Beruflich Dokumente
Kultur Dokumente
poly-Si,
Gate fin metal Gate
silicon thickness
oxide
oxide
w silicon
Source Drain
silicon
channel length
Dr.-Ing. J.Biba
Advanced MOSFETs and Novel Devices
AdMOS, 6-1
Chap 6 Advanced MOSFET Development
6.1 Overview
Feature size, pitch, node
180 nm node 130 nm node 100/90 nm node 70/65 nm node 50/45 nm node 35/32 nm node 22 nm node
140nm – 100nm 90nm – 70nm 65nm – 50nm 45nm – 35nm 30nm – 25nm 20nm – 17nm 14nm – 10nm
1999 - 2001 2002 - 2004 2004 - 2006 2006 - 2008 2008 - 2010 2010 - 2012 2012 - 2014
FinFET
Lateral GAA 450mm Wafer Monolithic 3D
3xNWs stacked Integration
16/14 nm node 11/10 nm node 8/7 nm node 6/5 nm node 4/3 nm node 3/2.5 nm node 2/1.5 nm node
24nm 18nm 14nm 10nm 10nm 10nm 10nm
2014 - 2016 2016 - 2018 2019 - 2020 2021 - 2023 2024 - 2026 2027 - 2029 2030 -
Emerging devices
TFETs, NanoCarbonTube, New materials
From economic view the shrinking of devices reduces fabrication costs dramatically (see chapter 2, economics).
Moore´s Law:
Industry is able to shrink the dimensions of MOSFETs
a factor S 1 0.7 every 3 years (one generation)
2
no problem
no imagination, how to do
http://www.itrs.net/
http://www.itrs2.net/
"Old way"
-> "More Moore"
Higher performance
by higher integration
-> shrinking
1) Generations were characterized by the so-called "minimum feature size F", which can be fabricated.
According to the fabrication process of MOSFETs, this minimum feature size was identically with the channel length.
Minimum Feature
Size 1 F pitch
MOSFET 1 MOSFET 3
MOSFET 2
2F plug
plug
Iso Gate
Gate Gate 1F
1F Iso 1F 1F
1F D 1F
Iso S Iso Iso Iso
4F 5F 5F
* all-over Isolation * fill in Isolation * open S/D plugs with 1F
* open 4F in Isolation with distance 2F
* put in 1F Gate (misalignment not critical) channel
* selfaligned S/D Implantation
3) On chips not single devices are fabricated, but functional structures (DRAM, inverter).
Therefore, to describe the number of functional blocks, not the channel length is important, but the density of blocks.
This is described by the "pitch" (distance between
Advanced repeating
MOSFETs structures).
and Novel Devices Prof.Dr. W. Hansch , Dr.J.Biba
Institut für Physik AdMOS, 6-7
6 Future MOSFETs Advanced MOSFET Development
6.1 Overview - Minimum Feature Size and Pitch
http://www.itrs2.net/
high performance
History year Future (MPU)
~2000
Long-channel MOSFETs Short-channel MOSFETs Nano-MOSFETs general purpose
(ASIC)
6.1 Overview
Feature size, pitch, node
• Fin width 8 nm
• Fin pitch 42 nm
• Replacement Gate
• High-k + Metal Gate
S: ~ 65mV/dec
DIBL: ~ 30mV/V
https://news.samsung.com/global/samsung-starts-industrys-first-mass-
production-of-system-on-chip-with-10-nanometer-finfet-technology
• Fin width: ?
• Fin pitch: ?
