III-V CMOS: What have we learned from HEMTs? J. A. del Alamo, D.-H. Kim 1, T.-W. Kim, D. Jin, and D. A. Antoniadis Microsystems Technology Laboratories, MIT 1 presently with Teledyne Scientific 23rd International Conference on Indium Phosphide and Related Materials Berlin, May 22-26, 2011 Acknowledgements: Sponsors: Intel, FCRP-MSD Labs at MIT: MTL, NSL, SEBL 1
Outline Why III-Vs for CMOS? What have we learned from III-V HEMTs III-V CMOS device design and challenges Conclusions 2
CMOS scaling in the 21 st century Si CMOS has entered era of power-constrained scaling : Microprocessor power density saturated at ~100 W/cm 2 Microprocessor clock speed saturated at ~ 4 GHz Pop, Nano Res 2010 Intel microprocessors 3
Consequences of Power Constrained Scaling Power = active power + stand-by power P A ~ f CV DD2 N N V DD clock frequency transistor capacitance #1 goal! operating voltage transistor count Transistor scaling requires reduction in supply voltage Not possible with Si: performance degrades too much 4
How III-Vs allow further V DD reduction? Goals of scaling: reduce transistor footprint extract maximum I ON for given I OFF 5
How III-Vs allow further V DD reduction? Goals of scaling: reduce transistor footprint extract maximum I ON for given I OFF III-Vs: higher electron velocity than Si I ON tight carrier confinement in quantum well S V DD 6
What have we learned from III-V HEMTs? State-of-the-art: InAs HEMTs Gate S D Oxide Kim, EDL 2010 t ins t ch Cap Etch stopper Barrier Channel Buffer - QW channel (t ch = 10 nm): InAs core (t InAs = 5 nm) InGaAs cladding - n,hall = 13,200 cm 2 /V-sec - InAlAs barrier (t ins = 4 nm) - Ti/Pt/Au Schottky gate - L g =30 nm 7
L g =30 nm InAs HEMT V GS = 2.0 Kim, EDL 2010 0.8 0.4 V 1.5 0.6 I D [ma/ m] 0.4 0.2 0.2 V g m [ms/ m] 1.0 0.5 0 V 0.0 0.0 0.2 0.4 0.6 0.8 V DS [V] V DS = 0.5 V 0.0-0.6-0.4-0.2 0.0 0.2 V GS [V] Large current drive: I ON >0.5 ma/µm at V DD =0.5 V V T = -0.15 V, R S =190 ohm.μm High transconductance: g mpk = 1.9 ms/μm at V DD =0.5 V 8
L g =30 nm InAs HEMT 40 Kim, EDL 2010 H 21 3 10-3 V DS = 0.5 V Gains [db] 30 20 10 U g MSG/MAG K 2 f T = 644 GHz f max = 681 GHz 1 0 K I D, I G [A/ m] 10-4 10-5 10-6 10-7 10-8 I D V DS = 0.05 V V DS = 0.5 V I G 0 V DS =0.5 V, V GS =0.2 V -1 10 9 10 10 10 11 10 12 Frequency [Hz] 10-9 V DS = 0.05 V -1.0-0.8-0.6-0.4-0.2 0.0 0.2 0.4 V GS [V] Only transistor of any kind with both f T and f max > 640 GHz S = 74 mv/dec, DIBL = 80 mv/v, I on /I off ~ 5x10 3 All FOMs at V DD =0.5 V 9
InAs HEMTs: Benchmarking with Si FOM that integrates short-channel effects and transport: I ON @ I OFF =100 na/µm, V DD =0.5 V IEDM 2008 InAs HEMTs: higher I ON for same I OFF than Si: Why? 10
Why high I ON? 1. Very high electron injection velocity at the virtual source E C v inj E V Kim, IEDM 2009 Liu, Springer 2010 v inj (InGaAs) increases with InAs fraction in channel v inj (InGaAs) > 2v inj (Si) at less than half V DD ~100% ballistic transport at L g ~30 nm 11
Why high I ON? 2. Quantum capacitance less of a bottleneck than previously believed InAs channel: t ch = 10 nm 40 Experiment (C G ) Jin, IEDM 2009 Capacitance [ff/ m 2 ] 30 20 10 C ins ( t ins = 4 nm) C Q1 (m * = 0.026m e ) C cent1 0-0.4-0.2 0 0.2 0.4 V G [V] C G ( 0.07 ) C G ( 0.05 ) C G (m * = 0.026m e ) Biaxial strain + non-parabolicity + strong quantization: m * C G n s I ON 12
