cmos future uni - ece.ubc.ca
TRANSCRIPT
Lecture 1
1
1
The Future of CMOSThe Future of CMOS
David Pulfrey
2
CHRONOLOGY ofCHRONOLOGY ofthe FETthe FET
• 1965 Commercialization (Fairchild)
• 1991 “The most abundant object
made by mankind’’ (C.T. Sah)
• 2003 The 10 nm FET (Intel)
• 1933 Lilienfeld’s patent (BG FET)
Lecture 1
2
3
SCALING: WHY DO IT?SCALING: WHY DO IT?
• Increase speed
• Increase density
• Reduce cost (?)
4
SCALING: SPEEDSCALING: SPEED
VDD=0.75V0.85V
Published data
Intel data
NMOS
1.45 THz2.63 THz
VDD=0.8V
Gate Length (nm)
Gat
e D
e lay
(ps)
10 100 1000
101
102
100
10-1
Morkoc04
CV/I
Lecture 1
3
5
SCALING: DENSITYSCALING: DENSITY
Perlmutter04, Raghavan00
P4 (130): 55M in 146 mm2
P4- Prescott (90): 125M in 112 mm2
P4- Dothan (90): 140M in 87 mm2
6
SCALING: COSTSCALING: COST
Lecture 1
4
7
SCALING: FACTORY COSTSCALING: FACTORY COST
Moore’s Law for Fabs!
MIT02
Intel 300mm
8
THE SHRINKING FETTHE SHRINKING FET
Leff reduced 30XBut devices are still “well tempered”
So urce
NCHW B
n+poly
NSUB
Dr ai n
oxide
Ln+ n+
ga te
xj
VDD
Lint
y
xCMOS 3 CMOS P18 CMOS P13 90NM
1987 2001 2002 2003
L nm 3000 180 130 100LINT nm 700 10 0 2.5XJ nm 1000 160 190 150TOX nm 85 4.1 2.8 2.3NCH cm-3 1.00E+16 3.90E+17 6.15E+17 8.37E+17VDD V 5.0 1.8 1.2 1.0VTHO V 0.95 0.47 0.35 0.24
Tox
Lecture 1
5
9
THE SIGNIFICANCE OF ETHE SIGNIFICANCE OF EXX AND EAND EYY
Ex < Ey = diode
Ex > Ey = transistor
• keeping Ex> Ey
• and avoiding the short- channel effect
NSUBx
VDD
TOX
XJLD
L
y
Well tempered means:
10 Pulfrey89
SHORTSHORT--CHANNEL EFFECT: CHANNEL EFFECT: VVTT depends on Ldepends on L
Charge under gate due to E from G, S and D
L′
L
ox
BT C
QV ∆=∆
• Geometrical constructionto estimate VT drop due to E encroachment
• Reduce junction depth
Lecture 1
6
11
Raised S and DRaised S and D
xj
Gargini02a
• Improves ION by 20- 30%
12
Current and Current and EEyy
1985 2003 200?
VDD,V 5 1 0.8
L, nm 3000 100 15
"Ey",mV/nm
1.6 10 53
factorbody 2
)( 2
=
−=
mmVV
CLZI TGS
oxDsat µ
1)(andLoftindependeni.e.
)(
TGS
TGSsatoxDsat
VVf
VVZCI
−
−= υ
Lecture 1
7
13
MOBILITYMOBILITY
• FETs don’t operate at high Ey all the time,
or over all of the channel.
