cmos beyond si: nanometer-scale iii-v mosfets slides.pdfintel microprocessors. intel...
Post on 26-Sep-2020
5 Views
Preview:
TRANSCRIPT
CMOS beyond Si:Nanometer-Scale III-V MOSFETs
J. A. del Alamo, X. Cai, J. Lin, W. Lu, A. Vardi, and X. ZhaoMicrosystems Technology Laboratories
Massachusetts Institute of Technology
IEEE Bipolar/BiCMOS Circuits and Technology MeetingMiami, FL, October 19-21, 2017
Acknowledgements:• Students and collaborators: D. Antoniadis, E. Fitzgerald• Sponsors: Applied Materials, DTRA, KIST, Lam Research, Northrop Grumman,
NSF, Samsung, SRC• Labs at MIT: MTL, EBL
Moore’s Law at 50: the end in sight?
2
3
Moore’s Law
4
?
How far can Si support Moore’s Law?
The problem: Transistor scaling Voltage scaling Performance suffers
5
Transistor current density:
What can we do about this?
Intel microprocessorsIntel microprocessors
Supply voltage:
6
Moore’s Law: it’s all about MOSFET scaling
1. New device structures with improved scalability:
2. New materials with improved transport characteristics:
n-channel: Si Strained Si SiGe InGaAsp-channel: Si Strained Si SiGe Ge InGaSb
Planar bulk MOSFET Thin-body SOI MOSFET
Nanowire MOSFETFinFET
7
III-V electronics in your pocket!
8
Historical evolution: InGaAs High-Electron Mobility Transistor
Transconductance (gm=dID/dVGS):
9
InGaAs MOSFETs vs. HEMTs
*
*inversion-mode
Transconductance (gm=dID/dVGS):
10
InGaAs MOSFETs vs. HEMTs
Lin, EDL 2016
What happened here?
*
*inversion-mode gm=3.45 mS/μm
Transconductance (gm=dID/dVGS):
11
Huang, APL 2005
ALD eliminates residual native oxides that pin Fermi level “Self cleaning”
Clean, smooth interface without native oxides
Atomic Layer Deposition (ALD)of gate oxide
• First with Al2O3, then with other high-K dielectrics• First in GaAs, then in other III-Vs
n-MOSFETs in Intel’s nodes at nominal voltage
Planar Si and InGaAs MOSFET Benchmark
12
Comparisons always fraught with danger…
Transconductance exceeds Si MOSFETs
InGaAs FinFETs
13
FinFETs
Bottom-up InGaAs FinFETs
14
Si
Waldron, VLSI Tech 2014
Aspect-Ratio TrappingFiorenza, ECST 2010
Epi-grownfin insidetrench
Sun, VLSI Tech 2017
InGaAs FinFETs @ MIT
15
Vardi, DRC 2014, EDL 2015, IEDM 2015
Key enabling technologies: BCl3/SiCl4/Ar RIE + digital etch
• Sub-10 nm fin width• Aspect ratio > 20• Vertical sidewalls
InGaAs FinFETs @ MIT
16
Vardi, VLSI Tech 2016
• Si-compatible process• Contact-first, gate-last process• Fin etch mask left in place double-gate MOSFET
Vardi, EDL 2016
InAlAs
InGaAs
n+-InGaAs
W/Mo Lg
SiO2
HSQHigh-K
InP
δ - Si
InP
Mo Mo
HSQ
High-K
InGaAs
17
Most aggressively scaled FinFETWf=7 nm, Lg=30 nm, Hc=40 nm (AR=5.7), EOT=0.6 nm:
Vardi, EDL 2016
At VDS=0.5 V:• gm=900 µS/µm• Ron=320 Ω.µm• Ssat=100 mV/dec -0.4 -0.2 0.0 0.2 0.4
0
200
400
600
800
1000
VDS=0.5 V
