0 Bewertungen0% fanden dieses Dokument nützlich (0 Abstimmungen)

18 Ansichten67 SeitenBJT Dc Parameter Estraction

Jun 21, 2016

© © All Rights Reserved

PDF, TXT oder online auf Scribd lesen

BJT Dc Parameter Estraction

© All Rights Reserved

Als PDF, TXT **herunterladen** oder online auf Scribd lesen

0 Bewertungen0% fanden dieses Dokument nützlich (0 Abstimmungen)

18 Ansichten67 SeitenBJT Dc Parameter Estraction

© All Rights Reserved

Als PDF, TXT **herunterladen** oder online auf Scribd lesen

Sie sind auf Seite 1von 67

Bipolar Transistors

Martin Linder

Department of Microelectronics and Information Technology

Device Technology Laboratory

Stockholm 2001

Bipolar Transistors

Martin Linder

Department of Microelectronics and Information Technology

Device Technology Laboratory

Stockholm 2001

in partial fulfillment of the requirements for the degree of Teknisk Doktor.

KTH, Royal Institute of Technology

Department of Microelectronics and Information Technology

Electrum 229

S-164 40 Kista

SWEDEN

ISRN KTH/EKT/FR-2001/2-SE

ISSN 0284-0545

TRITA - EKT

Forskningsrapport 2001:2

ISRN KTH/EKT/FR-2001/2-SE,

KTH, Royal Institute of Technology, Department of Microelectronics and

Information Technology, Stockholm 2001.

Abstract

This thesis deals with DC parameter extraction and modeling of bipolar

transistors (BJTs). Parameter extraction is performed using the dual base terminal BJT

test structure. Extraction procedures for compact model parameters such as the emitter

resistance RE and the intrinsic and the extrinsic base resistance (RBInt and RBExt

respectively) are presented. The dual base terminal test structure is also utilized for

DC extraction of the bias dependent Early voltages VEF and VER. From the extracted

Early voltage parameters, it is shown that the built-in potential of the base-emitter and

the base-collector junctions together with the grading coefficients of the junctions can

be extracted from DC measurements. Normally, extraction of these parameters

requires capacitance-voltage measurements. Further on, extraction of the zero bias

intrinsic base sheet resistance RS00 is demonstrated by considering the Early effect.

The concept of the dual base terminal test structure is also applied to the collector

region and a new type of test structure is proposed. The test structure uses two

terminals for the base and two terminals for the collector. It is shown that using the

new test structure, it is possible to directly measure the constant part of the collector

resistance RCC and to monitor the onset of the base push-out effect (Kirk effect).

Finally, a general model formulation for the distributed nature of the base-emitter

junction is presented. The proposed model includes both the Early effect and the

effect of charge storage in the base. Therefore, accurate calculation of the base

resistance and the potential, current and base charge distribution along the intrinsic

base region is enabled.

distributed model, base resistance, emitter resistance, collector resistance, Early

effect, base push-out, current crowding, base conductivity modulation.

Acknowledgements

First of all I would like to thank Prof. Mikael stling for giving me the

opportunity to pursue my Ph.D. degree at the department. He is always encouraging

and optimistic. I have had a good time working with you. I am also grateful to Dr. Jan

V. Grahn who, with his excellent lecturing, got me interested in the subject. Dr. Jan

V. Grahn also supervised my Master of Science thesis and I am glad that it worked

out so well that he offered me a Ph.D. student position in the bipolar group, regardless

of my background in material physics. One of the strongest impressions from the

beginning of my studies was the time when Dr. Shili Zhang and myself sat down to

design test structures to be manufactured in our in - house process BIP#2. I remember

we thought something may come out of it but I could never imagine that it would

turnout to be so rewarding. Maybe he did. During the years we have had many

discussions around device physics and mathematics. In short, he taught me a lot,

thanks.

During my second year I attended a summer school in France called MIGAS.

There I met Prof. Kjell Jeppson and Fredrik Ingvarson. We started a discussion

regarding the extraction of parasitic resistances. It turned into a long and fruitful

collaboration. Without them, I am sure that the scope of this thesis would have been

totally different. Many thanks to both of you!

Dr. Ping-Xi Ma, who spent a post doc year in our group, is acknowledged for

encouraging discussions and that he always was there to help out. Of course, the rest

of the Bipolar group are acknowledged: Martin Sandn (many thanks for Fig. 1 in

Paper 2, Fig. 3 in Paper 3 and Fig. 1 in Paper 4), Johan Pejnefors, Gunnar Malm,

Dr. Wlodek Kaplan, Erdal Suvar and Dr. Tord Karlin. It has been stimulating and fun

working with you. I think all the people that work within the EKT and FTE

departments deserve my sincere gratitude. It has been five very stimulating years,

thank you all!

Finally, I would also like to take the opportunity to thank my family for

encouraging me to pursue the Ph.D. degree. Most of all, I would like to thank my

fiance Jessica and our daughter Cajsa. Their inspiration, patience and support have

been invaluable during my time as a Ph.D. Student.

Martin Linder, Kista 2001

Appended Papers

1.

Effect of Emitter Overetch and Base Implantation Tilt on the Performance of

Double Polysilicon Bipolar Transistors, Physica Scripta, T79, p.246-249, 1999.

2.

M. Linder, F. Ingvarson, K.O. Jeppson, J.V. Grahn, S.-L. Zhang and M.stling,

Extraction of Emitter and Base Series Resistances of Bipolar Transistors from

a Single DC Measurement, IEEE Trans. Semicond. Manufact., 13, p.119-126,

2000.

3.

M. Linder, F. Ingvarson, K.O. Jeppson, J.V. Grahn, S.-L. Zhang and M.stling,

On DC Modeling of the Base Resistance in Bipolar Transistors, Solid-State

Electronics, 44, p.1411-1418, 2000.

4.

of the Intrinsic Base Region Sheet Resistance in Bipolar Transistors, In Proc.

Bipolar and BiCMOS Tech. Meeting, p.184-186, 2000.

5.

stling., A New Test Structure for Parasitic Resistance Extraction of Bipolar

Transistors, In Proc. IEEE Int. Conf. Microelectronic Test Structures, p.25-30,

2001.

6.

F. Ingvarson, M. Linder, K.O. Jeppson, S.-L. Zhang, J.V. Grahn and M. stling,

A Procedure for Characterizing the BJT Base Resistance and Early Voltages

Utilizing a Dual Base Transistor Test Structure, In Proc. IEEE Int. Conf.

Microelectronic Test Structures, p.31-36, 2001.

7.

in Bipolar Transistors Including the Early Effect, In manuscript.

8.

Grahn, S.-L. Zhang and M. stling, Extraction of Parasitic Resistances for

Silicon High Frequency Bipolar Transistors, In Proc. 1997 Gigahertz Meeting,

Oct 1997.

9.

Zhang and M. stling, Enhanced High Frequency Performance of IDP-Emitter

Bipolar Junction Transistors by Hydrogen Passivation, In Proc. 1997

Gigahertz Meeting, Oct 1997.

10.

Thin Films using Oxygen Radicals: Effect of Deposition Temperature, Journal

of Vac. Sc. and Tech., 16, p.2495-2500, 1998.

11.

Zhang, J. V. Grahn and M. stling, A Simplified High-Speed Bipolar Process

with Ti Salicide Metallization: Implementation of In Situ p-Doped Polysilicon

Emitter, Physica Scripta, T79, p.318-321, 1999.

12.

Doped Launcher Layer in SiGe Heterojunction Bipolar Transistors, In Proc.

RadioVetenskap och Kommunikation, 1999.

13.

H. H. Radamson, P. Jnsson, M. Jargelius, H. Fosshaug, M. Linder, G.

Landgren and M. stling, A High Speed SiGe HBT Process using

Non-Selective Epitaxy and In Situ Phosphorus Doped Emitter, In Proc.

European Solid-St. Dev. Research Conf., p.728-731, 1999.

14.

M. Linder, F. Ingvarson, K.O. Jeppson, J.V. Grahn, S.-L. Zhang and M.stling,

A New Procedure for Extraction of Series Resistances for Bipolar Transistors

from DC Measurements, In Proc. IEEE Int. Conf. Microelectronic Test

Structures, p.147-151, 1999.

