segmented mosfet technology for reduced variability
TRANSCRIPT
Segmented MOSFET Technology for Reduced Variability
T J Ki Li Xi S Ch h ShiTsu‐Jae King Liu, Xin Sun, Changhwan Shin, Nattapol Damrongplasit, Byron Ho, Reinaldo Vega Electrical Engineering and Computer Sciences DepartmentElectrical Engineering and Computer Sciences Department
University of California at Berkeley
September 30 2009September 30, 2009
IMPACT Project Seminar
Outline
• Introduction– Transistor scaling challenges– Advanced transistor structures
• Tri‐Gate Bulk/PD‐SOI FET Design
• Corrugated Substrate Technology
S• Summary
Modern MOSFET Structures(Intel Penryn© from www semiconductor com)(Intel Penryn©, from www.semiconductor.com)
N-channel MOSFETs P-channel MOSFETs
• 45nm CMOS technology features:– high‐permittivity gate dielectric and metal gate electrodes– strained channel regions– aggressive silicidation of n+ source/drain regions
3
Historical Voltage Scaling• Since the threshold voltage VTH cannot be scaled down
aggressively, the power‐supply voltage (VDD) has not beenaggressively, the power supply voltage (VDD) has not been scaled down in proportion to the MOSFET gate length:
V log IVDD
Low VTH
High VTH
log IDS
VDD – VTH
S
VGS0 VDDVTH
Source: P. Packan (Intel), 2007 IEDM Short Course
K. Bernstein et al., IBM J. Res. Dev. 50‐4/5, 2006: 4
The CMOS Power Crisis
Power Density vs. CMOS Scaling Power Density Prediction circa 2000
(W/c
m2 )
1E 00
1E+01
1E+02
1E+03Active Power Density
1000
10000
(W/c
m2)
Rocket NozzleSun’s Surface
er D
ensi
ty
1E-03
1E-02
1E-01
1E+00
40048008 8085
8086
Pentium® procP610
100
wer
Den
sity
Hot Plate
Nuclear Reactor
Core 2
Pow
e
1E-05
1E-04
1E-03
0.01 0.1 1G t L th ( )
Passive Power Density 80088080
8085286 386 486Pentium® proc
11970 1980 1990 2000 2010
YearPo
w
• Improved MOSFET design for lower leakage and reduced
Gate Length (μm)Source: B. Meyerson (IBM) Semico Conf., January 2004 Source: S. Borkar (Intel)
• Improved MOSFET design for lower leakage and reduced variability eventually will be needed to mitigate this crisis.
5
Sources of Variability• Sub‐wavelength lithography:
– Resolution enhancement Design
250nm250nm 180nm180nm
OPCOPC
90nm and Below90nm and Below
PSM
0°
PSMPSM
0°0°
techniques are costly and increase process sensitivity
Mask
Wafer
0°
180°
0°
180°
0°
180°
OPC0°
180°OPCOPCOPC
0°
180°
• Layout‐dependent transistor performance:– Process‐induced stress is dependent on layout
• Random dopant fluctuations (RDF):– Atomistic effects become significant
in nanoscale FETs
SiO2 GateA. Brown et al., IEEE Trans.Nanotechnology,p. 195, 2002
photoresist
in nanoscale FETs
p. 195, 2002
Source DrainA. Asenov, Symp. VLSI Tech. Dig., p. 86, 2007• Random Telegraph Noise (RTN) 6
Impact of Misalignment6-T SRAM Cell
Actual layout w/ lateral misalignment(gate length variations)
Desired layout(6‐T SRAM cell)
Lg reduced Lg increased
PD PGPU
Lg reduced Lg increased
PG PDPU
Actual layout(corner rounding)
Actual layout w/ vertical misalignment(channel width variations due to active jogs)
W reduced W increasedW reduced
7
Double Patterning of Gate6-T SRAM Cell
Actual layout after 1st gate patterningDesired layout(6‐T SRAM cell)
PD PGPU
PG PDPU
Actual layout after active patterning Actual layout after 2nd gate patterning(no gate length variation)
8
Impact of Variability on SRAM
• VTH mismatch results in reduced static noise margin.lowers cell yield and limits V scalinglowers cell yield, and limits VDD scaling
Circuit Schematic of 6‐T SRAM Cell Butterfly Curve
Y. Tsukamoto et al., Proc. IEEE/ACM ICCAD, p. 398, 2005
Immunity to short‐channel and narrow‐channel effects, aswell as RDF effects is needed to achieve high SRAM cell yieldwell as RDF effects, is needed to achieve high SRAM cell yield.
