of 51
7/31/2019 TahirGhani Intel Talk
1/51
1
Tahir Ghani
Intel Fellow and Director,
Transistor Technology and Integration,Intel Corporation
Innovative Device Structures and New
Materials for Scaling Nano-CMOS
Logic Transistors to the Limit
7/31/2019 TahirGhani Intel Talk
2/51
2
Key Messages End of traditional dimensional scaling era
New and rapid innovations in transistor structure and
materials are now key to sustaining Moores Law:
Uniaxial strained silicon and HiK + Metal Gate
Power Limited Era: New Transistor Architectures areneeded to meet the performance improvements whilekeeping within power budget
Nanoscale Design Rule Regime: Dimensional scaling doesnot mean better transistor performance.
This is the most exciting time to be doing transistor researchand development
7/31/2019 TahirGhani Intel Talk
3/51
3
Outline End of Traditional Scaling EraTraditional scaling limiters and implications
Intels Response
Uniaxial Strain (90nm and 65nm Nodes)
HiK + Metal Gate + Strain (45nm Node)
Challenges and Solutions Beyond 45nm Node
Uniaxial Strain: Ultimate limit of silicon mobility enhancement
Power Limitation: Implications on future transistor structures
Parasitics Dominated Era: How to address increasingnegative impact of parasitics?
New Channel Materials: III-V QW FETs at Vcc~0.5-0.7V
Key requirements for implementing
III-V channels into mainstream?
7/31/2019 TahirGhani Intel Talk
4/51
4
Geometric Dimensional Scaling Era
Gate Oxide Thickness Scaling
- Key enabler for Lgate scaling
Junction Scaling- Another enabler for Lgate scaling
- Improved abruptness (REXT reduction)
Vcc Scaling- Reduce XDEP (improve SCE)
- However, did not follow const E field
1990s: Golden Era of Scaling
Dramatic Vcc, Tox & Lg scaling. Increasing Idsat
R. Dennard et.al.IEEE JSSC, 1974
7/31/2019 TahirGhani Intel Talk
5/51
5
Gate Leakage
Mobility Degradation
Parasitic Resistance
SOURCE DRAIN
GATE
Gate leakage
Mobility Degradation
Top Traditional Scaling Limiters
Top Scaling Challenges faced by Intels90nm CMOS Research Team in 2000
Parasitic Resistance
7/31/2019 TahirGhani Intel Talk
6/51
6
1.E-03
1.E-02
1.E-01
1.E+00
1.E+011.E+02
1.E+03
1.E+04
0.5 1.0 1.5 2.0 2.5
TOX Physical [nm]
JOX
[A/cm2
]
180nmNitrided SiO2
SiO2[Lo et. al, EDL97]
130nm
Jox limit
T. Ghani et. al. VLSI Symp. 2000
Mobility(cm
2/(V
.s)
100
150
200
250
300
0 0.5 1 1.5
E EFF [MV/cm]
NA=
3x1017
1.3x1018
1.8x1018
2.5x1018
3.3x1018
Universal
Mobility
SiO2
Scaling and Mobility Reduction Trend
Gate Oxide Leakage:
Direct tunneling limited.
Running out of Atoms
Significant mobility reduction
due to channel ionized
impurity scattering
7/31/2019 TahirGhani Intel Talk
7/51
7
Parasitic Resistance Impact
Salicide interface resistancebecoming a significant
component of REXTdue to salicide area scaling
S/D doping close to solidsolubility in Si (Nsurf)
Solutions:Solutions:
Barrier height reduction
Higher dopant activation(Exceeding solid solubility )
Gate Gate
Salicide
)4
exp(*
surf
Bc
N
m
qh
SilicideSilicide
SiSi
EEFF
EEGapGap
EEVV
EECCBB
7/31/2019 TahirGhani Intel Talk
8/51
8
Outline End of Traditional Scaling EraTraditional scaling limiters and implications
Intels Response
Uniaxial Strain (90nm and 65nm Nodes)
HiK + Metal Gate + Strain (45nm Node)
Challenges and Solutions Beyond 45nm Node
Higher Strain: Ultimate limit of silicon mobility enhancement?