15.

Mohadjeri, J. Pejnefors, H. H. Radamson, M. Sandn, Y. B. Wang, G. Landgren

and M. stling, A Low-Complexity 62-GHz fT SiGe Heterojunction Bipolar

Transistor Process using Differential Epitaxy and In Situ Phosphorus-Doped

Poly-Si Emitter at Very Low Thermal Budget, Solid-State Electronics, 44,

p.549-554, 2000.

16.

H. H. Radamson, P. Jnsson, M. Jargelius, H. Fosshaug, M. Linder, G.

Landgren and M. stling, "A 62-Ghz fT SiGe HBT Process," In Proc. 5th

Symposium on Gigahertz Electronics, p.11-14, 2000.

17.

Model for Space-Charge Region Capacitance Based on Practical Doping

Profiles under Any Bias Conditions, Solid-State Electronics, 45, p.159-167,

2001.

18.

Modeling of the Planar and Peripheral Parts of the BaseEmitter Capacitance,

In manuscript, 2001.

19.

Early Effect in Bipolar Transistors from DC Measurements, In manuscript,

2001.

Table of Contents

1. INTRODUCTION ...............................................................1

2. THE BIPOLAR TRANSISTOR .............................................3

2.1. Fabrication of BJTs .............................................................. 3

2.2. Bipolar Transistor Operation ................................................ 7

2.3. Compact BJT Modeling...................................................... 14

3.1. Basic Concepts .................................................................. 20

3.2. Distributed Transistor Modeling.......................................... 22

3.3. Compact Models of the Base Resistance .......................... 25

3.4. Small Signal Modeling of the Base Resistance .................. 26

4.1. Intrinsic Parameters ........................................................... 30

4.2. Parasitic Resistances......................................................... 33

4.3. Dual Base Terminal Test Structure .................................... 38

4.4. Dual Collector Terminal Test Structure .............................. 39

6. REFERENCES ...............................................................45

7. SUMMARY OF THE APPENDED PAPERS ...........................51

Chapter 1. Introduction

1. Introduction

The tremendous market growth for the electronic industry in the

past years is predicted to continue [1]. Even though most of the

microelectronic applications are digital systems manufactured in CMOS

technology (Complementary Metal-Oxide-Semiconductor), properties

such as low noise performance, high frequency operation and high

linearity renders state-of-the-art Si and SiGe bipolar transistors an

attractive technology choice for high frequency analog and high-speed

digital applications [2]. Typical circuit elements utilizing bipolar

transistors are Voltage Controlled Oscillators (VCOs), Low Noise

Amplifiers (LNAs) and mixers for high-speed circuit operation.

During the last decade, the market for wireless applications and

high-speed integrated digital circuits has switched from purely military

and space applications to the consumer market. The low cost issue has

therefore become more important and silicon technology is the most costeffective choice. However, the competition from CMOS and III-V

technologies necessitates a continuous development of Si and SiGe bipolar

transistor technologies. As the technology is improved and the device

dimensions are scaled down, improved accuracy of the compact modeling

of the device behavior is required by the circuit designers [3-7]. This

1

coupling to the device physics and accurate determination of model

parameters. Further on, the increased complexity of the manufacturing

process inherently translates to increased production costs. Fast and simple

characterization techniques for automated wafer mapping are thus required

in order to enable yield control during manufacturing.

As the title suggests, this thesis deals with parameter extraction

and modeling of high performance bipolar transistors. In Chapter 2,

general aspects of bipolar transistor operation and compact modeling are

discussed. Models of the bias-dependent base resistance are reviewed in

Chapter 3, both regarding compact models and more complex model

formulations. Chapter 4 discusses parameter extraction for compact

bipolar transistor models, focusing on DC measurements. Finally, in

Chapter 5, concluding remarks are given together with suggestions for

future work.

In this chapter, the different fabrication technologies of bipolar

junction transistors (BJTs) are explained with emphasis on the double

poly-Si technology. Second order effects that occur in the BJT, especially

at high current levels, are discussed. Finally, a brief presentation of the

different compact models of the BJT is given.

Bipolar transistor process technologies are under constant

development. It can be noticed that the process schemes are not as unified

as for CMOS technology, where every company seems to follow the same

path of development. For bipolar transistors, a variety of different process

schemes exists, all showing both strengths and weaknesses.

The first technology reviewed here is the single polysilicon

process. One of the main advantages of this technology is the

a)

LOCOS

n+

n-

p+

n+

p+

b)

Intrinsic base implant

Resist

LOCOS

n+

n-

p+

n+

p+

c)

n+ poly-Si

LOCOS

p

n-

n+

n+

p+

d)

E

LOCOS

p+

C

B

B

p+

p+

n+

nn+

p+

BiCMOS process offers the advantages of having both bipolar,

NMOSFETs and PMOSFETs in the same process. The process flow of the

bipolar transistors is not very different from that of the MOSFET

transistors. A simplified process flow is described in Fig. 1.

The double poly-Si bipolar technology not only uses an n+ doped

poly-Si emitter but also allows the extrinsic base to be contacted with a p+

doped poly-Si layer. This feature enables full self-alignment of the active

a)

LOCOS

n-

p+

p+

n+

p

b)

n+ Collector plugg implant

Resist

LOCOS

n-

p+

p+

n+

p

c)

Intrinsic base implant

Si02

p+-poly-Si

LOCOS

+

p+ n

p+

n+

n+

p+

d)

E

B

p -poly-Si

LOCOS

p+

p+ n

p+

n+

n+

p+

layers of the transistor [8, 9]. A schematic process flow for a double polySi technology is shown in Fig. 2.

In order to minimize the fabrication time required to complete the

whole process, a simplified double poly-Si process was developed [10].

The process was simplified by omitting both trench isolation (as well as

junction isolation) and metallization. Consequently, the process is

completed using only four lithographic mask steps. The transistor is

electrically probed directly on the silicide that is formed on the emitter and

base poly-Si. However, the design of the intrinsic BJT is identical to the

full-scale process. As can be seen in Fig. 2(a), the first lithography step of

the process is the LOCOS mask, which defines the active areas and the

collector contacts. The second mask is the collector plug implantation

mask, Fig. 2(b). After deposition and boron implantation of the base polySi, an oxide is deposited which serves as an isolating layer between the

emitter and base poly-Si. Then the base poly-Si mask is exposed and both

the oxide layer and the base poly-Si layer are etched. Figure 2(c) shows

the structure after removal of the resist followed by a drive in anneal for

forming the extrinsic base regions. The intrinsic base is then implanted

through the emitter window. After spacer formation the emitter poly-Si is

deposited and implanted. The emitter poly-Si also serves as contact to the

collector window. The emitter poly-Si and the underlying oxide are then

etched after exposure and development of the last mask. Finally, the

uncovered areas of both the base and the emitter poly-Si are silicided.

The simplified process was used as a test vehicle for implementing

new process segments, e.g. in-situ phosphorous doped poly-Si

emitter [11]. Further on, the effect of the emitter window opening and the

intrinsic base implantation was investigated. As can be seen in Fig. 2, the

emitter window is opened by etching through the p+ poly-Si. Ideally, the

etching should stop at the interface between the p+ poly-Si and the n- epiSi. Since there is no possibility to detect this interface during etching, an

overetch is performed to guarantee that the n- epi-Si is reached. In addition

to this effect the intrinsic base implantation is usually tilted 5-7 in order

to avoid channeling of the implanted species. The relatively high sidewalls

of the emitter window can then shadow parts of the window from the

implantation. Together with the overetch, the shadowing may cause the

formation of a potential barrier between the extrinsic and the intrinsic base

at the shadowed side of the window. Naturally, this causes an increase in

the base resistance and consequently a decrease in the maximum

frequency of oscillation, fMAX. However, near the potential barrier between

the extrinsic and the intrinsic base regions, the intrinsic base will be

thinner than below the emitter. This raises the maximum cut-off frequency

fT,MAX since the base transit time is reduced. The effect of the overetch on

transistor is investigated in detail in Paper 1. In order to avoid the overetch

effects on the process stability, a thin oxide layer between the p+ poly-Si

and the n- epi-Si may be introduced [12]. This oxide is used as an etch stop

for the emitter window etch and the overetch is thus avoided. However,

formation of a low resistance base link between the extrinsic and the

intrinsic base regions becomes more elaborate.