• Replacement Gate
• High-k + Metal Gate
• Triple patterning
0.25 A/mm
/0.1 Vg
S: ~ 90mV/dec
DIBL: ~ 100mV/V
Ioff: 10-100nA/µm
Ion: ~ 1mA/µm gm: ~2500 mS/mm
0.2 A/mm
Ioff: ~ nA/µm
INTEL, IEDM 2008
/0.2 Vg Ion: ~ 1mA/µm
2009 Node: 32 nm INTEL, IEDM 2008 2014 Node (FinFET): 14 nm INTEL, IEDM 2014
Ioff: ~ 10nA/µm
Ioff: ~ 100nA/µm Ion: ~ 1mA/µm
Ion: ~ 1mA/µm
S: ~ 65mV/dec
S: ~ 100mV/dec DIBL: ~ 70mV/V
DIBL: ~ 130mV/V
0.3 A/mm
/0.1 Vg
S: ~ 65mV/dec S: ~ 65mV/dec
DIBL: ~ 70mV/V DIBL: ~ 30mV/V
Samsung Intel
10 nm 10 nm
0.3 A/mm xxx 34 nm
/0.1 Vg
64 nm 54 nm
48 nm 36 nm
gm: ~3000 mS/mm
0.04 um² ? um²
https://www.extremetech.com/computing/221532-tsmc-will-begin-10nm-production-this-year-claims-5nm-
by-2020
Vt roll-off: good above L > 70nm Vt roll-off: good above L > 35nm
! Make a test:
2000 -> 2009 = 3 generations
-> does scaling work ?
! Make a test:
0.059µm²
2009 -> 2014 = 2 generations
-> does scaling work ?
SRAM:
2014 Node (FinFET): 14 nm Intel's 6T SRAM options in their SoC technology, including
high density / low leakage (HDC), low voltage (LVC), and high
performance (HPC) [1
BL VDD VSS T1 T4 T5
WL T1 T4 T5
BL VDD VSS M1
T1 T4 T5
T2 T3 T6 WL Gate pitch
WL M1
VDD VSS BL = 70 nm
T2 T3 T6
VDD VSS BL M1
T2 T3 T6
fin pitch = 42 nm
poly-Gate
poly-Gate
Typically:
# 6-8 layers of interconnects
# Cu + ILD
# every layer separately optimized
2009 Node: 32 nm
RC delay for 1mm wire length, line capacity: ~ 250fF/mm for M1-M5
INTEL, 2017
Transistor
Density
MTr/mm²
• Fin pitch 34 nm
• Fin height 53 nm
Samsung Intel
10 nm 10 nm
xxx 34 nm
INTEL, 2017
64 nm 53 nm
48 nm 36 nm
INTEL, 2017
Through clever technological design (hyper scaling) INTEL seems to achieve a higher
transistor density
6.1 Overview
Feature size, pitch, node
Estimation:
If channel length: 100nm (Linear) Electric Field:
VDrain = 1 V 1V/100nm = 105 V/cm
v=µ*E
bulk mobility is reduced
d 2 (k ) 1
dk 2 m*
flat bands -> high masses
->low mobility
steep bands -> low masses
-> high mobility
bandstructure dependent on:
* material
wave vector k * crystal orientation
L <111> <100> X * strain
and combinations
Advanced MOSFETs and Novel Devices Prof.Dr. W. Hansch , Dr.J.Biba
Institut für Physik AdMOS, 6-32
6 Future MOSFETs
6.3 Mobility Enhanced MOSFETs Mobility Enhanced Channels - New Channel Materials
Due to interface states Dit from 1960 (first working MOSFET) until ~ 2007 the only possibility to create a proper working MOSFET
was using Si(100) and thermal oxidation for the Gate oxide
NV =1019
=1018 cm-3
Dit ~ 1010cm-2
2) How are the channel charges created ? How many in a 32nm MOSFET ?
Due to interface states Dit from 1960 (first working MOSFET) until ~ 2007 the only possibility to create a proper working MOSFET
was using Si(100) and thermal oxidation for the Gate oxide
1) Using Germanium in old days (1948-1954) -> the oxidation of Ge delivers a very hygroscopic thermal Gate oxide
-> no long-term stable MOSFET
-> standard technology with Ge is bipolar
2) Using Compound Semiconductors (GaAs, InP, ...) -> delivers inhomogeneous thermal Gate oxides
(because one element is oxidizing faster than the other)
-> inhomogeneous oxide thickness along channel
with different materials (work functions are changing)
-> no reproducible Vt for devices
Oxidfilme auf AIIIBV-Halbleitern
-> standard technology with III-V is bipolar
K. Löschkea, G. Kühna, H.-J. Bilz Unda and G. Leonhardta
Abstract
Thermal oxidation of GaP, InP, GaAs, GaSb and InSb was carried out at room temperature and at 300 °C in air. The optical constants and film thicknesses were measured ellipsometrically and the
same samples, including oxidized GaSb, were investigated by photoelectron spectroscopy. The concentration profiles of the elements in the films and their binding state were determined. It is shown
that the elements in oxide films of GaP are oxidezed perfectly, whereas films of the other compounds exhibit a very inhomogeneous structure. The thicknesses of the oxide films increase in the order
GaP ≈ InP < GaAs < GaSb, InSb. The interpretation of the ellipsometric measurements is complicated by the presence of a non-oxidized BV component, especially in the case of arsenides and
antimonides.