Power Limitation: Implications on future transistor structures
Parasitics Dominated Era: How to address increasingnegative impact of parasitics?
New Channel Materials: III-V QW channels at Vcc~0.5-0.7V
7/31/2019 TahirGhani Intel Talk
9/51
9
Innovations Introduced by Intel toOvercome Traditional Scaling Barriers
Uniaxial process induced strain innovations for dramatic
mobility enhancement starting at 90nm CMOS node
- Epitaxial SiGe S/D
- SiN Capping Layers
HiK gate insulator being introduced at 45nm CMOS
node to replace SiO2 to help address gate leakage
Metal Gate being introduced at 45nm CMOS node to
replace poly-silicon gate to enable Tox(e) scaling
7/31/2019 TahirGhani Intel Talk
10/51
10
Why is Low Field Mobility Important
for Nanoscale Transistors?M.M. LundstromLundstrom et. al., EDL 1997et. al., EDL 1997
Conventional theory assumes infiniteConventional theory assumes infinite
supply of carriers at the sourcesupply of carriers at the source
ss(0) and I(0) and IDSATDSAT limited by lower oflimited by lower ofthe two velocity termthe two velocity term UUltimatelyltimately
limited by thermal injection fromlimited by thermal injection from
source to channel (ballistic)source to channel (ballistic)
Best devices in production todayBest devices in production todayare ~ 60% ballisticare ~ 60% ballistic
Equal contributions by ballisticEqual contributions by ballisticand mobility termsand mobility terms
Low field mobility importantLow field mobility importanttoto nanonano--MOS transportMOS transport
)(0
11
(0)s
1
(0))]V(V[CWI STGSGDSAT
++=
=
Eeffinj
EcEc (x)(x)
EcoEco
Position (x)Position (x)Energy
Energy
llllllll
Source-Barrier
ttccrrcc
7/31/2019 TahirGhani Intel Talk
11/51
11
Uniaxial Strain Silicon Transistors
Intel: IEDM 2003PMOS NMOS
SiGeSiGe
These transistor structures introduced first at Intels 90nm
CMOS node. These structures have now become
industry standard for strain implementation
T. Ghani et. al. IEDM, 2003
7/31/2019 TahirGhani Intel Talk
12/51
12
PMOS Strain Implementation SiGe epitaxial S/D
Formed by Si recess etch
and selective StrainedSiGe epi growth
Strained SiGe induceslarge lateral compression
in channel Valence bands warpage
and LH-HH splitting
Dramatic mobility gain
SiGe S/D also improvesparasitic resistance by reducing
salicide interface resistanceLateral compression in channelLateral compression in channel
SiGeSiGe
UniaxiallyUniaxially StrainedStrained SiGeSiGe EpiEpi S/DS/D
T. Ghani et. al. IEDM, 2003
7/31/2019 TahirGhani Intel Talk
13/51
13
Strained SiGe S/D Reduces
Salicide Interface Resistance Strained SiGe has smaller Eg Smaller hole barrier height
at silicide interface
Exponential reduction ofinterface resistance on B
Higher boron activation inSiGe relative to Si ( Nsurf)
Dramatic reduction in sal interface
resistance with strained SiGe S/D
1.E-09
1.E-08
1.E-07
1.E-06
1.E-05