gm max=900 µS/µm
g m [µ
S/µm
]
VGS [V]
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.41E-9
1E-8
1E-7
1E-6
1E-5
1E-4
1E-3
VDS=50 mVDIBL=90 mV/V
I d [A/
µm]
VGS [V]
Ssat=100 mV/dev
VDS=500 mV
0.0 0.1 0.2 0.3 0.4 0.50
100
200
300
400
500
I d [µA
/µm
]
VDS [V]
VGS=-0.5 to 0.75∆VGS=0.25 V
Current normalized by 2xHc
18
0 100 200 300 400 500 600 7000
50
100
150
200
250 A: Al2O3, EOT=2.8 nm B:Al2O3/HfO2, EOT=1 nm C: HfO2, EOT=0.6 nm
S sat [
mV/
dec]
Lg [nm]
EOT↓60 mV/dec
0 100 200 300 400 500 600 700
0
200
400
600
800
1000
1200
1400
1600
EOT↓
VDS=0.5 VWf=20-22 nm
g m [µ
S/µm
]
Lg [nm]
Classical scaling with Lg and EOT
Lg and EOT scaling (Wf~20 nm)
Vardi, EDL 2016
100 10000
50
100
150
Wf=7 nm Wf=12 nm Wf=17 nm Wf=22 nm
S sat,m
in [m
V/de
c]Lg [nm]
60 mV/dec
Fin-width scaling (EOT=0.6 nm)
19
• Non-ideal fin width scaling• Sources:
• High Dit (~5x1012 cm-2.eV-1)?• mobility degradation? • line edge roughness?
Contaminated bygate leakage
0 100 200 300 400 500 600
0
200
400
600
800
1000
1200
1400
1600
7
∆Wf= 5 nm
g m m
ax [µ
S/µm
]
Lg [nm]
Wf=22 nm
Vardi, EDL 2016
0 20 40 600
5
10
15
20
0.452.2
0.321
1.8
0.570.8
5.73.3
2.3
1.8InGaAs FinFETs
5.3
4.3
IntelSi FinFETs (VDD=0.8 V)
g m/W
f [m
S/µm
]
Wf [nm]
0.630.6
0.230.66 1 0.18
InGaAs FinFET benchmarking
20
gm normalizedby fin width
• First InGaAs FinFETs with Wf<10 nm• Doubled gm over earlier InGaAs FinFETs• Short of Si FinFETs sidewall quality?
Our work (VDS=0.5 V)Wf
Target
channelaspect ratio
See latest at IEDM 2017!
21
InGaAs Vertical Nanowire MOSFETs
VNW MOSFET
22
Vertical nanowire MOSFET: ultimate scalable transistor
Vertical NW MOSFET: uncouples footprint scaling from Lg, Lspacer, and Lc scaling
Lc
Lg
Lspacer
Key enabling technologies:• RIE = BCl3/SiCl4/Ar chemistry• Digital Etch (DE) = self-limiting O2 plasma oxidation + H2SO4 or HCl oxide removal
23
• Radial etch rate=1 nm/cycle• Sub-20 nm NW diameter• Aspect ratio > 10• Smooth sidewalls
RIE + 5 cycles DE
Zhao, IEDM 2013Zhao, EDL 2014Zhao, IEDM 2014
InGaAs Vertical Nanowires @ MIT
24
InGaAs VNW-MOSFETsby top-down approach @ MIT
Top-down approach: flexible and manufacturable
n+ InGaAs, 70 nm
i InGaAs, 80 nm
n+ InGaAs, 300 nm
Starting heterostructure:
n+: 6×1019 cm-3 Si doping
VNW-MOSFET I-V characteristics: D=40 nm
25
Single nanowire MOSFET: • Lch= 80 nm• 3 nm Al2O3 (EOT = 1.5 nm)
-0.2 0.0 0.2 0.4 0.610-10
10-9
10-8
10-7
10-6
10-5
10-4
10-3 Vds=0.5 V
Vgs(V)
I s (A
/µm
) Vds=0.05 V
0.0 0.1 0.2 0.3 0.4 0.50
50
100
150
200
250
300 Vgs=-0.2 V to 0.7 V in 0.1 V step
Vds (V)
I s (µ
A/µm
)
-0.2 0.0 0.2 0.4 0.60
100200300400500600700800
Vgs(V)
g m (µ
S/µm
)
Vd= 0.5 V
gm,pk=720 μS/μm
Slin = 70 mV/decSsat = 80 mV/decDIBL = 88 mV/V
Zhao, CSW 2017
Benchmark with Si/Ge VNW MOSFETs
26
