch Exchange (PBX): An in-house telephone switching system used in enterprise networks to interconnect telephone extensions to each other and to the public switched telephone network. Telco Switch: Also called a class 5 switch or CO switch, these are the primary carrier access devices used in legacy voice networks. These large-scale systems contain thousands of line cards that connect directly to customers. Digital Cross Connect (DXC/DCC/DCS): Network device used by telecom carriers and large enterprises to switch and multiplex low-speed signals onto high-speed lines and vice versa. They are often used to aggregate several T1/T3 lines into higher speed optical SONET lines. SONET Add/Drop Multiplexer (ADM): Optical device used in fiber-based telecom networks to enable new SONET signals to come in and existing signals to go out. As signals pass through the device, some are "dropped" by splitting them from the line, others can be "added" and directed to other destinations. ADMs are placed throughout the SONET ring and serve as on/off ramps to other rings. Note that SONET ADMs are increasingly being replaced by multiservice provisioning platforms (MSPPs), which add digital cross connect, switching, and routing functions.

217

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Telecom/Datacom Telecom/Datacom
Types of WAN Equipment - Data Networks
Remote Access Concentrator (RAC) & Analog Modem CPE Equipment used to send data traffic over POTS twisted pair copper wire in an analog format at speeds of up to 56Kbps. The RAC is housed at the central office while the modem CPE resides at the customer. Digital Subscriber Line (DSL) Access Multiplexer (DSLAM) and DSL Modem CPE - Equipment used to send data traffic over POTS twisted pair in a digital format at speeds of 1 Mbps or more. Cable Modem Termination System (CMTS) and Cable Modem CPE - Equipment used to send data traffic over coaxial cable in a digital format at speeds of up to 10 Mbps. Passive Optical Networking (PON) Optical Line Terminal (OLT) and Optical Network Terminal (ONT) - Equipment used to send data over fiber-to-the-home networks at speeds of 50-100 Mbps+. The ONT resides at the customer with the OLT in the carrier network. Access Router - Intermediate-speed router used at the edge of the network, both for enterprise network access and within service provider points-of-presence for routing in the metro network. Core Router - High-speed router used in the core of the network. WAN Switch - Network device that switches data, voice, and video traffic, used in carrier and large enterprise network backbones. Most WAN switches use the ATM protocol, which uses a circuit-oriented connection and encapsulates information in fixed-length cells.
Types of Data Networking Equipment Remote Access Concentrator (RAC) and Analog Modem Customer Premise Equipment (CPE): Access equipment used to send data traffic over POTS twisted pair copper wire in an analog format at speeds up to 56Kbps. Digital Subscriber Line (DSL) Access Multiplexer (DSLAM) and DSL Modem CPE: Access equipment used to send DSL data traffic over the POTS copper twisted pair infrastructure in a digital format at speeds of 1 Mbps or more. Cable Modem Termination System (CMTS) and Cable Modem CPE: Access equipment used to send DOCSIS cable modem traffic over coaxial cable video networks in a digital format at speeds of up to 10 Mbps or more. Passive Optical Networking (PON) Optical Line Terminal (OLT) and Optical Network Terminal (ONT): Equipment used to send data over fiber-to-the-home networks at speeds of 50-100 Mbps or more. The ONT resides at the customer with the OLT in the carrier network. Access and Core Routers: Service provider routers move traffic around the wide area data network. Access routers (1-10 Gbps) are used at the edge, both for enterprise network access and within service provider points-ofpresence for routing in the metro network. Core routers (10 Gbps+) are used in the core of the network. WAN Switch: Sometimes called a multi-service switch, this device switches data, voice, and video traffic used in carrier and large enterprise network backbones. They are multiprotocol devices, usually incorporating ATM.

218

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Telecom/Datacom Telecom/Datacom
Carrier Cap-ex Digesting the Bubble Spending on telecom/datacom is heavily tied to carrier cap-ex, usually pegged as a percentage of revenue. In the late 1990s, however, telecom deregulation and the growth of the Internet sparked a sharp increase in spending. As price competition heated up and revenue growth stalled, carriers quickly pulled in spending plans, and have been digesting those investments ever since.
Service Provider Revenue
$1,400 40% $300

Service Provider Cap-ex


40%

Wireless Services Revenue


$1,200

Wireless Cap-Ex
30%
Service Provider Cap-Ex Revenue in Billions

Wireline Services Revenue Wireless YoY Growth Wireline YoY Growth

Wireline Cap-Ex
$250

35%

Wireless Capital Intensity Wireline Capital Intensity


30%
Capital Intensity (Cap-ex / Sales)

Service Provider Revenue in Billions

$1,000

20%

$200 25%

$800

10%

YoY Growth

$150

20%

$600

0%

15% $100 10% $50 5%

$400

-10%

$200

-20%

$00 2000 2001 2002 2003 2004 2005

-30%

$00 2000 2001 2002 2003 2004 2005

0%

Source: Company reports and CIBC World Markets

Source: Company reports and CIBC World Markets

Spending on telecom and datacom equipment is heavily tied to service provider capital expenditures, which in turn is typically pegged as a percentage of revenuehistorically around 15%-20% of revenue. In the late 1990s, however, telecom deregulation and the growth of the Internet sparked a sharp increase in carrier spending, as heavily funded start-up telecom providers began to build competitive networks and incumbents increased their spending in order to keep up. In 2000, spending by carriers worldwide peaked at over 28% of industry revenueclearly an unsustainable level of spending. Note that the cap-ex ramp was especially severe for U.S. service providers, which produced large negative free cash results despite record revenue. In 2001, carriers began to aggressively pull in capital spending plans as it became clear that revenue growth, despite the array of new data and wireless services being provided, would not be sufficient to warrant the level of investment of 2000. The situation worsened in 2002 as price competition increased, and the cuts in capital spending became even more urgent. Also in 2002, the decline in wireline capital spending spread to wireless spending, which had been strong through 2001. Both wireline and wireless spending took another step down in 2003, as carriers sought a longterm sustainable level of capital spending as a percentage of sales. Cap-ex finally grew in 2004 and 2005 as carriers re-entered investment mode to build next-generation networks to provide triple play services.

219

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Telecom/Datacom Telecom/Datacom
Service Providers Major Issues
Technologies driving the industry Focus remains on consolidating voice/data/video onto one network Next generation access technologies EPON in Asia; BPON/GPON in the U.S.; ADSL2+, VDSL, and VDSL2 worldwide Moving intelligence from the core of the network to the edges IP networks with the customers controlling much features and functionality Wireless will continue to drive the industry Cable remains the biggest threat MSOs preparing for broader rollouts of voice-over-cable The cable companies and telcos have much to lose from aggressive price wars and both sides have some advantages Triple Play, bundling and flat rate pricing are focuses Triple play includes voice, video, and data service Increases revenue per household Lowers cost Increases the TAM over time Consolidation and Regulatory issues Heavy consolidation in the U.S.: Verizon-MCI, SBC-AT&T-Bellsouth, SprintNextel, Level 3-Wiltel-Broadwing & others Inter-carrier compensation especially for VoIP Universal Service Funding

The telecom industry remains in a state of transition as carriers address increasingly data-centric connectivity demands and attempt to fight off competition from cable and satellite. All service providers are moving toward triple play services, looking to provide voice, video, and data. The major issues facing carriers include: Technology issues: Service providers are squarely focused on consolidating voice/data/video onto one network in order to lower the cost for existing services and branch into new services. This has required the adoption of next generation access technologies, including EPON in Asia, BPON/GPON in the U.S., and next-generation DSL worldwide. Behind the scenes, carriers continue to move their networks to more flexible IP-based equipment, and are moving intelligence from the core of the network to the edges. Wireless is still viewed as the big growth area for the industry. Threats: The biggest threat facing carriers is from cable voice, as cable MSOs prepare for broader rollouts of VoIP. Note that the cable companies and RBOCs have much to lose from aggressive price wars and both sides have some advantages, which should prevent more aggressive behavior. Non-traditional service providers like Vonage and Skype present additional challenges; the effects will be less widespread but will hurt on the margins. Market factors: Carriers remain focused on ARPU, in particular adding services to increase the TAM over time. On the cost side, cap-ex frugality has eased but maintenance and other variable costs remain an important focus area. Consolidation and regulatory issues: There has been massive consolidation in the U.S. market. On the regulatory front, inter-carrier compensation and Universal Service Funding are still issues as carriers try to fend off VoIP.

220

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Telecom/Datacom Telecom/Datacom
Capital Spending Outlook With competition still fierce and the revenue growth outlook modest, carriers have trended spending down toward the mid-teens as a percentage of revenue. However, providers have moved beyond conserving cash and are looking to position their business models and networks for long-term competitiveness and profitability. With triple-play clearly the goal, we are seeing a shift within the budget toward next-generation broadband, VoIP and the early stages of IPTV. 3G wireless rollouts are well under way in developed markets. Metro and core networks are also being upgraded to support increased levels of traffic.
Carrier Cap-ex Forecast
$300 40% $250
Service Provider Cap-Ex Revenue in Billions

Wireless Cap-Ex Wireline Cap-Ex Wireless Capital Intensity Wireline Capital Intensity

35%

30% $200 25%


Capital Intensity (Cap-ex / Sales)

$150

20%

15% $100 10% $50 5%

$00 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

0%

Source: RHK, CIBC World Markets Corp.

Carriers have completely reworked their strategies in the post-bubble period, and have been trending capital spending down toward the mid-teens as a percentage of revenue. The rush to ramp up network capacity has largely abated, but so has the focus on cash conservation. At this point, carriers are focused on migrating their networks (and their business models) to support more sustainable growth and margins over time. Wireline carriers are challenged with balancing (1) declining voice revenues as a result of wireless substitution and growth in VoIP alternatives with (2) the need to expand their networks capacity to handle the growing demand for data and video (IPTV) services. These challenges are typically manifested with an internal shift in CapEx spending from old legacy equipment to next generation IP networks. Net net, growing cash flow constraints and increasing margin pressures cap spending beyond legacy substitution. Wireless carriers in developed markets are challenged with lack of subscriber growth and are forced to manage a transition to higher capacity 3G networks within increasingly constrained capital spending budgets. In emerging markets, subscriber trends are still positive, but stiff OEM competition enables carriers to get more for less and increase budgets slowly. We see upside to our estimates only if network traffic shows stronger than expected growth. In terms of our forecast, we believe global carrier cap-ex will grow 7% in 2006, followed by a flat year in 2007 and then declines of 2%-3% in 2008-1010. Overall, we forecast cap-ex spending to be flat over the forecast period, hovering near $220 billion.

221

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Telecom/Datacom Telecom/Datacom

Five Key Markets:


Line cards used in large telco switches to connect residential voice customers to the public voice network.

Central Office Line Cards Modems

Data access devices used to connect individual residential customers to the Internet over POTS copper wire infrastructure or coaxial cable networks. We also include faxes due to their similarity to analog modems.

Fiber access technology used to connect residential customers to the Internet using optical fiber.

PON

A broad category of infrastructure equipment used for switching, aggregation, routing, and transport in voice and data networks.

Communications Infrastructure Voice-over-IP

Equipment used to send voice traffic over an IP-based data network.

We divide telecom and datacom equipment into five key segments. The first three are access technologies with transparent applications: POTS for voice access and modems and PON for data access. The fourth is a broad category we term communications infrastructure, which includes switching, aggregation, routing, and transport equipment used in voice and data network infrastructure. The final category is VoIP, which includes equipment layered onto the other categories to provide voice service over data-centric networking equipment. Central Office (POTS) Line Cards: The least common denominator in voice networks, POTS line cards sit in telco switches at the central office and interface directly with customer phone lines. Modems (includes fax): Datacom access devices used to connect individual users to the Internet and data WANs. Analog and DSL modems use the POTS telephone system; cable modems use coaxial cable networks. PON: Fiber access technology used to connect residential customers to the Internet using optical fiber. Communications Infrastructure: A broad category of equipment used for switching, aggregation, routing, and transport in voice and data networks. Note that equipment included in this category is used both in service provider networks and in enterprise networks for trunking, backhaul, and access to the public network. We chose to include all infrastructure equipment here rather than in the networking section due to its similarity to service provider equipment and the difficulty in distinguishing between multi-use equipment. Voice-over-IP: Specialized equipment used to send voice traffic over IP-based networking equipment.

222

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Central Office Line Cards Central Office Line Cards


Central Office Line Card Unit Forecast
Central office line cards are a declining business; normal churn and emerging market growth are more than offset by secular declines in developed regions. Wireless replacement of wireline, broadband penetration replacing modem lines, Internet replacing fax lines, and VoIP all contribute to the secular decline. In total, we expect central office line cards to decline at a CAGR of 13%, from 29.4 million in 2005 to 15.0 million ports in 2010.
100,000 90,000 80,000 70,000
Unit Shipments (000s)

Central Office Switch Suppliers


Others 23 % Alcatel 22 %

Siemens 12 %

Lucent 16 %

Nortel 12 % Ericsson 15 %
Source: Dataquest, CIBC World Markets Corp.

Central Office Line Cards

60,000 50,000 40,000 30,000 20,000 10,000 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: Dataquest, CIBC World Markets Corp.

Central Office Line Card Unit Forecast: The market for central office line cards, also called POTS (plain old telephone system) line cards, is driven by the number of net new phone lines added plus replacement of depreciated equipment (churn). In theory, this market should grow roughly at the rate of population, but certain trends accelerated this growth leading up to 2000, such as emerging market growth, fax machines, analog modems and installations by emerging competitive carriers following telecom deregulation. POTS line card unit shipments peaked at 180 million units in 2000, but then declined to a fraction of this number as the telecom bubble burst. Looking ahead, we believe the market will continue to decline and expect a negative five-year CAGR of 13%, with shipments falling from 29.4 million units in 2005 to 15.0 million units in 2010. A number of factors are now driving the decline. Residential users are canceling second phone lines as they install DSL or cable modem service, which also offers voice in addition to data. Increased Internet usage is also reducing demand for separate fax lines. Cellular phones have eliminated the need for additional phone lines in many households. More advanced enterprise phone systems are also eliminating a huge constituency usually terminated at the central office. Residential VoIP services are now widely available from telecom, cable, and third party providers and eliminate the need for a POTS line altogether. Finally, multi-port solutions are reducing the units required to serve the same number of users. Central Office Switch Suppliers: The largest supplier of central office switches in 2005 was Alcatel with 22% revenue share. Lucent (acquired by Alcatel in 2006) was second with 16%, followed by Ericsson with 15%, Nortel with 12%, and Siemens with 12%. Other vendors include NEC, Fujitsu, Hitachi, Huawei, Datang, and ZTE.

223

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Central Office Line Cards Central Office Line Cards


The ringing IC generates the telephone ring on the user telephone when an incoming call is received on that line

A microcontroller is used to control the line card

Ringing IC Telephone Line

MCU

SLIC
Source: Dataquest, CIBC World Markets Corp.

PCM Codec

Switch Backplane

This SLIC is a line driver that interfaces with the telephone line, sending and receiving the analog signals to and from the telephone user

The codec performs the analog-to-digital and digital-to-analog conversion to feed the call into and out of the PSTN (public switched telephone network)

POTS signals come off a telephone line and interface with a SLIC, which transmits and receives the voice signal to and from the telephone user. A ringer IC also sits near the SLIC and generates the ringing of the telephone set. Once received, a PCM codec converts the signal to a digital format to be sent to the switch backplane. SLIC: The SLIC performs the analog signal reception and transmission and interfaces with the telephone line. POTS line card SLICs are specialized devices that support standard BORSCHT functions (battery feed, over-voltage protection, ringing, signaling, coding, hybrid and testing), as well as key power management technologies, including thermal management, battery switching and switching regulation. Ringing IC: The ringing IC generates the telephone ring on the user telephone when an incoming call is received. Ringing SLIC (not pictured in our block diagram): A ringing SLIC combines the functions of a SLIC and a ringing IC. Codec: The codec performs the analog to digital conversion of the telephone signals and sends them into the PSTN. Microcontroller: A standard MCU is used to control the line card.

224

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Central Office Line Cards Central Office Line Cards


Central Office Line Card Semiconductor Forecast
We expect the market to decline at an 11% CAGR, from $232 million in 2005 to $131 million in 2010. ICs should slightly outpace units, however, as the market moves to denser solutions which carry higher ASPs.
$400 SLIC/Line Driver $350 Codec

Central Office Line Card IC Competitors


Winbond 3% NEC 3% Others 8%

Freescale 4% IDT 6% Infineon 42 %

STMicro 11 %

$300
Semiconductor Revenue ($M)

$250

$200

Legerity 24 %
Source: Dataquest, CIBC World Markets Corp.