The BJT is essentially two pn diodes connected back to back, see

Fig. 3. From basic pn diode theory it is well known that the current

passing through the diode is given by:

VT

I = IS e

1

(1)

VT (= kT/q 0.026 V) is the thermal voltage and IS is the saturation

current. Ideally, the base and collector currents should be similarly

dependent upon the applied voltages. However, in reality a large number

of effects occur which causes a deviation from the ideal theory. This is

clearly illustrated in Fig. 4, where the base and collector currents are

plotted versus the base-emitter voltage. The dotted lines are the ideal

behavior according to the simple theory and the solid lines represents a

real transistor.

E

n+

B

n-

n+

the equivalent circuit of the ideal case.

The deviation of the measurements from the ideal behavior in the high

current regime, shown in Fig. 4, is due to various second order effects.

These second order effects will be described in the following paragraphs.

-2

10

-3

10

-4

10

-5

10

-6

10

IC

-7

10

-8

IB

10

-9

10

-10

10

-11

10

-12

10

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Figure 4. A typical Gummel plot for a BJT. The solid lines are real

transistor data and the dotted lines represent the ideal case.

When changing the bias of the transistor, the width of the depletion

layers of the base-emitter and base-collector junctions will change. This

modulates the thickness of the quasi-neutral base region, and thus affects

the current transport of the transistor. This effect is called the Early effect

[13] or base width modulation and is observed as a non-zero output

conductance. Typical output characteristics are shown in Fig. 5. The

extrapolated intercept of the collector current IC data with the

collector-emitter voltage VCE axis is often referred to as the forward Early

voltage VAF. Similarly, in reverse mode of operation the intercept of the

emitter current IE(VEC) plot is called the reverse Early voltage (or Late

voltage) VAR. VAF is an important figure of merit for the BJT, and should

be as high as possible. A simple way to increase VAF is to increase the base

Gummel number, i.e. to increase the doping and/or the base width which

unfortunately reduces the DC current gain . Therefore, the VAF product

is often referred to, as a more relevant figure of merit for a bipolar

technology. The Early voltages are also important parameters for compact

0.7

0.5

IB = 5.0 A

IB = 4.0 A

0.4

IB = 3.0 A

0.3

IB = 2.0 A

0.6

0.2

IB = 1.0 A

0.1

0.0

Figure 5. Output characteristics of a BJT. The reason for the VCE

dependence of IC is the Early effect.

9

Intrinsic transistor

REPoly

RBPoly

RBpoly-Epi

LOCOS

p+

Rn-epi

RCPlug

n-

p+

RBuriedC

n+

n+

different parasitic resistances.

The intrinsic transistor only constitutes a small part of the layout as

can be seen in Fig. 6. The extrinsic parts, such as the extrinsic base and the

buried collector, contribute to parasitic resistances that the terminal

currents must pass through. Figure 6 depicts the different resistances

present in the double poly-Si BJT.

For

large

terminal

current

densities,

typically

above

1-10105 A/cm2, the voltage drop over the parasitic resistances must be

considered.

10

The

effects

of

these

resistances

are

shown

-1

VBE

10

10

10

10

-3

-5

-7

-9

0.50

0.75

1.00

1.25

1.50

Figure 7. IB plotted against VBE. The dotted line represents IBIdeal and the

straight line is the measured IB.

10

p+

in Fig. 7, where the base current IB is plotted vs. the base-emitter voltage

VBE. The deviation from the ideal behavior of IB is VBE = IBRBTot +

IERE, where RBTot is the total base resistance and RE is the emitter

resistance. As will be discussed later, the voltage drop over the collector

resistance may force the transistor into quasi saturation.

The distributed nature of the base-emitter junction causes the

potential of the intrinsic base region to vary laterally. This is due to the

lateral voltage drop over the intrinsic base resistance, caused by the base

current. Since the base current is injected from the sides of the intrinsic

base, the base-emitter junction is more forward biased near the edges of

the emitter window and more electrons will be injected there as compared

to the center region of the emitter. Figure 8 shows the potential

distribution along the intrinsic base region. The base current is injected

from the right hand side in Fig. 8. In the transistor behavior, this effect

appears as a lowering of the intrinsic base resistance and will be discussed

in detail in Chapter 3.

1.06

1.04

1.02

1.00

0.98

0.96

0.94

0.92

0.90

0.88

0.0

0.2

0.4

0.6

0.8

1.0

Distance, x (m)

Figure 8. Schematic potential distribution along the intrinsic base region

in forward mode of operation.

11

Further on, the current crowding effect causes other high injection effects

to appear locally for smaller currents than can be expected when assuming

uniform current distribution.

When the concentration of the injected minority carriers in the base

region is comparable with the background doping, the conductivity is

modulated according to:

=

q = p p 0

1

= q ( p p + n n ) = b

q p p 0 qb .

p >> n

(2)

For low currents, qb = 1 and = qpp0. However, when the base current

increases, the hole concentration in the base increases and p > p0.

Therefore the conductivity in the base is increased. Since the intrinsic base

resistance is reduced by this effect, the voltage drop that causes the emitter

current crowding is reduced. Consequently, the base conductivity

modulation effect and the current crowding effect counteract each other.

In Paper 3, the current dependence of the base resistance reduction is

investigated for both of these effects. It is shown that the two effects cause

different current dependence of the base resistance, this will be discussed

further in Chapter 3.

The Kirk effect [5, 14, 15], also called base push-out, occurs when

the concentration of the injected electrons from the base-emitter junction

into the collector region is comparable with the doping level in the lowdoped collector region. As IC increases, the electron concentration in the

epitaxial collector exceeds NC,EPI. Thus the positive space charge in the

collector side of the base-collector depletion region is overcompensated by

electrons injected from the emitter. The electrical base-collector junction

is then pushed to the n- - n+ junction in the collector. Holes are injected

into the n- collector region to fulfill quasi neutrality, resulting in an

increased IB due to recombination currents. Therefore, the current gain is

degraded. The Kirk effect is demonstrated in Fig. 9,

12

Potential (V)

0.4

0.70 V

0.2

0.80 V

0.85 V

0.0

-0.2

0.90 V

-0.4

1.00 V

0.00

0.25

0.50

Collector

0.75

1.00

1.25

1.50

1.75

2.00

Distance (m)

Figure 9. The potential through the transistor for VBE = 0.7, 0.8, 0.85, 0.9

and 1 V.

showing the electrostatic potential for different VBE, simulated using the

Silvaco software [16]. The collector current density when the Kirk effect

starts (JC,KIRK) may be defined as when the electron concentration in the

epitaxial collector (nC,EPI) becomes 10% of NC,EPI, implying that:

JC,KIRK 0.1 q vs NC,EPI

(3)

depletion region. From Fig. 9, it is clear that the Kirk effect causes the

internal

base-collector

junction

to

become

forward

biased

for

0.85 V < VBE < 0.9 V. This behavior is similar to what is observed when

the transistor is operated under quasi saturation conditions.

Quasi Saturation

This effect occurs when the voltage drop over the collector

resistance (IC RC) is large enough to cause the base collector junction

to be forward biased, even though it is externally reverse biased.

Consequently, holes are injected into the collector region, IB increases and

decreases. The effect of quasi saturation is indeed similar to the Kirk

effect and no physical difference can be found in the intrinsic transistor

[5]. However, the quasi saturation effect is clearly evident in the output

13

0.20

IB = 1.0 A

0.15

0.10

0.05

0.00

Figure 10. IC(VCE) characteristics at IB = 1 A for three different ncollector thicknesses, wn-. Squares represent wn- = 0.5 m, circles wn- =

1.0 m and triangles wn- = 1.5 m.

is shown for three cases. The simulated structures are similar except for

the thickness of the n- epitaxial collector layer, wn-. A larger wn- results in a

higher collector resistance RC. Thus, the quasi saturation effect becomes

more pronounced as wn- is increased. The slower increase in IC for larger

wn- that can be observed in Fig. 10 is typical for the quasi saturation effect.