• InGaAs competitive with Si• Need to demonstrate VNW MOSFETs with D<10 nm
0 20 40 60 80 1000
200
400
600
800
1000
1200
1400
1V1 V
1
1 V
1.2 V
1.2 Vg m,p
k (µS
/µm
)
Si/Ge InGaAs
Diameter (nm)
Peak gm of InGaAs (VDS=0.5 V), Si and Ge VNW MOSFETs
Target
Our work (VDS=0.5 V)
InGaAs VNW Mechanical Stability for D<10 nm
27
Yield = 0%
Broken NW
Difficult to reach 10 nm VNW diameter due to breakage
8 nm InGaAs VNWs after 7 DE cycles:
InGaAs VNW Mechanical Stability for D<10 nm
28
Yield = 0%
Broken NW
Difficult to reach 10 nm VNW diameter due to breakage
Water-based acid is problem:
Surface tension (mN/m):• Water: 72 • Methanol: 22 • IPA: 23
Solution: alcohol-based digital etch
Alcohol-Based Digital Etch
29
10% HCl in IPAYield = 97%
10% HCl in DI waterYield = 0%
Alcohol-based DE enables D < 10 nm
Broken NW
Radial etch rate: 1.0 nm/cycle Radial etch rate: 1.0 nm/cycle
8 nm InGaAs VNWs after 7 DE cycles: Lu, EDL 2017
D=5.5 nm VNW arrays
30
90% yield10% H2SO4 in methanol
• H2SO4:methanol yields 90% at D=5.5 nm!• Viscosity matters: methanol (0.54 cP) vs. IPA (2.0 cP)
InGaAs Digital Etch
31
First demonstration of D=5 nm diameter InGaAs VNW (Aspect Ratio > 40)
32
0.0 0.1 0.2 0.3 0.4 0.50
50
100
150
200 Vgs= 0 V to 0.6 V in 0.1 V stepRon= 5500 Ω⋅µmMo contactD = 15 nm300 oC N2 RTA
Vds (V)
I d (µ
A/µm
)Latest! D=15 nm InGaAs VNW MOSFET
Single nanowire MOSFET: • Lch= 80 nm• 2.5 nm Al2O3 (EOT = 1.3 nm)
Zhao, IEDM 2017
Benchmark with Si/Ge VNW MOSFETs
33
Most aggressively scaled VNW MOSFET ever
Peak gm of InGaAs (VDS=0.5 V), Si and Ge VNW MOSFETs
Target
Our latest work (VDS=0.5 V)
Even better results at IEDM 2017!
InGaAs Vertical Nanowires on Si by direct growth
34
Selective-Area Epitaxy (SAE)
Au seed
Vapor-Solid-Liquid (VLS) Technique
InAs NWs on Si by SAE
Riel, MRS Bull 2014
Riel, IEDM 2012
VNW MOSFETs: path for III-V integration on Si for future CMOS
35
Vertical nanowire MOSFET for 5 nm node
Yakimets, TED 2015Bao, ESSDERC 2014
Vertical NW: power, performance and area gains w.r.t. Lateral NW or FinFET
5 nm node
30% area reduction in 6T-SRAM19% area reduction in 32 bit multiplier
Conclusions1. Great recent progress on planar, fin and nanowire
InGaAs MOSFETs
2. Device performance still lacking for 3D architecture designs
3. III-V Vertical-Nanowire MOSFETs: most likely architecture for future integration with Si
4. Many, MANY issues to work out:
sub-10 nm fin/nanowire fabrication, self-aligned contacts,
device asymmetry, introduction of mechanical stress, VT control, sidewall roughness, device variability, BTBT and parasitic HBT gain, oxidetrapping, self-heating, reliability, NW survivability, co-integration on n- and p-channel devices on Si, interface states, metal routing, contact resistance < 10-9 Ω.cm-2, off-state leakage, TDDB, etc.…
36
top related