1.E+19 1.E+20 1.E+21
Doping Concentration (cm-3)
Resistivity(ohm-cm2) BB = 0.5eV= 0.5eV
BB = 0.4eV= 0.4eV
BB = 0.3eV= 0.3eV
BB = 0.2eV= 0.2eV
SiGe
B
SiGe (Nsurf)
EEFF
BB
SilicideSilicide Si orSi orSiGeSiGe EECC
EEVV SiSi
EEVV SiGeSiGeEEVV
)4
exp(*
surf
Bc
N
m
qh
7/31/2019 TahirGhani Intel Talk
14/51
14
Uniaxial Strain Performance Gain (Intel)
65nm CMOS Node
00
IdlinIdlin
PMOSPMOS
2020
4040
6060
8080
DriveGain(%)
DriveG
ain(%)
00
55
1010
1515
2020
25253030
Drive
Gain(%)
Drive
Gain(%) IdsatIdsat
IdlinIdlin
NMOSNMOS
IdsatIdsat
Ref: Unstrained Silicon
Uniaxial strain has demonstrated dramatic PMOS and NMOS
performance improvement on 90nm & 65nm CMOS nodes
2.2x Mobility 1.5x Mobility
7/31/2019 TahirGhani Intel Talk
15/51
15
Innovations Introduced by Intel toOvercome Traditional Scaling Barriers
Uniaxial process induced strain innovations for dramatic
mobility enhancement starting at 90nm CMOS node
- Epitaxial SiGe S/D
- SiN Capping Layers
HiK gate insulator being introduced at 45nm CMOSnode to reduce gate leakage
Metal Gate being introduced at 45nm CMOS node toreplace poly-silicon gate to eliminate poly depletion:Scale Tox(e)
7/31/2019 TahirGhani Intel Talk
16/51
16
IEEE Spectrum October, 1969
28
Thermal Oxidation and Poly Silicon Gate:
KEY TO MICROELECTRONIC REVOLUTION
SiO2
Growth Technology: Enabled MOS transistor to become a reality
Poly Silicon Gate: Key to Self Alignment Device Scaling
Poly /SiO2 gate stack was the foundation on which
IT revolution has been built. Served well for 40y BUT
7/31/2019 TahirGhani Intel Talk
17/51
17
1E1E--66
1E1E--44
1E1E--22
1E+01E+0
1E+21E+2
1E+41E+4
55 1010 1515 2020 2525
EOT (A)EOT (A)
JJ
oxox
(A/cm
(A/cm22)) SiOSiO22 dielectricdielectric
HiHi--KK
dielectricdielectric
BENEFIT:BENEFIT:
Significant gate leakageSignificant gate leakage
reduction at a given EOTreduction at a given EOT
SiO2SiO2
1.0x1.0x
1.0x1.0x
HighHigh--kk
1.6x1.6x
< 0.01x< 0.01x
Capacitance:Capacitance:
LeakageLeakage::
Gate Leakage Reduction withGate Leakage Reduction with HiKHiK
SiliconSiliconsubstratesubstrate
GateGate
electrodeelectrode
HighHigh--kk
SiliconSilicon
substratesubstrate
1.2nm SiO1.2nm SiO22
Poly Si GatePoly Si Gate
electrodeelectrode
Intel 65nm NodeIntel 65nm Node HiKHiK Gate DielectricGate Dielectric
M. Radosavljevic et. al., Intel Corp. DARPA CMOSNano, 01/12/04
7/31/2019 TahirGhani Intel Talk
18/51
18
Metal Gate Eliminates Poly Depletion
-2.0 -1.0 0 2.01.0
Vgs (V)
Cgate
Change with Tox
True Tox change
Capacitance benefit everywhere
Both Idsat and SCE improve
Inversion
MG eliminates poly dep (inversion)
Increases gate E-field
Larger QinvHigh Idsat
Tox (inv) scales significantly
BUT! Tox(e) does not impact SCE
-2.0 -1.0 0 2.01.0
Vgs (V)
Cgate
Change with MG
Inversion
Poly depletion
improvement
No Ioff
benefit
7/31/2019 TahirGhani Intel Talk
19/51
19
Transistor Drive Current (rel.)