For efficient circuit design, accurate transistor models are crucial.

The history of compact bipolar transistor models starts with the EbersMoll model [17]. It is a simple model of two back-to-back diodes. This

model is still the basis of all the modern bipolar models. The evolution

continued with the development of the Gummel-Poon [18] model. The

normalized majority base charge qb was introduced and consequently the

14

base width modulation effects (Early effect) [13] are included in the

model. Most of the following models are based on the qb concept. More

recent models, such as the VBIC95 [3] and the MEXTRAM [4] model

also includes quasi saturation effect [15], Kirk effect [14] and base

conductivity modulation. The models for describing the non-linearity in

the different parasitic capacitances and resistances have also been further

improved. In addition, careful modeling of the parasitic pnp transistor

(base-collector-substrate) is included. The resulting transistor models are

complex, consisting of a large number of model parameters. However, the

ability to predict transistor performance is greatly improved compared to

earlier models. This chapter briefly reviews the different compact BJT

models that can be found in the literature and that has been used or

referred to in the appended papers.

Gummel and Poon [18] improved the Ebers-Moll model by

introducing the integral charge control concept. By using the normalized

majority base charge qb, the model inherently considers the base width

modulation (Early effect). Further on, a current-dependent base resistance

RB(IB) was introduced. A detailed discussion about modeling of the base

resistance is given in Chapter 3. Later, the model was extended to include

quasi-saturation or the Kirk effect [14, 15]. However, the model of these

effects is not very smooth which may cause problems with convergence.

Further on, the modeling of the cut-off frequency and output conductance

is not very accurate since they are given by the derivatives of the currents.

Figure 11 shows the equivalent circuit of the Gummel-Poon model.

15

C

RC

VBC

nCLVT

C4 I S e

1

RB

CSUB

C

IEC

R

CjC

CDC

ICT=ICC-IEC

V B E

n V

C 2 I S e EL T 1

ICC

F

CjE

CDE

E

RE

E

Figure 11. Equivalent circuit of the Gummel Poon model.

V B 'C '

V B 'E '

I SS VT

e VT

I CT =

e

qb

(4)

I SS =

qADn ni2

(5)

xC

N A (x )dx

xE

with xC and xE given at zero bias. The normalized base charge qb is defined

as

2

q

q

qb = 1 + 1 + q 2

2

2

(6)

where q1 models the base width modulation effect (Early effect) and q2 the

high injection effects. By assuming constant junction capacitances, i.e.

that the depletion charges are linearly dependent upon voltage, q1 is given

by

16

V

V

q1 = 1 B ' E ' B ' C '

V AR

V AF

(7)

Equation (7) is actually not the correct formulation of q1. A more correct

formulation is [20]

V

V

q1 = 1 + B ' E ' + B 'C ' .

V AR

V AF

(8)

The reason that Eq. (7) was used in the Gummel-Poon model is that the

resulting current expression then was consistent with the Ebers-Moll

model [20]. In the more recent versions of the Gummel-Poon model

Eq. (8) is used for calculating q1. The high injection base charge q2 is

modeled according to

V B 'E '

I SS N F VT

1 +

q2 =

e

I KF

V B 'C '

I SS N RVT

1 ,

e

I KR

(9)

where IKF and IKR are the forward and reverse knee currents, respectively.

The VBIC95 Model

The VBIC model (Vertical Bipolar Inter Company) was developed

in a joint project including many of the largest bipolar manufacturers in

the world [3]. It is based on the Gummel-Poon model [18]. However,

some major improvements are presented. The q1 formulation is not the one

based on the linearized depletion charges used in the Gummel-Poon

model. In VBIC95, q1 is given by:

q1 = 1 +

q je

VER

q jc

VEF

(10)

where qje and qjc are functions describing the voltage dependence of the

base-emitter and the base-collector depletion charges, respectively. The

parameters VER and VEF are referred to as the bias-dependent reverse and

forward Early voltages, respectively, in contrast to VAR and VAF which are

constant model parameters. It is notable that qje, qjc, VER and VEF are

dimensionless parameters. By using functions for describing the depletion

charges the prediction of output conductance and cut-off frequency is

17

the modeling of the quasi saturation effect is improved.

The MEXTRAM model (Most EXquisite TRAnsistor Model) is a

very complex compact model [4]. The model was developed by Philips

and is a subject of continuous development [e.g. 21]. The latest version,

MEXTRAM 504, requires close to 90 model parameters. In its

formulation, this model is much similar to the VBIC95 model. However,

major improvements are made in the smoothing functions used for

describing the high injection effects in the collector epilayer [5]. These

improvements serve to both increase the model accuracy and the

convergence of the model since the derivatives are continuous functions.

As in VBIC95, the MEXTRAM 504 model for the intrinsic base resistance

considers only base conductivity modulation. However, the model uses an

additional base emitter diode, connecting to the base between the

extrinsic and the intrinsic part of the base resistance. Thus, one could

argue that MEXTRAM 504 is a first order distributed model and therefore

current crowding effects are included in the model. Further on, the

MEXTRAM 504 model for the intrinsic base resistance includes both

current crowding and base conductivity modulation effects.

18

The base resistance has a large impact on device performance, both

for DC and high frequency operation. Therefore, the accuracy of a

compact transistor model is much determined by its ability to predict the

base resistance of the device. Modeling of the non-linear current

dependence of the base resistance has drawn much attention in the

literature [e.g. 19, 22-26]. In this chapter, the different definitions of the

base resistance are first reviewed. Then, the distributed models found in

the literature are discussed and the model presented in Paper 7 is

described. In compact BJT models, the base resistance models should

preferably be direct functions of the current and/or voltages of the device.

Therefore, the distributed models are not suitable for implementation in

circuit simulators. The last section of this chapter deals with the different

compact base resistance models that are used in BJT models. Also, a brief

description of the compact base resistance model presented in Paper 3 will

be given.

19

RBTot(IB) ()

175

150

125

100

75

50

25

-6

10

-5

10

10

-4

-3

10

10

-2

Figure 12. Schematic IB dependence of RBTot.

The total base resistance RBTot, consists of two parts, one extrinsic

part RBExt and one intrinsic part RBInt. A typical RBTot vs. IB plot is shown in

Fig. 12. The decrease in RBTot is due to the decrease of the intrinsic part

RBInt. As discussed earlier, the reason for the current dependence of RBInt is

mainly due to current crowding and base conductivity modulation.

In the literature, many different definitions of the intrinsic base

resistance can be found [19, 26, 27]. One way of defining RBInt(IB) is to

state that the power dissipation in the intrinsic base region PBase should be

the same as the power dissipated in RBInt(IB) by IB. Thus, RBInt is given by

PBase ( I B ) =

Power

I B RBInt

(I B )

WE

I 2 ( x)

R B 0 dx

qb ( x ) W E

(11)

region for small IB and RSq. is the intrinsic base region sheet resistance.

Historically, the power dissipation definition has been the most widely

used, due to the fact that it was used in the derivation of the Gummel-Poon

model of RBTot(IB) [19, 28]. However, as is mentioned in [29] and shown in

Paper 7, this definition is not compatible with the definition used when

20

according to Fig. 13. Consequently, RBInt is given by

Compact

RBInt

(I B )

Eff

VBE VBE

=

IB

I

VBE VT ln B

IS

=

IB

(12)

Eff

where VBE

is the effective base-emitter voltage. A third definition that is

useful for extraction of the base resistance parameters is the dual base

terminal definition. This definition is convenient when measuring the base

resistance using the dual base terminal test structure and will be discussed

in detail in Chapter 4.3. RBInt(IB) is defined as the voltage drop over the

whole intrinsic base region divided by IB. Thus, it reads

Dual Base

RBInt

(I B ) =

V (WE ) V (0)

IB

(13)

where V(WE) is the applied voltage at one side of the base region and V(0)

is the resulting voltage at the other side.