$150

$100

$50

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Central Office Line Card IC Market Trends


Discrete SLIC and ringing ICs are being integrated as ringing SLICs. The codec remains separate, as the SLIC requires high-voltage CMOS. Multi-port SLICs and codecs are becoming industry standard.

Source: Dataquest, CIBC World Markets Corp.

Central Office Line Card Semiconductor Forecast: We expect central office line card semiconductors to decline slightly less than units, as the market moves to denser solutions that carry higher ASPs. Still, we expect revenue to decline at an 11% rate, from $232 million in 2005 to $131 million in 2010. This excludes MCU content. Central Office Line Card Semiconductor Competitors: Only two major vendors remain in the CO line card IC market. Infineon, with its strong ties to Siemens, was the top supplier with 42% market share in 2005. Legerity was second with 24%. Other vendors include STMicroelectronics, IDT, Freescale, NEC, Winbond, Silicon Labs, Datang Microelectronics, and Oki Electric. Trends in Central Office Line Card Semiconductors Discrete SLIC and ringing ICs are increasingly being integrated as ringing SLICs. Integration is taking place at the front end, with ringers increasingly being integrated into the SLIC. We note that most vendors are leaving out the codec function, as the SLIC requires more expensive high-voltage CMOS while the codec does not. Multi-port SLICs and codecs are becoming industry standard. With per-port bill of materials cost and board space in the central office switches being key issues for service providers, OEMs have turned to IC vendors to supply multi-port devices.

225

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
Modems are used to connect individual PCs or small consumer/SOHO networks to the Internet. Modem CPE (consumer premise equipment) on the client side connects via copper twisted pair that has historically provided voice service or via coaxial cable that has historically provided video service. On the service provider side, a modem concentrator enables the data service and interfaces with the public data network.
Customer Premise
Copper Twisted Pair or Coaxial Cable

Service Provider
Uplink to Public Data Network/ Internet

Modem (Analog, DSL, Cable)

Modem Concentrator (RAC, DSLAM, CMTS)

Source: CIBC World Markets Corp.

Modems are the most common remote data access device, used to connect individual PCs or small consumer or SOHO networks to the public data network and the Internet. Unlike pure data networking equipment like Ethernet, modems are designed to send data traffic over networks that were originally intended for either voice or video. Modem CPE (consumer premise equipment) on the client side connects via copper twisted pair (from the traditional voice network) or coaxial cable (from the traditional cable video network) to a modem concentrator in the service provider network, which interfaces with the public data network. The service provider can be a traditional telecom carrier (Verizon, AT&T), a cable MSO (Comcast, Cablevision), or a specialized Internet service provider (America Online, Microsoft Network, Juno).

226

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
Three key types of modems exist:
Analog Modems - Analog modems use a traditional telephone line to place a call, which is received by the service provider modem, and transmits data in an audible analog format (like a fax machine). Transmission speed is typically 56 Kbps. Cable Modems - Cable modems use the existing coaxial cable network to transmit data in a digital format. Transmission rates vary with user congestion, but can reach 10 Mbps downstream and 1 Mbps upstream. DSL - Digital Subscriber Line uses the existing POTS infrastructure, but transmits data in a digital format using a higher, inaudible frequency on top of the traditional analog voice signal. Symmetric DSL (SDSL) transmits and receives data at the same rate, while Asymmetric DSL (ADSL) carries faster downstream rates than upstream rates. Newer, faster asymmetric protocols are referred to as VDSL (very high bit rate DSL). DSL speeds vary widely between a 1-10 Mbps for SDSL and ADSL and 50+ Mbps for VDSL/VDSL2.
Three key types of modems exist: Analog Modems: Analog modems use a traditional POTS phone line to place a call, which is received by the service provider modem housed in a remote access concentrator (RAC). Analog modems transmit data in an audible analog format, which is why you can hear it if you pick up the phone line, similar to a fax machine. Top transmission speed is typically 56 Kbps. Phone lines are designed to transmit 64 Kbps of voice (the 8 Kbps difference accounts for overhead). Analog modem CPE is usually embedded on the motherboard within the PC. Cable Modems: Cable modems use the existing coaxial cable network to transmit data in a digital format using the DOCSIS protocol, over the same infrastructure used to transmit video. Transmission rates vary with user congestion (slower when more users are online), but can reach 10 Mbps downstream and 1 Mbps upstream. DSL: Digital subscriber line technology uses the existing POTS infrastructure, but transmits data using a digital format in a higher, inaudible frequency on top of the traditional analog voice signal. On the service provider side, lines connect to a DSL access multiplexer (DSLAM), which connects directly to the public data network, usually using ATM or IP as a backhaul protocol. Two basic forms of DSL exist: symmetric, which transmits and receives data at the same rate; and asymmetric, which carries faster downstream rates. Newer, faster asymmetric protocols are referred to as VDSL (very high bit rate DSL). DSL speeds vary widely between 1-10 Mbps for SDSL and ADSL and 50+ Mbps for VDSL/VDSL2.

227

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
Broadband Growth Broadband service is rapidly displacing analog modems. Both DSL and cable modem service providers view data service as strategic to their business models, and have dropped prices to be extremely competitive vs. most phone-based Internet plans. Consumers are attracted by much higher data rates and always-on service.
Broadband Subscribers
200,000 180,000 160,000 140,000
Subscribers in 000s

Internet Connections
100% 90% 80% 70% 60% 50% 40% 30% VDSL/FTTP/Ethernet 20% 10% 0% 2000 ADSL / G.SHDSL Cable Modem

VDSL/FTTP/Ethernet ADSL / G.SHDSL Cable Modem

120,000 100,000 80,000 60,000 40,000 20,000 2000 2001 2002 2003 2004 2005

Analog Modem

2001

2002

2003

2004

2005

Source: DellOro, CIBC World Markets Corp.

Source: DellOro, CIBC World Markets Corp.

The past few years have seen a rapid shift to broadband Internet service, which includes cable modem and DSL service. In 2005, we estimate that broadband subscribers totaled 193 million worldwide, including 46 million cable modem subscribers, 130 million ADSL/SDSL subscribers, and 17 million VDSL and other high bit rate subscribers (includes FTTH/PON and metro Ethernet, though these are discussed in other sections). This was up over 35% from 2004 and represents a nearly fifteen-fold increase from the 13 million broadband subscribers in 2000. Wide availability and falling prices have been a huge part of broadband growth. Both DSL and cable modem service providers view data service as strategic to their business models and have dropped prices to be extremely competitive with most phone-based Internet plans, while providing a service that is far superior and doesnt occupy the users telephone line. Consumers are attracted by much faster data rates and always-on service. Also note that several government programs, especially in Asia, have subsidized broadband deployment, driving penetration higher in certain countries. In terms of penetration, in 2005 we estimate that broadband penetration was 63% on a worldwide basis, up from 53% in 2004 and 40% in 2003 (note that penetration back in 2000 was just 11%). Of the three largest geographies, broadband penetration is highest in EMEA at 68%, followed by Asia/Pacific at 63%, and North America at 59%.

228

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
Analog Modem and Fax Unit Forecast
We expect analog modems and fax units to stay essentially flat; shipments should peak in 2006 at 181 million units and decline to 172 million units in 2010. PC modem penetration should decline as broadband becomes more pervasive and as ISPs stop giving subsidies to PC OEMs. Multi-function peripherals should generate some incremental growth to offset declining PC penetration.
250,000
Remote Access Concentrators Embedded Analog Modem Fax and MFP Notebook Analog Modem Desktop Analog Modem

Analog Modem and Fax Suppliers


Desktops
Dell HP Lenovo Fujitsu/F-S Lenovo Gateway

Notebooks
Dell HP Toshiba Lenovo Acer Fujitsu/F-S

MFPs
HP Lexmark Epson Dell Canon Brother

Fax
HP Brother Panasonic Sharp Lexmark Canon

200,000

Unit Shipments (000s)

150,000

100,000

50,000

2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, DellOro, Dataquest, CIBC World Markets Corp.

Lucent/Alcatel Nortel 3Com Cisco UTStarcomm


Source: IDC, Dataquest, DellOro, CIBC World Markets Corp.

RAC

Analog Modem and Fax Unit Forecast: We expect the market for analog modems to face pressure as PC penetration rates continue to decline. Analog modem penetration in PCs has fallen steadily over the past few years, and should continue to drop as broadband becomes more pervasive and as ISPs stop subsidies to PC OEMs (e.g., AOL began scaling back its subsidies of analog modems in 2006). Outside of PCs, analog modems and faxes should see modest growth, primarily from multi-function peripherals which incorporate printer, scanner, copier, and fax functions in a single device. On the infrastructure side, port shipments have come down dramatically, as few service providers install modem concentrators except for maintenance. We estimate infrastructure ports will represent less than 1% of total in 2010. In total, we expect the market to remain roughly flat, declining from 175 million units in 2005 to 172 million units in 2010. Shipments probably peaked in 2006 at 181 million. Analog Modem and Fax Suppliers: Analog modems ship into a variety of applications including PCs, MFPs, faxes, and RACs. From a unit perspective, PCs remain the largest end market for analog modems. Hence, the largest modems suppliers are PC makers Dell, HP, Lenovo (Lenovo purchased IBMs PC business in 2005), Acer, Fujitsu/Fujitsu-Siemens, Toshiba, NEC, Apple, Gateway, and Sony. Outside of PCs, MFPs and faxes are also big consumers of modems and top suppliers include HP, Lexmark, Epson, Dell, Canon, Brother, Panasonic, and Sharp. On the infrastructure side, top suppliers of RACs include Lucent/Alcatel, Nortel, 3Com, Cisco, and UTStarcomm.

229

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems

Analog Modem

RJ-11 Phone Line

AFE / Codec Line Driver


MCU PCI Data Pump

Host System

Interfaces with the telephone line, sending the analog data signals between the modem and the RAC over the telephone line

Performs the analog-to-digital and digital-to-analog conversion to feed the modem data into and out of the host system (in a modem) or backplane (in a RAC)

RJ-11 Phone Line

Line Driver

AFE / Codec
MCU Data Pump

Remote Access Concentrator

Port 2 Port 3 Port N

System Backplane: interface to switching, routing, WAN trunking, and control/management cards

Source: CIBC World Markets Corp.

Note: Our modem block diagrams, semiconductor forecast, and associated discussion include all semiconductors included in modem CPE, in embedded modem/fax subsystems within MFPs, set-top boxes, fax machines, and other embedded fax/modems, as well as those modem-specific ICs found on remote access concentrator line cards. We do not include those devices used in switching, routing, WAN trunking, and control/management cards, as these are not protocol dependent and can vary greatly depending on the functionality of the system. We discuss these ICs separately in the communications infrastructure section. Line Driver: This analog SLIC interfaces with the RJ-11 copper twisted pair (phone line), performing transmission and reception functions. Analog Front End (AFE)/Codec: The AFE/codec performs the analog-to-digital conversion, interfacing between the line driver and the host system (in a PC modem) or switch fabric (in a RAC). Modem codecs are specialized processors and typically include a data pump, embedded microcontroller and PCI interface. Fax codecs are similar but will often incorporate different software and features.

230

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
Analog Modem and Fax Analog Modem and Fax Semiconductor Forecast Semiconductor Competitors
We expect analog modem and fax ICs to decline over the next 5 years as unit growth slows and IC price declines chip away at market revenue. PCs have already adopted softmodems; as embedded applications slowly follow suit codec/DSP revenue should suffer. Overall, we expect IC revenue to decline from $552 million in 2005 to $403 million in 2010, a -6% CAGR.
$700 Analog Modem and Fax ICs $600

Analog Devices 2% Austria Microsystems 4%

Others 4%

Agere 16 % Conexant 50 %

$500
Semiconductor Revenue ($M)

Silicon Laboratories 24 %
Source: Dataquest, CIBC World Markets Corp.

$400

$300

$200

$100

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, DellOro, Dataquest, CIBC World Markets Corp.

Analog Modem and Fax Semiconductor Forecast: We expect analog modem and fax ICs to decline over the next five years as unit growth stagnates and semiconductor price declines chip away at market revenue. The move to softmodems in some embedded applications (we saw this happen in 2006 in set-top boxes) should further depress codec/DSP revenue; this transition is already behind us in the PC market. Overall, we expect IC revenue to decline at a 6% rate, from $557 million to $403 million in 2010. Analog Modem and Fax Semiconductor Competitors: The market for analog modem and fax ICs is highly mature and consolidated. Conexant remains the market leader with 50% share in 2005, followed by Silicon Labs with 24% and Agere with 16%. Other analog modem vendors include Austria Microsystems, Analog Devices, STMicroelectronics, Winbond Electronics, Freescale, and Oki Electric.

231

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
Cable Modem Unit Forecast
We expect cable modem unit shipments to grow 3%, from 22.3 million ports in 2005 to 26.5 million ports in 2010. Growth is slowing as penetration rates plateau; competing technologies such as VDSL and FTTX may also play a factor. Cable modems incorporating WLAN and VoIP will drive some incremental churn. DOCSIS 3.0 is still in development and will be more of a factor on the CMTS side when service rollouts begin towards the end of the decade.
30,000
CMTS Cable Modems

Cable Modem Suppliers


Cable Modems
Others 16 % D-Link 1% Cisco 2% Motorola 43 % Thomson 10 %

Arris 12 % Scientific Atlanta/Cisco 16 %

CMTS (Ports)
BigBand 4% Motorola 19 % Others 0%

25,000

20,000
Unit Shipments (000s)

15,000

10,000

Cisco 54 %
5,000

Arris 21 %
2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: DellOro, IDC, Dataquest, CIBC World Markets Corp.

Source: DellOro, CIBC World Markets Corp.

Cable Modem Unit Forecast: Cable modem demand has surged in the past few years as cable MSOs added broadband data to their video service offerings. From under one million units shipped in 1998, the market reached 22.3 million units in 2005 (up 29% from 2004). Going forward, we expect more moderate growth through 2010 as cable modem service penetration peaks (note that cable is primarily a U.S. phenomenon). Competing technologies such as VDSL and FTTX may also play a factor in muting growth prospects as U.S. telecom providers bolster the performance of their broadband offerings, which have consistently lagged cable to date in terms of bandwidth. In total, we expect modems to grow at a 3% CAGR, from 22.3 million units in 2005 to 26.5 million units in 2010. On the infrastructure side, we measure CMTS ports in terms of upstream ports, with each port able to support 100 or more individual subscribers depending on how the CMTS is configured. We expect a more robust growth on the CMTS side as opposed to CPE, largely driven by churn to support DOCSIS 3.0 scheduled for rollouts toward the end of the forecast period. In total, we expect ports to rise from 293,000 ports in 2005 to 530,000 ports in 2010, a 13% CAGR. Cable Modem and Infrastructure Suppliers: In 2005, Motorola led the cable modem market with 43% unit share. Scientific Atlanta (now Cisco) followed with 16% and Arris and Thomson held 12% and 10%, respectively. Other major suppliers include Cisco and D-Link, each with less than 5% of the market. On the infrastructure side, Cisco was the top vendor with 54% of CMTS ports shipped in 2005. Arris and Motorola followed with 21% and 19%, respectively. BigBand held 4% share.

232

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
Integrated Cable Modem
QPSK/QAM Modulator QAM Receiver Ethernet Controller USB Controller

Cable Modem
Interfaces with the coaxial cable and tunes the incoming and outgoing signals

Ethernet Cable USB Cable VoIP Block WLAN Block


In VoCable modems, a VoIP DSP and SLICs enable voice functionality

Cable Tuner

DOCSIS MAC

Processor

Flash DRAM SRAM

Performs all cable modem front end, protocol, and interface processing, including transmission and reception, MAC (protocol) functions, control plane functions, and Ethernet and USB protocol processing

DRAM, SRAM and flash memory are used by the cable modem chips embedded MAC and MIPS processor

Transmits downstream data traffic in the cable signal, interfacing with the head end

In Wi-Fi enabled cable modems, an 802.11 chipset enables wireless LAN functionality

CMTS

Cable Head End/ Combiner

QPSK/QAM Modulator

QAM Receiver

CMTS DOCSIS MAC DRAM

Traffic Provisioning ASIC SRAM

System Backplane: interface to switching, routing, WAN trunking, and control/management cards

Receives upstream data traffic in the cable signal from cable modem subscribers (by way of the head end)

Performs cable modem protocol processing, applying and removing cable modem DOCSIS overhead

Source: Broadcom, Texas Instruments, RHK, IDC, CIBC World Markets Corp.