>25%Higher Drive
65 nm 45 nm
10
100
1000
0.5 1.0 1.5
LeakageCurrent
(nA/um)
Source: Intel Internal
High-k+Metal Gate
Performance / Power BenefitsTransistor Performance vs. S/D LeakageTransistor Performance vs. S/D LeakageTransistor Performance vs. S/D LeakageTransistor Performance vs. S/D Leakage
>25% Idsat gain demonstrated for
45nm CMOS vs. 65nm CMOS at fixed Ioff
7/31/2019 TahirGhani Intel Talk
20/51
20
Extensive R&D done at IntelExtensive R&D done at Intel
to successfully address theto successfully address the
significant Material,significant Material,Integration andIntegration and
Manufacturing challengesManufacturing challenges
in implementingin implementing HiKHiK + Metal+ Metal
GateGate CMOS Technology.CMOS Technology.
Right Metal Gate MS electrodeswhich are HiK compatible
Bulk & interface traps:Poor reliability (Need better thanSiO2 reliabilty due to higher E)
New scattering modes:Poor mobility
Technology Integration
Yield / Manufacturability
Top Issues with Hi-k + Metal Gate
CMOS Technology
7/31/2019 TahirGhani Intel Talk
21/51
21
HiK Mobility Challenge
Metal Gate recovers ~50% of the
degradation. Further stack optimization
is required for mobility improvement
Model: High-k dipoles vibrate !!
Mobility degradation due to
to scattering with soft optical
vibrational modes of dielectric
Very high charge density of MG
screens dipole vibrations.
NMOS mobility with high-k
degrades ~ 40% from
SiO2 / poly stack
ELO
Gate
High-k dielectric
Si channel
e-
+ -
-+Eg
ELOEg
ETOT=ELO+ Eg
M. Fischetti et. al. J App Phys, 2001
R. Kotlyar et. al. (Intel)
IEDM 2004
7/31/2019 TahirGhani Intel Talk
22/51
22
Metal-Hf based Oxide system susceptibleto oxygen vacancy sites Efficient
electron traps located in upper half of
HfO2
bandgap: Well documentedin literature
These traps are responsible for NMOShysteresis, BT and TDDB
Key to reliability is passivating Vo sites
HiK/Metal Gate intrinsic reliabilityrequirement more stringent than best
SiO2 because they need to withstandhigher E-Field
K. Torii, IEDM, 2004
J. Mitard, IRPS, 2006
Effective Solutions to Bias-Temp
and TDDB are Key to
HiK + Metal Gate Implementation
High-K Reliability Challenge
7/31/2019 TahirGhani Intel Talk
23/51
23
HiK - Dual Metal Gate Integration
Gate FirstGate First Gate LastGate Last
- Stability Right MS N & P metalswhich survive high thermal anneal
- Dual metal gate stack patterning
+ Standard process flow
+ Metals deposited afterhigh Dt More MG options
- Non-std process flow
Si
Poly
HiHi--KK
MetalMetal
NN PP
7/31/2019 TahirGhani Intel Talk
24/51
24
Intels 45nm Node HiK/MG Transistors
Low Resistance
Layer
Integrated 45 nm
CMOS process
High performance
Low leakage
Meets reliability
requirements
Manufacturable
in high volume
High-k DielectricHafnium based
Silicon Substrate
Work Function MetalDifferent for NMOS and PMOS
45nm HiK + Metal Gate CMOS technology meetsmeets performance, yield and reliability goals
The implementation of highThe implementation of high--k and metal gate materials marks the biggest changek and metal gate materials marks the biggest change
in transistor technology since the introduction of polysilicon gin transistor technology since the introduction of polysilicon gate MOS transistorsate MOS transistors
in the late 1960s.in the late 1960s. Gordon MooreGordon Moore
7/31/2019 TahirGhani Intel Talk
25/51
25
Worlds First Working 45 nm CPUwith HiK + Metal Gate
Worlds first working 45 nm CPU
Intel Penryn: 45nm CPU
Jan 2007
45nm SRAM Test Vehicle
Jan 2006
> 1 Billion Transistors
45nm SRAM Test Vehicle has >1B transistors
On track to ship 45nm CPUs in 2007
7/31/2019 TahirGhani Intel Talk
26/51
26
Outline End of Traditional Scaling EraTraditional scaling limiters and implications
Intels ResponseUniaxial Strain (90nm and 65nm Nodes)
HiK + Metal Gate + Strain (45nm Node)
Challenges and Solutions Beyond 45nm Node
Higher Strain: Ultimate limit of silicon mobility enhancement?