IB

B

RBInt(IB)

Figure 13. The base-emitter branch of a compact BJT model. The emitter

resistance RE and the extrinsic base resistance RBExt are left out for clarity.

21

RBInt(IB) ()

200

150

100

50

0

-6

10

Compact Model def.

Power Dissipation def.

10

-5

10

-4

10

-3

10

-2

Figure 14. Three different definitions of RBInt, simulated with the

distributed model presented in Paper 7.

In Fig. 14, the three definitions of RBInt are plotted versus IB. Data are

calculated using the distributed model presented in Paper 7. The physical

resistance RB0 in the calculations is 400 . The low current value for the

dual base terminal definition (Eq. (13)) corresponds to 1/2RB0. For the

other definitions this value is 1/3RB0 which is known from textbooks [e.g.

30]. This difference has to be corrected for when using the dual base

terminal test structure for extracting the base resistance model parameters.

It is also clear that the power dissipation definition is not compatible with

the compact model definition.

In compact models, two-dimensional effects (such as current

crowding) are not always easy to implement with a clear coupling to the

physical layout. In order to model these effects accurately, the distributed

nature of e.g. the base emitter junction must be considered. The

equivalent circuit of a distributed model is shown in Fig. 15.

22

IT

RE

E

VBEi(x)

B

x

RBInt dx

WE

RBExt

x=WE

dV ( x) =

R BInt

RB 0

I ( x) dx =

I ( x) dx .

WE

W E qb ( x )

(14)

dI ( x) =

V ( x)

IS

dx

exp

WE

VT

(15)

RB 0 I S

V ( x) dq b ( x) dV ( x)

1

d 2V ( x)

exp

=

qb ( x) WE2

V

dx

dx

dx 2

T

(16)

dV ( x)

dx

=0

(17)

x =0

V (0) = V0 .

(18)

is required. Earlier studies have either neglected the base conductivity

modulation, i.e. qb = 1 [19, 26, 27], or used a simplified formulation of qb

that is not very correct for large voltages [25]. No distributed model that

includes the Early effect has previously been presented. In Paper 7, such a

model is presented and analysed in detail. By including the Early effect in

the qb formulation of the model, the VCE dependence of qb(x) and RBInt(IB)

is described, see Fig. 16.

23

qb(x)

2.00

VCE = 1 V

VCE = 3 V

VCE = 5 V

1.75

1.50

1.25

1.00

0.0

qb = 1

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Normalized Distance, x

Figure 16. The VCE dependence of the qb distribution along the intrinsic

base region. The calculations are for VAF = 10 V and VAR = 5 V, the other

parameters are those presented in Paper 7.

By neglecting qb, i.e. qb(x)=1, both the Early effect and base

conductivity modulation are neglected. Equation (16) then corresponds to

the model presented in [19]. The solution of Eq. (16) is then [19]

cos[Z (1 x WE )]

V ( x ) = V (WE ) 2VT ln

.

cos( Z )

(19)

Z tan( Z ) =

RB0 I B

.

2VT

(20)

written

tan( Z ) Z

.

RBInt ( I B ) = RBInt (0)

Z tan 2 ( Z )

(21)

The RBTot model used in the SPICE version of the Gummel-Poon model is

based on Eq. (21). However, a simplified expression for Z(IB) is used,

which will be treated in the following section.

24

In the Ebers-Moll model, the base resistance, RBInt was modeled as

a constant resistance. The current dependence of RBInt was first included in

the Gummel-Poon model. The RBInt(IB) model was a simplified

formulation of the distributed model derived by Hauser [19], shown in Eq.

(21). However, in order to make the RBInt(IB) expression an explicit

function of IB, Z(IB) is simplified compared to the formulation in [19] (Eq.

(20)) and reads:

Z (I B ) =

1 + 1 + 144 I B / 2 I rB

(24 / )

2

(22)

I B / I rB

where IrB is defined as the base current at which RBInt(IB) = 1/2 RBInt(0).

Since the model is based on a distributed transistor model it accounts for

current crowding effects. However, the effect of base conductivity

modulation is neglected. Further on, the derivation of Eq.(21) is based on

the power dissipation definition of RBInt [19] which was shown earlier and

in Paper 7 to not be compatible with the compact model definition.

In the base resistance models included in MEXTRAM and

VBIC95, the current dependence of RBInt is solely due to the base

conductivity modulation. RBInt(IB) is then given by:

RBInt ( I B ) =

RBInt 0

,

qb ( I B )

(23)

where RBInt0 is the low current value of the base resistance. In [3], the need

for a compact model of RBInt(IB) that considers both current crowding and

base conductivity modulation was pointed out. Such a model is presented

in Paper 3. By adding a new parameter to the qb formulation used in

Eq.(23), the model for RBInt(IB) in VBIC95 is modified to also include the

effects of current crowding. The new qb(VBE), qb,(VBE) for large currents,

including the new parameter, may be written as

V

I

1

qb, (VBE ) = 1 + 1 + 4 S exp BE

2

I KF

N F VT

(24)

25

parameter, allowed to vary between 0.5 and 1, it is shown in Paper 3 that

Eq. (24) also covers the case of current crowding.

The intrinsic base resistance in the linearized small signal

transistor model (from now on rB) is given by the differential resistance rB

which is defined as [19, 26, 27, 31, 32]:

rB = RBInt +

dRBInt

IB

dI B

(25)

Eq. (25), the resulting small signal model of rB is:

rB =

2 x (1 + 2 x ) 1

2 RBInt 0

1 + (1 + 2 x )

1 + (1 + 2 x )

(26)

2 x (1 + 2 x ) 1

rB = RBInt 1

1 + (1 + 2 x )

(27)

This small signal model considers both emitter current crowding ( close

to 1) and base conductivity modulation ( close to 0.5) as causing the

decrease in RBInt at high currents, as shown in Paper 3. In Fig. 17, the

small signal model is compared to the large signal model for three

different values of . Since dR BInt dI B in Eq. (25) always is negative, rB

is smaller than RBInt.

26

Normalized RB, rB

1.0

(a)

0.8

0.6

= 0.5

0.4

0.2

Normalized RB, rB

0.0

1.0

(b)

0.8

0.6

= 0.75

0.4

0.2

Normalized RB, rB

0.0

1.0

(c)

0.8

0.6

= 1.00

0.4

0.2

Large Signal RB

Small Signal rB

0.0

-4

-3

-2

-1

0

10

10 10

10

10

IB / IRB

10

10

10

Figure 17. Comparison between the small signal model (dashed lines) and

the large signal model (solid lines) of the intrinsic base resistance for (a)

= 0.50, (b) = 0.75 and (c) = 1.00.

27

28

For a transistor model to be accurate it is important that the model

parameters are correct. First we have to define what is meant by correct. It

is not always true that the correct value of a physical parameter gives the

best accuracy of the transistor model. This is due to two reasons. First, the

model formulation is based on some assumptions and simplifications that

make the parameters difficult to interpret physically. Second, many

parameters are coupled to each other, meaning that a small error in one

parameter may require another parameter to take non-physical

proportions. However, since the modern parameter extraction routines [33]

utilize a global optimization scheme, it is extremely important that the

initial model parameters are close to the correct ones. Otherwise, the

optimization routine may be trapped in a local minimum.

In the following section, extraction of the intrinsic model

parameters is described. Extraction of the parasitic resistances is presented

in section 4.2. The dual base terminal test structure is presented in section

4.3. As shown in Paper 2, Paper 3, Paper 4 and Paper 6, this test structure

enables extraction of both resistances and Early voltage parameters. In the

final section of this chapter, the dual collector terminal test structure,

presented in Paper 5, is discussed.

29

The intrinsic model parameters e.g. the Early voltages VAF, VAR, the

saturation currents IS and the ideality factors NF and NR, are of crucial

importance for accurate BJT modeling. Different extraction schemes have

been proposed in the literature [18, 20, 34, 35]. For an extraction

procedure to be accurate it is important that it is based on the physics of

the model. For example, VAF and VAR extraction must be coupled since

both the emitter-base and the base-collector depletion widths are

modulated simultaneously. Typically, the extraction procedure is based on

forward and reverse output and Gummel characteristics.