9
Why New Transistor Structures?
• Leakage must be suppressed as Lg is scaled down
• Leakage occurs in the region away from the channel surface
Let’s get rid of it! Lg
G t
Thin-BodyMOSFET:
Gate
DrainSource
Gate
Source Drain
Gate
“Silicon-on-Insulator”
Buried Oxide
Substrate
Insulator (SOI)Wafer
10
Thin‐Body MOSFETs
• Leakage can be suppressed by using a thin body (tSi < Lg)
Ch l d i i d d hi h i bili– Channel doping is not needed higher carrier mobility– Aggressive gate‐oxide scaling is not needed
• The double‐gate structure is more scalable (to L <10nm)The double‐gate structure is more scalable (to Lg<10nm)
Ultra-Thin Body (UTB)L
Double-Gate (DG)
Gate
Lg
Gate
Buried Oxide
Source Drain
Gate
tSi
Gate
Source Drain tSi
Substrate
11
Vertical Double‐Gate “FinFET”
Planar DG-FET D. Hisamoto et al., IEEE Int’l Electron Devices Meeting, 1998
FinFET
Gate
Lg N. Lindert et al., IEEE Electron Device Letters, p.487, 2001Drain
GateLg
Source Drain
Gate
tSiSource
Gate
Fin Width WSi = tSi
Fin Height Hfin = W/2
GATEDRAIN
20 nm
10 nm Y.-K. Choi et al., IEEE Int’l Electron Devices Meeting 2001
15nm Lg FinFET:
SOURCE
12
Thin‐Body MOSFET Challenges• “There are various problems associated with a move to multi‐gate field‐effect
transistors (MuGFETs)…challenges with parasitic resistance, parasitic capacitance and vertical topology will make MuGFETs difficult to implement ”
K. Kuhn (Intel Fellow, Director of Advanced Technology) in EE Times,12/15/08 capacitance and vertical topology will make MuGFETs difficult to implement.
F. Boeuf (ST Microelectronics), 2009 Symp. VLSI TechnologyShort Course
A better solution is needed…
13
The Tri‐Gate Bulk FETM Kito et al (Toshiba Corp ) 2005 Symp VLSI Technology
XTEM images
M. Kito et al. (Toshiba Corp.), 2005 Symp. VLSI Technology
• A tri‐gate structure is easily achievedpoly‐Sigate
poly‐Sigate
Si Si
A tri gate structure is easily achieved by slightly recessing the isolation oxide (or by selective epitaxial growth) just prior to gate‐stack formation
Measured I‐V Characteristics
prior to gate stack formation
• Superior electrostatic integrity is achieved with the tri‐gate structure
reduced impact of process‐reduced impact of processinduced variationsfacilitates voltage scaling
14
Outline
• Introduction
• Tri‐Gate Bulk/PD‐SOI FET Design
• Corrugated Substrate Technology
• Summary
Bulk vs. SOI Tri‐Gate FET DesignX Sun et al IEEE Electron Device Letters Vol 29 pp 491 493 May 2008
tri-gate bulk tri-gate SOItri-gate bulk tri-gate SOIOFF
Current Contour Plots
Required Si Channel Dimensions
X. Sun et al., IEEE Electron Device Letters Vol. 29, pp. 491‐493, May 2008.
A/cm-2A/cm-2
(μm
)
(μm
)
A/cm-2A/cm-2
(μm
)
(μm
)State
Wider/taller stripes can be used with bulk ( ) ( )( ) ( )
used for 2D cuts
used with bulk design tri-gate bulk tri-gate SOI
(μm
)
(μm
)
tri-gate bulk tri-gate SOI
(μm
)
(μm
)
ON State
(μm) (μm)(μm) (μm)
[16] J.G. Fossum et al., IEDM Tech. Dig., pp. 613-616, 2004.