Power Limitation: Implications on future transistor structures
Parasitics Dominated Era: How to address increasingnegative impact of parasitics?
New Channel Materials: III-V QW channels at Vcc~0.5-0.7V
7/31/2019 TahirGhani Intel Talk
27/51
27
How Far Can Uniaxial Strain
Extend Si Performance Gains?
Significant headroomSignificant headroom leftleftto increase PMOS mobilityto increase PMOS mobility
in future (> 5x)in future (> 5x)
Mobility gain driven by hole meff
reduction due to band warpage !
Limited Max Mobility GainLimited Max Mobility Gain
for NMOS (~ 2x).for NMOS (~ 2x).
Maximum gain limited by
fundamental physics
0
1
2
3
4
5
6
0 1000 2000 3000
Channel Stress (Mpa)
MobilityEnha
ncem
entRati
PMOS
NMOS
Source: Intel(100)/
Channel=Si
EEFF
=1MV/cm
Today
7/31/2019 TahirGhani Intel Talk
28/51
28
Implications of Significantly Higher
PMOS Mobility Enhancement
Expect NMOS and PMOS
mobility values toapproach each other
PMOS device drive strengthto approach NMOS in future
N/P ~ 1 N/P Symmetry
Device sizing in circuitsDevice usage model
1.0
1.5
2.0
2.5
3.0
0 1 2 3 4 5 6 7
Technology Node (nm)
N/PIdsatratio
250 180 130 90 65
Source: Intel
?
7/31/2019 TahirGhani Intel Talk
29/51
29
Outline End of Traditional Scaling EraTraditional scaling limiters and implications
Intels ResponseUniaxial Strain (90nm and 65nm Nodes)
HiK + Metal Gate + Strain (45nm Node)
Challenges and Solutions Beyond 45nm Node
Higher Strain: Ultimate limit of silicon mobility enhancement?
Power Limitation: Implications on future transistor structures
Parasitics Dominated Era: How to address increasingnegative impact of parasitics?
New Channel Materials: III-V QW channels at Vcc~0.5-0.7V
7/31/2019 TahirGhani Intel Talk
30/51
30
CPU Transistor Count Trend
2x Transistors Every 2 Years
In Line with Moores Law
1.E+03
1.E+04
1.E+05
1.E+06
1.E+07
1.E+08
1.E+09
1.E+10
1970 1980 1990 2000 2010
1.E+03
1.E+04
1.E+05
1.E+06
1.E+07
1.E+08
1.E+09
1.E+10
40048008
8080
8086
286 386
TM
486TM
Pentium CPU
Pentium II CPU
Pentium III CPU
Pentium 4 CPU
Itanium 2 CPU
2x increase every2 years
Penryn QC
7/31/2019 TahirGhani Intel Talk
31/51
31
Negative Consequence of
CPU Transistor Count Trend
Right Hand Turn: Power Dissipation Limited to ~100W
BUT increased transistor count neededBUT increased transistor count neededin Multiin Multi--Core CPU Era !!!Core CPU Era !!!
CPU Power (W)
10
100
1000
1990 1995 2000 2005 2010 2015
CPU Power (W)
10
100
1000
1990 1995 2000 2005 2010 2015
7/31/2019 TahirGhani Intel Talk
32/51
32
Multi-Core CPU Power Limited EraP = Switching Power + Leakage Power + ..
~ (fCgateVCC2 ) * N
RelativeV
RelativeV
CCCC
Relative Transistor CountRelative Transistor Count1x1x 1.5x1.5x 2x2x
Fixed PowerFixed Power
1x1x
VVCCCC scaling required for continued increase in transistorscaling required for continued increase in transistorcount in power limited worldcount in power limited world
Key Issue withKey Issue with VccVcc Scaling: Performance loss !!!Scaling: Performance loss !!!