Early Voltages

Historically, the forward Early voltage VAF has been extracted from

the intercept with the VCE axis of the extrapolated IC(VCE) plot for constant

IB or VBE, as shown in Fig. 18. The reverse Early voltage VAR has similarly

been extracted as the intercept with the VEC axis of the extrapolated

IE(VEC) plot for constant IB or VBC [35]. This extraction technique

decouples the two effects (forward and reverse Early effect). This means

that when extracting VAF the reverse Early effect is neglected and vice

versa. In [34], a technique for extracting both VAF and VAR in a coupled

manner was presented. It is based on a procedure where normalized data

are used. As discussed previously, when modeling the Early effect with

constant Early voltages, the output conductance is assumed to be constant.

In the more recent compact models like VBIC95 and MEXTRAM, the

Early voltages are modeled as bias-dependent quantities and the prediction

of the output conductance is much improved. A technique for extracting

the bias-dependent Early voltage parameters are presented in [20]. Base

emitter and base collector capacitance data are used to extract the builtin potential Pe and Pc and the grading coefficient mje and mjc for

30

5

4

3

2

1

0

VAF = 28 V

-30

-25

-20

-15

-10

-5

Figure 18. Traditional extraction of VAF.

the respective junction. These parameters are used in the models for the

junction charges qje(VBE) and qjc(VBC). The bias-dependent Early voltage

parameters VEF and VER are then calculated from forward and reverse

output data for low biases. In Paper 6, an alternative extraction procedure

for the bias-dependent Early voltage parameters is presented. The

procedure is based on DC measurements on the dual base terminal test

structure. Since the capacitance parameters are extracted from DC

measurements, no C-V measurements are required. This extraction

procedure is discussed further in section 4.3.

Current Parameters

When the Early effect parameters are known, the current

parameters ISS, IKF, IKR, F and R are extracted. Typically, forward and

reverse Gummel characteristics are used, see Fig. 19. Since qb is known,

ISS and the forward ideality factor NF can be extracted by fitting

IC =

VBE

VT

I SS e

q b (VBE )VBC =0

(28)

31

-3

10

-5

10

-7

10

-9

10

Extraction of IS

-11

10

-13

10

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Figure 19. Region used for extraction of the saturation current IS by

fitting Eq. (28) to measurement data. Measurement is performed

for VBC = 0 V.

at VBC = 0 V. Typically, the curve fitting is performed in the voltage range

0.4 V < VBE < 0.8 V [36], see Fig. 19. The forward knee current IKF is

extracted from the F (= IC / IB) versus log(IC) plot as the value of IC when

F has dropped to half its peak value, see Fig. 20. The reverse knee current

IKR is extracted in the same way for data in reverse mode of operation. In

modern extraction software [e.g. 33], these parameters are then optimized

after the resistances are extracted in the high current operation regime.

32

250

200

150

100

50

0

-6

10

-5

10

-4

10

-3

10

Figure 20. Extraction of the forward knee current, IKF.

For large currents, the parasitic resistances of the model are

important since the voltage drops over these resistances are significant.

High frequency transistors are usually operated in this regime, making

these parameters very important for accurate prediction of transistor

behavior. It is possible to calculate the resistances out of the doping levels

and geometry. However, current distribution and small geometries makes

it difficult to achieve satisfactory agreement between measurements and

simulations using the calculated resistances. Extraction of these

parameters may be performed from e.g. S-parameter [37], AC [22], noise

[38] and DC measurements [e.g. 39-41]. DC extraction techniques are

generally preferred due to the fact that they are both time and cost

effective. This in turn enables automated statistical wafer mapping to be

performed for increased accuracy of the model parameters as well as

increased process control. In this section, the most widely used DC

extraction techniques will be reviewed.

33

Ning-Tang method

The Ning-Tang method allows extraction of RE, RBExt and RBInt

[39]. From IB(VBE) data at large voltages, the deviation from the ideal

behavior VBE is written as

VBE = RE I E + RBTot I B

(29)

can be rewritten to read

VBE

= RE + (RE + RBTot ) 1

IC

(30)

proportional to , the plot of VBE / IC versus -1 will be a straight line

with a slope equal to RE + RBExt and the intercept with the y-axis equal to

RE + RBInt / . In Fig. 21, extraction according to this method is

demonstrated. The slope ( 300 ) roughly corresponds to RBExt since RE

is normally much smaller then RBExt. The intercept with the y-axis ( 30

) is more difficult to interpret due to the RBInt / term. However, by

using different emitter geometries separation of RE and RBInt / is enabled.

It is worth to mention that the assumption that RBInt is proportional

to is only valid when current crowding effects are negligible.

500

40

30

300

20

200

10

0

0.010

100

0.015

0.020

0

0.025

Figure 21. Resistance extraction according to the

Ning-Tang method [39].

34

Slope ()

VB1E / IC ()

400

directly proportional to . In that case, as is shown in Paper 2, the VBE /

IC versus -1 plot will not be linear and thus the parameter extraction

ambiguous.

Weng method

Weng et al. proposed a method for extraction of RBExt and RBInt

[23]. The method is based on a test structure consisting of a transistor with

two separated base terminals, B1 and B2, respectively. This test structure

will be discussed in detail in Chapter 4.3. In [23], the extraction procedure

included measurements of forward operation data while the additional

base terminal senses the voltage. The sense voltage VB2 is assumed to be

that of the intrinsic base. The total base resistance RBTot (= RBExt + RBInt) is

then calculated according to:

V V

RBTot ( I B ) = B1 B 2

I B1

(31)

where VB1 is the biasing voltage and IB1 is the current supplied by terminal

B1. RBInt(IB) is formulated to read

RBInt ( I B ) =

W 2d

1

RSq. E

3

LE

(32)

where RSq. is the sheet resistance of the intrinsic base, WE and LE is the

drawn emitter width and length, respectively and d is the deviation of the

real WE from the drawn. By plotting the extracted RBTot vs. WE for various

VB1E the point where the lines intercept corresponds to WE = 2d and

RBTot = RBExt. However, measurement data reveals that this intercept is

bias-dependent, implying that RBExt varies with bias, which is nonphysical. The reason for this is that when assuming Eq. (32) to be valid,

current crowding is neglected. All of the current dependence of RBInt is

then included in RSq.. Consequently, base conductivity modulation is

assumed to be the only cause for the decrease in RBInt(IB). If current

crowding is present, RBInt should be formulated to read

RBInt ( I B ) = f CC ( I B ) RSq.

WE

LE

(33)

35

where fCC(IB) is a function that describes the current crowding effect. The

fact that current crowding is neglected results in a bias dependence of the

intercept observed in Fig. 22. Further on, the factor of 1/3 in Eq. (32),

although accurate for compact models, should not be used when RBInt is

measured according to Eq. (31). As is pointed out in Paper 4, a prefactor

of 1/2 should be used instead.

Fly-back method

This method was first proposed by Kulke and Miller in [40]. Later,

Giacoletto [41] proposed a similar procedure. The method allows for

extraction of either RE or RC. For extraction of RE, VCE is measured as a

function of IB with the collector terminal left open, i.e. IC = 0 A. This

means that the transistor is operating in saturation. The saturation voltage

VCE,sat can be expressed as

VCE , sat = VC ' E ' + I C RC + I E R E

(34)

where VC'E' is the internal voltage between the collector and emitter node.

By using the Ebers-Moll model formulation for VC'E' , Eq. (34) can be

14

VBE = 0.80 V

12

10

VBE = 0.85 V

8

VBE = 0.90 V

6

4

2

0

0

2

3

4

5

Emitter Width, WE (m)

Figure 22. Measured RBTot as a function of drawn WE for VBE = 0.8, 0.85

and 0.9 V.

36

written as

I

1 C

I B

VCE , sat = VT ln R

1 + (1 ) I C

R

IB

+ I C RC + I E RE

(35)

IB, Eq. (35) is reduced to

VCE , sat = C + I B RE

(36)

VCE,sat(IB) plot, as is shown in Fig. 23. RC can be extracted in a similar

manner by leaving the emitter terminal open during measurements.