Bulk design achieves better layout efficiency
A/cm-2 A/cm-2A/cm-2 A/cm-2
efficiency
(μm) (μm)(μm) (μm)
16
Idealized MOSFET Designs Planar bulk Tri-gate bulkxy
0 Planar bulk Tri-gate bulkxy0
Gate GateSource DraintSi
tSiWz
toxW Gate Gate
Source DraintSi
tSiWz
toxW
Si Substrate Si Substrate
Ground-plane
Source DrainSi
Si Substrate Si Substrate
Ground-plane
Source DrainSi
tSi
Double-gate SOIxz
y
0
tSi
Double-gate SOIxz
y
0
ping
con
c.
Ground-plane Npeak≥1019cm-3
ping
con
c.
Ground-plane Npeak≥1019cm-3
GatetSi
W GatetSi
WC
hann
el d
optSi
NA
Cha
nnel
dop
tSiNA
BOXBOXDepth from top surface z
C
Depth from top surface z
C
17
Tri‐Gate Bulk FET: Scale Length DerivationX Sun et al to appear in IEEE Trans Electron Devices
Poisson’s equation:Si
AqNdzd
dyd
dxd
ε=
Φ+
Φ+
Φ2
2
2
2
2
2
X. Sun et al., to appear in IEEE Trans. Electron Devices
+ Sidzdydx ε
parabolic potential approximations+ )()(2 φφd+appropriate boundary conditions
where λ is the scale length:
0)()(22
2
=−λφφ x
dxxd
2
2
2
1Siox
leak
Si
ox
Si
leak
ox
leak
Si
ox
ttz
tz
tz
−−+=
εε
εε
λ
2
2
2
2
2228SiSioxSi
ox
Siox
leak
Si
ox
Si
leak
ox
leak
Si
ox
tttttz
tz
tz
W++⎟⎟
⎠
⎞⎜⎜⎝
⎛−−
=
εε
εε
εε
λ
SiSi
ox
ox
Sileak
tttz 221
2 +=
εεwhere is the location of the highest leakage current density
18
Scale Length ComparisonTri‐Gate Bulk FET vs Planar Bulk FET vs FinFETTri‐Gate Bulk FET vs. Planar Bulk FET vs. FinFET
50double-gate SOI
50double-gate SOI
Scale length contour plots
40
double-gate SOI ground-plane tri-gate bulk
m) λ = 10nm
40
double-gate SOI ground-plane tri-gate bulk
m) λ = 10nm
tox = 1nm
20
30
t Si (n
m λ = 10nm
λ = 7nm20
30
t Si (n
m λ = 10nm
λ = 7nm
10 20 30 40 50 6010
20 λ = 7nm
10 20 30 40 50 6010
20 λ = 7nm
• For given values of W and tSi, the scale length is smallest
10 20 30 40 50 60W (nm)
10 20 30 40 50 60W (nm)
Sifor the tri‐gate bulk FET design.
X. Sun et al., to appear in IEEE Trans. Electron Devices 19
Sensitivity to tox VariationsX Sun et al to appear in IEEE Trans Electron Devices
3 double gate SOIth
λ = 7nm at tox = 1nm
X. Sun et al., to appear in IEEE Trans. Electron Devices
123 double-gate SOI
ground-plane tri-gate bulk, W=20nm
cale
leng
-10
nge
of s
c
0 9 1 0 1 1-3-2
% c
han
0.9 1.0 1.1 tox (nm)
• Tri‐gate and planar ground‐plane bulk FETs are lessTri gate and planar ground plane bulk FETs are less sensitive to tox variation as compared with the FinFET.
