Upload
vokhue
View
218
Download
0
Embed Size (px)
Citation preview
1 © 2010 Toshiba Corporation 1
New channel engineering for the low power CMOS technology
東芝グループは、持続可能な
地球の未来に貢献します。
Akira Nishiyama Corporate R & D Center, Toshiba Corporation
2 © 2010 Toshiba Corporation 2
SMART Community: 3 Major Streams
* Intelligent Transport System
Smart Grid → Stream of Energy
Cloud Computing →Stream of Information
Intelligent Transport System Smart logistics
Contribution of Semiconductor Industry ?
→ Stream of people, goods
3 © 2010 Toshiba Corporation 3
PhotographNewspaper
Phone(1hr)
Music(1hr)
TV(1hr)
HD-TV(1hr)2k*4k
4K*8k
100
1000
10000
100000
1E+06
1E+07
1E+08
1E+09
1E+10
1E+11
1E+12
1E+13
1E+14
1E+15
1E+16
Media (without comperssion)
Dat
a[bi
t]
Explosion of Information
1000 words
3D TV
Larger Data Files X More Frequent Access ll • High Speed Processing • Large data capacity with low cost
Data Center of Apple (50,000m2) (P.30, Bussinessweek Sept.6, 2010)
Social Demands
4 © 2010 Toshiba Corporation 4
Power Consumption increase for IT facilities
Large Power Consumption by IT facilities → Measures for Dramatic Energy Saving Requested.
Power Consumption (x1E8 kWh) x
World IT facilities 15% of world
total power
Source: METI Japan http://www.meti.go.jp/committee/materials/downloadfiles/g80520c03j.pdf
5 © 2010 Toshiba Corporation 5
Sensor Network Market inside Japan
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
1 2
健康管理
その他
設備管理
防災・環境
交通
セキュリティー
2007 2015
Security
Environment Facility
Health Care
Nikkei Electronics 2009 December 28
~13B$ market in 2015 (Japan Only)
Mar
ket
[T¥]
Transportation system
10B$
6 © 2010 Toshiba Corporation 6
Sensor Network and Sensor Node
S
S
S
S
S コイン電池1個で何年間も動作させることが必要。
Power Manager + Battery
ProgramMemory
DataMemoryCPU
MAC
ADC
RFsignal
sensor
コイン電池1個で何年間も動作させることが必要。
Power Manager + Battery
ProgramMemory
DataMemoryCPU
MAC
ADC
RFsignal
sensor
Vast numbers of sensors, active for several years with a tiny battery or a solar cell (ex.10mW level、100kbps 10m Wireless Network)
Sensor network for Social Security and Information Grid
Sensor Node
S
7 © 2010 Toshiba Corporation 7
• Power Saving - Low power , long battery life • Data processing speed – high speed
processing with very low power and limited memory capacity
• Large band width – high speed network • Cost - smaller and cheaper ! • Flexibility with universal protocol • Information security – Certification between
nodes
Issues for Sensor Nodes
Basic requirement
8 © 2010 Toshiba Corporation 8
Dennard’s Scaling Law
Source: Intel (http://www.intel.com), from presentation at ISSCC 2003
パラメータ 記号 ScalingElectric Field E ∝ V/x 1
Tr density 1/WL K2
Current I ∝ WV2/(LTox) 1/K
Capacitance Cload ∝ LW/Tox 1/K
Speed τ ∝ Cload*V/I 1/KPower density P ∝ VI/LW 1
High speed processing and low power consumption was satisfied by scaling at
the same time.
9 © 2010 Toshiba Corporation 9
ITRS device trend: Device Speed vs Power Consumption
0
1E+12
2E+12
3E+12
4E+12
5E+12
6E+12
7E+12
1000 1500 2000 2500 3000 3500 4000
Power [kW/cm2]
Spe
ed
[GH
z]
系列1
High-k/ MetalG
UT SOI
Bulk Conv.
HP
UT SOI Bulk Conv.
LSTP
Power Consumption per unit Area (A.U.)