• High mobility still very desirable to increase drive current
• Get high µ from strained-silicon channel
14
SiSi on on SiGeSiGe: Tensile strain: Tensile strain
IBM04
Lecture 1
8
15
Strained Strained SiSi: breaking the symmetry: breaking the symmetry
• 6 equivalent directions
• Intervalley scattering
• 2 sub- bands lowered in energy• Reduced intervalley scattering• Decreased effective mass (horizontal)
16
Strained Strained SiSi: Relaxed sub: Relaxed sub--layerslayers
IBM04
Lecture 1
9
17
18
Current and VCurrent and VTT
Decreasing (VDD - VT ) means a loss of gate overdrive
Morkoc04
• Ultimate VT ≈ 0.2 V
• Determined by ID,subt
10-2
10-1
100
101
101 102 103
VDD
VT
Channel length (nm)
Pow
er s
uppl
y (V
),Th
r esh
old
Volta
ge (V
)
ECE1
E0
Quantization important
Lecture 1
10
19
S
B
G
CB
ICOX
:
1
factorbodythem
mV
CC
VV GS
ox
B
GSSIB =
+=≡ψ
It’s done by capacitive control of the source- channel barrier height
Control of Control of IID,subtD,subt
Ec
⎟⎠⎞
⎜⎝⎛
mqVT
OFF
THRESHOLD
SOURCE DEPLETION
⎥⎦
⎤⎢⎣
⎡−==
t
TthreshDGSsubtD mV
VIVI exp)0( ,,
This sets lower limit to VT, e.g., 0.2Vi.e., ION/IOFF ≈104
20
KVmVmdV
IdS
t
GS
D
300at 060.0303.2
log1
10
==
⎟⎟⎠
⎞⎜⎜⎝
⎛=
−
Aox
B
ox
ox
s
ox
B
NsmallandtsmallNeed
Wt
CC
mcall
:
11:Re ⋅+=+=εε
It’s the VGS needed to reduce ID by 10X
SubSub--Threshold SlopeThreshold Slope
T reduce ∴=q
kTVt
VT compromiseGate leakage
Lecture 1
11
21
(b) FN tunneling(c) direct tunneling
de Broglie wavelength:KEm
hmvh
.2==λ
For an electron in Si at KE=φox/2 : λ = 6.1 nm
Electron could be either side of the barrier!
2.3nm
Taur98
ColdCold--Electron TunnelingElectron Tunneling
22
λπa
AA
Tinc
trans 4expy probabilit Tunneling 2
2 −≈=
= 0.0002 for 180nm= 0.0031 for 130nm= 0.0088 for 90nm
What is a tolerable gate current?
Taur98
Tunneling FactsTunneling Facts
Lecture 1
12
23
Ultimate SubUltimate Sub--Threshold CurrentThreshold Current
• S → D tunneling
<V t
Gate controlledbarrier
Source
Drain
Short channel,
moderate drain bias
direct
Short channel,larger drain bias
Gate controlledbarrier ?
Source
Drain
<Vt
Zener
• Expected to occur at L ≈ 10 nmMorkoc04
24 Chau03
Lecture 1
13
25 Chau03
26 Frank02
Power constrained scaling limitsPower constrained scaling limits
It is power dissipation, rather than scaling, that will be the limiting factor
e.g., can scale big servers more aggressively than portables and SRAMs
Lecture 1
14
27
HighHigh--k dielectricsk dielectrics
ox
oxox T
Cε
=• High TOX needed to reduce gate leakage• High COX needed for ID and S• Resolve conflict by increasing ε
Wong02
28
HighHigh--k dielectrics: tunnelingk dielectrics: tunneling
λπa
AA
Tinc
trans 4exp yprobabilit Tunneling 2
2 −≈=
∫ −−
=a
dxExVmh
T0
])([24expy probabilit Tunneling π
oxhigh a Need φ∴
Lecture 1
15
29
HighHigh--k dielectrics: contendersk dielectrics: contenders
Wong02
• must withstand
poly activation (950C)
• or use metal gate
Also:
30
Metal gate: selfMetal gate: self--alignmentalignment
Poly gates made self- alignment possible
• Perhaps use sacrificial poly gate,
• then deposit metal.
• Co- evaporation of metals (Ti and Ni)
to obtain different work functions,
• i.e., different VT’s for NMOS and PMOS
or for different blocks on same wafer.