20
Tri‐Gate Bulk FET AdvantagesX Sun et al to be published
• 3‐D device simulations of transistors designed for minimum intrinsic delay for a given off‐state leakage specification:
X. Sun et al., to be published
y g g p– Lg=25nm (EOT=1nm): IOFF=8nA/µm; Lg=20nm (EOT=0.9nm): IOFF=18nA/µm
The tri‐gate bulk FET is less sensitive The delay advantage of the tri‐gate
1 5
2.0
Planar bulk Tri-Gate bulk FinFET SOI
s)
15
) 86% reduction
to WSTRIPE variation than the FinFET.
81% reduction
bulk FET increases with scaling.
34% reduction37% d ti
1.0
1.5
ic D
elay
(ps
10
WSi| (
mV/
nm)37% reduction
0 0
0.5
Intr
ins
0
5|d
V T/dW
0.028nm22nm18nm34nm27nm21nm
20nm 12nm20nm15nm25nm25nmWSiLeff LG=20nmLG=25nm
028nm22nm34nm27nm
20nm 12nm15nm25nmWSiLeff
LG=20nmLG=25nmVDD=0.8V VDD=0.7V VDD=0.8V VDD=0.7V
21
Impact of WSTRIPE VariationsX Sun et al to be published
Lg = 20nm, EOT = 0.9nm, HSTRIPE = tSi = XJ = 14nm
X. Sun et al., to be published
0.36
0.32
0.34
V T_LI
N (V)
14 16 18 20 22 24 260.30
0.32
• Smaller WSTRIPE stronger gate control smaller VTH
14 16 18 20 22 24 26WSTRIPE (nm)
– ΔVTH < 10mV for 10% variation in WSTRIPE
22
Impact of Gate‐Length VariationsX Sun et al to be published
0 39
EOT = 0.9nm, WSTRIPE = 20nm, HSTRIPE = 7nm, tSi = 14nm, XJ = 7nm
X. Sun et al., to be published
0.33
0.36
0.39
V)
0.27
0.30
0.33
Planar
V T_LI
N (V
14 16 18 20 22 24 260.24
Planar Tri-Gate Bulk
LG (nm)
• The tri‐gate bulk FET is less sensitive to gate‐length i ti d t i d t t l
(Channel‐ and source/drain‐doping profiles are identical for the planar and tri‐gate MOSFETs)
variations, due to improved gate control.• Benefit is equivalent to >6Å reduction in tox
23
Impact of Gate Line‐Edge Roughness3-D Device Simulation Study
Device Simulation3D StructureGeneration
Sample measured LER data using MATLAB code
(200 cases)Tri-gate bulk FET
Use MATLAB to process LER data to be inputted into Sentaurus Structure Editor
P f 3 D d i i l ti i
Tri gate bulk FET with gate LER:
Perform 3-D device simulation using Sentaurus Device
Assumption:Assumption: – LWR in Leff is the same as that for
the gate electrode. (worst case scenario relevant for
Doping contour plot:
(worst case scenario, relevant for ultra-shallow junction technology)
24
Simulated ID‐VG CharacteristicsX Sun et al to be published
• 200 gate-LER cases were simulated for each structure.• I V curves for a nominal structure (without gate LER)
X. Sun et al., to be published
• ID-VG curves for a nominal structure (without gate LER) are shown in black.
VDD = 0.7 VPlanar Bulk MOSFETs Tri-Gate Bulk MOSFETs
σVTH = 22.26 mVσVTH = 42.68 mV
Lg = 20 nmEOT = 0.9 nm
Lg = 20 nmEOT = 0.9 nm
WSTRIPE = Weff = 20 nmX 7 6
WSTRIPE= 20 nmHSTRIPE = 12 nmWeff = 44 nm
TSi = 12 nmXJ = 7.6 nm XJ = 15 nm
25
Impact of Random Dopant FluctuationsX Sun et al IEEE Electron Device Letters Vol 29 pp 491 493 May 2008
Monte Carlo simulations, 100 cases eachLg = 20nm, EOT = 0.9nm, WSTRIPE = 20nm, HSTRIPE = 14nm, tSi = 14nm, XJ = 14nm