DRAM and SRAM memory are used by the CMTS MAC and Traffic provisioning ASIC

This block, usually a set of ASIC and FPGAs, performs traffic provisioning before sending packets to the system backplane

Note: As in the analog modem section, our cable modem block diagrams, semiconductor forecast, and associated discussion include all application-specific semiconductors found in cable modem CPE and on CMTS line cards. We do not include those devices used in switching, routing, WAN trunking, and control/management cards, as these are not protocol dependent and can vary greatly depending on the functionality of the system. We discuss these ICs separately in the communications infrastructure section. Also note that in most block diagrams in this report, we separate out IC functionality as if there were no integration to better illustrate what functions designers need in order to implement a specific application. Cable modems, however, are almost entirely based on integrated cable modem chip designs. Therefore, we picture the modem as such above. Cable Modem CPE Cable Tuner: The tuner tunes the cable RF frequency for transmission and reception to and from the cable system. Older models use a can tuner, while newer designs use silicon tuners or a tuner block integrated into the cable modem chip.

233

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Integrated Cable Modem: The integrated cable modem includes front end processing (QPSK/QAM modulation for transmission and QAM demodulation for reception), the DOCSIS MAC (media access controller), an embedded processor and interface controllers (usually an Ethernet MAC/PHY and a USB controller). Newer cable modem chips sometimes include an integrated tuner. SRAM, DRAM, flash: Memory used by the embedded processor and DOCSIS MAC within the cable modem chip. Ethernet PHY (not shown in diagram): Some older cable modem SoC designs incorporate only the Ethernet MAC, requiring an external PHY IC to interface between the cable modem system and the Ethernet cable. Ethernet Switch (not shown in diagram): Though the basic cable modem needs only one Ethernet port to interface with the PC or wired network, newer residential gateways that include voice and wireless LAN functionality alongside the modem often include multiple Ethernet ports. These are usually implemented with a single-chip Ethernet switch, which includes Ethernet PHY, MAC, and switch functions; sometimes these functions are included in the cable modem SoC. Most gateways have 4-5 ports of 10/100 Ethernet, though this can vary.

VoIP Block (VoCable Modem Only) VoIP DSP (not shown in diagram): A VoIP packet processor DSP performs the TDM-to-IP conversion on IP-based phone calls. Integrated codecs perform the analog-to-digital and digital-to-analog conversion. Note that newer cable modem SoC designs often include a VoIP DSP. Line driver/SLIC (not shown in diagram): This SLIC is a line driver that interfaces with the telephone line, sending and receiving the analog signals to and from analog telephones.

Wireless LAN Block (Wi-Fi Enabled Cable Modem Only) Power Amp (not shown in diagram): The power amp sits in front of the antenna and boosts signal strength. RF/IF Converter (RF) (not shown in diagram): The RF front end controls the radio wave generation on the transmit side and receives the radio wave signal on the receive side, translating between low-frequency analog signals used by the system (specifically by the IF transceiver) to high-frequency RF signals used in radio communications. In newer implementations, the RF converter and IF are integrated in a single direct conversion transceiver, as shown in the block diagram above. I/Q Modem (IF) (not shown in diagram): The IF performs all quadrature modulation of I and Q baseband signals, converting the analog radio signal to a digital stream to be processed by the baseband and vice versa. In most implementations, the IF is integrated with the RF in a single direct conversion transceiver, as shown in the block diagram.

Cable Modem Termination System (CMTS) QAM Receiver/Demodulator: The QAM demodulator receives upstream data traffic in the cable signal from cable modem subscribers. QPSK/QAM Modulator: The QPSK/QAM modulator transmits downstream data traffic in the cable signal. CMTS DOCSIS MAC: The MAC performs protocol processing, assigning and removing DOCSIS overhead on the data packet. Traffic Provisioning ASIC: This block, usually a series of ASICs and FPGAs, performs traffic provisioning before forwarding packets to the system backplane. DRAM and SRAM: DRAM and SRAM memory is used both by the DOCSIS MAC and the traffic provisioning ASIC.

234

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
Cable Modem Cable Modem Semiconductor Forecast Semiconductor Competitors
We expect cable modem ICs to grow at a healthy rate as more functionality is integrated into the cable modem chipset. VoIP and WLAN will drive growth, though integration will limit the opportunity for discrete ICs while benefiting the SoC. Infrastructure ICs should grow nicely, but will remain <20% of total. In total, we expect cable IC revenue to grow at a 7% CAGR, from $433 million in 2005 to $613 million in 2010.
$700 Discrete VoIP ICs Discrete Wireless LAN ICs CMTS ICs Cable Modem ICs $600

Conexant 2% Microtune 6%

Others 5%

Texas Instruments 25 % Broadcom 62 %

Semiconductor Revenue ($M)

$500

$400

Note: Market share data excludes discrete WLAN and VoIP components as well as ICs for the CMTS. Source: Dataquest, IDC, Linley, CIBC World Markets Corp.

$300

$200

$100

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Note: WLAN and VoIP ICs used in cable modems are also captured in our semiconductor forecasts for those categories. Source: IDC, Dataquest, DellOro, RHK, CIBC World Markets Corp.

Note: We have included wireless LAN and VoIP ICs that sell into cable modems in our forecast, even though these are discussed in their own respective sections of this report. We believe these functions will be integrated into the cable modem SoC over time, and we wanted to show the effect on our forecast specifically that the cable modem SoC will benefit even while the system BOM declines. Cable Modem Semiconductor Forecast: We expect cable modem ICs to grow at a healthy rate over the next few years as IC suppliers integrate more functionality into cable modem chipsets, increasing the total semiconductor BOM. Increasing VoIP and WLAN penetration should be the primary growth drivers (VoIP going from 28% of units in 2005 to 100% in 2010 and WLAN from just 3% in 2005 to 80% in 2010); note that integration trends will limit growth prospects for discrete components while benefiting the cable modem SoC. Infrastructure ICs should also grow nicely (an 8% CAGR), but will remain less than 20% of total semiconductor revenue. In total, we expect the market for cable modem semiconductors to grow at a 7% rate, from $433 million in 2005 to $613 million in 2010. Cable Modem CPE Semiconductor Competitors: The market for cable modem semiconductors has remained very concentrated, with Broadcom commanding a 62% share of cable modem IC revenue (excludes WLAN, VoIP, and CMTS) as of 2005. Broadcoms flagship customer is Motorola, though the company also supplies Scientific Atlanta (now Cisco), Thompson and Samsung as well as several ODMs. TI held the No. 2 position with 25% share; TI has strong positioning in VoCable products. Microtune followed with 6% share due to its position in tuners, and Conexant held 2% share. Other cable modem IC players include Terayon, Toshiba, Sharp, and ALPS Electronics.

235

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
DSL Modem Unit/Port Forecast
We expect DSL modem units/ports to grow at a 1% CAGR through 2010, from 120 million in 2005 to 127 million in 2010. We expect shipments peaked in 2006 but should remain much larger than cable due to the availability of copper infrastructure worldwide. VDSL started to ramp in 2005 and should grow nicely through 2010, driving a new leg of growth beyond the forecast period.
140,000
VDSL

DSL Modem Suppliers


DSL Modems
Thomson 15 % Others 33 % Siemens 12 %

NEC 3% Huawei 3%

Zyxel 10 % 2Wire 5 % Westell 9%

D-Link 10 %

120,000

SDSL ADSL

DSLAM (Ports)
Others 17 %

100,000
Unit Shipments (000s)

80,000

60,000

40,000

Zyxel 3% Ericsson 3% UTStarcomm 3% Marconi 3%


2002 2003 2004 2005 2006 2007 2008 2009 2010

Alcatel 31 %

20,000

NEC 6% ECI Telecom 6% Huawei 12 % Lucent 8% Siemens 8%

Source: DellOro, IDC, Dataquest, CIBC World Markets Corp.

Source: DellOro CIBC World Markets Corp.

DSL Modem Unit/Port Forecast: DSL demand has surged as voice telephony providers worldwide have embraced residential broadband service and now position data at the center of their service offerings. From less than half a million units/ports shipped in 1998, the market reached 120 million ports in 2005, up 24% from 2004. Looking ahead, we expect an intermediate-term peak in 2006 after which shipments should begin to decline, particularly on the DSLAM side. Still, DSL should enjoy a larger long-term unit opportunity than cable due to the availability of copper infrastructure worldwide, and could see another wave of growth beyond the forecast period as VDSL ramps up. In total, we expect DSL ports to grow at a 1% CAGR, from 120 million in 2005 to 127 million in 2010. In terms of DSL standard, we expect ADSL to peak in 2006, showing a negative 5% CAGR 2005-2010, though it should still represent more than 65% of all DSL shipments in 2010. SDSL should peak in 2007 and flatten thereafter, showing a 9% CAGR 2005-2010, still small at less than 2% of DSL connections. VDSL should grow through the end of the forecast period at a 37% rate, reaching 32% of connections in 2010. DSL Modem CPE and Infrastructure Suppliers: The DSL modem market is highly fragmented. The largest suppliers in 2005 were Thompson with 15% unit share and Siemens with 12%. Zyxel and D-Link each held 10% and Westell held 9%. Other major suppliers included 2Wire with 5% and Huawei and NEC, each with 3%. On the infrastructure side, Alcatel was the top DSLAM supplier with 31% share on a port basis, followed by Huawei with 12%, and Siemens with 8%. Other suppliers include Lucent, ECI, NEC, Marconi, UTStarcomm, and Ericsson.

236

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
In Wi-Fi enabled DSL modems, an 802.11 chipset enables wireless LAN functionality

WLAN Block

VoIP Block

In VoDSL modems, a VoIP DSP and SLICs enable voice functionality

Ethernet interface driver (PHY) and protocol processor (MAC)

DSL CPE

RJ-11 Phone Line

DSL Controller Line Driver


AFE

MCU

DSP Data Pump

Ethernet Controller USB Host

RJ-45 Ethernet Cable USB Cable


Interfaces with the USB cable to connect up to the PC

SRAM
Interfaces with the telephone line, sending the analog data signals between the modem and the RAC over the telephone line

Flash
This DSL Controller performs DSL protocol processing, adding DSL overhead on the transmit side and removing it on the receive side. Usually incorporates DSP data pump, AFE and microcontroller

SRAM, Flash memory used by the DSL controller

RJ-11 Phone Line Port 2

Line Driver Line Driver

AFE AFE

DSLAM Controller
DSP Data Pump MCU

DSLAM
Port N

System Backplane: interface to switching, routing, WAN trunking, and control/management cards

AFE

Line Driver

Flash

SRAM

Source: IDC, CIBC World Markets Corp.

SRAM, Flash - memory used by the DSLAM controller processor

Note: As in the other modem sections, our DSL block diagrams, semiconductor forecast, and associated discussion include all semiconductors included in DSL modem CPE, as well as those on DSLAM line cards. We do not include those devices used in switching, routing, WAN trunking, and control/management cards, as these are not protocol dependent and can vary greatly depending on the functionality of the system. We discuss these ICs separately in the communications infrastructure section. Line Driver: The line driver interfaces with the RJ-11 phone line. DSL/DSLAM Controller: This controller includes a DSP data pump and analog front end. It also typically includes an embedded microcontroller or microprocessor for system control. DSLAM controllers include the same functions but can include multiple AFEs and a more robust DSP. SRAM and Flash: SRAM and flash memory are used by the DSL/DSLAM controller. Ethernet Controller: An Ethernet controller provides the interface between the modem and the PC or network. USB Host: Found in USB-enabled modems, the USB host interfaces between the modem and the host PC. WLAN and VoIP block: See the cable modem diagram on page 233 for a description of the devices found in these blocks.

237

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
DSL Semiconductor Forecast
We expect DSL IC revenue to remain roughly flat through 2010 at $1.1 billion. A shift to higher-ASP VDSL should help offset some of the normal ASP erosion in this cost-sensitive market. Increasing VoIP and WLAN penetration will also help mitigate pricing pressure, though integration trends will limit the opportunity for discrete components.
$1,500 VoIP ICs Wireless LAN ICs Total VDSL ASSPs Total SDSL ASSPs Total ADSL ASSPs

DSL Semiconductor Competitors


STMicro 5% Centillium 6% Others 7% Conexant 26 %

Ikanos 9%

$1,200

Infineon 13 %

Texas Instruments 17 %
Broadcom 17 %

Semiconductor Revenue ($M)

$900

$600

Note: Market share excludes discrete WLAN and VoIP ICs. Source: IDC, Linley, CIBC World Markets Corp.

$300

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Note: WLAN and VoIP ICs used in DSL modems are also captured in our semiconductor forecasts for those categories. Source: IDC, Dataquest, DellOro, CIBC World Markets Corp.

Note: As in the cable modem section, we have included wireless LAN and VoIP ICs that sell into DSL modems in our forecast, even though these are discussed in their own respective sections of this report. We believe these functions will be integrated into the DSL modem SoC over time, and we wanted to show the effect on our forecastspecifically that the SoC will benefit even while the system BOM declines. DSL Modem Semiconductor Forecast: We expect DSL IC revenue to remain roughly flat at $1.1 billion through 2010. A steady decline in ADSL (negative 6% CAGR through 2010) should be offset by a shift to higher-ASP VDSL, which should grow at a 23% CAGR to represent just over 25% of the market in 2010. Similar to cable, increasing wireless LAN and VoIP penetration should help mitigate pricing pressure, though integration trends will limit the opportunity for discrete components. DSL Semiconductor Competitors: The top DSL semiconductor vendor in 2005 was Conexant with 26% share, followed by Texas Instruments with 17% of the market. Broadcom gained roughly eight percentage points of share in 2005 as it ramped new ADSL2+ products with infrastructure leader Alcatel and a number of key CPE OEMs. Infineon also picked up share and was the No. 4 vendor in 2005 with 13% share. Rounding out the top five was start-up Ikanos, which dominated VDSL in Japan and branched into ADSL with the acquisition of Analog Devices DSL business in 2006. Other major suppliers include Centillium, which held 6% share, and STMicroelectronics with 5% (STMicroelectronics share in DSL fell six percentage points in 2005 as it lost share at Alcatel to Broadcom).

238

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Modems Modems
Modem Semiconductor Market Trends
Pure play broadband semiconductor vendors are all but squeezed out. Wireless LAN and VoIP integration are becoming more important.
Analog Modem and Fax Semiconductors PC analog modem penetration continues to decline. Embedded modems are increasingly being implemented in software. Silicon DAAs are gaining share vs. transformer-based DAAs. Multi-function peripherals are a standout area for growth. Cable Modem Semiconductors DOCSIS 2.0 is standard in modems and is driving churn in CMTS. Development continues on wideband cable modems for DOCSIS 3.0. VoCable has reached critical mass and is driving integration; wireless LAN is also starting to gain traction. DSL Modem Semiconductors ADSL has transitioned to ADSL2+; VDSL is accelerating worldwide after a successful rollout in Japan. VDSL2 and ADSL2+ CPE solutions are likely to converge in the future. SDSL has consolidated around g.shdsl, but deployments remain limited. Multi-port data pumps are enabling higher density DSLAM line cards. DSLAMs are moving away from ATM backhaul toward IP. VoIP got a faster start in cable, and is only now showing up in DSL.
Pure-play broadband semiconductor vendors are all but squeezed out. The major modem vendors are adopting a diversified approach that includes DSL, cable, wireless LAN, and VoIP. There are many examples: DSL market leader Globespan first bought Intersils wireless LAN business and then merged with Conexant, which was itself a DSL and cable modem IC supplier. TI moved from DSL to cable modem to wireless LAN and dominates VoIP silicon. Broadcom moved from cable modems to wireless LAN and DSL and also supplies VoIP ICs. Centillium, which struggled as a pure-play DSL vendor, has branched out into VoIP. The only real pure-plays left are either analog modem vendors in harvest mode (e.g., Agere, Silicon Labs) or start-ups focusing on nextgeneration standards (e.g., Ikanos). VoIP integration is becoming more important. Voice-over-IP has started to gain steam as a residential service, particularly as cable MSOs look to voice to drive additional revenue. Most implementations to date have required that the modem (or home router) be connected to a VoIP terminal adapter, which interfaces with traditional analog telephones and provides voice service. Newer modem designs, however, incorporate VoIP functionality directly into the modem. Most modem IC vendors have therefore aligned their chipset offerings to include an easy interface to VoIP ICs or integrated VoIP capability. Wireless LAN is increasingly found in modems and will see some integration over time. Wireless LAN technology has proven enormously popular for home networking, as consumers seek to share their broadband connection among multiple PCs. Most implementations today require that the modem be connected to a wireless

239

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

router, which distributes the signal to the PC. Newer modem designs, however, are incorporating wireless LAN functionality directly into the modem. Most modem IC vendors have therefore aligned their chipset offerings to include an easy interface to wireless LAN chipsets. Some newer designs are sucking wireless LAN functionality directly into the modem IC. Trends in Analog Modem and Fax Semiconductors PC analog modem penetration continues to decline. Broadband has taken a big bite out of analog modemsaccelerated by aggressive pricing from service providers. Still, the majority of units are tied to PCs and penetration should decline as broadband becomes even more pervasive and as ISPs stop subsidies to PC OEMs. Modems are increasingly being implemented in software. While some analog modem implementations (primarily in the embedded market) still include a discrete codec or controller IC, many new designs mimic the PC softmodem model, perform processing in software, leaving only the line driver/DAA external. Silicon direct access arrangements (DAAs) are gaining share vs. transformer-based DAAs. DAAs perform signal isolation, reducing signal voltage from the high voltage current used on the telephone line to the low voltage used in a PC. Newer silicon DAAs perform this function with fewer components than traditional transformer-based DAAs and also reduce bill-of-materials cost for modems. Multi-function peripherals are a standout area for growth. One area where modem IC vendors are seeing unit growth is MFPs, which combine printer, copier, scanner, and fax functions in a single piece of equipment.