However, it is worth to mention that the measured RC only corresponds to

the part of the collector resistance originating from the buried collector

and contact resistance, i.e. the constant part RCC. The variable part RCV

may not be measured using this method. Recently, improvements of this

extraction method were proposed which also includes the substrate effects

[42].

10

1/RE

4

5.20

5.25

5.30

Figure 23. Extraction of RE using the fly-back method. Data are from

Silvaco simulations [16].

37

The different trade-offs regarding the design of the dual base

terminal test structure is presented in Paper 5. A schematic layout of the

test structure is shown in Fig. 24. Parameter extraction is done by using

one of the base contacts as a voltage tap, i.e. setting the current to zero and

measuring the voltage [23, 27]. RBTot(IB) is given directly according to

V VB 2

RBTot ( I B ) = B1

.

I B1

(37)

according to the dual base terminal definition of RBInt. Therefore, a

conversion is required if the compact model parameters are searched for.

A conversion scheme is presented in Paper 3.

The dual base terminal test structure also allows for extraction of

RE from the same measurement data. The voltage difference between the

sense terminal B2 and the emitter can be written as

Eff

VB 2 E = VBE

+ RE I E .

(38)

Eff

where VBE

is the effective base emitter voltage. If current crowding

Eff

dominates, VBE

becomes pinned to a certain value, even though VB1E is

further increased [27]. RE may then be extracted from the slope of the VB2E

versus IE plot. However, if base conductivity modulation occurs, the VB2E

E

B1

Contact holes

LOCOS

B2

Emitter Poly-Si

Figure 24. Simplified layout of a single poly-Si dual base terminal test

structure.

38

elaborate. In Paper 2, a procedure for extracting RE under these

Eff

circumstances is proposed. Using a simple diode model, VBE

in Eq. (38)

is calculated from

I

I B1 = S

F

Eff

exp VBE

N F VT

1 .

(39)

The sheet resistance of the intrinsic base region RSq. can also be

extracted using the dual base terminal test structure. A procedure for

extraction of RSq. is presented in Paper 4. The measurement procedure is

the same as for the pinched base measurements. However, by considering

the Early effect, the zero bias sheet resistance is extracted with large

accuracy. Further on, in Paper 6 extraction of the bias-dependent Early

voltage parameters is demonstrated using only DC measurements on the

dual base terminal test structure. By measuring the current between

terminals B1 and B2 for different collector and emitter biases, the Early

voltage can be calculated for each bias point. Thus, the bias-dependent

Early voltages are extracted. This in turn allows for calculation of the

built-in potentials and grading coefficients for both the emitter-base and

the base-collector junctions. Consequently, the capacitance measurements,

otherwise needed for extraction of these parameters, are not necessary.

The measurement effort is thus considerably reduced.

In Paper 5, a new type of test structure is proposed which adopts

the basic ideas from the dual base terminal test structure. A second

collector terminal C2 is added on the adjacent side of the emitter to C1.

Figure 25 shows a schematic layout of the proposed test structure. The C2

terminal is used for sensing the voltage in the buried collector. RCC is

given directly under forward biasing conditions from

RCC =

VC1 VC 2

.

I C1

(40)

39

C1

B1

C2

B2

Figure 25. A simplified layout of the dual collector terminal test structure.

All measurements and extraction procedures developed for the dual base

terminal test structure are possible to transfer to the dual collector terminal

test structure. However, since the collector region is not modeled in the

same way as the base region, only extraction of RCC is directly applicable

to compact BJT modeling.

In Fig. 26, RCC extracted according to Eq. (40) is plotted for three

different emitter widths as a function of the collector current. As shown in

Paper 5, the dramatic drop that occurs for high currents is due to the Kirk

effect [14]. It was observed in SILVACO simulations [16] that the

collector current takes a shortcut along the field oxide and thus never

passing through the n+-epi layer.

Similar to the pinched base measurements, the additional collector

terminal allows for pinched collector measurements. In such a way, the

sheet resistance of the buried collector layer can be determined. However,

knowledge of the drawn dimensions of the buried collector layer is

required.

40

28

26

WE = 3 m

24

WE = 1.2 m

22

WE = 0.6 m

20

18

-4

10

-3

10

10

-2

10

-1

Figure 26. RCC extracted according to Eq. (40) for

WE = 0.6, 1.2 and 3.0 m.

41

42

5. Concluding Remarks

modeling of bipolar transistors have been treated. Emphasis is given to the

dual base terminal BJT test structure which is shown to enable extraction

of the emitter resistance RE, the extrinsic and intrinsic base resistance RBExt

and RBInt and the sheet resistance of the intrinsic base region RS00 from DC

measurements. DC extraction of parameters traditionally requiring

capacitance measurements is also shown. These parameters are the biasdependent Early voltages VER and VEF, the built-in potentials Pe and Pc and

the grading coefficients mje and mjc of the base-emitter and the basecollector junctions, respectively. A compact model of the currentdependent base resistance is proposed. The formulation is compatible with

state-of-the-art compact BJT models. Further on, a new test structure is

proposed which uses two separated base terminals and two separated

collector terminals. Measurements on this test structure enable direct

extraction of the constant part of the collector resistance and monitoring of

the base push-out effect. Finally, a general formulation of a distributed

model of the base-emitter junction is presented. By including the Early

effect in the model formulation, more accurate calculations of the base

resistance and the current, voltage and base charge distributions along the

intrinsic base region is enabled, compared to earlier formulations.

43

the industry is relatively conservative regarding the use of test structures

for extraction of compact model parameters. This is mostly due to the fact

that there is always a physical difference between the test structure and the

actual transistor. Even though this thesis has shown that accurate

parameter determination is feasible using the dual base and dual collector

terminal test structure, I doubt that the method will extensively be utilized

by industry for extraction of parameters for compact BJT modeling.

However, the fact that DC measurements on the test structure yields

information about the collector layer, the extrinsic and intrinsic base

regions, the emitter region, the built-in potentials and the grading

coefficient of the pn-junctions renders it very suitable for process control

in

an

industrial

environment

where

automated

wafer

mapping

Looking into the future, an interesting continuation of Paper 7

would be to include the distributed b-e capacitance in the model. Thus, the

AC debiasing of the base-emitter junction may be calculated as a function

of frequency. The same effect is also a serious concern for RF applications

of deep submicron CMOS circuits, where the distributed gate resistance

and capacitance debias the gate electrode.

Regarding the test structure, the separation of the extrinsic and the

intrinsic base resistance could be improved. One idea would be to extract

the intrinsic part using the method proposed in [43] and then subtract this

resistance from the measured total base resistance. The result is then the

extrinsic base resistance. This would however require additional

measurements. Further on, only one method has been proposed for DC

extraction of the bias-dependent collector resistance RCV [44]. The result

from this method is difficult to interpret. The physical location of RCV (the

quasi-neutral part of the epitaxial collector) makes it difficult to assess

directly from measurements. Therefore, further work is required since

determination of RCV is highly needed for accurate modeling of future high

frequency BJTs.

44

Chapter 6. References

6. References

1.

2.

Integrated Circuits," John Wiley & Sons, Inc., 1993.

3.

Foisy, I. Getreu, M. McSwain, S. Moinian, J. Parker, D. J.

Roulston, M. Schrter, P. van Wijnen and L. F. Wagner, "VBIC95,

the Vertical Bipolar Inter-Company Model," IEEE Journal of

Solid-State Circuits, vol. 31, p. 1476, 1996.

4.

Transistor Model level 504," Philips Electronics, Unclassified NLUR 2000/811, June 2000.

5.

Graaff, "Improved Modeling of Output Conductance and Cut-off

Frequency of Bipolar Transistors," In Proc. 2000 Bipolar/BiCMOS

Circuits and Technology Meeting, p. 62, 2000.

6.

Model for High-Speed Bipolar Transistors at High Current

Densities - Part I: One-Dimensional Model," IEEE Trans Electron

Dev., 34, p. 1741-1751, 1987.

45

7.

Model for High-Speed Bipolar Transistors at High Current

Densities - Part II: Two-Dimensional Model and Experimental

Results," IEEE Trans Electron Dev., 34, p. 1752-1761, 1987.