X. Sun et al., IEEE Electron Device Letters Vol. 29, pp. 491‐493, May 2008.
tri-gate bulk MOSFETstri-gate bulk MOSFETsplanar MOSFETsplanar MOSFETs
VDD = 0.8 V
N+ N+N+ N+N+ N+N+ N+
P-subP-subP-subP-sub
• The tri‐gate bulk MOSFET is less sensitive to RDF effects.26
Impact of HSTRIPE VariationsX Sun et al to be published
LG = 20nm, EOT = 0.9nm, WSTRIPE = 20nm
X. Sun et al., to be published
0.39
0.42 TSi = XJ,EXT = HSTRIPE TSi = XJ,EXT = 14nm
0.33
0.36
V T_LI
N (V)
12 13 14 15 160.27
0.30
• If tSi is fixed, VTH is not sensitive to HSTRIPE variation.
12 13 14 15 16HSTRIPE (nm)
If tSi is fixed, VTH is not sensitive to HSTRIPE variation.
27
VTH Adjustment ApproachesC Shin et al IEEE 2008 Silicon Nanoelectronics Workshop
• VTH of a tri‐gate bulk MOSFET can be adjusted by tuning either the dose (Npeak) or the depth (tSi) of the retrograde doping profile.
C. Shin et al., IEEE 2008 Silicon Nanoelectronics Workshop
( peak) p ( Si) g p g p
– 200 atomistic simulations were run for each nominal design.
• VTH adjustment via tSi tuning provides for less variation, and li i t th t d ff ith h t h l t leliminates the trade‐off with short‐channel control.
DIBL values for each design are indicated
28
Outline
• Introduction
• Tri‐Gate Bulk/PD‐SOI FET Design
• Corrugated Substrate Technology– SegFET fabrication process g p– SRAM performance benefits– Advanced channel materials
• Summary
Segmented Bulk MOSFET (SegFET)
• The channel is digitized into stripes of equal width, isolated by very shallow trench isolation (VSTI) oxideisolated by very shallow trench isolation (VSTI) oxide– The effective channel width is adjusted by adjusting the number of stripes.p
– Each stripe is a tri‐gate bulk MOSFET.
• Note that the source/drain regions are contiguous./ g g
Cross‐Sectional ViewsA ‐ A’ C ‐ C’B ‐ B’gate
electrode
Plan Viewfield region (STI)
Lgactive device
regionC’ H
WSTRIPEvery shallow trench
isolationregions
pn+n+XJ
gate electrode
WSTRIPEXVSTI SiO2
A A’
B B’
C HSTRIPE
TSi
gp‐Si p‐Sip‐Si
A AC
T.‐J. King Liu and L. Chang, “Transistor Scaling to the Limit,” in Into the Nano Era, H. Huff ed. (Springer), 2008.
30
T J King and V Moroz U S Patent 7 265 008
SegFET Fabrication ProcessT.‐J. King and V. Moroz, U.S. Patent 7,265,008
1. Start with corrugated substrate
2. Define active areas
3. Fill trenches to formSTI; Implant wells
4. Slightly recess theisolation oxide (optional)
5. Implant channels;Form gate stack
6. Form S/D extensions, then sidewall spacers
7. Grow epitaxial materialin S/D regions (optional)
8. Dope S/D regions;Form silicide
• Precise control of channel width is achieved.LG
WSTRIPE
Precise control of channel width is achieved.(Layout dependencies and sensitivity to misalignment are reduced.)