How to maintain high performance at scaled VHow to maintain high performance at scaled VCCCC??
Constant
7/31/2019 TahirGhani Intel Talk
33/51
33
Multi-Gate Transistor ArchitectureV=0
V=0
V=1V=0 WsiDrain
Lgate
Gate
Gate
Source
Wsi=20nm
Wsi= 10nm
Si
AqN
yx
=
+
2
2
2
2
M. Stettler, 2006 SINANO Device Modeling
MultiMulti--Gate Transistors have better SCE:Gate Transistors have better SCE:
+ Gates in close proximity+ Gates in close proximity
reduce spread ofreduce spread ofVVdraindrain+ Small+ Small WWSiSi desired to minimize SCEdesired to minimize SCE
+ At very thin+ At very thin WWSiSi, channel, channel
potential impervious topotential impervious to dopantsdopants
MutigateMutigate transistors have highertransistors have higher
mobility due to:mobility due to:
+ Lower channel doping+ Lower channel doping+ Lower+ Lower EEeffeffin channelin channel
7/31/2019 TahirGhani Intel Talk
34/51
34
Multi-Gate Transistors Enable Vcc Reduction
MUGFET
Planar Si
NMOSR. Chau et al., ICSICT 2004
DG (Midgap gate)
D. Antoniadis, NIST Workshop, 2001Leakage Power Improvement
at Iso Performance
Inverter
Voltage
%I
mpro
vement
Larger improvementat lower Vcc
Simulation: Intel
Multi-Gate Transistors show superior
DIBL & Mobility: Lower Vt at a given Ioff:
Better gate overdrive vs Vcc.
Higher mobility vs Planar:
More so at lower lower Vcc
Power-performance tradeoff scalesbetter (vs Planar) as Vcc is reduced
7/31/2019 TahirGhani Intel Talk
35/51
35
Multi-Gate Transistors:
Implementation and Design
FinFET / Tri-Gate Transistor:
++ Self Aligned structure-- Non-Planar structure
Tri-Gate Transistor
7/31/2019 TahirGhani Intel Talk
36/51
36
Tri-Gate / FinFET
Value PropositionPerformance / Power:
+ Scale better at lower Vcc: Reduce Active Power OR
Better mobility & lower Vt+ Operate at lower Leakage Reduce Standby Power
+ Lower Channel Doping Lower BTBT: Lower IJUNCTIONLower Cjp: Performance gain
Reduce Standby Power
Random Dopant Fluctuation:
+ Lower Channel Doping Lower RDF. Better SRAM Vmin?
Multi-Gate Transistor is a serious contenderfor post-45nm CMOS nodes due to its
many fundamental advantages
7/31/2019 TahirGhani Intel Talk
37/51
37
Process Challenges in Fabricating
Tri-Gate / FinFET Transistors
J. Kavalieros et. al. (Intel)
VLSI Symp 2006
A. Dixit, K. Anil et al.,
Solid State
Electronics, 2006
NonNon--PlanarityPlanarity
Implementing high levelImplementing high levelof channel strain:of channel strain:
-- Planar Ref= Highly strainedPlanar Ref= Highly strained
and optimized deviceand optimized device
HigherHigherRextRext::-- SelectiveSelective epiepi S/DS/D-- Minimize spacerMinimize spacer
Process control:Process control:
-- Fin width controlFin width control-- Poly sidewall profilesPoly sidewall profiles
These concerns need to be successfullyThese concerns need to be successfully
addressed foraddressed for TriGate/FinFETTriGate/FinFET TransistorsTransistors
to become mainstream.to become mainstream.
