Department of Electrical and Computer Engineering (i.e., A brief history of sand) M. Fischetti October 2, 2009 A (partial, biased?) history of the MOSFET

Department of Electrical and Computer Engineering

((i.e.i.e., A brief history of sand), A brief history of sand)

M. Fischetti

October 2, 2009

A (partial, biased?) history of the MOSFET A (partial, biased?) history of the MOSFET from a physicist’s perspectivefrom a physicist’s perspective

2Department of Electrical and Computer Engineering MVF October 2, 2009

This talk: Void where prohibited, limitations and restrictions applyThis talk: Void where prohibited, limitations and restrictions apply

Technology (i.e., how do we make them?) vs. electronic operation (i.e., how do they work and how do we make them better?)

Too much to cover Talk about what I know Major omissions (that is, a disclaimer*):

• Doping: Diffusion (theory and technology), ion implantation, high-doping effects• Lithography, possibly a “technology enabler”: Optical, contact, phase-shift, X-ray…• Metallization: Deposition/growth, DAMASCENE, electromigration• Etching: Wet vs. dry, RIE, plasma• Film growth: Epitaxy, CVD, PE-CVD, MBE, ALD,…• Contacts: Silicides, salicides, FUSI,…• Layout issues: Isolation (deep/shallow trenches), cross-talk, latch-up, design rules,

DRAM/SRAM design, power,…• …


History of the MOSFET? What’s that?History of the MOSFET? What’s that?

Thanks to Jiseok Kim for having put me on the spot…. It’s OK… I wish him good luck in getting his PhD wherever ELSE he may wish to get it….

I’m not sure what he meant by “history”… So, let’s start from the beginning….


The main character of our story: The MOSFETThe main character of our story: The MOSFET

No other human artifact has been fabricated in larger numbers (except perhaps nails?) “…some consider it one of the most important technological breakthroughs in human

history…” (Wikipedia, the source of all human knowledge)


Timeline ITimeline I

1925: Julius Edgar Lilienfeld’s MESFET patent1935: Oskar Heil’s MOSFET patent194?: Unpublished Bell Labs MESFET1947: Ge BJT (Bardeen, Brattain, Shockley, Bell Labs)1954: Si BJT (Teal, Bell Labs)1960: MOSFET (Atalla&Khang, Bell Labs)1961: Integrated circuit (Kilby, TI)1963: CMOS (Sah&Wanlass, Fairchild)1964: Commercial CMOS IC (RCA)1965: DRAM (Fairchild)1968: Poly-Si gate (Faggin&Klein, Fairchild)1968: 1-FET DRAM cell (Dennard, IBM)1971: UV EPROM (Frohman, Intel)1971: Full CPU in chip, Intel 8008 (Faggin, Intel)1974: Digital watch1974: Scaling theory (Gänsslen&Dennard, IBM)1978: Use of ion implanter1978: Flotox EEPROM (Perlegos, Intel)1980: Ion-implanted CMOS IC1980: Plasma etching1984: Scaling theory <0.25 μm (Baccarani, U. Bologna)1986: 0.1 μm Si MOSFET (Sai-Halasz, IBM)1991: CMOS replaces BJT also at high-end1993: DGFET scalable to 30 nm (theory, Frank et al.)2007: Non-SiO2 (HfO2–based) MOSFET (Intel)

1955: Si, Ge conduction band (Herring&Vogt) Deformation-potential, high-field (Bardeen&Shockley)1957: BTE in semiconductors – impurities (Luttinger&Kohn), phonons (Price, Argyles)1964: Band structure calculations (Hermann) Monte Carlo for semiconductors (Kurosawa) 1965: Linear-parabolic oxidation model (Deal&Grove)1966: Observations of 2DEG (Fowler, Fang, Stiles, Stern,..)1967: Conductance technique (Nicollian&Goetzberger)

1974: DDE device simulator (Cottrell&Buturla)1975: Quantum Hall Effect predicted (Ando)

1979: Quantum Hall Effect observed (von Klitzing)1981: Identification of native Nit: Pb-centers (Poindexter) Full-band MC (Shichijo&Hess) 1982: Fractional QHE observed (Störmer&Tsui, Laughlin)1988: Full-band MC device simulator (MVF&Laux)1992: NEGF device simulator (Lake, Klimeck, et al.)