reduced variation in MOSFET performance31
Layout Area Efficiency ComparisonHigh‐Performance Design
2100planar w/o STIk 3 9 ( l )
High Performance DesignLg = 30nm, EOT = 1nm, WSTRIPE = 30nm, tSi = XJ = 15nm, VDD = 1V
1500
1700
1900
A/µ
m)
k=3.9 (planar)k=23 (planar)k=3.9 (tri-gate)k=23 (tri-gate)
Tri-gate
1100
1300
500
I ON/S
P (µ
Segmented planarContinuous planar
700
900
40 45 50 55 60
Segmented planar
Stripe Pitch (nm) @ Wstripe=30 nm
• A segmented planar FET can achieve layout efficiency comparable to that of a conventional planar FET, if WVSTI is very small (<10 nm)to that of a conventional planar FET, if WVSTI is very small ( 10 nm) or if a high‐k dielectric is used as the VSTI material.
• The tri‐gate FET achieves superior layout efficiency. 32
Impact of Channel Width on Strain ProfileCapping‐layer‐induced strain along the channel
Segmented FET
pp g y gContact etch stop liner is assumed to be a 30nm‐thick silicon nitride with
Narrow ChannelWide Channel
(bulk tri‐gate)
• SegFET parameters:
thick silicon nitride with 2GPa tensile stress
Planar
WSTRIPE = 20nmWSPACING = 20nmHSTRIPE = 10nm
Planar• LG = 20nm• EOT = 0.9nm• T = 40nm• TGATE = 40nm• LSPACER = 20nm
• More stress is induced in SegFET More mobility enhancement• Reduced variation with Weff for SegFET Reduced μ variation
X. Sun et al., to be published 33
6‐T SRAM Cell DesignC Shin et al presented at the 2009 SISPAD
• 6‐T SRAM cells with comparable α and β ratios were designed based on 22nm CMOS technology design rules
C. Shin et al., presented at the 2009 SISPAD
designed, based on 22nm CMOS technology design rules.–For the SegFET SRAM cell: WSTRIPE = 20nm, WSPACING = 15nm–WSPACER = 15nm (limited by gate‐to‐contact pitch), XJ ~ 10nm
Simulated Planar SRAM Cell(isolation oxide not shown for clarity)
Simulated Tri‐gate bulk SRAM Cell(isolation oxide not shown for clarity)
Simulated SegFET SRAM Cell(isolation oxide not shown for clarity)
PG PUPD
PGPD
PU
PGPG
PUPUPGPG
PUPUPDPD
PG
PDPU PG
PU
PD
PGPG
PUPUPGPG
PUPUPDPD
PG
PD
PU PG
PUPD
34
Simulated Read and Write MarginsC Shin et al presented at the 2009 SISPAD
• SNM is largest for the SegFET cell: 241 mV vs 221mV for the planar bulk MOSFET cell
C. Shin et al., presented at the 2009 SISPAD
vs. 221mV for the planar bulk MOSFET cell
• Iw for the SegFET cell is also superior to that for the planar b lk MOSFET cell
1.0
Seg
Butterfly curves
the planar bulk MOSFET cell.
120
Seg
Write‐N‐curves
0.6
0.8
2 [V
]
Seg Tri-gate Planar
60
80
100
[uA
]
Seg Tri-gate Planar
0.2
0.4
V n2
SNM|Seg= 241.2mV
SNM|TG= 226.4mV
SNM| = 221 2mV
20
40
60
I n1 [
Iw Seg= 14.8 μA
Iw TG= 12.7 μA
I 10 2 A
0.0 0.2 0.4 0.6 0.8 1.00.0
Vn1 [V]
SNM|PL= 221.2mV
0.0 0.2 0.4 0.6 0.8 1.00
Vn1 [V]
Iw PL= 10.2 μA
35
Single‐Event‐Upset ComparisonFull 3‐D transient simulations
• When a heavy‐ion beam impinges on the high (“1”) storage node at time t = timpact, the generated e‐/h+ can turn on a
Full 3‐D transient simulations
impact, g /parasitic thyristor, shorting the drain node to the source node.
• The SegFET cell is more robust to a given ion‐beam strike.ibl l i ll d i li h l l i l– Possible explanations: stronger pull‐up device, slightly larger internal
capacitance, or smaller body effect.