Trends in Cable Modem Semiconductors DOCSIS 2.0 is standard in modems and is driving churn in CMTS. DOCSIS 2.0 has been a standard feature in modems for some time, as the cost of future-proofing has been minimal. On the infrastructure side, however, MSOs have been slow to install the equipment. The faster upstream performance of DOCSIS 2.0 is less of a priority for MSOs than video and voice, and we therefore expect moderate churn of CMTS ports for DOCSIS 2.0. Development continues on wideband cable modems, which will be included in DOCSIS 3.0. Wideband cable modems will use a bonded approach to boost cable modem data rates to match (or exceed) those of VDSL and PON. Products are unlikely to hit the market until 2008, and will probably be focused on the business market, as MSOs do not need a faster data pipe to offer video like telecom providers. VoCable has reached critical mass and is driving integration; wireless LAN is also starting to gain traction. As discussed above, cable MSOs have been rolling out VoIP to increase their revenue base from video and data. The cost of implementation has come down as new VoCable modems have come to market using single-chip solutions from Broadcom or TI. Wireless LAN is less of a priority (as MSOs cant really charge for this), but we are seeing uptake accelerate to drive churn.

Trends in DSL Semiconductors

ADSL has transitioned to ADSL2+. DSL has consistently lagged cable modem in terms of bandwidth, and service providers were quick to upgrade their pipes to support the faster ADSL2+ (though in most cases they are charging extra for more bandwidth). ADSL2+ is now the worldwide standard for ADSL equipment. VDSL is accelerating worldwide after a successful rollout in Japan. Japan and Korea were first to bump their broadband speeds with ADSL2/2+, and have been early adopters of VDSL as well. We expect carriers in the U.S. and Europe to follow, with notable deployments by AT&T, Deutsche Telekom, and Telecom Italia in 2007. VDSL2 and ADSL2+ CPE solutions are likely to converge in the future. Given the similarities in the standards and common base of service provider deploying these services, we expect mainstream VDSL modems to converge on SoC solutions that support both ADSL2+ and VDSL over time. SDSL has consolidated on the g.shdsl standard, but service deployments will be limited. SDSL has been less than successful in winning customers away from T1, and faces competition from metro Ethernet as well. Though we forecast some growth, SDSL is unlikely to grow beyond a niche service for the small business market. Multi-port data pumps and DSL PHYs are enabling higher density DSLAM line cards. These high-density cards are enabling lower-cost DSLAMs. DSLAMs are moving away from ATM backhaul toward IP. This trend is discussed in the communications infrastructure section. VoIP got a faster start in cable, and is only now showing up in DSL. Since most DSL service providers are telco carriers, there has been less of a rush to add VoIP. We are seeing this trend accelerate in emerging countries where POTS service has a smaller footprint.

240

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PON PON
Passive Optical Networking (PON) is a fiber-based access technology that allows service providers to deploy super-fast broadband access over optical fiber, usually referred to as Fiber-To-The-Home (FTTH). Passive splitters can be used in the field to split the optical signal into 16, 32, or even 64 channels without active switching components, dramatically lowering the cost vs. traditional metro Ethernet services. FTTH Service Using PON
Customer Premise
POTS
O

Service Provider
pt ic al Fi be r

Ethernet

ONT
IPTV RF Overlay

Optical Fiber

Splitter (passive)

Uplink to Public Network/ Internet

OLT

Source: CIBC World Markets Corp.

Passive optical networking (PON) is a high-speed access technology that enables broadband access over optical fiber, sometimes called fiber-to-the-home (FTTH). The optical signals used in fiber-based networks are capable of achieving much higher data rates than the electrical signals that use the copper network (both the traditional POTS and coaxial cable networks). This allows for super-fast data access with a roadmap for IP-based video down the road. In a PON system, access units called OLTs (optical light terminals) are installed in the service provider network, usually at the central office. Fiber runs from the OLT to a passive splitter out in the field, which splits the optical signal into 16, 32, or even 64 channels. Each channel is meant for an individual subscriber. Individual strands of fiber run from the splitter to customer premise units called ONTs (optical network terminals), which are installed on the exterior of the customers homes or apartment building. The ONT extracts the data (as well as TDM phone calls) from the PON signal and distributes it to the home PC or LAN over Ethernet (and POTS for the voice signals). The key to PON technology is the use of passive splitters. These allow service providers to split off signals to individual subscribers without active switches, dramatically lowering the cost vs. traditional switched metro Ethernet services. It does, however, mean that bandwidth must be shared between all subscribers connected to the OLT. A major driver for FTTH is the ability to provide video over the telecom network. This ideally is accomplished via IPTV, which would require an IP set-top box that is connected to the ONT. A less revolutionary option is an RF overlay; traditional RF cable signals are transmitted over the PON, and cable set-top boxes are connected to the ONT.

241

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PON PON
Some service providers are hesitant to roll out full FTTH, but still want to bring fiber closer to their customers to boost data rates. In these Fiber-To-The-Curb (FTTC) or Fiber-To-The-Node (FTTN) networks, PONs extend the fiber network to within a few hundred feet of the customer, and high-speed DSL makes the final connection. FTTC/FTTN Service Using PON
Curb/ Building Customer Premise
L DS
O

Service Provider
pt ic al Fi be r

Ethernet

DSLAM

ONU

Optical Fiber

IPTV

DSL Modem

Splitter (passive)

Uplink to Public Network/ Internet

OLT

Source: CIBC World Markets Corp.

Some service providers are hesitant to roll out full FTTH because of the high cost of ONTs and laying fiber around the neighborhood. They still want to bring fiber closer to their customers to boost data rates, however, and are able to leverage PON technology using a slightly different topology. In these fiber-to-the-curb (FTTC) or fiber-to-the-node (FTTN) networks, PONs extend the fiber network to within a few hundred feet of the customer, and high-speed DSL makes the final connection. Instead of ONTs at individual subscriber premises, the service provider installs ONUs (optical network units) in the neighborhood, at the curb, or even in the basement of a multiple dwelling unit. These ONUs interface with a small DSLAM, which distributes the signal to individual customers. Service providers are currently looking at VDSL2, ADSL2+, and bonded ADSL to provide this final connection, and OEMs are coming to market with equipment that merges the ONU and the DSLAM in a robust, flexible platform. Note that FTTC/FTTN networks are capable of supporting IPTV video service. They cannot, however, support the RF video overlay like in an FTTH network, since the RF overlay needs to be extracted at the customer premise.

242

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PON PON
There Are Three Main PON Standards:
EPON/GE-PON Ethernet/Gigabit Ethernet PON is based on Ethernet LAN technology. Upstream and downstream speeds can reach 1.25 Gpbs and can be split into 16 channels. EPON is now being deployed aggressively in Japan; Korea, Taiwan, and China are also likely to deploy EPON. BPON Broadband PON is the most common ATM-based PON technology in place today and is set by the ITU. BPON supports upstream and downstream speeds of 155 Mbps or 622 Mbps, and can be split into 32 channels. BPON can also support an RF overlay for video. BPON is seeing deployment in the U.S.; Japan had BPON service but has since moved to EPON. GPON Gigabit PON is the successor to BPON. GPON supports downstream speeds of 1.24 Gbps and 2.5 Gbps and upstream speeds of 155 Mbps, 622 Mbps, 1.2 Gbps, and 2.5 Gbps. GPON can be split into 64 channels, and also supports an RF overlay.

There are three main standards of PON being deployed or developed: EPON/GE-PON Ethernet/Gigabit Ethernet PON is based on Ethernet LAN technology and is controlled by the IEEE, the same standards body that controls Ethernet and wireless LAN. Downstream speeds can reach 1.25 Gps and use the 1500-nm wavelength. Upstream speeds can also reach 1.25 Gps and use the 1310-nm wavelength. An EPON signal can be split into 16 channels with a maximum range of around 12 miles; it does not support an RF overlay. EPON is now being deployed aggressively in Japan; Korea, Taiwan, and China are also likely to deploy EPON. BPON Broadband PON is the most common ATM-based PON technology in place today and is set by the ITU. BPON supports downstream speeds of 155 Mbps or 622 Mbps, and use wavelengths of 1480-1500-nm. Upstream speeds are 155 Mbps or 622 Mbps, and use wavelengths of 1260-1360-nm. BPON can be split into 32 channels with a range of about 12 miles; it can also support an RF overlay for video using the 1550-nm wavelength. BPON is seeing deployment in the U.S.; Japan had BPON service but has since moved to EPON. GPON Gigabit PON is the successor to BPON. GPON supports downstream speeds of 1.24 Gbps or 2.5 Gbps, and like BPON uses wavelengths of 1480-1500-nm. Upstream speeds are 155 Mbps, 622 Mbps, 1.24 Gbps or 2.5 Gbps, and use wavelengths of 1260-1360-nm. GPON can be split into 64 channels with a range of about 37 miles; like BPON it can support an RF overlay for video using the 1550-nm wavelength. GPON will likely be deployed in the U.S. once the equipment is ready for commercial production.

243

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PON PON
PON Port Forecast
We expect total port shipments to grow at a 30% CAGR, rising from 2.7 million in 2005 to 9.9 million ports in 2010. The most aggressive rollouts have taken place in Japan with EPON; Japan has abandoned its initial BPON effort. Korea, China, and Taiwan are conducting trials now with EPON and are set to ramp in 2007. The U.S. is ramping BPON now and will move to GPON once equipment becomes available in volume.
12,000
EPON GPON BPON

PON Equipment Suppliers


BPON/GPON
Tellabs Alcatel Motorola Calix Terawave Wave7 Entrisphere Flexlight Hitachi

EPON
Mitsubishi Sumitomo UTStarcomm Fujitsu NEC Oki Melco Alloptic
2004 2005 2006 2007 2008 2009 2010

10,000

8,000
Port Shipments (000s)

6,000

4,000

2,000

2002 2003

Source: IDC, Ovum-RHK, CIBC World Markets Corp.

Source: DellOro, IDC, CIBC World Markets Corp.

PON Port Forecast: The market for PON equipment has strong growth prospects as carriers have become more aggressive with fiber deployments. The technology should enable telecom carriers to offer triple play services (voice, data, video), with IPTV being the killer application. Most growth to date has come from Japan, as NTT, Softbank, and KDDI have been aggressive with PON rollouts; these service providers originally deployed BPON but quickly switched to EPON in 2004. Note that PON deployments in Japan began to decelerate in 2006 and future growth will likely be driven from other regions. EPON trials are currently well underway in Korea, China, and Taiwan and deployments are expected to ramp in 2007. In the U.S., Verizon and SBC/BellSouth are working with BPON today, though deployments are likely to accelerate once GPON becomes available in volume in 2007 and beyond. In total, we expect PON port shipments to grow at a 30% CAGR, from 2.7 million units in 2005 to 9.9 million units in 2010. EPON should decline as a percentage, from 85% in 2005 to 62% in 2010, with GPON representing the bulk of the rest. From an equipment perspective, we expect the ONT/OLT ratio to rise from 6.9:1 in 2005 to 10.1:1 in 2010. PON Equipment Suppliers: There are currently two PON standards being deployed today, EPON for the Asian market and BPON/GPON for U.S. deployments; with a discrete set of suppliers for each. The top suppliers of EPON equipment are Mitsubishi, Sumitomo, UTStarcom, Fujitsu, NEC, Oki, Melco, and Alloptic. The top suppliers of BPON/GPON equipment are Tellabs, Alcatel, and Motorola, as well as several start-ups including Calix, Terawave, Wave7, Entrisphere, and Flexlight. Hitachi also supplied BPON equipment to Japan, though these networks have switched to EPON.

244

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PON PON
A SerDes provides the interface between the optical module and the PON MAC

SRAM is used by the PON MAC

SRAM, Flash memory used by the comm processor

Ethernet interface driver (PHY) and protocol processor (MAC) provide the interface between the ONT and the PC/LAN

ONT
Optical Fiber

SRAM

SRAM

Flash Ethernet MAC Ethernet PHY SLIC SLIC PC or Ethernet LAN Analog phone lines Coax/Video (BPON/GPON)

Optical Module

SerDes

PON ONT MAC

Comm Processor

This PON MAC performs protocol processing, adding PON overhead on the transmit side and removing it on the receive side

The communications processor controls overall system functioning and also performs data path functions

Standard linear ICs provide the interface for standard analog telephones

OLT
Optical Fiber

Optical Module

SerDes

PON OLT MAC SRAM

Comm Processor

ATM Backplane (BPON/GPON) Ethernet MAC Ethernet PHY Ethernet Backplane (EPON)

A SerDes provides the interface between the optical module and the PON MAC Source: Ovum-RHK, CIBC World Markets Corp.

SRAM

Flash
Ethernet interface driver (PHY) and protocol processor (MAC); EPON OLTs use Ethernet as the interface to the network

SRAM is used by the PON MAC

SRAM, Flash - memory used by the comm processor

The devices found in PON equipment are not dissimilar from those in traditional broadband modems. The big exception is the optical module, which interfaces with the optical fiber. PMD ICs (inside optical module, not shown in diagram): Physical Medium Dependent (PMD) ICs like laser drivers, limiting amps, and transimpedence amps are found inside the optical module and control the interface. SerDes: The SerDes provides the interface between the optical module and the PON MAC. In newer ONT designs the SerDes is integrated with the MAC. PON MAC: The PON MAC performs basic protocol processing, adding PON overhead on the transmit side and removing it on the receive side. Comm. Processor: The comm. processor controls overall system functioning and some data path functions. SRAM and Flash: Memory used by the communications processor; the PON MAC also uses some SRAM. Ethernet MAC and PHY: In an ONT, the Ethernet MAC and PHY provide the interface between the ONT and the host PC or Ethernet LAN. The Ethernet MAC is sometimes integrated with the PON MAC/processor. In an OLT, the Ethernet MAC and PHY feed traffic into the WAN (they are only used in EPON systems). SLIC (ONT only): Standard linear ICs interface with analog telephone lines, allowing traditional telephones to be used over the PON.

245

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PON PON
PON Semiconductor Forecast
We expect the market to grow at a 16% CAGR, from $76 million in 2005 to $161 million in 2010. EPON ICs represented most of the growth in 2005 and 2006 as PON ramped in the Japanese market. BPON is just starting to ramp in the U.S. as carriers begin FTTH deployments; it will eventually give way to GPON starting in 2007. In 2010, we expect the split to be 52% EPON, 47% GPON, and 1% BPON.
$180 EPON ICs $160 $140
Semiconductor Revenue ($M)

PON Semiconductor Competitors


Others 9% Freescale 6%

BroadLight 10 %
PMC/Passave 55 %

Teknovus 19 %

GPON ICs BPON ICs

Source: Linley, CIBC World Markets Corp.

$120 $100 $80 $60 $40 $20 $0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, Ovum-RHK, CIBC World Markets Corp.

PON Semiconductor Forecast: PON semiconductor revenue growth nearly tripled in 2004 and grew 23% in 2005, driven primarily by aggressive PON deployments in Japan. Note that market revenue was expected to be down slightly in 2006 as deployments in Japan decelerate, but should pick back up in 2007 as PON ramps in the U.S. and in other parts of Asia. Overall, we forecast a CAGR of 16%, from $76 million in 2005 to $161 million in 2010. Note that semiconductor revenue growth significantly trails port shipment growth as OEMs put pressure on their suppliers to drive costs down to help drive PON CPE equipment into the mainstream. By standard, we expect GPON to start to ramp meaningfully in 2007 and grow to 47% of revenue in 2010; EPON should grow more modestly (6% CAGR through 2010) to 52% of total. BPON should represent the balance. PON Semiconductor Suppliers: The top supplier of PON semiconductors in 2005 was Passave (acquired by PMCSierra), which held 55% share, primarily attributed to EPON with NTT DoCoMo in Japan. Passave also serves other EPON carriers in Asia and is targeting GPON as well. Teknovus was second with 19% share; the company focuses exclusively on EPON and serves carriers in Japan (besides NTT) and Asia. Broadlight and Freescale followed with 10% and 6% share, respectively; both supply BPON and are moving to GPON. Other suppliers include Centillium for EPON and Conexant and AMCC for GPON. PMD suppliers for PON optical modules include Mindspeed, Maxim, and Vitesse. Passave/PMC recently indicated that it intends to compete in this area as well.