7/31/2019 TahirGhani Intel Talk
38/51
38
Outline End of Traditional Scaling EraTraditional scaling limiters and implications
Intels ResponseUniaxial Strain (90nm and 65nm Nodes)
HiK + Metal Gate + Strain (45nm Node)
Challenges and Solutions Beyond 45nm Node
Higher Strain: Ultimate limit of silicon mobility enhancement?
Power Limitation: Implications on future transistor structures
Parasitics Dominated Era: How to address increasingnegative impact of parasitics?
New Channel Materials: III-V QW channels at Vcc~0.5-0.7V
7/31/2019 TahirGhani Intel Talk
39/51
39
Density Scaling on Track (Gate Pitch)
0.1
1
1997 1999 2001 2003 2005 2007 2009
Contacted
Gate Pitch
(micron)
Source: Intel
Pitch
0.7x scaling
every 2 years
Gate pitch scaling continues to follow Moores Law
showing 2x transistor increase per area every 2 years
D i C t D d ti
7/31/2019 TahirGhani Intel Talk
40/51
40
GateGate
SpacingSpacing
SS DD
Shared Source DrainShared Source Drain
IOFF = Fixed
Beyond 45nm node, gate pitch scaling dramatically drive currentdramatically due to increased resistance (shrinking S/D contact area) Dramatic performance gains expected if salicide interface resistance
can be reduced.
Past: Yield vs. density tradeoffFuture: Transistor performance vs. density trade-off (NEW PARADIGM)
Drive Current Degradation
with Gate Pitch Scaling
0.60
0.80
1.00
1.20
1.40
1.60
240nm 180nm 120nm 90nm 65nm 40nm 25nm 25nm,
Rsal-
Int~0Gate-to-Gate Spacing
Idlin (normalized)
Idsat (normalized)
45nmNode
90nmNode
Shared Source DrainSpacer=10nm
7/31/2019 TahirGhani Intel Talk
41/51
41
Innovative Solutions for
Salicide Resistance Reduction
1 Increase S/D dopant electrical activation
above solid solubility:
Non-Equilibrium regime
2 S/D bandgap engineering to reduce barrier
height: Example: Strained SiGe S/D
3 Explore new salicides with reducedbarrier height: Dual Salicide
2 + 3 Key Challenge:
Interface states dominate band
alignment (Fermi level pinning).
Need to develop effective interface
passivation techniques
)4
exp(*
surf
Bc
N
m
qh
Silicide
Source / Drain
SilicideSilicide
SiSi
EEFF
EEGapGap
EEVV
EECC
BB
F i L l Pi i
7/31/2019 TahirGhani Intel Talk
42/51
42
Fermi Level Pinning
Yeo, King and Hu, JAP, 15 Dec 2002 )()1()( CNLCsMn EESS +=
)( sMn =
)( CNLCn EE =
Ideal
Pinned
Fermi Level Pinning: Barrier height insensitive to metal work function
Suppress Fermi level pinning by passivating dangling bondsEnables dual-metal work function materials with M near Ec and Ev
OR Effective barrier pinned close to desired level (Ev or Ec)
)()1()( CNLCsMn EESS +=
ECNL
n
)( sMn =
)( CNLCn EE =
Ideal
Pinned
Ec
Ev
7/31/2019 TahirGhani Intel Talk
43/51
43
Outline End of Traditional Scaling EraTraditional scaling limiters and implications
Intels ResponseUniaxial Strain (90nm and 65nm Nodes)
HiK + Metal Gate + Strain (45nm Node)
Challenges and Solutions Beyond 45nm Node Higher Strain: Ultimate limit of silicon mobility enhancement?
Power Limitation: Implications on future transistor structures
Parasitics Dominated Era: How to address increasingnegative impact of parasitics?