Possibilities:
Lecture 1
16
31
Metal gate and NMetal gate and NSUBSUB
• If VT controlled by metal, perhaps can use undoped Si substrate.
• This would remove the problem of dopant fluctuations.
Gargini02
32
Beyond Planar CMOSBeyond Planar CMOS
Planar CMOS:Planar CMOS:• 10 nm prototypes demonstrated• raised source and drain• strained Si• high- k dielectric• metal gate• limitation is power dissipation
• Double gate CMOS• SOI CMOS
Further improvements:Further improvements:
Lecture 1
17
33
DoubleDouble--Gate CMOSGate CMOS
• SCE controlled by device geometry, not doping
• Can use undoped channel - reduces statistical fluctuations and Zener BD
• Increased Cox improves ION and SWong02
• Design flexibility - different VG’s and Tox’s
34
SCE: VSCE: VTT RollRoll--offoff
Philips04, MIT02
Note: benefit of
shrinking dSi
DIBL
Lecture 1
18
35
ox
B
CC
mq
kTmScall +== 1303.2:Re
DG: Improved ON/OFF ratioDG: Improved ON/OFF ratio
DG doubles without reducing Tox
Tends to zero (small dSi and inversion from top and bottom)
• For same IOFF, set VT 60mV lower, get more ION
36
DG example: FINFETDG example: FINFET
• DG is a deeply scalable FET, but fabrication is difficultWong02
Lecture 1
19
37
• More easily fabricated• Ultra thin body• No leakage through substrate• Very low Cj
• Good device isolation for RF• Technology of choice for SOC • Not as deeply scalable as DG
SOI CMOSSOI CMOS
Gargini02a
SIMOX
38
SOI: stateSOI: state--ofof--thethe--artart
• Small L, xj, dSi
Chau03
• Fully depleted• Raised S and D
Lecture 1
20
39
ConclusionConclusion
• Planar CMOS: 10nm - THz operation - millions of transistors
• DG CMOS: reduced SCE - best sub- threshold slope
- high performance digital
• SOI CMOS: reduced leakage and parasitic C - RF capable
• Do we, or our children, need anything more?
CMOS
40
ReferencesReferences
Chau03 - ftp://download.intel.com/research/silicon/Chau%20DRC%20062303%20foils.pdfDavid04 - ftp://download.intel.com/research/silicon/Ken_David_GSF_030604.pdfFrank02 - Frank D.J, IBM J. R&D, v46, 235, 2002Gargini02 - ftp://download.intel.com/research/silicon/PaoloISSUS0102.pdfGargini02a - ftp://download.intel.com/research/silicon/Paolo%20M2S2%200902.pdfIBM03 - http://www- 3.ibm.com/chips/services/foundry/offerings/sige/5hp/IBM04 http://www.research.ibm.com/resources/press/strainedsilicon/MIT02 http://ocw.mit.edu/NR/rdonlyres/Electrical-Engineering-and-Computer-Science/6-720JIntegrated-Microelectronic-DevicesFall2002/4E1C74EA-38FB-41CF-B654-3C5AD913B2E9/0/lecture33.pdMorkoc04 - Morkoc H., WOCSDICE, Slovakia, 2004Perlmutter04 - ftp://download.intel.com/research/silicon/Perlmutter053104.pdfPulfrey89 - Pulfrey D.L., N.G.Tarr, “Introduction to Microelectronic Devices”, Prentice-Hall, 1989Raghavan00 - Raghavan G. et al., IEEE Spectrum, v37(10), 47, 2000Rucker03 - Rucker R. et al., IEDM, paper 5.3, 2003Taur98 - Taur Y., T.H.Ning, “Fundamentals of Modern VLSI Devices”,Cambridge University Press, 1998Taur99 - Taur Y., IEEE Spectrum, 25, July 1999Wong02 - Wong H-S.P., IBM J. R&D, v46, 133, 2002Yeo02 - Yeo K-S. et al., “CMOS/BiCMOS VLSI”, Prentice-Hall, 2002