3) For deposited Gate dielectrics the interface state density is for all materials close to 1012 /cm²
GaAs metal-oxide-semiconductor devices with a complex gate oxide composed of SiO2 and GaAs oxide grown using a photoelectrochemical oxidation method
Hsin-Ying Lee et al 2010 Semicond. Sci. Technol. 25 015005 (4pp)
Abstract. In this study, a SiO2/GaAs oxide bi-layer layer was used as the gate oxide in GaAs-based metal-oxide-semiconductor (MOS) devices.
The GaAs oxide layer of the bi-layer layer was directly formed on the GaAs surface by using the photoelectrochemical (PEC) oxidation method.
Some samples were thermally treated at 200 °C and 300 °C in O2 ambience for 30 min.
The surface state density of the oxide/GaAs interface with and without GaAs oxide thermal treatment was 7.2 × 1011 and 7.9 × 1011 cm−2 eV−1, respectively.
The GaAs MOS field effect transistors (MOSFETs) with the PEC-deposited GaAs oxide thermally treated showed an output current of 152 mA mm−1 at VDS = 2.4 V and VGS = 0 V
and an extrinsic transconductance of 89 mS mm−1.
Solution:
Ge lattice is about 4.2% larger than Si So let´s try to use Ge in the channel
In the SiGe Hetero-Bipolar Transistor (HBT), Ge is selectively introduced into the base region of a Si-bipolar transistor
The bandgap of the SiGe-base arranges linearly between the bandgap of pure Si (1.12eV) and pure Ge (0.67eV) (called Vegard´s law).
The band-offset is almost in the valence band.
SiGe (HBT) technology: Transistor HF-performance: Frequency divider: Low Noise Amplifier (LNA):
(digital application) (analogue application)
Input
Output
80mV
The smaller base bandgap enhances electron injection, producing a higher current gain compared to Si for the same base doping level.
In addition with varying vertical Ge-content in the base an intrinsic drift-field can be incorporated to improve the transistor frequency
n-MOSFET p-MOSFET
CMOS with 100nm gate length (University research)
Source:UCLA, IEDM2000
!!! x
Ioff < 0.1nA/µm S: ~100mV Threshold voltage is shifted to positive values for SiGe
DIBL: <10mV
Holes Electrons
The biaxial stress (lateral and vertical) breaks down the existing
cubic lattice symmetry. To create strain different approaches exist
Creation/Measurement of strain
Depending on wafer and pit orientation
tensile or comressive stress, and also
uniaxial and biaxial stress can be created
tensile stress
Example of measurement
compressive stress
Advantage:
50 % current gain !