246

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

PON PON
PON Semiconductor Market Trends
IC vendors are racing to market with GPON solutions for the U.S. MAC solutions are evolving into PON SoCs with integrated SerDes, communications processors, and memory. Service providers deploying BPON and GPON will be quicker to market with video by using an RF overlay; IPTV will come later. FTTC networks will use cabinet solutions that integrate a PON ONU with a DSLAM.

IC vendors are racing to market with GPON solutions for the U.S. This includes those coming from the BPON world (e.g., Broadlight and Freescale), the EPON world (e.g., PMC/Passave), and new players like Conexant and AMCC. It is still unclear who will win designs for GPON; many of these decisions will be made in 2007. MAC solutions are evolving into PON SoCs with integrated SerDes, communications processors, and memory. Though earlier designs were built with discrete MAC, SerDes, and processor ICs, the major PON IC vendors are now in the market with SoC solutions that include these functions on a single chip. Service providers deploying BPON and GPON will be quicker to market with video by using an RF overlay; IPTV will come later. BPON and GPON support an RF overlay, which allows a video stream to be sent over fiber in its native format (i.e., without a conversion to IP). Verizon in the U.S. is currently installing traditional cable video equipment in its network to offer video service over PON, which will leverage existing settop box designs. This will allow the company to get video to market quicker than if it went with an IP-TV solution. Down the road, we expect service providers to look to IPTV once the equipment and service offerings mature. FTTC networks will use cabinet solutions that integrate a PON ONU with a DSLAM. OEMs targeting service providers that are adopting a FTTC approach (e.g., BellSouth) are developing cabinet systems that integrate a PON ONU with a DSLAM (ADSL2+ or VDSL/VDSL2). This will leverage the speed of PON without requiring a fiber connection all the way to the home.

247

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


This section covers communications infrastructure equipment, a broad category that includes equipment used to switch, route, and manage voice and data traffic in both service provider and enterprise networks. Included are a wide variety of dedicated infrastructure equipment, such as routers, digital cross connects, add/drop multiplexers, and multi-service switches. We also include here the switching, routing, traffic management, and control plane functions (and their semiconductors) that are embedded in data networking, access, and wireless infrastructure devices that are also discussed in other sections of this report. Key examples include routing functions in layer 3+ switches, aggregation, routing, and WAN interface functions in modem concentrators, PON OLTs, and central office switches, and backhaul for wireless basestations.

This section covers communications infrastructure equipment, a broad category that includes equipment used to switch, aggregate, route, and manage voice and data traffic in both service provider and large enterprise networks. Included are a wide variety of dedicated infrastructure equipment, such as routers, digital cross connects, ATM switches, multi-service provisioning platforms, and optical transport equipment. We also include here the switching, aggregation, routing, traffic management, and control plane functions that are embedded in data networking, access, and wireless infrastructure devices that are also discussed in other sections of this report. Key examples of this include routing functions in layer 3+ switches, backhaul for wireless basestations, and WAN interface functions in modem concentrators, PON OLTs, central office switches, and PBXs. One notable exception is the switching function used in Ethernet LAN equipment; we include those ICs in the Ethernet section instead of here, as LAN switching is highly specialized for that application. Note that a unit forecast is somewhat irrelevant; we therefore display our forecast for infrastructure equipment (both for carrier and enterprise) in revenue terms. On the semiconductor side, we forecast revenue for specific ICs, e.g., SONET framers and PHYs, ATM UNIs and SARs, NPUs, and switch fabrics. Note that in all cases, the forecast excludes all protocol-dependent ICs that are discussed in other sections (e.g., DSLAM PHYs, POTS line card SLICs, PON OLT MACs, baseband ASICs); while the other sections exclude all ICs used in switching, routing, traffic management, and system control (e.g., framers, network processors, control plane processors). Again, the one notable exception is Ethernet LAN switching, where we include switch fabric ICs in the Ethernet section instead of here.

248

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


The Primary Functions of Infrastructure Equipment Though communications infrastructure equipment vary widely in application and capability, the functions they can perform are common across device types:
Transport moving traffic between network nodes, over network links external to the device. Switching forwarding traffic between ports within a given piece of equipment. Aggregation combining multiple smaller, slower streams into larger, faster streams. Routing forwarding traffic by first determining the optimal path through the network. Traffic Management prioritizing the movement of traffic based on content type, service level agreements, or other rules/algorithms. Control Plane management of the overall system including sub-systems listed above.

Communications infrastructure equipment can vary widely in terms of capability and application. The functions they can perform, however, are pretty common across device types. These include: Transport moving traffic between network nodes, over network links external to the device. Most of the equipment discussed in this section use optical fiber for transport, though copper can be used at lower speeds. Switching forwarding traffic between ports within a given piece of equipment. This can be done either over a backplane or within a separate switching sub-system. Aggregation (also called multiplexing) combining multiple smaller, slower streams into larger, faster streams. Routing determining the optimal path through the network, and directing traffic accordingly. Traffic Management prioritizing the movement of traffic based on content type, service level agreements, or other rules/algorithms. Traffic management is related to QoS (quality of service); infrastructure equipment can only insure a certain level of quality of service by intelligently managing the flow of traffic based on what is present in the data stream. Control Plane management of the overall system, including sub-systems that perform the above-mentioned transport, switching, aggregation, routing, and traffic management.

249

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Line Card and Chassis Systems Infrastructure equipment is often implemented in a chassis/line card configuration. Individual cards can be added to support transport links, switching, routing, traffic management, and system control.

Telecom and datacom equipment OEMs design their carrier and large enterprise offerings to be robust, redundant, and extremely flexible. Carriers need to adjust the voice and data capabilities of their network nodes depending on local factors, and need to be able to change these capabilities as those factors change. On the enterprise side, IT managers need to have a great deal of control over the allocation of network resource among users; they also need a network that is scalable, secure, and easy to upgrade and manage. In either case, system flexibility, scalability, and upgradeability can be just as important as performance and general robustness of the system. This has prompted OEMs to design their large enterprise carrier grade systems in a chassis/line card configuration instead of enclosed discrete devices like modems or access routers. Using this architecture, carriers install a chassis, which is essentially an empty metal box equipped with a backplane, power supply, and cooling system; and populate the system with line cards to suit the needs of the network node. Cards can be installed and removed at will, protecting the initial investment in the platform while allowing for future upgradeability. Line Cards There are four main types of line cards found in infrastructure equipment. Each line card contains a backplane interface used to communicate with the other cards within the overall system and to pass traffic back and forth. I/O Line Cards: These cards include one or more ports used to send traffic back and forth over the broader network. These can include fiber or copper ports for POTS, DSL, DOCSIS, T/E, SONET, Ethernet, or PON.

250

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Some I/O line cards support shorter-range technologies like ATM UNI or Infiniband for communication with other equipment within the network node. Switching Cards: These cards include switch fabric arrays, and are used to switch traffic between the I/O cards in the chassis. WAN trunking Cards: For systems that interface with individual users (like modem aggregators or class 5 switches), a WAN trunking card provides the high-speed uplink into the larger network. The WAN trunking card can contain routing functionality on it, though sometimes routing functions are integrated on individual line cards. Management/Control Cards: These cards are used to control the overall system. They can also perform special functions on the data path when there are exceptional packets.

Besides I/O, switching, trunking, and management, other line cards or modules can be added to perform periphery functions. Examples include security line cards, monitoring cards, LAN interface cards, and modules with optical drives embedded in them. The type of line cards used in each system will depend on the functionality of the system and its location in the network. An add/drop multiplexer will likely have several high-speed line cards and a switch card, along with the management card. A digital cross connect will likely have a variety of line cards of varying speeds along with a set of robust switch cards, and the management card. A class 5 switch will have hundreds of individual line cards for POTS access, a switch card, a WAN trunking card for interface to the PSTN, and the management card. A DSLAM will have several multi-port DSL I/O cards, a switch card, a WAN trunking card, and the management card. A wireless base station will have several transceiver cards (essentially I/O line cards, but wireless), a WAN uplink card, a management card, and several cards that are specific to wireless (like baseband processing and power amplifier cards). A core router will have several high-speed I/O cards, several switch cards, and a robust set of management/control cards. In terms of our forecast, we include all the content on switching cards, WAN trunking cards, and management cards. We also include the content on line cards for SONET, T/E, and ATM. We exclude, however, content on line cards for DSL, DOCSIS (cable modem), PON, Ethernet LAN, and wireless basestations, as these are all included in other sections of this report. Note that in some cases, our forecast includes communications infrastructure ICs sold into applications other than communications infrastructure equipment. Examples would include discrete communications processors sold into wireless LAN access points and switching ICs sold into blade servers. The impact to our forecast is small but relevant.

251

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


T/E Carrier T-carrier is the first level of aggregation calls go through when being transported over the PSTN
1 2 3 DS-0 DS-1 1.544 Mbps 1 DS-2 6.312 Mbps DS-3 44.736 Mbps DS-4 274.176 Mbps

M01 M UX

2 3 4

M12 MUX

24 64 kbps DS-1C 3152 Kbps

3 4 5 6 7

M23 MUX

2 3 4 5 6 7

M34 MUX

Signal Level DS-0 DS-1/T1/J1 E1 DS-1C DS-2/T2/J2 E2 J3 E3 DS-3/T3 DS-3C DS-3C E4/DS-4E DS-4/T4 J4 E5

Number of DS-0 Channels 1 24 30 48 96 120 480 480 672 1,344 1,440 1,920 4,032 5,760 7,680

Transmission Rate 64 kbps 1.544 Mbps 2.048 Mbps 3.152 Mbps 6.312 Mbps 8.448 Mbps 32.064 Mbps 34.368 Mbps 44.736 Mbps 91.053 Mbps 97.728 Mbps 139.264 Mbps 274.176 Mbps 397.200 Mbps 565.148 Mbps

Geography North America, Europe and Japan North America and Japan Europe North America and Japan North America and Japan Europe Japan Europe North America North America Japan Europe North America Japan Europe

Note: Each DS-0 channel represents a single 64 Kbps phone line Source: CIBC World Markets Corp.

T-carrier is the first level of aggregation that telephone calls or data traffic go through when being transported over the PSTN. A basic DS-1/T1 signal multiplexes 24 DS-0 (each DS-0 is a single phone call, which is a 64 Kbps signal) channels into a single 1.544 Mbps signal, called a DS-1 or T1. The next level up is DS-2/T2, which multiplexes 4 DS1s together in a 6.312 Mbps stream (in practice, DS-2 is rarely used), followed by DS-3/T3, which multiplexes 28 DS1 signals into a 44.736 Mbps DS-3 or T3 signal. DS-1 and DS-3 signals are used by service providers when sending voice traffic around the metro or wide area network. Businesses also typically lease T1 (or T3) lines from the service provider that can provide a constant 1.544 Mbps (or 44.736 Mbps) channel, enough for 24 (or 672) simultaneous phone calls, which will be aggregated at the customer premise and sent as a digital signal (the DS in DS-1 and DS-3) to the PSTN. The bandwidth can also be used for the corporate data network in its interaction with the public WAN and the Internet. Note: Although we mostly focus on signal rates found in North America, it is important to note that the European digital hierarchy, represented with an E rather than a DS or T, multiplexes a different number of DS-0 signals at each level of the hierarchy. For example, an E1 signal multiplexes 30 DS-0s instead of 24. From the E1 signal rate and higher, each subsequently higher signal is formed by multiplexing four signals from the preceding digital signal level; an E2 signal consists of four multiplexed E1 signals and an E3 signal consists of four multiplexed E2 signals.

252

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


SONET/SDH SONET/SDH is the primary optical transport protocol used in telecommunications throughout the world
OC - 1
OC - 1 OC - 1 OC - 1
SONET MUX

OC - 3

OC - 12

OC - 48

OC - 192

OC - 768

2 3 4

SONET MUX

2 3 4

SONET MUX

2 3 4

SONET MUX

2 3 4

SONET MUX

Optical Carrier Level (OC-N) OC-1 OC-3 OC-12 OC-48 OC-192 OC-768

SONET STS Level (STS-N) STS-1 STS-3 STS-12 STS-48 STS-192 STS-768

SDH STM Level (STM-N) N/A STM-1 STM-4 STM-16 STM-64 STM-256

Transmission Rate 51.84 Mbps 155.52 Mbps 622.08 Mbps 2488.32 Mbps 9.953 Gbps 39.812 Gbps

Equivalent Channels DS-3 DS-1 DS-0 1 28 672 3 84 2,016 12 336 8,064 48 1,344 32,256 192 5,376 129,024 768 21,504 516,096

Note: Each DS-0 channel represents a single 64 Kbps phone line Source: CIBC World Markets Corp.

Synchronous optical networking (SONET) and synchronous digital hierarchy (SDH) are the primary protocols used by carriers for optical transport (SONET is in North America, SDH in Europe). SONET can reach much higher transmission rates than T/E-carrier, as it is specifically designed to run over optical fiber. SONET equipment can handle a wide range of speeds and therefore is used throughout the service provider network. Lower-speed SONET is used to send traffic within the metropolitan area between central offices, while higher-speeds are used to send traffic between metro areas or even countries. Voice calls and data traffic are aggregated into SONET frames that travel around the network, picking up and dropping signals as calls are placed, all at high speeds. The basic building block in SONET networks is the STS-1 frame, also called an OC-1. The frame has 51.84 Mbps of bandwidth, enough to hold an entire DS-3 signal plus the SONET overhead. When STS-1s are aggregated, they are transmitted at faster rates, such as 155 Mbps (OC-3), 622 Mbps (OC-12), 2.5 Gbps (OC-48), 10 Gbps (OC-192), or 40 Gbps (OC-768). Most metro area SONET equipment uses OC-3 or OC-12, while long-haul uses OC-48 or OC-192. Optical networking is fast, reliable and scalable, but it is also very expensive. In order to leverage the installed fiber and equipment infrastructure, OEMs and service providers have developed DWDM (dense wave division multiplexing) equipment, which can put multiple lambdas (wavelengths of light; each lambda can carry a SONET signal) on each fiber. This is strictly an optical technology, and does not affect the electrical properties of the signal. DWDM equipment is slowly replacing standard SONET equipment in most long-haul and metro networks. Although some equipment must be replaced to use DWDM, compatibility with legacy SONET is retained for the most part.

253

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


ATM (Asynchronous Transfer Mode) ATM is a switching protocol used primarily in service provider data and multi-service networks. The topology uses switches that establish a logical circuit from end-toend, guaranteeing quality of service (QoS), but unused bandwidth can be appropriated when needed.
ATM uses fixed-length 53-byte cells, and can switch voice, video, or data traffic at high speeds. Routers and modems feed traffic into the carrier network, which uses a series of ATM switches to handle the traffic and send it back out. ATM is a layer 2 protocol. Cells can travel between line cards or even between switches in a single network node using point-to-point links, but need to sit inside a layer 1 protocol such as SONET or T-carrier to travel over the WAN.

T/E and SONET area both efficient transport protocols for the WAN, but they cannot perform the important switching functions that direct traffic around the network. Most service providers employ the switching technology ATM (asynchronous transfer mode) in the core and edge of the network in order to switch voice, data, and even video traffic. Based on a cell-switching and multiplexing technology, ATM is a connection-oriented protocol for the transfer of fixed length cells, which can carry many different types of traffic. ATM equipment can specify different levels of quality of service for different cells; this is a key feature in making it the preferred choice for multi-service networks. It is highly scalable and supports transmission speeds equivalent to those of SONET. Unlike other data link layer protocols, ATM units are not variable in length, but consist of fixed 53-byte cells. ATM switches establish a logical circuit from end to end, which guarantees quality of service. However, unlike true circuitswitched protocols that dedicate circuits end to end, unused bandwidth in ATM's logical circuits can be appropriated when needed. In this way, ATM combines the benefits of circuit switching, including constant network latency/delay and guaranteed bandwidth, with those of packet-switching, including flexibility and better network utilization. The small, fixed cell size is ideal for the transfer of voice, video and other real-time traffic, as these types of traffic are not tolerant of the long download delays associated with long, variable length packets.