1.0 1.5 2.0 2.5 3.0 3.5 4.0
7
Spe
ed
(Gat
e D
elay
) [T
Hz]
6
3
2
1
0
5
4
From: ITRS* Roadmap 2009
(2009- 2024 trend)
Ideal situation is not expected for further scaling ! Why??
ideal
ITRS: International Technology Roadmap for Semiconductors
10 © 2010 Toshiba Corporation 10
Power Voltage Trend Forecast in ITRS
0
0.2
0.4
0.6
0.8
1
1.2
0 10 20 30 40 50 60
系列1
LSTP
HP
2009
2024
Half Pitch [nm]
Pow
er V
olta
ge V
dd [V
]
Insufficient Power Voltage Reduction
Scaling Law Requirement
Major Barriers for Power Voltage Reduction
• Small Ion (Device Speed
Degradation)
• Off Leakage Current Ioff Increase (Standby Power Consumption Increase)
• Relative Threshold Voltage Fluctuation Increase
From: ITRS* Roadmap 2009
11 © 2010 Toshiba Corporation 11
Approach: Non-planer structure + Ion Booster
① Large Id ⇒ Stress engineering , High-mobility channel material ② Steep subthreshold ⇒Multi-gate
Multi-gate + Ion Booster Multi-gate Bulk Planer
Vg
Log(
I d)
0 VDD
I on
I off
①
②
Multi-Gate
Channel Cross Section
Gate
New channel
strain Gate
Si BOX
Id
12 © 2010 Toshiba Corporation 12
Low power consumption CMOS Development
0
1
2
3
4
5
6
2008 2012 2016 2020
Pow
er
Consu
mpt
ion
(A
.U.)
0
0.2
0.4
0.6
0.8
1
1.2
Vdd
(V
)
New Concept Device
Green CMOS Si Nanowire
High Mobility Channel
Power= α・f・Cload・Vdd2 ・n + Ioff・Vdd・n
Active power Standby Power
α: Activation Ratio, f: Clock Freq., Cload:Load Capacitance,
Vdd:Supply Voltage, Ioff, n: Amount of devices
ITRS trend
13 © 2010 Toshiba Corporation 13
Si NW: Short Channel Effect Suppression
1x10-12
1x10-10
1x10-8
1x10-6
1x10-4
0.01
-1 -0.5 0 0.5 1 1.5 2 2.5 3
Dra
in C
urre
nt (A
/µm
)
Gate Voltage (V)
Lg = 26nmWnw = 16nm H
nw = 24nm
nMOSFET Vd = 1V
Vd = 10mV
w/ S/D epi
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.01 0.1 1
DIB
L (V
/mV)
Lg (µm)
nMOSFET Vd : 1V
Wnw : 101,31, 21,11nm
0
30
60
90
120
150
0.01 0.1 1
S Fa
ctor
(mV/
dec)
Lg (µm)
nMOSFET
Wnw : 101,31, 21,11nm
Vd : 1VWire width reduction leads to steeper subthreshold swing
Gate
Source Drain
Tox=3nm
15 © 2010 Toshiba Corporation 15
Si NW: Parasitic Resistance Reduction SiN
TEOS sidewallPoly Gate
BOX
Epi-Si(25nm)
Source Drain
SiN
TEOS sidewallPoly Gate
BOX
Epi-Si(25nm)
Source Drain
(111)面
BOX
+25nm
+20nm<001>
<110>
(111)面
BOX
+25nm
+20nm<001>
<110>
Cross sectional View of NW after Epigrowth on S/D
200nm
Source
Drain
Gate
NW channel
Enlarged by Epi-growth
200nm
Source
Drain
Gate
NW channel
Enlarged by Epi-growth
16 © 2010 Toshiba Corporation 16
NW: Parasitic Resistance Reduction
1x10-11
1x10-10
1x10-9
1x10-8
1x10-7
1x10-6
1x10-5
1x10-4
0 100 200 300 400 500 600Ion (µA/µm) @ Vg = 0.7 V
I off (A
/µm
) @ V
g = -0
.3 V
Vd = 1 V
x 2.4
nMOSFETWnw : 20nm
w/o S/D Epi
w/ S/D EpiSOI
BOX
GateTEOSSW
Silicide
NW
Rsw
Repi
RcSOI
BOX
GateTEOSSW
Silicide
NW
Rsw
Repi
Rc
Ion increase by a factor of 2.4 was obtained. ⇒Parasitic resistance reduction is important
especially for NW transistors.
17 © 2010 Toshiba Corporation 17
3D stress engineering for NW MOSFETs
• In addition to longitudinal stress, transversal and vertical stress is effective for NW MOSFETs
BOX
Si substrate
SOI Fin
Vertical stress Longitudinal stress
Transversal Stress
18 © 2010 Toshiba Corporation 18
Stress engineering to nanowire channel is highly effective for the performance improvement of nanowire transistors.