Disadvantage:
very expensive
Courtesy: AMD
presented at Analyst Meeting 2002
Advanced MOSFETs and Novel Devices Prof.Dr. W. Hansch , Dr.J.Biba
Institut für Physik AdMOS, 6-45
6 Future MOSFETs
6.3 Mobility Enhanced MOSFETs
Mobility Enhanced Channels - Strained Channels
metal
from Liner gate from Liner from Liner from Liner
spacer
tensile stress
compressive stress
INTEL CMOS:
65 nm node
L = 35 nm
tox (SiON) = 1.2nm
FUSI: NiSi
n-MOSFET p-MOSFET
complexe technology
-> more fluctuations ! best Ion / Ioff ratio as reported
higher strain
<100> <110>
(100)
<110> (100)
surface
(110)
in <110> direction
flat
in <110> direction
Combined technology
+ strain
We need to have detailed knowledge of bandstructure
Both types of handling wafers are possible, but today best developed wafers (with 300mm) are (100)-wafers
Advantage: Advantage:
+ well established (100)-wafer in 300mm (cheap)
+ low temperature wafer fabrication + better performance with n-FET on SOI
+ well established (100) selective epitaxy and p-FET in substrate
Disadvantage: Disadvantage:
- non-established (110)-wafer bonding - high-temperature wafer bonding
- n-Fet in substrate requires additional - difficult (110) selective epitaxy
technology steps (halo) to avoid SCE
Smart Cut
SIMOX (Separation by implanted oxygen)
Depending on company´s technology experience and risk willingness various combinations of the mobility enhancement techniques
are possible:
6.1 Overview
Feature size, pitch, node
In contacts of materials always a contact potential exists which depends more or less on the workfunction of the various materials
This contact potential causes a depletion zone depending on charge carrier density
Advantages: Technological: * self-adjust S/D formation because of high temperature stability (1100°C), no reaction with SiO2
Electrical: using p+ and n+ poly gates symmetrical VT on doped Si (P-MOS and n-MOS) can be realized
important for good CMOS-inverter characteristics
Disadvantages 1) high resistivity of gate electrode even with silicide on -> additional design and technological effort on Gate lines
in connection with thin gate oxides (3-1)nm as required from scaling roadmap
2) penetration of boron doping atoms from poly-Si Gate in/through oxide causes shift in threshold voltage VT
3) because poly-Si is just high doped (1020 cm-3) and not a metal (1022 cm-3) a depletion width exist of about 1 nm
at the interface of poly-Si to SiO2.
With scaling to tox ~1nm this depletion zone contributes in Gate capacitor to electrical oxide thickness,
this reduces gate capacitance and therefore drive current Ion
Possible solutions: 1) extremely high doping (ev.Ge) -> but creating charged defects -> mobility in channel and Ion decreases
-> no help for boron diffusion
2) opposite doping of poly-Gate -> poly-Si in accumulation, if MOS is in inversion
-> but additional processes are needed (increasing cost, decreasing yield)
3) new Gate material -> Metal -> what about adjustment of VT ?
-> what about manufacturing ?
these solutions are addressed in this section
Conventional CMOS: Poly-Si Gate allows selfadjusting process with nanometer resolution: S/D extensions doping,
spacer formation,
S/D doping,
Various processes for poly-Gate are possible: Halo doping
n-doping p-doping
Standard process (1970 - 2007):
1) deposit undoped poly-Si for all devices
isolation
2) Gate patterning (1 mask)
Si 3) S/D doping for n-MOSFETs and p-MOSFETs (2 masks)
n-doped
Sometimes used:
1) deposit single-doped poly-Si for all devices
2) Gate patterning (1 mask)
Si 3) S/D doping for n-MOSFETs and p-MOSFETs (2 masks)
poly-Si
c-Si
Reduction of Gate capacitance in On-state (inversion) Voltage drop in Gate depletion zone
-> increase of Short Channel effects (SCE) -> lower Gate voltage on channel -> lower drive current Ion
20 % Capacity Gain
INTEL CMOS: with Fully Silizided (FUSI)
65 nm node Gate metal (NiSi)
L = 35 nm
tox (SiON) = 1.2nm 15 % Current Gain n-MOS
FUSI: NiSi
with Fully Silizided (FUSI)
Gate metal (NiSi)
as measured VT normalized
* better S
* lower Ioff
Basic value of work function may be choosen from reported bulk values
but: 1) scaling VT will need different metal MS every future node -> no chance for technology try to adjust workfunction
2) measured workfunction may be dependent on process, charges (doping) and dielectrics with one metal
not
shown
here
published data for NiSi, exact values are very sensitive to fabrication process
frequently used process:
doping poly-Si and then silicidation which results in rearrangement of doping Source: Hosoi et al., RCNS, 2004
The workfunction values of the fully silicide gates are very sensitive to fabrication process.
The workfunction of the metal gate is tailored by engineering the metal/dielectric interface. This leads to the usage of capping
layers.