New Channel Materials: III-V QW channels at Vcc~0.5-0.7V
7/31/2019 TahirGhani Intel Talk
44/51
44
High Mobility n-Channel Materials
Source: A. Pethe (Stanford)
7/31/2019 TahirGhani Intel Talk
45/51
45
Ultimate Channel: Ballistic Transport
CCGG
Quantum Capacitance
important at thin TOX
)(0
11
(0)s
1
)V(V(0)WCI TGSSGDSAT
++=
=
Eeffinj
*
TGSGDSAT
2kT
)V(VWCI
t
inj
inj
m
=
=
EcEc (x)(x)
EcoEco
Position (x)Position (x)
Energy
Energy
z
0
VVGG
TTOXOX
NNINVINV z
0
VVGG
TTOXOX
NNINVINV
~t
*
1111
minj
Ballistic
2
*2
~h
mqC DOSINV
High Performance in Ballistic Regime:
1. Low mt along channel direction High inj Maximize IDSAT
2. High mDOS High CGATE High QINV Maximize IDSAT
**
**
eff
High Mobility III-V Channel= High Performance?
7/31/2019 TahirGhani Intel Talk
46/51
46
High Mobility III-V Channel= High Performance?
Weaklyquantized
Stronglyquantized
III-V materials (GaAs, InSb, InAs) beinginvestigated due to small -valley m* inj IDSAT
However, lower m* leads low DOS QINV IDSAT
-valley lifts up due to confinement (1/m*)Charge transfers into X & L valleys with high m*
inj IDSAT
Small EG (InAs, InSb): High BTBT leakage
Tailor bandgap by QW confinement
Higher:::: Higher sub-T slope (poor SCE)
Projecting III-V NMOS performance based
on simplistic models could lead to erroneous
performance assessment. Need detailed
physics modeling + fabricate devices
K. Saraswat,
INMP 2005
Requirements for Building a Competitive
7/31/2019 TahirGhani Intel Talk
47/51
47
Requirements for Building a Competitive
III-V Channel Transistor Technology(VCC~ 0.5-0.7V)
Integrate III-V layers on large Silicon wafers
Develop HiK dielectric compatible with III-V channels
Determining PMOS material to go with NMOS
Insertion 15nm node or beyond. Meet LG< 20nm.III-V devices may need to be Tri-Gate / FinFET structure.
It is expected to be scalable beyond first node.
III-V channel materials will have to simultaneously meetmultiple requirements to be serious contenders as
replacement for Strained-Si channel transistors
Transistor Feature Set Mapping
7/31/2019 TahirGhani Intel Talk
48/51
48
Transistor Feature Set Mapping
to CMOS Nodes: Potential Roadmap
Process inducedstrain+ (2nd gen)
Gate Oxide withpoly-Si Gate
NiSi
Hik + MG (Intel)
Process inducedstrain ++
NiSi
HiK + MG(Intel: 2nd Gen.)
Process inducedstrain +++
NiSi
Alternative waferorientation?
Dopant super-activation?
HiK + MG (3rd
gen) Process induced
strain ++++
Alternative waferorientation?
Dopant super-activation?
Multi-Gate FETswith strained Si?
Next generation
silicide /contacts?
65nm Node65nm Node 45nm Node45nm Node 32nm Node32nm Node 22/15nm Nodes22/15nm NodesTODAYTODAY
Challenging but feasible roadmap for scaling
logic CMOS technology down to 15nm
CMOS Node with Si Channel.
7/31/2019 TahirGhani Intel Talk
49/51
S mmar / Ke Message
7/31/2019 TahirGhani Intel Talk
50/51
50
Summary / Key Message Transistor structure and material innovations pioneered at
Intel such as uniaxial strained silicon and high-k/metal gate
have enabled Intel to scale planar CMOS beyond 90nm node.
Achieving high performance at low Vcc is critical in a powerlimited world and will play important role in transistor
architecture and front-end feature set selection.
Multi-Gate transistors have potential to improve performancevs. power tradeoff and enable lower Vcc on products
Improving transistor parasitics is as important as improving
intrinsic transistor performance. Needs higher focus!!
Roadmap for scaling CMOS technology during next 10 yearsis quite challenging but feasible
7/31/2019 TahirGhani Intel Talk
51/51
51
THANK YOU!