Why high I ON? 3. Sharp subthreshold swing due to quantum-well channel Subtreshold swing [mv/dec] 90 80 70 60 In 0.7 Ga 0.3 As HEMTs: t ch = 13 nm InAs HEMTs: t ch = 10 nm InAs HEMTs: t ch = 5 nm t ins = 4 nm, L side = 80 nm state-of-the-art Si 40 80 120 160 200 L g [nm] Kim, IPRM 2010 Dramatic improvement in short-channel effects with thin channel Thin channel does not degrade v inj at L g ~40 nm (Kim, IPRM 2011) 13
Limit to III-V HEMT Scaling: Gate Leakage Current 10-3 10-4 10-5 InAs HEMT t ins = 10 nm L g = 30 nm t ins = 7 nm t ins = 4 nm t ch = 10 nm I D I D, I G [A/ m] 10-6 10-7 10-8 t ins =4 nm t ins =7 nm I G 10-9 10-10 t ins =10 nm V DS = 0.5 V -1.00-0.75-0.50-0.25 0.00 0.25 0.50 V GS [V] t ins I G Further scaling requires high-k gate dielectric 14
III-V CMOS: device design and challenges Modern III-V HEMT vs. modern Si MOSFET: III-V HEMT Intel s 45 nm CMOS ~2 m What do we preserve? What do we change? 15
III-V CMOS: HEMT features worth preserving Quantum-well channel: key to scalability Undoped channel: InAs-rich channel: Buried-channel design: for high mobility and velocity Raised source and drain regions: essential for scalability Undoped QW channel in extrinsic regions: key to low access resistance 16
III-V CMOS: HEMT features to change Schottky gate: need MOS gate with very thin high-k dielectric T-gate: need rectangular gate Barrier under contacts: need to eliminate Alloyed ohmic contacts: change to refractory ohmic contacts Source and drain contacts: need self-aligned with gate Footprint: need to reduce by 1000 X! n + n + HEMT QW-MOSFET 17
III-V CMOS: other critical needs p-channel MOSFET: with performance >1/3 that of n-mosfet Co-integration of n-fet and p-fet on Si: compact, planar surface n-mosfet p-mosfet Silicon 18
III-V CMOS: other designs n + n + Etched S/D QW-MOSFET Regrown S/D QW-MOSFET FinFET Gate-all-around nanowire FET 19
The high-water mark: Intel s InGaAs Quantum-Well MOSFET Radosavljevic, IEDM 2009 Direct MBE on Si substrate (1.5 m buffer thickness) InGaAs buried-channel MOSFET (under 2 nm InP barrier) 4 nm TaSiO x gate dielectric by ALD, L g =75 nm First III-V QW-MOSFET with better performance than Si 20
XOI InAs MOSFET G More recent notable work S InAs SiO 2 Si D InAs Nanoribbon MOSFETs on Insulator (UC Berkeley) Ko, Nature 2010 Aspect Ratio Trapping (Amberwave) Fiorenza, ECS 2010 Self-aligned QW-FET (MIT) Kim, IEDM 2010 Gate Source Fin-Channel Fin-Channel Drain EXT. Fin-channel Al 2 O 3 /InGaSb QW- MOSFET (Stanford) Nainani, IEDM 2010 Ge p-type QW-MOSFET (Intel) Pillarisetty, IEDM 2010 p InP p+ InP InGaAs FinFET (Purdue, Intel) Wu, IEDM 2009 Radosavljevic, IEDM 2010 21
Conclusions III-V HEMTs suggest strong potential for III-V CMOS: InAs electron injection velocity > 2x that of Si at 1/2x V DD Quantum capacitance less of a bottleneck than previously believed Quantum-well channel yields outstanding short-channel effects Impressive recent progress on III-V CMOS Sub-100 nm InGaAs MOSFETs with I ON > than Si at 0.5 V demonstrated Lots of work ahead Demonstrate ~10 nm III-V N-MOSFET that is better than Si P-channel MOSFET N-channel + P-channel cointegration on Si 22