254

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Carrier Infrastructure Equipment
The carrier infrastructure market should be relatively stable with pockets of strength reflecting the continued evolution of carrier networks. IP-based switches and routers should displace legacy gear; broadband and video infrastructure should also grow nicely. Wireless infrastructuremore than half total market revenueshould peak in 2006 as 3G deployments slow. In total, we forecast a 2% CAGR, from $94 billion in 2005 to $102 billion in 2010.
140,000

Enterprise IT Infrastructure Equipment


The IT infrastructure market should outpace the carrier market, driven by strong growth in IT security, IP PBXs, Fibre Channel storage, and increasing penetration of enterprise WLAN. The market should remain highly diversified with only switching and IT security representing >10% of 2010 revenue. In total, we expect the market to grow at a 5% CAGR, from $33 billion in 2005 to $44 billion in 2010.
60,000

120,000

100,000

Wireless Equipment Video Infrastructure Broadband Infrastructure Carrier and Managed Voice Optical Transport ATM Switching IP/Ethernet Switching IP Routing

50,000

40,000 80,000

Other Infrastructure Equipment IT Security Fibre Channel Switches IP and Hybrid PBX Traditional PBX/Key System WLAN Switch/Server/Appliances Switches Routers

Revenue ($M)

Revenue ($M)

30,000

60,000

20,000 40,000

20,000

10,000

2003 2004 2005 2006 2007 2008 2009 2010

2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, DellOro, RHK, Infonetics, In-Stat, IDC, CIBC World Markets Corp.

Source: IDC, Dataquest, CIBC World Markets Corp.

Carrier Infrastructure Equipment Revenue Forecast: After several years of sluggish spending following the telecom bubble, the carrier infrastructure equipment market grew 13% in 2004 and another 13% in 2005 as service providers began spending again, with a focus on migrating their wireless networks to 2.5G and 3G, building out their broadband footprints, preparing for triple play services, and removing network bottleknecks. Wireless infrastructure far outpaced wireline over the past few years, and has grown to well over half of total equipment spending. Going forward, we expect mixed trends as wireless infrastructure peaks in 2006 as global 3G deployments slow. On the wireline side, IP switching and routing will likely see healthy growth as they replace legacy gear, and broadband and video infrastructure will grow nicely to support next-generation access and telco TV services. In total, we expect carrier infrastructure equipment to grow at a 2% CAGR, from $94 billion in 2005 to $102 billion in 2006. Enterprise IT Infrastructure Equipment Revenue Forecast: We expect the enterprise IT infrastructure market to outpace the carrier market, driven by strong growth in IT security (16% CAGR), IP PBXs (9% CAGR), Fibre Channel storage (7% CAGR), and increasing penetration of enterprise WLAN (31% CAGR). Note that strength in switches and IP PBX will be slightly offset by declines in basic enterprise routers and traditional PBXs (routing functionality is being incorporated into layer 3+ switches and VoIP is replacing traditional PBXs). The market should remain highly diversified with only switching and IT security representing greater than 10% of revenue in 2010. In total, we expect the market to grow at a 5% CAGR, from $33 billion in 2005 to $44 billion in 2010.

255

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Carrier Infrastructure Equipment Suppliers
DWDM & Optical Cross Connect
Others 20 % Nortel 16 % Cisco 6% Tellabs 5% NEC 6% Cisco 6% Siemens 6% Ericsson/ Marconi 7% Huawei 13 %

MSPP & ADM


NEC 6% Others 7% Alcatel 15 %

Alcatel 14 %

Nortel 8% Huawei 11 % Ciena 7% Lucent 10 %

Tellabs 12 %

Lucent 8%

Fujitsu 8%

Siemens 10 %

Multi-Service WAN Switch


Ciena 2% Others 5%

Service Provider Routers


Alcatel 3% Huawei 4% Redback 3% Others 4%

Marconi 5% Cisco 8% Nortel 38 %

Lucent 16 %

Juniper 29 %

Cisco 57 %

Alcatel 25 %

Source: DellOro, IDC, CIBC World Markets Corp.

This chart above (and the two following it) include market share figures for several classes of communications infrastructure equipment. Though not an exhaustive list of equipment included in this category, it gives a good sense of the key OEM players. Note that all market share figures are in revenue terms as opposed to ports, lines, or units. DWDM & Optical Cross Connect Suppliers: The top DWDM and OCC suppliers in 2005 were Nortel with 16% revenue share, Alcatel with 14%, Huawei with 11%, and Lucent with 10%. Other major suppliers include Ciena, Siemens, Cisco, NEC, and Tellabs. Multi-Service Provisioning Platform & Add/Drop Multiplexer Suppliers: The major suppliers of MSPP and ADM equipment in 2005 were Alcatel with 15% market share, Huawei with 13%, Tellabs with 12%, and Siemens with 10%. Fujitsu, Lucent and Siemens each held 8% and Ericsson/Marconi held 7%. Cisco and NEC followed, each with 6%. Multi-Service WAN Switch Suppliers: The top suppliers of multi-service WAN switches (includes ATM switches) in 2005 were Nortel with 38% revenue share, Alcatel with 25%, Lucent with 16%, and Cisco with 8%. Other major suppliers include Marconi, Ciena, and Tellabs. Service Provider Router Suppliers: Cisco was the No. 1 supplier of service provider routers in 2005 with 57% share, though its share is far lower than its overall router share (closer to 72%). Juniper was a strong No. 2 with 29% share (17% of total routers). Huawei held 4%, while Alcatel and Redback each held 3%.

256

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Wireless and Wireline Access Equipment Suppliers
Wireless Infrastructure
Others 9% Lucent 7% Ericsson 30 %

Central Office Switch


Others 23 % Alcatel 22 %

Alcatel 8% Siemens 12 %

Nortel 9%

Lucent 16 %

Motorola 10 %

Siemens 12 %

Nokia 14 %

Nortel 12 % Ericsson 15 %

DSLAM
Others 18 % Alcatel 33 % Arris 20 %

CMTS
BigBand 4% Others 1%

UTStarcomm 3% Zyxel 3% Sumitomo 3% Marconi 4% ECI Telecom 5% Lucent 7% Huawei 8%

Cisco 54 %

Siemens 9% NEC 7%

Motorola 21 %

Source: DellOro, Dataquest, CIBC World Markets Corp.

Wireless Infrastructure Suppliers: Ericsson was the top supplier of wireless infrastructure in 2005, with 30% revenue share. Ericsson remains the incumbent in GSM/GPRS, where it holds nearly 40% share overall; the company has a slightly lower share in W-CDMA. Nokia and Siemens followed with 14% and 12%, respectively, also mostly from GSM; note that in 2006 Nokia and Siemens announced that they would combine their wireless infrastructure equipment divisions. Motorola was fourth with 10% share and is strong in CDMA, though it does have a significant presence in GSM as well. Rounding out the top five was Nortel with 9% share from both CDMA and GSM. Alcatel was next with 8% share (mostly GSM/GPRS and W-CDMA), followed by Lucent with 7% share (almost entirely CDMA); in 2006 Alcatel and Lucent agreed to merge. Other vendors include Samsung, Huawei, ZTE, UTStarcomm, and Datang. Central Office Switch Suppliers: The top suppliers of central office switching equipment in 2005 were Alcatel with 22% revenue share, Lucent with 16%, Ericsson with 15%, and Nortel and Siemens, each with 12%. DSLAM Suppliers: Alcatel was the dominant supplier of DSLAMs in 2005 with 33% revenue share, followed by Siemens with 9% share and Huawei with 8%. NEC and Lucent each held 7% share, followed by ECI Telecom at 5% and Marconi at 4%. Other significant suppliers include Sumitomo, Zyxel, UTStarcomm, and Ericsson. Note that the market share data listed here differs from the modem section of this report, which deals in port shipments. CMTS Suppliers: The top CMTS supplier in 2005 was Cisco with 54% revenue share. Motorola and Arris followed with 21% and 20%, respectively. BigBand held 4%. Market share is for revenue and differs from the modem section.

257

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Enterprise IT Infrastructure Equipment Suppliers
Enterprise Routers
Nortel Juniper 1 % Others 4% Enterasys 1% Extreme 2% Foundry 2% 3Com 3%

Ethernet Switches
F5 1%

Others 11 %

Huawei 3%

3%

HP 4% Nortel 5%

Cisco 89 %

Cisco 71 %

Fibre Channel Switches


QLogic 8% Others 1% Others 31 % Cisco 20 % Brocade 40 %

PBX
Avaya 19 %

Siemens 15 %

Cisco 7% McData 32 % NEC 7% Alcatel 8% Nortel 13 %

Source: DellOro, IDC, CIBC World Markets Corp.

Enterprise Router Suppliers: The enterprise router market is dominated by Cisco with 89% revenue share. Huawei, with its footprint in emerging markets, was next with 3%. Juniper, which has struggled to move into the enterprise, was next, also with 3%. Other vendors include Nortel, Adtran, Allied Telesyn, 3Com, and Lucent. Ethernet Switches: Cisco was the dominant supplier of Ethernet switches in 2005 with 71% revenue share; note that Ciscos revenue share far exceeds its port share of 53% reflecting its positioning at the high end and the price premium it commands over competitive offerings. Nortel was a distant second at 5% share, followed by HP at 4% and 3Com at 3%. Other vendors include Foundry, Extreme, Enterasys, F5 Networks, Alcatel, and Allied Telesyn. Note that the share data listed here differs from the Ethernet section of this report, which deals in port shipments. Fibre Channel Switches: Brocade is the top supplier of Fibre Channel switches with 40% revenue share in 2005. McData was No. 2 with 32% share; McDatas revenue share exceeds its port share of 25% due to its positioning at the high end. Note that in 2006 Brocade announced that it would acquire McData. Cisco was the No. 3 player in 2005 with 20% share, almost entirely from chassis solutions. QLogic was next with 8%; most of its share was derived from fixed-port solutions. Market share is for revenue and differs from the storage section which deals in port shipments. PBX Suppliers (includes traditional, IP, and converged PBX): The top supplier of PBX equipment in 2005 was Avaya with 19% revenue share. Siemens held 15%, followed by Nortel at 13%. Alcatel was next with 8%, followed by NEC and Cisco, each with 7% share. Other vendors include Mitel, Aastra/EADS, and Inter-Tel.

258

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Comm Infrastructure Semiconductor Forecast
We expect communications infrastructure semiconductor revenue to grow at a 5% CAGR, from $3.1 billion in 2005 to $4.1 billion in 2010. Line card ICs should grow at a 3% CAGR, led by growth in SONET; T/E should grow more modestly and ATM should decline through 2010. Higher-layer and system-level ICs should grow 7%, led by communications processors; switching and network processing ICs should each grow 6%.
$4,500 $4,000 $3,500 $3,000
Unit Shipments (000s)

Line Card ICs T/E-carrier growing 2%, from $311 million in 2005 to $348 million in 2010. SONET/SDH growing 6%, from $451 million in 2005 to $594 million in 2009 ATM ICs declining 8%, from $146 million in 2005 to $96 million in 2010. Higher-Layer and System-Level ICs Comm processors growing 7%, from $857 million in 2005 to $1.2 billion in 2010. Network processing ICs growing 6%, from $822 million in 2005 to $1.1 billion in 2010. Switching ICs also growing 6%, from $520 million in 2005 to $682 million in 2010.
Note: each of the six sub-segments above is discussed separately in the cahrts that follow.

Switch Fabric and SerDes Network Processing IC Communication Processor ATM SONET/SDH T/E

$2,500 $2,000 $1,500 $1,000 $500 $2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, Dataquest, CIBC World Markets Corp.

Communications Infrastructure Semiconductor Forecast: We estimate that the market for communications infrastructure semiconductors totaled $3.1 billion in 2005. Nearly 30% of this total was derived from line card ICs, which we define as protocol-dependent ICs usually found on individual line cards and, to a lesser extent, WAN trunking cards. The remaining 70%+ was derived from higher-layer and system-level ICs, which we define as all processing and switching ICs that are not protocol dependent. This segment includes communications processors, network processing ICs, and switch fabrics, which are found on switch, control/management, and WAN trunking cards and to a lesser extent on individual line cards. Looking ahead, we expect the market to grow at a 5% CAGR, rising from $3.1 billion in 2005 to $4.1 billion in 2010. Line card ICs should grow at a 3% CAGR, led by SONET. T/E should grow more modestly as carriers continue to focus on optical, and ATM should decline through 2010 as IP makes a push into access and metro networks. Higher-layer and system-level ICs should grow 7%, benefiting from the additional intelligence OEMs are seeking to pack into their product offerings at each node in the network. Growth should be led by communications processors at a 7% CAGR, though network processors and switching ICs should also grow nicely, each with 6% CAGRs. Note: each of the six IC sub-segments is discussed separately in the pages that follow. The slide above is meant as a summary to size the opportunity for communications infrastructure ICs. In practice, the design competencies required to address each sub-segment of the market are highly specialized and thus require separate attention.

259

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


T/E Carrier Line Card
High-speed SRAM is the primary memory used in telecom and networking applications

SRAM Line Interface Unit (LIU)


Source: CIBC World Markets Corp.

Copper/ Coaxial Cable

Framer

HDLC Controller (optional)

Backplane

Interfaces with the copper or coaxial cable and performs the analog-to digital conversion to send signals up to the service provider network, and vice versa

Performs protocol processing, removing T/E overhead on the receive side and applying it on the transmit side

An HDLC controller is sometimes used to aggregate user data and manage transmission over multiple T/E links, and encapsulates user data in an HDLC frame

T-carrier semiconductors transmit and receive T/E-based signals over the PSTN and implement the T-carrier protocol. The LIU receives data from the transmission mediumusually copper twisted pair or coaxial cableand converts it to a digital format for processing. The framer removes or applies T-carrier overhead (recall that overhead is the additional bits of data attached to the signal payload that helps transport the data, but is not part of the original signal). The system can also run the data through an HDLC controller for further channel management. Line Interface Unit (LIU): The LIU interfaces with the transmission medium and converts it to and from the digital format for processing. Framer: The framer applies and removes T-carrier overhead, preparing the payload data for transmission over the network and extracting it from an encoded form on the receive side. HDLC Controller: An HDLC controller is sometimes used to aggregate user data and manage transmission over multiple T1/E1 or T3/E3 links. It encapsulates user data into HDLC (high-level data link control) frames and prepends header information that is used to manage network connections. Note that an HDLC controller is usually implemented as a communications processor; though we show it in the diagram above, the revenue associated with it is included in the forecast for communications processors. SRAM: High speed SRAM is the most commonly used memory in wireline applications.

260

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


T/E Carrier Semiconductor Forecast
The market for T/E carrier ICs is expected to grow at a modest 2% CAGR, from $311 million in 2005 to $348 million in 2010. The market enjoys consistent demand from wireless basestations, corporate data access, and metro/access aggregation and transport, but will see some displacement by optical links. By device, framers should grow to nearly 80% of revenue by 2010; PHYs should represent the balance.
$400 LIU $350 Framer

T/E Carrier Semiconductor Competitors


Others 10 % Exar 5% Freescale 6% Infineon 33 %

Mindspeed 9%

Maxim/Dallas 10 % Agere 13 % PMC-Sierra 14 %

$300
Semiconductor Revenue ($M)

Source: IDC, Dataquest, CIBC World Markets Corp.

$250

$200

$150

$100

$50

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, Dataquest, CIBC World Markets Corp.

T/E carrier Semiconductor Forecast: T/E carrier semiconductors, as the basic building block of service provider metro and access networks, enjoy consistent demand from applications such as wireless basestations, corporate data access, and aggregation/transport links around carrier access and metro networks. Over time, however, we expect to see increased displacement of lower-speed copper T/E links with higher-speed optical SONET links, limiting growth in this market. We also expect to see continued ASP declines, as IC vendors provide denser multi-port solutions and higher levels of integration. Overall we expect T/E semiconductors to grow at a five-year CAGR of 2%, from $311 million in 2005 to $348 million in 2010. In terms of individual devices, we expect framers (already 75% of the content) to slightly outgrow LIUs, as some newer designs migrate to integrated solutions which suck the LIU into the framer. T/E carrier Semiconductor Competitors: The top supplier of T-carrier semiconductors in 2005 was Infineon with 33% market share. PMC-Sierra was second with 14% share, followed by Agere with 13%. Maxim and Mindspeed followed with 10% and 9%, respectively. Freescale held 6% and T3/E3 specialist Exar held 5%. Other T/E suppliers include Cortina/Intel (Cortina purchased Intels T/E and SONET business in 2006), TDK, and IDT.

261

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


SONET/SDH Line Card
Laser Driver TIA Limiting Amp CDR

Optical Fiber

Mux FEC Demux Framer/Pointer Processor Framer Backplane

Physical Medium Dependent (PMD)


Source: CIBC World Markets Corp.