Si Nanowire Transistors: Stain Technique
30nm
SOI
Lg : 15nm Epi Si : 25nm Spacer :
10nm
<110>
HNW : 15nm
WNW :11nm
Stress Memorization Technique (SMT) of Poly-Si Gate
0 200 400 600 800 1000I of
f @ V
g = -0
.4V
(A/µ
m)
Ion @ Vg = 0.6V (µA/µm)
+58 %
<110> NW nFETWNW ~ 15nm
w/ SMT
w/o SMT
Vd = 1V
Tinv = 3.8nm
10-5
10-6
10-7
10-8
10-9
10-10
10-11
10-12
19 © 2010 Toshiba Corporation 19
Strain is especially effective for thinner NW.
Dependence on wire width
Cross-sectional View
Planer view
Planar(W=1µm)
1
1.1
1.2
1.3
1.4
1.5
1.62.7
3
0 20 40 60 80 100WNW (nm)
NW nFETL ~ 70nmVg - Vth = 1 V
I dlin
Enh
ance
men
t Rat
io b
y SM
T
<110><100>
30nm
ナノワイヤ
Lg : 15nmEpi Si
Gate電極
Spacer
30nm
ナノワイヤ
Lg : 15nmEpi Si
Gate電極
Spacer
Wire Width [nm]
20 © 2010 Toshiba Corporation 20
Performance of SMT NW CMOS
CMOS performance is greatly improved by NW .
Ieff-1 : Measure of CMOS inverter delay
0.5
0.6
0.7
0.8
0.9
1I ef
f,nM
OS-1
+ I ef
f,pM
OS-1
w/o SMT
Ioff
= 100nA/µm
<110> NW CMOS
w/ SMT
15nm
w/ SMTnMOS :pMOS : w/o SMT w/ SMT w/o SMT w/o SMT
w/ SMT
24nmHNW :
Large Ion(nMOS)Increase by SMT
Slight Ion(pMOS)Decrease by SMT
µ Increaseby Tall NW
21 © 2010 Toshiba Corporation 21
High-mobility channels for CMOS
Material Strain
Normalized mobility (vs. unstrained Si) Remarks n p
Si biaxial 2 [1] 1.4 [1]
uniaxial 1.7 [2] 3 - 4 [2, 10]
Si1-xGex biaxial 2.3 [3] 6~10 [4]
x=0.42 x=0.92
Ge
No 1.4 [5] 3 [7]
2 [6] with high-k theory
Uniaxial biaxial
4.1 [7] 10~20 [8,9]
theory buried channel
GaAs No 6.3 [2] 0.8 [2] Low field/bulk
InAs No 13-20 [2] 0.2-0.9 [2] Low field/bulk
[1] K.Rim et al., IEDM 2003, [2] S. E. Thompson, IEDM 2006, [3] T.Tezuka et al., IEDM 2001, [4] T.Tezuka et al., 2004 Symp. on VLSI Tech. [5] H. Shang et al., IEDM 2002, [6] C. Chui et al.,IEDM 2002, [7] Y.-J. Yang et al., APL 91,102103 (2007).[8] M. L. Lee et al., IEDM 2003, [9] T. Irisawa et al., APL 81, 847 (2002). [10] S. Mayuzumi et al., 2009 Symp. on VLSI Tech.