0.9
SegFET SRAM Cell0 9
Planar MOSFET SRAM Cell
oltage [V
]
0.9
Vn1:Red dot‐line
Vn2:Blue solid‐line
oltage [V
]
0.9
Vn1:Red dot‐line
Vn2:Blue solid‐line
Vo
0
Critical LET:
Vo
0
Critical LET:
C. Shin et al., presented at the 2009 SISPAD
Time [ns]timpact
1 3 75
Critical LET:0.35pC/μm
Time [ns]timpact
1 3 75
Critical LET:0.24pC/μm
36
Advanced Channel Materials• High‐mobility semiconductor
materials can potentially provide
Si Ge GaAs
Electron mobility (cm2/Vs)
1500 3900 8500materials can potentially provide for improved performance:– Ge for PMOS
(cm /Vs)
Hole mobility (cm2/Vs)
450 1900 400
Lattice constant (Å) 5.431 5.646 5.653
– (In)GaAs for NMOS Band gap (eV) 1.12 0.66 1.424
Dielectric constant 12 16 13
• Selective epitaxial growth directly on Si is facilitated by
Ge on Si GaAs on Si
p g y ythe use of a corrugated substrate:
J.‐S. Park et al., Appl. Phys. Lett. 90 052113, 2007J. Z. Li et al., Appl. Phys. Lett. 91 021114, 2007
37
Ge Integration Options
• Homogeneous semiconductor stripes:R l d ( t i d) h l
GATESi• Relaxed (unstrained) channel
• High BTBT leakage due to small Eg• Significant DIBL due to large εr
GeSiO2 SiO2
Si
Significant DIBL due to large εr
• Heterogeneous semiconductor stripes:
Si
GATE• Heterogeneous semiconductor stripes:
• Thin, strained channel
• Reduced BTBT leakageGe Si
g
• Reduced DIBLSi
SiO2 SiO2
38
Outline
• Introduction
• Tri‐Gate Bulk/PD‐SOI FET Design
• Corrugated Substrate Technology
• Summary
Summary
• Power density and variability now limit transistor scalingK (ITRS) l ti l ti d th f tl– Known (ITRS) solutions are revolutionary and therefore costly
• Tri‐gate bulk MOSFET (corrugated substrate) technology offers an evolutionary (low‐cost) pathway to lower VDD, reduce VTH variability, and extend transistor scaling
tili ti l ( t bli h d) IC f b i ti t h i• utilizes conventional (established) IC fabrication techniques
• is compatible with technologies developed to date (e.g. metal‐gate/high‐k dielectric, strained‐Si) for bulk/PD‐SOI CMOSg / g , ) /
… that is almost transparent to end users• Established compact models; zero design cycle time impactEstablished compact models; zero design cycle time impact• Minimal (non‐disruptive) impact on design style
40
The Age of Ubiquitous ComputingR PCs (1 person/computer)
Mainframes (>1 persons per computer)Technology, Device & Circuit
Innovations,Heterogeneous Integration
Transistor Scaling
Acknowledgement: Mark Weiser
UbiComp (>1 computers per person)
LES
($)/Y
R ( p p )
Lower Power,Lower Cost
Heterogeneous Integration
Investment Higher Performance,Lower Cost
today
SA
TIME• Information technology will befor better quality‐of‐life
Market Growth
Environment– pervasive
– embedded
– human‐centered
$370 BillionTotal U.S. Annual Energy Costs
200%Increase in U.S. Electricity Consumption
– solving societal‐scale problems
Healthcare Disaster response
EnergySensatex
human centeredsince 1990
40%Total U.S. Energy Consumptionfor Buildings
72%
Philips Transportation
Total U.S. Electricity Consumption for Buildings
55%Total U.S. Natural Gas Consumptionfor Buildings 41
Acknowledgements
• FLCC & IMPACT project corporate sponsorsFLCC & IMPACT project corporate sponsors
• UC Discovery Grants Program
• Semiconductor Research Corporation