Physical Layer (PHY)

The PMD layer interfaces with the optical fiber Laser Driver - drives the optical laser to transmit the signal TIA and Limiting Amp - receives the signal from the optical receiver

The physical layer performs transmit and receive functions CDR - retimes the incoming signal Demux - receives the incoming signal and converts it to a digital parallel stream Mux - transmits the outgoing signal and converts it to a serial analog stream

The framers applies and removes SONET overhead Framer - terminates and removes SONET overhead on the receive side, applies it on the transmit side FEC - a specialized framer optimized to reduce errors over long distances

SONET line cards are found in high-speed optical transport and switching systems used in metro, edge, and core networks. In a typical line card implementation, the optical signal comes off the fiber and interfaces an optical module containing a block of ICs known as the physical medium dependent (PMD) block, which controls the optical receptor and sends the analog signal into the system. Physical layer (PHY) devices sitting behind the optical module then convert the signal to a digital bit stream for processing. The framer removes SONET overhead and sends the signal to the network processing block, which classifies, modifies and manages the packet before switching. On the transmit side, the process is reversed, with the framer applying overhead, the PHY converting the signal to analog and transmitting it through the PMD devices in the optical module over the fiber. Physical Medium Dependent (PMD) Transimpedence Amplifier (TIA) and Limiting Amp: The transimpedence amp and limiting amp receive the signal from the optical receiver. Laser Driver: On the transmit side, the laser driver controls the optical laser that sends the SONET signal across the fiber.

262

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Physical Layer (PHY) Clock and Data Recovery (CDR): On the receive side, the CDR retimes the signal from the PMD layer and prepares it for analog processing by the transceiver. Demultiplexer (Demux): The demultiplexer acts as the receiver and converts an incoming signal from an analog serial bit stream to a digital parallel stream for processing by the system. Multiplexer (Mux): The multiplexer acts as the transmitter and converts the signal from the digital parallel stream used by the system to the analog serial stream used by the optical components. Transceiver (not shown in diagram): A transceiver is an integrated physical layer device that performs clock and data recovery, mux, and demux functions on a single chip. Integrated transceivers are typically found supporting the slower, more mature speeds (OC-3 and OC-12), while faster next-generation speeds (OC-48, OC-192, OC768) require discrete components to perform physical layer functions.

Framer Layer Framer: The framer terminates and removes SONET overhead on the receive side and applies it on the transmit side. The framer also breaks out the SONET stream into smaller bandwidth signals for more granularity in processing (e.g., it can break an OC-48 into four OC-12s or twelve OC-3s, etc.). Mapper (not shown in diagram): A mapper is similar to a framer but includes the capability to translate between speeds or protocols. Mappers are often used in systems which provide transport for other protocols, such as ATM or Ethernet, over a SONET link. Forward Error Correction (FEC): Also called a digital wrapper, the FEC is a specialized framing device that uses the g.709 protocol to encapsulate the packet in an additional FEC frame for more reliable transmission over longer distances. FEC devices are only found in those systems designed to support g.709, generally higher speed long-haul equipment.

263

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


SONET/SDH Semiconductor SONET/SDH Competitors Semiconductor Forecast
We expect SONET ICs to grow at a 6% CAGR, rising from $451 million in 2005 to $594 million in 2010. Framers should remain about 55% of IC content, while FEC devices should quickly grow to greater than 10% of IC revenue by 2010. The remaining 30%-35% is split roughly evenly between PMDs and PHYs. Though a small percentage of ports, OC48 is the largest IC revenue segment; it should be surpassed by OC-192 by 2010.
$700 Framer $600 FEC PHY $500 PMD

Others TranSwitch 5 % 5%

Cortina/Intel 6% Broadcom 7%

Agere 24 %

Infineon 7%

PMC-Sierra 19 % AMCC 13 % Vitesse 14 %


Source: IDC, Dataquest, CIBC World Markets Corp.

Semiconductor Revenue ($M)

$400

$300

$200

$100

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, RHK, CIBC World Markets Corp.

SONET Semiconductor Forecast: We expect SONET semiconductor revenue to grow at a 6% CAGR, from $451 million in 2005 to $594 million in 2010. The market should continue to benefit from the incremental shift from copper to optical in the metro network, as well as a renewed focus on the core and edge of the network where bottlenecks have appeared. A shift to higher speed devices should also boost growth and help to offset ASP erosion. In terms of the device split, we expect framers to continue to represent about 55% of IC content, while FEC devices should quickly grow to more than 10% of IC revenue by 2010. The remaining 30%-35% is split roughly evenly between PMDs and PHYs. Note that SONET line cards will often have network processing, control plane, and even switching ICs; these are excluded from this forecast and discussed later in this section. SONET Semiconductor Competitors: The top SONET IC vendor in 2005 was Agere with 24% market share, followed closely by PMC-Sierra with 19%. Vitesse was third with 14% and was followed closely by AMCC with 13%; both vendors are positioned at the core and edge of the network. Infineon and Broadcom followed, each with 7% share; neither of these vendors puts much focus on SONET any longer, having refocused on broadband access and wireless technologies. Other significant vendors include Cortina/Intel (Cortina purchased Intels optical networking business in 2006), which held 6% share in 2005, and Transwitch, with 5% shareTranswitch is heavily focused on Ethernet-over-SONET. Other smaller suppliers include LSI Logic, Mindspeed, and Exar (through its 2006 acquisition of an OC-768 product line from Infineon and some internally developed products).

264

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


ATM (Asynchronous Transfer Mode) Line Card

Optical Fiber/ Copper Twisted Pair

Transceiver

ATM UNI

ATM Layer Device(s) ATM Layer

SAR

AAL

Backplane

Physical Layer
Source: CIBC World Markets Corp.

ATM Adaptation Layer

The physical layer performs transmit and receive functions Transceiver - interfaces with the transmission mediumdevice type varies based on the transport protocol used (i.e. SONET, T/E etc.) ATM UNI - converts traffic from the transport protocol to ATM, and vice versa

The ATM layer manages ATM traffic ATM layer devices can have a variety of functions performed on the cell stream, including address translation, cell appending, policing, cell counting, and OAM requirements for virtual concatenation

ATM adaptation devices encapsulate packets in ATM cells SAR - converts data packets into ATM cells, and vice versa AAL - deals with different types (classes) of traffic that can be captured in ATM cells, such as voice, video, audio, e-mail, FTP, web, IP, etc.

ATM-based equipment is a bit different than other types of carrier network equipment in that it applies the ATM protocol inside of another transport protocol (such as T/E or SONET). The transceiver can therefore be a variety of devices: a SONET transceiver, a T/E LIU, etc. On the receive side, the signal is passed from the transceiver to an ATM UNI (User Network Interface), which takes the digital signal from the transceiver and implements the ATM physical layer interface. It then goes to the ATM layer, where traffic management functions are performed, and then to the ATM adaptation layer, where traffic is translated from ATM cells into data packets and sent to the backplane for switching. The data packets are then sent back to a switching card, where they are switched, transformed back into ATM cells and sent out to another port.

Physical Layer Transceiver: The transceiver is the primary interface between the ATM line card and the transmission medium. The transceiver in an ATM system can be any number of devices depending on the transport protocol, including SONET/SDH transceivers, T1/E1 or T3/E3 LIUs. Alternatively, pure ATM traffic can be sent over point-to-point links between switches using standard ATM PHYs, though these links are usually used for intra-switch or switchto-switch signaling.

265

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

ATM UNI (User Network Interface): The ATM UNI converts traffic from the transport protocol (T/E, SONET, etc.) to the ATM protocol and vice versa.

ATM Layer ATM Layer Device(s): ATM systems can have a variety of functions performed on the cell stream performed at the ATM layer. This can include VPI/VCI address translation, cell appending, policing, cell counting and OAM requirements for virtual connections. IMA: (Inverse Multiplexer for ATM) (not shown in diagram): IMAs are used in ATM uplinks to manage ATM traffic over multiple lower-speed T/E and DSL physical links.

ATM Adaptation Layer (AAL) SAR (Segmentation and Reassembly): The SAR converts data packets into ATM cells and vice versa. AAL (ATM Adaptation Layer): The AAL device deals with the different types of data traffic that can be captured in ATM cells, also called Classes-of-Service (CoS). Classes of service can include compressed and uncompressed voice, video, audio, e-mail, FTP, World Wide Web, IP, etc.

266

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


ATM Semiconductor Forecast
We expect ATM semiconductors to decline at an 8% CAGR, from $146 million in 2005 to $96 million in 2010. ATM is actively being displaced by IP in broadband access equipment, and should be displaced by IP in basestations in the near future. ATM will retain a steady but limited role in multi-service WAN switches. PHY and UNI ICs should remain roughly 25% of revenue, the balance should be SAR and other ICs.
$250 SAR/Others PHY/ATM UNI $200

ATM Semiconductor Competitors


TranSwitch 4% Mindspeed 5% Others 5%

Zarlink 6%

PMC-Sierra 38 %

NEC 7%

Fujitsu 11 %

Agere 24 %
Source: IDC, Dataquest, CIBC World Markets Corp.

Semiconductor Revenue ($M)

$150

$100

$50

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, RHK, CIBC World Markets Corp.

ATM Semiconductor Forecast: We expect the market for ATM semiconductors to decline from $146 million in 2005 to $96 million in 2010, a negative 8% CAGR. Though data traffic growth continues, ATM is being challenged by Ethernet and pure-IP switching, which is far cheaper to implement (though less robust and therefore not appropriate for all applications). A key example of this is DSLAMs, which are transitioning from ATM backhaul to IP. Nextgeneration basestations are also likely to exhibit this trend, though the transition is in just the early stages today. We expect to see ATM retain a steady but limited role in multi-service WAN switches, where legacy compatibility is a necessary factor. Within our ATM semiconductor forecast, we expect PHY/ATM UNI devices to remain roughly 25% of revenue, with the balance in SAR and ATM layer devices. Note that ATM line cards will often have network processing, traffic management, control plane, and even switching ICs; these are excluded from this forecast and discussed later. ATM Semiconductor Competitors: The two dominant vendors of ATM semiconductors are PMC-Sierra and Agere. PMC-Sierra continued to lead the category in 2005 with 38% overall share; PMC is strong in WAN switches, wireless backhaul and broadband infrastructure, especially DSLAMs. Agere, with its strength in multi-protocol processing in edge and metro applications, was second with 24%. Fujitsu and NEC followed with 11% and 7% share, respectively. Other major suppliers include Zarlink with 6% share, Mindspeed with 5%, and TranSwitch with 4%.

267

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Communications Processors Equipment used in communications infrastructure usually includes specialized microprocessors which perform control plane functions, both on individual line cards and on management or supervisor cards. Usually an x86-, PowerPC-, or MIPS-based device, a communications processor sits outside the data path and performs control plane functions, managing the functionality of the line card or the entire system. Note that in addition to their function as the control plane processor, communications processors are often used to perform functions within the data path. Sometimes referred to as Integrated Communications Processors (ICPs), these devices are essentially fullyprogrammable network processors and are generally found on lower speed line cards.

Communications infrastructure equipment usually includes specialized microprocessors which perform control plane functions, both on individual line cards and on management or supervisor cards. Usually an x86-, PowerPC-, or MIPS-based device, a communications processor sits outside the data path and performs control plane functions, managing the functionality of the line card or the entire system. As they are fully programmable, communications processors are ideal for overall system control, which needs to be extremely flexible to support a constantly evolving set of features as well as individual customer preferences. In addition to their function as the control plane processor, communications processors are often used to perform functions within the data path. Sometimes referred to as Integrated Communications Processors (ICPs), these devices are essentially fully-programmable network processors (discussed in the following section) and are generally used on lower speed line cards. These ICPs often include specialty network processing logic around the processor core, and tend to look like SoCs. Note that this section includes standalone microprocessors used in large telecom and enterprise systems. It does not address processors used in modem CPE, wireless LAN access points, and other consumer networking devices, though the processors used in those applications are very similar to those found in infrastructure equipment. The market for standalone processors sold into CPE devices is quite small anyway, as processor cores are usually embedded into SoCs designed specifically for those markets.

268

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Communication Processor Forecast
We expect communications processors to grow at a 7% CAGR from $857 million in 2005 to $1.2 billion in 2010. X86-, MIPS-, and PowerPC-based processors should all continue to see wide adoption in telecom and enterprise infrastructure equipment. Processors continue to perform many critical data path functions, and have not lost much ground to NPUs or other dedicated ASSPs.
$1,400 Communication Processor $1,200

Communication Processor Competitors


IDT NEC 3 % IBM 3 % 3% Toshiba 3% Renesas 3%

Others 5%

PMC-Sierra 3% Broadcom 6% AMCC 7% Intel 14 %


Note: Excludes processors sold into networking CPE as well as into enterprise storage equipment. Source: IDC, Dataquest, CIBC World Markets Corp.

Freescale 49 %

Semiconductor Revenue ($M)

$1,000

$800

$600

$400

$200

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, RHK, CIBC World Markets Corp.

Communications Processor Forecast: Communications processors are pervasive in infrastructure equipment, included both on line cards and management cards that run the overall system. In 2005, we estimate that the market totaled $857 million. Looking ahead, we expect OEMs to continue to pack additional functionality into their high-end offeringsmuch of which gets implemented in softwaredriving demand for increased processing. Comm processors continue to perform many critical data path functions, and have not lost much ground to NPUs or other dedicated ASSPs. In total we expect a 7% CAGR through 2010, with the market growing to $1.2 billion. Communications Processor Competitors: The market for communications processors is served by vendors of x86-, PowerPC- and MIPS-based processors. Freescale dominates the category with its PowerQUICC family of PowerPC-based processors, and held 49% market share in 2005. Intel, which supplies both x86 as well as ARM-based processors, followed with 14% share. AMCC entered the market in 2004 after its acquisition of IBMs PowerPC 405 and 440 processor families, and held 7% share in 2005. Broadcom was fourth with 6% share; the company has been successful in ramping its MIPS-based SiByte family in routers. PMC-Sierra, with its MIPS-based offering for routers, was next with 3%; note that PMC significantly pared back its R&D efforts in MIPS router business in order to focus on printers, but is still achieving revenue as legacy products ramp down. Other vendors include Renesas, Toshiba, IBM, NEC, IDT, Fujitsu, Samsung, and Wintegra. Note that our forecast and market share figures include core logic and memory interface ICs associated with communications processors. These ICs are supplied by Marvell, IBM, and Tundra.

269

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Network Processing ICs Communications infrastructure equipment incorporates ASICs and specialized network processing devices to perform data path processing functions. These devices primarily relate to routing and traffic management and can be included on line cards, management cards, switch cards, or WAN trunking cards.
Network Processor (NPU) - A standard programmable processor optimized for packet classification, modification, and traffic management. Packet Classifier - A co-processor designed to classify packets based on content, usually helping to implement routing tables and QoS. CAM - A specialized SRAM-like co-processor designed to quickly look up routing addresses. Traffic Manager - A co-processor designed to manage traffic flow and implement QoS. Security Co-processor - A co-processor designed to accelerate networking security protocols, such as IPSec. ASICs Custom ASICs can perform one or all of the above functions.

Communications infrastructure equipment incorporates specialized network processing devices to perform data path processing functions. These devices primarily relate to routing and traffic management and show up on control cards, switch cards, WAN trunking cards, or individual line cards, depending on the system. Key types of devices include: Network Processor (NPU): A network processor is a standard programmable processor designed specifically to perform packet classification, modification and traffic management. Software is still a major component. Packet Classifier: A packet classifier is a co-processor designed to classify packets based on content, usually helping implement routing tables and QoS. CAM: A CAM is a specialized SRAM-like device designed to quickly look up routing addresses. CAMs are sometimes referred to as network search engines or IP co-processors. Traffic Manager: A traffic manager is a co-processor designed to manage traffic flow and implement QoS. First used in ATM-based equipment, traffic management is now a common feature in most network devices. Security Co-processor: A security co-processor accelerates networking security protocols such as IPSec. These processors often occupy their own line cards within larger systems or servers to implement security protocols. Network Processing ASICs: Custom ASICs can perform one or all of the above functions.

270

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Network Processor and Co-processor Forecast
Network processors are expected to grow at a 6% CAGR, from $822 million in 2005 to to $1.1 billion in 2010. ASICs should decline at a 3% CAGR as they are slowly displaced by NPUs, coprocessors, and comm processors/ICPs, shrinking from 42% of IC revenue today to 26% in 2010. NPUs, CAMs, and other co-processors should each grow double-digits.
$1,200
Other co-processor CAM

Network Processor and Co-processor Competitors


Others 24 % IBM 21 %

Intel 4% AMCC 4% Hifn 4% Cypress 6%

Netlogic 10 %

Fujitsu 10 % LSI Logic 7% IDT 9%

$1,000

NPU Network Processing ASIC

$800
Unit Shipments (000s)

Note: Includes ASICs, NPUs, and co-processors. Source: IDC, Linley Group, CIBC World Markets Corp.
$600

$400

ASICs
IBM Fujitsu LSI Logic NEC Texas Instruments Agere Toshiba

$200

$2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, RHK, Linley Group, CIBC World Markets Corp.