Capping layer AlN is shifting the workfunction depending on the gate dielectric
n+-Si p+-Si
Source: J.A. Carraveo-Frescas et all, Experimental and theoretical investigation of the effect of SiO2 content in gate
dielectrics on work function shift induced by nanoscale capping layers, Applied Physics 101, 2012
Capping layer AlN/Al2O3 are suitable for PMOS Capping layer La2O3 is suitable for NMOS
Source: Koji Kita et all,Intrinsic Origin of Electric Dipoles Formed at High-k/SiO2 Interface, IEDM 2008 Koji Kita et all,Intrinsic Origin of Electric Dipoles Formed at High-k/SiO2 Interface, IEDM 2008
• For some oxides there is a positive VFB shift • Al2O3 and HfO2 have a higher density of O2 than SiO2
• For some oxides there is a negative VFB shift • Y2O3 and La2O3 have a lower density of O2 than SiO2
VFB shift due to the formation of a dipole layer To adjust the oxygen atom density at the interface,
at high-k/SiO2 interface oxygen atoms move from higher density material to
lower density material
Voltage drop at the interface layer for HfO2 leads to higher workfunction
Voltage drop at the interface layer for La2O3 leads to smaller workfunction
Source:H.N. Alshareef et all, Work function engineering using Source:H.N. Alshareef et all, Work function engineering using
AlN interfacial layers, Applied Physics 89, 2006 lanthanum oxide interfacial layers, Applied Physics 89, 2006
Source: J.A. Carraveo-Frescas et all, Experimental and theoretical investigation of the effect of SiO2 content in
gate dielectrics on work function shift induced by nanoscale capping layers, Applied Physics 101, 2012
Through annealing, the capping metal reaches the SiO2 /HfO2 interface layer creating the dipole
The higher the amount of oxygen at the interface is the higher is the influence of the dipole
Disadvantage is a low mobility of the carriers due to higher scattering caused through the dipole (especially for PMOS)
FGA
• HfO2, TiN, Al ,TiN are deposited • To high N in TiAlN leads to positive VFB
• Forminggas annealing (450°C) diffusion • To high amount of aluminum leads to
of Al towards the interface of TiN and HfO2 positive VFB
Dipole • Annealing with O2 => Al2O3 which leads to
Shift to negative Flatband Voltage positive VFB
NMOS aluminum has to diffuse to the interface high-k/TiN to become a negative shift of VFB
Is the threshold voltage still symmetrical with the new capping layers?
For a quick calculation we take the values of exercise 3 and change the workfunction accordingly.
TiN TiAlN
PMOS TiN
Exercise 3
𝐸𝑔
ΦMS = Φ𝑀 − Χ𝑠𝑖 + − |Ψ𝐵 | = 4.8V – 4.05V – 0.56V + 0.45V = 0.64V
2𝑞
Exercise 3
′′
𝑄𝐷𝑒𝑝𝑙
𝑉𝑇𝑝 = − ′′ − |2Ψ𝐵 | + Φ𝑀𝑆 = -0.34V – 2*0.45V + 0.64V = -0.6V
𝐶𝑜𝑥
Channel doping must be eliminated to avoid dopant fluctuations and increase drive current Ion
Poly-Si Gate depletion must be eliminated to avoid lowering of Gate capacitance and increasing short channel effects
PMOS PMOS
Samsung/TSMC/Intel are using TiN for PMOS and TiAlN for NMOS as workfunction layer/capping layer
Most companies used until 20nm technology node gate-first with fully silicide gate and capping layers
Since the 20 nm, technology node all companies are using replacement-gate and capping layers
6.1 Overview
Feature size, pitch, node
Vertical MOSFETs offer the possibility Stacking devices results in 3-dim integration
to extremely reduce channel length without Vgate and further reduces the footprint.