8.

G. C. Feth, S. K. Wiedmann, "Self-Aligned Bipolar Transistors for

High-Performance and Low Power Delay VLSI," IEEE Trans

Electron Dev., ED-28, p. 1010, 1981.

9.

trench isolated bipolar technology for analogue and digital

applications," In Proc. IEEE Custom Integrated Circuits Conf.,

22.2.1, 1988.

10.

Malm, S.-L. Zhang, J. V. Grahn and M. stling, "A Simplified

High-Speed Bipolar Process With Ti SALICIDE Metallization:

Implementation of In Situ P-Doped Polysilicon Emitter," Physica

Scripta, vol. T79, pp. 318-321, 1999.

11.

"Characterisation of In Situ Phosphorus-Doped Polycrystalline

Silicon Films by Disilane-Based Low-Pressure Chemical Vapor

Deposistion," J. Electrochem Soc, 144, p.3952, 1997.

12.

and A. Chantre, "A Performance Comparison Between 0.35 m

Self-Aligned and Quasi-Self-Aligned Double-Polysilicon Bipolar

Transistors," In Proc. European Semicond. Dev. Research Conf.,

p.412-415, 1997.

13.

Transistors," Proceedings of the IRE, 11, p.1401, 1952.

14.

High Current Density," IRE Trans. Electron Dev., ED-9, p.164,

1962.

46

Chapter 6. References

15.

and H. Dirks, "A Unified Circuit Model for Bipolar Transistors

Including Quasi-Saturation Effects", IEEE Trans Electron Dev,

ED-32, p.1103-1113, 1985.

16.

(1997).

17.

Transistors," Proc IRE, 42, p. 1761-1772, 1954.

18.

Model of Bipolar Transistors," The Bell System Technical Journal,

vol. 49, pp. 827-852, May-June 1970.

19.

Current Injection Density and Effective Base Resistance for Stripe

Transistor Geometries," IEEE Trans Electron Dev, ED-11, p. 238242, 1964.

20.

SPICE," IEEE Journal of Solid State Circuits, 31, p.136-138,

1996.

21.

Comprehensive Bipolar Avalanche Multiplication Compact Model

for Circuit Simulation," In Proc. 2000 Bipolar/BiCMOS Circuits

and Technology Meeting, p. 172, 2000.

22.

Measurement of the Base and Emitter Resistances of Bipolar

Transistors, IEEE Journal of Solid-State Circuits, vol. SC-7, p.

492-498, 1972.

23.

Base Resistance of Bipolar Transistors, IEEE Electron Device

Letters, vol. 13, p. 158-160, 1992.

24.

the Base Series Resistances in Bipolar Transistor in Presence of

Current Crowding, IEEE Journal of Solid-State Circuits, vol. 31,

p. 132-135, Jan. 1996.

47

25.

Analytical Model and the Impact of Base Charge Storage on Base

Potential Distribution, Emitter Current Crowding and Base

Resistance," Proc. 1992 International Electron Device Meeting.

p.573, 1992.

26.

Bipolar Transistors," IEEE Trans Electron Devices, ED-32,

p.2503, 1985.

27.

Bipolar Transistor for Extracting Parasitic Series Resistances,

IEEE Trans Electron Dev, vol. 38, p. 2139-2154, 1991.

28.

with SPICE," McGraw-Hill, 1993.

29.

M. Schrter, "A Survey of Present Compact Models for HighSpeed Bipolar Transistors," Frequenz, 47, p. 178-190, 1993.

30.

Wiley & Sons, Inc., 1988.

31.

T. Fuse and Y. Sasaki, "An analysis of small-signal and largesignal base resistances for submicrometer BJT's," IEEE Trans

Electron Dev, vol. 42, p. 534-539, 1995.

32.

its Application in the Prediction of the Emitter-Base Diode

Characteristic, Base Impedance, and Pulse Response of the

Device," IEEE Trans Electron Dev, ED-12, p.513, 1965.

33.

34.

Algorithm for Intrinsic Gummel-Poon BJT Model Parameters, in

Proceedings of the International Conference on Microelectronic

Test Structures, p. 159-164, 1998.

35.

36.

Oritz-Conde and S. Moinian, "Comparison of the new VBIC and

conventional Gummel-Poon bipolar transistor models ," IEEE

Trans Electron Dev, 47, p.427-433, 2000.

48

Chapter 6. References

37.

Technique for Small-Signal Equivalent Circuit of Polysilicon

Emitter Bipolar Transistors, IEEE Trans Electron Dev, 41, pp.

233-238, 1994.

38.

Sources in Submicrometer Bipolar Transistors," IEEE Trans

Electron Dev, 39, p.1501, 1992.

39.

and Base Series Resistances of Bipolar Transistors, IEEE Trans

Electron Dev, ED-31, p. 409-412, 1984.

40.

collector series resistances in transistors, in Proc. IRE, 45, p. 90,

1957.

41.

Resistances, IEEE Trans Electron Dev, ED-19, p. 692-693, 1972.

42.

C. Kelly, Inclusion of Substrate Effects in the Flyback method for

BJT Resistance Characterisation,

in

Proceedings

of

the

194-199, 1999.

43.

Base Resistance of Bipolar Transistors and its Dependence on

Current and Geometry," IEEE Trans Electron Dev, 38, p.538,

1991.

44.

"Measurement of Collector and Emitter Resistances in Bipolar

Transistors," IEEE Trans Electron Dev, 38, p.365, 1991.

49

50

In this chapter, the main achievements presented in the appended

papers are briefly summarized. Further on, the contributions of the author

are specified.

Paper 1.

The effect of the emitter window overetch in a double poly-Si BJT

process is investigated. The results indicate that an increased overetch

raises the maximum cut-off frequency fT,max on the expense of a reduced

maximum frequency of oscillation fMAX. The effect of the overetch is

increased when using a tilted geometry when performing the intrinsic base

implantation.

Contributions: The author participated in process development and

processing of the devices, performed the simulations and wrote the

manuscript.

51

Paper 2.

A new procedure for extraction of the base and emitter resistances

of a BJT is presented. The procedure only requires one measurement in

forward mode of operation on a dual base terminal test structure.

Contributions: The author designed and measured the double poly-Si test

structures. In addition, most of the writing of the manuscript was

performed by the author.

Paper 3.

transistors

A compact model of the intrinsic base resistance is presented. By

introducing a new model parameter, the existing model in e.g. the VBIC95

BJT model is improved so that both current crowding and base

conductivity modulation are considered. A scheme for extraction of the

base resistance parameters is also presented.

Contributions: The author came up with the idea, performed some of the

measurements and most of the writing of the manuscript.

Paper 4.

Bipolar Transistors

This paper presents a procedure that allows for extraction of the

zero bias intrinsic base sheet resistance, RS00 from pinched base

measurements. The extraction procedure considers base conductivity

modulation, current crowding and base width modulation (Early effect).

Contributions: The author performed some measurements, contributed in

the development of the method and assisted in writing the manuscript.

52

Paper 5.

Bipolar Transistors

By using an additional collector terminal, direct extraction of the

constant part of the collector resistance is enabled. Also, monitoring of the

Kirk effect using the additional collector terminal is demonstrated.

Contributions: The author came up with the idea, performed the design of

the double poly-Si test structures, all the simulations, some of the

measurements and wrote the manuscript.

Paper 6.

A simple way of extracting the bias-dependent Early voltage

parameters from pinched base measurements is presented. Separation of

the intrinsic and extrinsic parts of the base resistance is discussed.

Contributions: The author performed some measurements, contributed in

the development of the method and assisted in writing the manuscript.

Paper 7.

A distributed model of the base-emitter junction is presented. The

model accounts for current crowding, base conductivity modulation and

base width modulation (the Early effect). In addition, a complete review of

previous work on distributed transistor models are presented.

Contributions: The author came up with the idea, performed all the

calculations and measurements. Further on, the author wrote the

manuscript.

53

54

## Viel mehr als nur Dokumente.

Entdecken, was Scribd alles zu bieten hat, inklusive Bücher und Hörbücher von großen Verlagen.

Jederzeit kündbar.