AMCC Intel Agere Wintegra Mindspeed Hifn Vitesse Bay Micro

NPUs

Co-processors
NetLogic IDT Cypress Hifn Cavium Broadcom SafeNet Freescale Renesas

Network Processor and Co-processor Forecast: The market for data path network processing ICs is driven not only by port/unit shipments of wireline equipment, but also by the level of intelligence required in these devices. Simple transport or switching equipment do not require much outside of physical layer and protocol processing; the addition of routing, QoS, traffic management, security, or other higher layer functionality, however, requires significant incremental silicon content. With most OEMs moving toward more intelligent edge and access designs, we expect the market for network processors to see healthy growth, with double-digit growth in next-generation ASSP segments. In total, we expected the market to grow at a 6% CAGR, from $821 million in 2005 to $1.1 billion in 2010. Note that a significant portion of the network processing IC market is still ASIC-based. In 2005, we estimate that 42% of network processing revenue was derived from ASICs. The remainder was split between standard network processors (21% of total), CAMs (26% of total), and other co-processors (11% of total). Going forward, we expect a continued migration from ASIC solutions to ASSPs, with ASICs shrinking to 26% of the overall market in 2010. Network Processor and Co-Processor Key Competitors: The largest suppliers of network processing silicon in 2005 were ASIC and CAM suppliers. IBM, the market leader in ASICs, held the top position with 21% share followed by CAM leader Netlogic with 10% share. Fujitsu, also strong in ASICs, was third with 10% share, followed by IDT, the No. 2 supplier of CAMs, with 9% share. LSI, the third largest ASIC supplier, was next with 7% share, and was followed by Cypress, the third largest CAM supplier, with 6%; note that Cypress exited the CAM market in 2006, selling the majority of the business to Netlogic. Leading security co-processor vendor Hifn and standard NPU vendors

271

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

AMCC and Intel rounded out the major vendors, each with 4% share. A myriad of other players also participate, including ASIC suppliers NEC, Texas Instruments, Agere, and Toshiba; NPU vendors Agere, Wintegra, Mindspeed, Vitesse, and Bay Microsystems; and co-processor vendors Cavium, Broadcom, SafeNet, Freescale, and Renesas. Looking at the segments individually, we estimate that IBM held 49% of the network processing ASIC market in 2005. Fujitsu followed with 23% and LSI Logic held 14%. Other vendors include Texas Instruments, Agere, and Toshiba. Note that PLD vendors Altera and Xilinx also participate in this market with structured ASICs based on their PLDs. On the ASSP side, the market for standard programmable NPUs was originally dominated by a handful of start-ups. Most of these have been acquired, while others sold their assets after becoming insolvent. Today, most of these products are produced by the larger comm IC vendors. In 2005, AMCC was the market leader with 20% market share followed by Intel and Agere with 18% and 13%, respectively. Other significant NPU vendors include Wintegra, Mindspeed, Hifn, Vitesse, and Bay Microsystems. Note that Vitesse and Mindspeed have significantly pared back their R&D efforts in NPUs, but are still achieving revenue as legacy products ramp down. In the co-processor market, we segment the market between CAMs and all other co-processors (note that the latter category is mostly security co-processors, as most packet classifier and traffic manager ICs have been folded into the larger network processing units or chipsets which are included in the NPU segment). In CAMs, two vendors were about even in 2005: Netlogic was the market leader with 39% share, followed by IDT with 36%. Cypress was the third largest supplier of CAMS with 23% share. Note that Cypress exited the CAM market in 2006, selling the majority of the business to Netlogic. In the security segment, Hifn, Cavium, Broadcom and SafeNet were each significant players in 2005. Other co-processor vendors include Vitesse, Tarari, AMCC, and Renesas. The exhibits below display market shares for network processing ASICs, NPUs, and co-processors in 2005. 2005 Network Processing ASIC Market Share 2005 Network Processor (NPU) Market Share
Others 10 % Bay Micro 5% Vitesse 7% IBM 49 % AMCC 20 %

Others 14 %

LSI Logic 14 %

Hifn 8%

Intel 18 %

Mindspeed 9% Fujitsu 23 % Wintegra 11 % Agere 13 %

2005 CAM Market Share


Others 3% Cypress 23 % Netlogic 39 %

2005 Other Co-processor Market Share


Others 9% Tarari 7% Hifn 24 %

Vitesse 9%

Safenet 14 %

Cavium 22 %

IDT 36 %

Broadcom 15 %

Source: IDC, CIBC World Markets Corp.

272

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Switching ICs Communications infrastructure equipment requires specialized switching components to direct traffic between line cards, usually over a backplane. Some platforms also include dedicated switching cards.
TDM Switches Time division switches divide the bandwidth of a shared transmission path into time slots, one time slot per input port. Data is addressed and sent over the bus or backplane in that time slot. Crosspoint Switches Crosspoint switches provide a unique transmission path from each input to each output by using a rectangular matrix of crosspoints. Packet and Cell Switches Fabrics These switches are used to switch individual packets or cells rather than complete streams. ASICs Switching is often implemented with custom ASICs. Backplane SerDes Backplane interface ICs are used on each card in the system to interface with the backplane. Standard Interconnect Newer advanced switching systems leverage standard interconnects like PCI-Express, RapidIO, and HT.
Note: Ethernet LAN switches are specifically excluded from this discussion, as they are protocol specific devices and are included in our discussion of Ethernet in the networking section.

Communications infrastructure equipment often includes a switching component to direct traffic between line cards. This function is usually implemented with a series of switch fabric devices, with some element on each individual line card and a central fabric located on the backplane or on a separate switch card. Key switching IC types include: TDM Switches Time division switches divide the bandwidth of a shared transmission path into time slots, one time slot per input port. Data is addressed and sent over the bus or backplane in that time slot. Crosspoint Switches Crosspoint switches provide a unique transmission path from each input to each output by using a rectangular matrix of crosspoints. Crosspoints are used in smaller fabrics than TDM switches. Packet and Cell Switches Fabrics These switches are used for packets or cells rather than complete streams. ASICs Switching is often implemented with custom ASICs. Backplane SerDes Backplane interface ICs are used on each card in to interface with the backplane. Standard Interconnect Newer advanced switching systems leverage standard interconnects like PCI-Express, RapidIO, and Hypertransport.

Note that Ethernet LAN switches are specifically excluded from this discussion, as they are protocol specific devices and are included in our discussion of Ethernet in the networking section.

273

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Switch Fabric Switch Fabric Semiconductor Forecast Semiconductor Competitors
We expect switch fabric and related ICs to grow at 6% CAGR through 2010, from $520 million in 2005 to $682 million in 2010. ASICs will continue to be heavily used, limiting growth prospects for ASSPs; both segment will grow low single digit through 2010. Standard interconnect should outgrow the market as OEMs leverage open standards like PCI-Express, RapidIO, and Hypertransport.
$800 Backplane ICs $700 Standard Interconnect Switch Fabric ASSPs Switch ASICs $600
Semiconductor Revenue ($M)

Others 20 %

IBM 14 %

Fujitsu 12 % Agere 6%

LSI Logic 7% Zarlink 7% Vitesse 7% AMCC 8% Tundra 9%

PLX 11 %

Source: IDC, Dataquest, Linley, CIBC World Markets Corp.

$500

$400

$300

Switching ASICs
IBM Fujitsu LSI Logic Agere NEC Toshiba TI

Switching ASSPs
AMCC Vitesse Zarlink Agere PMC-Sierra Dune TranSwitch

$200

$100

$0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: IDC, RHK, CIBC World Markets Corp.

Switching IC Forecast: We estimate that the market for switch ICs totaled $520 million in 2005, split roughly 36% switching ASICs, 27% switching ASSPs, 28% standard interconnect, and 9% backplane ICs. We expect the market to grow at a 6% CAGR, from $520 million in 2005 to $682 million in 2010. Unlike in the network processing market, OEMs continue to favor ASICs for switching; we forecast a 4% CAGR through 2010. This will limit growth prospects for switch ASSPs, which we expect to grow only 3%. Backplane ICs should experience healthy growth with a 7% CAGR, driven more by line card shipments than any factor relating to switching capacity. Standard interconnect ICs (PCI, PCI Express, RapidIO, HyperTransport, ASI, etc.) are expected to outperform the market with a 9% CAGR. These products leverage open standards and over time should breed low cost, highperformance solutions that will also simplify system design, displacing proprietary solutions that dominate the market today. For this reason, we also include standard interconnect ICs for backplanes in the emerging technologies section of this report. Note that in addition to communications infrastructure, these products find heavy usage in server (especially blade server) and storage applications. Switching IC Key Competitors: The top two vendors of switch fabric ICs are ASIC suppliers IBM and Fujitsu. In 2005 IBM held 14% market share; note that IBMs market share excluded the PRS product line that it sold to AMCC, who holds the No. 1 spot on the switching ASSP side. Fujitsu was second with 12% share. Standard interconnect leaders PLX and Tundra followed with 11% and 9% share, respectively. Next were switch ASSP vendors AMCC,

274

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Vitesse, and Zarlink, followed by LSI Logic and Agere (note that in 2006 LSI Logic and Agere agreed to merge). A number of players have smaller positions, including PMC-Sierra, NEC, Mindspeed, Broadcom/Sandburst, Intel, Cypress, TranSwitch, IDT (through its Zettacom acquisition), NEC, Toshiba, Texas Instruments, and Samsung, along with start-ups Dune, StarGen and Erlang. Looking at the individual segments, switch ASICs are supplied by the same three major vendors that supply network processing ASICS: IBM (39% share), Fujitsu (32% share), and LSI Logic (12%). On the merchant ASSP side, AMCC was the No. 1 vendor in 2005 with 23% share, but in 2006 the company announced its intention to halt future product development in this area. Vitesse was the No. 2 vendor with 21% share, with a portfolio of crosspoint switches, backplane interface ICs, and other fabrics. TDM switch leaders Zarlink and Agere were next with 20% and 14% share, respectively. Other vendors include PMC-Sierra, Dune, TranSwitch, Mindspeed, Broadcom/Sandburst, Stargen, IDT, and Erlang. Standard interconnect ICs are dominated today by PLX Technology and Tundra, with 36% and 31% share, respectively. Intel also participates, as does start-up Stargen. IDT has also made some major product announcements in this area and expects revenue in 2007. The exhibits to the right and below display market share data for the major segments of the switching market. 2005 Switch Fabric ASSP Market Share
TranSwitch 4% Dune 5% PMC-Sierra 9% Intel 20 % Agere 14 % Others 5% AMCC 23 % Stargen 4% Fujitsu 32 %

2005 Switch Fabric ASIC Market Share


Others 12 %

LSI Logic 16 %

IBM 39 %

2005 Standard Interconnect IC Market Share


Others 9%

PLX 36 %

Vitesse 21 %

Zarlink 20 %

Tundra 31 %

Source: IDC, Dataquest, Linley, CIBC World Markets Corp.

275

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Communications Infrastructure Communications Infrastructure


Comm Infrastructure IC Market Trends
Line Card ICs OEMs are flocking to multi-port T/E devices at the lower speeds. IC vendors are increasingly integrating the T/E LIU with the framer. Integrated SONET transceivers are becoming industry standard. Framers are offering increasing granularity and aggregation support. Intelligent mappers enabling Ethernet-over-SONET (EoS). SONET FEC devices are gaining increasing adoption. ATM has replaced Frame Relay in most carrier switching networks, but is starting to be challenged by IP-based solutions. DSLAMs are moving from ATM to IP, basestations should follow soon. Higher-Layer and System-Level ICs Communications processors are integrating advanced networking logic. NPUs are gaining traction, especially in access and metro applications. CAMs continue to penetrate the LAN backbone and are being used in service provider networks as well. Security processors are gaining traction in routers and LAN backbones. IPv6 is driving the need for more advanced network processing ICs. Switch fabric vendors are focusing on granularity and switching capacity. Standard interconnects are gaining traction as OEMs leverage open standards like PCI-Express, RapidIO, and Hypertransport. AdvancedTCA is starting to gain momentum.
T/E-Carrier Semiconductor Market Trends OEMs are flocking to multi-port devices at the lower speeds. Over the past few years, OEMs have increasingly chosen multi-port LIUs and framers for their T1/E1 line cards, with configurations of up to 8 or 12 ports per device. T3/E3 designs are also going this direction but the design is more challenging. IC vendors are increasingly integrating the PHY with the framer. Several vendors offer integrated PHY/framers for T1/E1 and T3/E3, reducing bill-of-material costs and saving board space. SONET designs will be more challenging and will likely be relegated to the lower speeds.

SONET Semiconductor Market Trends SONET PHY and framer ASICs continue to get squeezed out; the trend toward outsourcing is moving steadily up the layer stack. Optical networking OEMs traditionally competed on performance and have therefore chosen to do much of their silicon as custom-designed ASICs. As the market has matured and cost and time-to-market have become more important, however, OEMs have steadily moved to standard products. Today, the majority of PHY devices and framers are implemented as standard products. Higher layer devices such as network processors and switch fabrics continue to remain primarily ASIC-based, though even this is starting to change. The availability of larger, faster and more extensively libraried FPGAs is also cannibalizing ASICs.

276

Semiconductors: Technology and Market Primer 4.0 - January 17, 2007

Integrated SONET transceivers are becoming industry standard. With each new speed deployment, IC vendors enter the market with discrete CDR, mux and demux chips, which they then follow up with integrated transceivers that perform all physical layer functions. At present, integrated OC-192 transceivers are standard in most new designs. Framers are becoming more advanced, offering increasing granularity and aggregation support. Most framers sold today as merchant products provide basic SONET overhead framing, with the ability to step down one speed grade, e.g., from an OC-48 to four OC-12s, etc. Recent product introductions have seen a greater focus on granularity, i.e., the ability to step down even further. Several vendors offer framers that provide granularity from OC-48 all the way down to DS-3 or even further. Intelligent mappers are enabling Ethernet-over-SONET (EoS). Recognizing that carriers would like to offer Ethernet-based data services but still use SONET for transport, OEMs and IC vendors are trying to make it easier to translate between these protocols. Ethernet-over-SONET products are now available from a number of incumbent and start-up IC vendors supporting speeds of OC-3, OC-12, and OC-48. Platforms designed around these products are now ramping to significant volume. SONET FEC devices are gaining increasing adoption. FEC digital wrappers ensure a more reliable signal transmission, especially over long distances. OEMs have been quick to adopt these devices since the g.709 standard was ratified, and systems utilizing these FECs are now ramping to significant volume. Note that for the next generation, IC vendors are offering integrated framer/FEC devices, especially at higher speeds where FEC is more likely to be implemented.

ATM Semiconductor Market Trends ATM has replaced Frame Relay in most carrier switching networks, but is starting to be challenged by IP-based solutions. Frame Relay is a legacy protocol originally designed to interconnect LANs and to be a highspeed WAN protocol. As data networks have scaled, however, Frame Relay has been replaced by Ethernet in the LAN and by ATM in the WAN. Though ATM will remain the mainstream WAN switching choice for some time, Ethernet WAN is beginning to eat into ATM especially for access and metro equipment. DSLAMs are moving from ATM to IP for backhaul. ATM has been a natural fit for DSL backhaul, as it was fast, scalable, and able to handle both the voice and data traffic sent over the copper wire. ATM equipment also operates at the same speeds as SONET, making for an easy interface to the public WAN. With the growth of IP in the WAN, however, the major DSLAM vendors are migrating to IP for backhaul, eliminating ATM from the DSLAM. As with all technology shifts in carrier networks, this trend is taking several years time to play out, but it is a major driver behind the negative CAGR in our ATM forecast. 2.5G and 3G basestations are using ATM for backhaul; IP will come later. As in the DSLAM market, ATM has been an effective solution for backhaul in wireless basestations, especially to support 2.5G and 3G services which mix voice and data. The increased focus on data services at wireless carriers should provide some incremental demand for ATM ICs for several years, helping to buffer the effects of the transition away from ATM in DSLAMs, which is currently in progress. We do expect IP to begin to replace ATM in basestations as well sometime in the next few years, however, which will eventually contribute to the decline of ATM.

Communications Processor, Network Processor and Switching Market Trends Communications processors are beginning to integrate more advanced networking logic, blurring the line between comm processors and NPUs. Freescale has been very aggressive on this front, evolving its PowerQUICC comm processor line to include Integrated Communications Processors (a term it coined when it was part of Motorola). AMCC, which acquired two PowerPC processor lines from IBM in 2004, is evolving its product line to include IP from the companys network processors (AMCC is actually still No. 1 in NPUs and has very mature IP blocks). Wintegra, the most successful NPU start-up the market has seen in the past few years, has been very successful with an architecture that combines a MIPS core with network processing engines driven by software. NPUs are gaining traction, especially in access and metro applications. Merchant solutions for network processing have not lived up to original rosy expectations, especially at the high-end where they were originally targeted. In recent years, however, IC vendors have refocused on the lower speeds, and have been successful in displacing ASIC