sophisticated lithography: Vdd
Vsd The difficulties with vertical interconnects increase
GATE
n out
wafer
Vsd Vgate DRAIN
SOURCE
DRAIN
DRAIN
Vgate
channel length defined by layer deposition p
(nm-resolution), not by optical lithography
SOURCE SOURCE
Single MOSFET
Vdd
1F
Vdd
p-MOSFET
5F
Using wraped gates the MOSFETs Vin
8F
n-MOSFET Vss
5F
Vss
Planar: 40F2 Vertical: 16.5 F 2
Planar Inverter Vertical Inverter
Reduction by ~2
Reduction by ~5
Historical:
deposition poly-Source
deposition nitride
Drain
1E-3
VDS = 1.5 V 0.25
tox = 2.8nm
1E-4
0.1 V 1
1E-5 Planar
ID / WC (A/µm)
ID / WC (mA/µm)
0.1
1E-6 25 Å Oxide
0.01
IG (A/cm )
1E-7 0.15 1.6 V
2
1E-3
1E-8 DIBL = 90 mV 1E-4
1.4 V
1E-9 S = 105 mV/dec 0.10 1E-5
1E-10 1.2 V 1E-6
1E-11 0.05 Planar
1.0 V 1E-7
32 Å Oxide
1E-12 0.8 V 1E-8
0.0 0.5 1.0 1.5 0 1 2 3 4
0.00
VGS (V) 0.0 0.5 1.0 1.5 VG (V)
VDS (V)
VRG vs Planar:
• Similar packing density
• Up to double drive
• Parallel more dense
• Series less dense
Overlap Capacitance
• Dominated by true overlap and local fringing
• Doubled width compensated by doubled drive current
Substrate Capacitance
• Source: minimal
• Drain: reduced coefficient
IEEE 2008
LG = 150 nm
D = 25 nm
6.1 Overview
Feature size, pitch, node
In 1997, when the first "Roadmap" was created, the companies started to look several years in the future
It was recognized:
Taur et al. (IBM): "CMOS devices below 0.1µm: How high will performance go ?",
IEDM 1997, p.215-218
Electrical properties:
Advantage: important:
with HALO-doping the depth of Source/Drain sharp doping profiles (nm/dec) are needed
-S ~ 200mV/dec
junctions is not very critical
- -> for Ion/Ioff ~ 105 1V swing is needed
-questionable from implant process
--> VDD=1V and Vth=0.2 V
- leaving scaling rules (statistics in range straggling, under-implant)
- Source/Drain resistance keeps low -flash heating (1000°C for a few sec is needed)
- S/D extensions (if any) with low resistance (recrystallization of implant damage, but no diffusion)
Feasibility of 25nm channel length (proposed for year 2012) planar bulk MOSFETs was shown in 1998
Halo doping is used till 22/20 nm technology node (Samsung/TSMC planar technology)
For fully depleted channels like FinFET Halo doping not necessary
6.1 Overview
Feature size, pitch, node
Advantage of SOI:
radiation tolerance
Classical SOI-MOSFETs show various advantages in performance compared to classical bulk-MOSFET, but are more expensive
1018 O2/cm²
100keV
T ~ 500°C
2) BESOI: Bonded etch-back silicon on insulator MOSFETs fabricated on SOI
N doping EF
The position of Fermi-Level can be
calculated (chap. 3.13):
bulk kT ln Ei = 0
ni
wdep
The barrier in a pn-junction can be calculated:
Vbi
Ei = 0
kT N D N A kT N A N D p
Vbi n p ln ln ln 2 EF
q ni ni q ni
n
2 0 Si Vbi Vext
and the width of the (one-side abrupt) w
depletion zone will be: qN dop
Drain VDS
Short-channel
Calculating the threshold voltage, we receive:
Drain Vbi
SCE and DIBL will be small, if: Lel ->large of course, because we have a long channel device
tox -> small better gate control -> less reduction of threshold voltage
fully depleted
oxide
tSi2 t
• Fin width 8 nm tox ,el Si 2
• Fin pitch 42 nm EI 0.5 1 24
Lel el Lel
L
• Assumption:
• Lel = 14 nm ( 8 nm )² 8 nm
4 0.8 nm 2
• tox,el = 0.8 nm EI 0.5 1 14 nm 14 nm 0.0088
( 14 nm )²
Calculation of DIBL
INTEL, IEDM 2014
S: ~ 65mV/dec
DIBL: ~ 70mV/V DIBL 2.5 EI VDS
DIBL 2.5 0.0088 0.7V 15.4 mV / V
6.1 Overview
Feature size, pitch, node
Double-gate
Single-gate SOI Double-Gate, ground-plane
2 equal gates
2 different gates
In planar technique the fabrication of a back-gate (oxide with nm thickness and good interface quality) is not easy
Simulations indicate:
Lgate = 25nm
Tox,eq = 12Å
Threshold voltage roll-off for double-gate and single –gate Threshold voltage roll-off for double-gate and ground-plate
MOSFET MOSFET is similar
short channel effects are reduced in thinner silicon layers -> fully depleted
Concepts
single gate
Gate 1
poly-Si,
metal Gate
Gate fin
silicon thickness
oxide
oxide Source Drain
w silicon
Source Drain Gate 2
silicon
channel length
2 identical gates 2 independent gates
(just by etching down)
FinFET
contact
contact
fin
Channel length = LG
gate width LG
Channel width = 2*fh + fw
wG contact Gate
dielectric
fin width fw
fin f fin height
h
S
i
15nm
50nm
Source 30nm Drain
50 nm
Vds=0.05, 1.0 V
10-6
Vg=1.5 V
10-8 1.25 V
5 5µA with w=50nm
10-10 1.0 V
-> 100µA/µm
In early days several geometrical designs due to special process flow were developed and various names were proposed,
since some of the early names were protected by patent laws:
Drain Gates
w/o connected
metal lines
Source Multi-FET
Gate Fin
looking for the cross section various other names were invented:
silicon
isolator
low leakage
Single-Gate-FET
SRAM
the tri-gate architecture provides more device width for a given cell footprint compared to the standard planar transistor,
thus providing a higher read current because total current is a direct function of the total device width
Samsung Intel
Advanced MOSFETs and Novel Devices Prof.Dr. W. Hansch , Dr.J.Biba
Institut für Physik AdMOS, 6-117
6 Future MOSFETs Fig.1
6.8 Double Gate MOSFETs
Multi-Gate FETs
Si3N4 ?
A1 for self-adjusting contact
metal, W ?
barrier, TiN ?
top of fin
S/D contacts,
A2 etched groove and filled
with SiGe
zoom
Fig.4
cross-section of NMOS
Dummy-Gate Dummy-Gate
at end of fin at end of fin
A1 A2
Fin etched fin (Si) Fig.2: cross section along the fin
(this cross-section is a different structure (only 5 gates) compared
Si-substrate to the top view above (Fig.1) with 6 gates
Advanced MOSFETs and Novel Devices Prof.Dr. W. Hansch , Dr.J.Biba
Institut für Physik AdMOS, 6-118
6 Future MOSFETs
6.8 Double Gate MOSFETs
Multi-Gate FETs
Gate
Drain/Source
Source: Chipworks
W
The fill has been changed from TiAl in the earlier parts to tungsten. It is
more prominent in the NMOS gates than the PMOS, because the
PMOS structure includes both work-function metals, whereas the TiN
has been etched out of the NMOS gates. At the 45-nm node Intel used
tensile tungsten in the contacts to apply channel stress – have they
TiAlN TiN transposed this to the gates in the 22-nm process?
high-k
With FinFET proper working devices with acceptable SCE and DIBL down to channel length 7 nm (year 2019) look reasonable
ultra-thin silicon < 5nm must be vertically patterned -> oxide quality on etched, vertical sidewall ?