p&nFET
pFET
nFET
22 © 2010 Toshiba Corporation 22
SiGe Nano Wire formed by Ge condensation process
Higher Ge fractions for the narrower mesas Formation of Ge-rich and narrow SiGe wires Compressive stress is imposed at the same time
20
30
40
50
60
70
80
90
100
0 100 200 300 400 500 600 G
e fra
ctio
n (%
)
Wire Width (nm)
From EDX
Before 3D Condensation RIE
BOX(SiO2) Si1-xGex
1st-Condensation
SiO2
2nd-Condensation Si1-yGey (y>x)
950°C, 2h
23 © 2010 Toshiba Corporation 23
SGOI-wire pFETs
On/off ratio >104
Higher gm by a factor 3 due to higher Ge fraction, uniaxial compressive strain and {110}-like sidewall
10-12
10-11
10-10
10-9
10-8
10-7
10-6
10-5
10-4
-20 -10 0 10 20
Vg (V)
Id /
W (A
/µm
)
Vd = -0.05 VLg = 4 µm
Ge = 92%
79%
65%
53%46%
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
-20 -10 0 10 20
Vg (V)gm
/ W
(mS
/µm
)
Vd = -0.05 VLg = 4 µm Ge = 79%
65%
53%46%
92%0%
Higher x and ε
24 © 2010 Toshiba Corporation 24
SiGe-Trigate pFET Lg=50nm
•Mobility enhancement x4.9 •Id increase 45% (Power reduction 39% by overdrive decrease)
0
1 10-5
2 10-5
3 10-5
4 10-5
5 10-5
-2 -1.5 -1 -0.5 0
SOISGOI (x=0.65)
I d / C
inv (
A µ
m/ f
F)
Gate voltage Vg-Vth (V)
Lg=50 nm
0
100
200
300
400
500
600
1012 1013
SGOI (x = 0.65)Si (100)/<110>Si (110)/<110>
Hol
e m
obili
ty µ
eff (c
m2 /V
s)
Surface carrier concentration N
s (cm-2)
30nm
BOX
NiSi
TaNHfO2
xGe=0.65一軸圧縮ひずみ:2.7%
SiGe
30nm
BOX
NiSi
TaNHfO2
30nm
BOX
NiSi
TaNHfO2
xGe=0.65一軸圧縮ひずみ:2.7%
SiGe
Uniaxial strain
25 © 2010 Toshiba Corporation 25
How about NMOS?
2.3~4.1 times higherthan unstrained Si
Tensile strain over 1 %
2.3~4.1 times higherthan unstrained Si
Tensile strain over 1 %
• Higher electron mobility of Ge than Si Si Ge
µe 1500 3900 µh 450 1900
• Tensile strain in Ge … µe enhancement
Possibility of strained Ge-nMOS in combination with high-speed Ge-pMOSFETs
Y.-J. Yang et al., APL 91,102103 (2007).
27 © 2010 Toshiba Corporation 27
Tensile strain vs Ge comp. of stressors
290 295 300 305 310
X=0.7X=0.8X=0.9Ref.(Ge)
Inte
nsity
(a.u
.)
Wavenumber [cm-1]
Ge-Ge in Ge
Ge-Ge in SiGe
0
0.002
0.004
0.006
0.008
0.01
0.012
0.65 0.7 0.75 0.8 0.85 0.9 0.95
Uni
-axi
al T
ensi
le S
train
Ge Composition, x
Max.; ~1.15 % @ x=0.7
Lattice mismatch between Ge and Si0.3Ge0.7 ~1.3%
Tensile strain
Tensile uni-axial strain over 1 % was introduced @ x=0.7
Raman spectra @ Gate edges
28 © 2010 Toshiba Corporation 28
Lg dependence of strain and Gm increase
-0.002
0
0.002
0.004
0.006
0.008
0.01
70 80 90 100 110 120 130 140
GeSiGe (x=0.8)SiGe (x=0.7)
Uni
-axi
al T
ensi
le S
trai
n
Lg + 2L
sw [nm]
Gmmax increased by a factor of 2 by Lg scaling
Average value along channel
29 © 2010 Toshiba Corporation 29
Normalized Gm max gain by 1% strain
0
5
10
15
20
40 60 80 100 120 140
ControlGe EpiSiGe(x=0.8) EpiSiGe(x=0.7) Epi
Gm
max
(L) /
Gm
max
(L=1
µm)
Gate Length [nm]
Vd = 1V
0
5
10
15
20
40 60 80 100 120 140
ControlGe EpiSiGe(x=0.8) EpiSiGe(x=0.7) Epi
Gm
max
(L) /
Gm
max
(L=1
µm)
Gate Length [nm]
Vd = 0.05V
About 60% gain for Gmmax at Vd=0.05V About 30% gain for Gmmax at Vd=1V
30 © 2010 Toshiba Corporation 30
Graphene : Ultimate material ?
A.H.Castro Neto, et al., Rev.Mod.Phys. B. 81(2009), 109.
Mobility > 200,000cm2/Vs K.I.Bolotina, et al., solid state commun. 146(2008), 351.
31 © 2010 Toshiba Corporation 31
Summary
• Cloud computing and storage data explosion lead to the large power consumption of IT facilities worldwide. Sensor Network system requires vast numbers of low power sensor nodes. Measures for the dramatic reduction of power consumption of LSIs are quite important.
• Novel approach such as 3D channel structure, strain introduction, parasitic resistance reduction as well as application of new channel material is effective to meet the demands through the reduction of the power supply voltage of LSIs.
• Further research and development of this field is indispensable for the evolution of very low power electric system throughout the globe.