Electrical problems:
-4
Electrons
Log |x|
BB Gener
Signed Log
10
20
18
Vds=1V, Vgs=0V, m=5eV, channel: p1e17 23.72
22 -6
16 20 10
Id / A/m
0.00
14 18
12 16
-8
10
Band to band tunneling 14
10
Energy (eV)
8 12
6 10
4 8 -10
Tunneling 3.448 7.171
10
-1.00
-12
Tox=3nm,
10 Drain: n1e20
20nm
Due to band to
Source
10
Drain
-2.00
Source: Anisotropic Nanomaterials, Preparation, Properties, and Applications, Li Q, 2015, Springer Verlag
Top-down fabrication
IEEE 2009
CMOS-Inverter
IEEE 2008
LG = 350 nm
D = 3 nm
Fabrication of CMOS-circuits
Transistors show good IV-characteristic
High fluctuation of ION and VT
Reducing short channel effects
Tuning of VT
Advanced MOSFETs and Novel Devices Prof.Dr. W. Hansch , Dr.J.Biba
Institut für Physik AdMOS, 6-125
6 Future MOSFETs
6.8 Double Gate MOSFETs
Lateral Gate-All-Around Silicon FET-2016
IMEC, 2016 Symposia on VLSI Technology & Circuits
http://phys.org/news/2016-06-imec-gate-all-around-mosfets-lateral-silicon.html
First time vertically stacked horizontal silicon nanowires at scaled dimensions are shown
(Lateral pitch 45 nm, 20 nm vertical separation)
http://semiengineering.com/uncertainty-grows-for-5nm-3nm/
IEEE 2008
LG = 150 nm
D = 25 nm
M. Vinet et all, Monolithic 3D Integration: a powerful alternative to classical 2D Scaling, IEEE 20014
Top Wafer
Substrate Wafer
• FDSOI-MOSFET
• High-k/Metal-Gate • Low temperature
bonding
• Dielectric deposited • Initial substrate removal
• Planarization with CMP
4. Top MOSFET 5. Contact process
IEEE 2016
RF devices required for high-frequency analog or mixed signal circuits integrated on top of MOSFETs
Advanced MOSFETs and Novel Devices Prof.Dr. W. Hansch , Dr.J.Biba
Institut für Physik AdMOS, 6-131
6 Future MOSFETs
6.8 Double Gate MOSFETs
Summary
With FinFET proper working devices with acceptable SCE and DIBL down to channel length 7 nm (year 2019) look reasonable
IBM 2015
IEEE 2013
IEEE 2016
• CMOS Technology
• High-k/Metal-Gate
• Replacement Gate Process
• Dual Strained channels (SiGe)
• Channel length 15 nm
• PFET SiGe channel
IEEE 2016
6.1 Overview
Feature size, pitch, node
https://www.micron.com/products/solid-state-storage/3d-nand-client-ssds
14
The Flash-NAND Memory consists out of Floating-Gate MOSFETs which are linked together in a NAND circuit.
Input Output
A B
0 0 1
1 0 1
0 1 1
1 1 0
log |ID|
VT1
10-3 A/µm
VT2 To store 1 the electrons have to tunnel into the floating gate
10-6 A/µm
Stored electrons in the floating gate lead to a higher Threshold Voltage VT2
10-9 A/µm
10-12 A/µm Only when both transistors are on, the output in the NAND circuit changes
- Vgate + Vgate
Advanced MOSFETs and Novel Devices Prof.Dr. W. Hansch , Dr.J.Biba
Institut für Physik AdMOS, 6-138
6 Future MOSFETs
6.9 Advanced Memory Devices
Working principle of 2D-Flash-NAND Memory
http://panasonic.net/avc/sdcard/industrial_sd/img/lineup_img01.jpg
MLC can save 2 bits in one single Floating Gate, TLC can save 3 bits in one Floating Gate
Due to stacking,
change from single
crystalline substrate
to poly crystalline
devices
http://www.theregister.co.uk/2013/07/23/sandisk_takes_the_bics_route_to_3d/
NAND is coming to an end because of reliability and degrading 128 Gbit memory
performance
• Two states:
• HRS high-resistance state
• LRS low-resistance state
• Set process:
Switching from HRS to LRS
• Reset process:
Switching from LRS to HRS
• Read process:
Small voltage which has no
H.-S. Philip Wong et. all, Metal–Oxide RRAM, IEEE 2012
affect on the state
• Device structure is metal insulator(i.e. HfO2) metal
• Formation process:
• High electric field: oxygen atoms drift to the anode
• Electrode/oxide interface behaves like an oxygen reservoir
• Localized deficiency of oxygen leads to a conductive filament
• LRS State
• Two states:
• HRS high-resistance state (“0”) https://www.theregister.co.uk/2016/08/31/nram_dev_nantero_signs_fujitsu/, 2017
• High-resistance state
• Carbon nanotubes are not in contact with each other
• Low-resistance state
• Carbon nanotubes are in contact with each other
1 Transistor with one RRAM cell 1 Transistor with several RRAM cells
www.crossbar.com
Four memory
layers
Connecting
metallization
layers
New channel materials instead of the polysilicon channel Most promising candidate is the RRAM
6.1 Overview
Feature size, pitch, node