Technology Physics/Technology Physics/SimulationsSimulations


Timeline IITimeline II

1975: 20 μm (tOX≈250 nm)

1980: 10 μm (tOX≈150 nm)

1985: 5 μm (tOX≈70 nm)

1990: 1 μm (tOX≈15 nm)

1995: 0.35 μm (tOX≈8 nm)

2000: 0.18 μm (tOX≈3 nm)

2005: 65 nm (tOX≈1.4 nm)

2010: 32 nm (tOX≈1.2 nm?)

SiO2 growth and instability: Ions, traps, interface

SiO2 instability during operation: electron trapping, NBTI

Hot electron effects: oxide trapping, VT shift, breakdown

Scaling: Short-channel effects (SCE), oxide, dopants

….life is good…

Scaling: SCE, insulatorLeakage: InsulatorPower: Alternative devices

Feature size Main ProblemsFeature size Main Problems

↑

↓


Timeline IIITimeline III

1975: 20 μm

1980: 10 μm

1985: 5 μm

1990: 1 μm

1995: 0.5 μm

2000: 0.25 μm

2005: 63 nm

2010: 32 nm

2015: 16 nm ?

Feature size Transport PhysicsFeature size Transport Physics

Drift-Diffusion

Hydrodynamic/Energy transport

Boltzmann

Quantum?

↓

↕

↓


Transistor prehistoryTransistor prehistory

1935 Heil’s patent 1947 First BJT 1960 Atalla’s MOSFET Bardeen, Shockley, Brattain (Bell Labs)


IC PrehistoryIC Prehistory

1961 Kilby’s first IC 1962 Fairchild IC 1964 First MOS IC (RCA)


Moore’s law prehistoryMoore’s law prehistory

Gordon Moore 1965: Cost vs time Moore’s law 1960-1975


Moore’s lawMoore’s law

Number of transistors/die & feature size vs time


Microprocessor prehistoryMicroprocessor prehistory

1965: Federico Faggin 1968: Fairchild 8-bit μP 1971: Intel 8080 μP


Memory prehistory: DRAM and EPROMMemory prehistory: DRAM and EPROM

Bob Dennard (1-FET DRAM cell, 1968; 1971 Frohman’s UV-erasable EPROM scaling theory with Fritz Gänsslen,1974) (written by avalanche injection)


More historical trendsMore historical trends

J. Armstrong (ca.1989)


Timeline II once moreTimeline II once more

1975: 20 μm (tOX≈250 nm)

1980: 10 μm (tOX≈150 nm)

1985: 5 μm (tOX≈70 nm)

1990: 1 μm (tOX≈15 nm)

1995: 0.35 μm (tOX≈8 nm)

2000: 0.18 μm (tOX≈3 nm)

2005: 65 nm (tOX≈1.4 nm)

2010: 32 nm (tOX≈1.2 nm?)

SiO2 growth and instability: Ions, traps, interface

SiO2 instability during operation: electron trapping, NBTI

Hot electron effects: oxide trapping, VT shift, breakdown

Scaling: Short-channel effects (SCE), oxide, dopants

….life is good…

Scaling: SCE, insulatorLeakage: InsulatorPower: Alternative devices

Feature size Main ProblemsFeature size Main Problems

↑

↓


SiOSiO22 growth and instability growth and instability

Ionic contamination (K, Na): Unrecognized source of early problems Fixed traps (oxygen vacancies?), especially near Si-SiO2 interface Growth kinetics: Deal & Grove model: linear (reaction-limited) and parabolic (diffusion-limited)

regions; dry and wet oxidation rates Interface-state passivation: Al (with H) Post Metallization Anneal (PMA, Peter Balk):

H2O → H+ + OH-

Si- + H+ → Si-H

Andrew Grove (left), Bruce Deal (center) and Ed Snow (left)

Ed Snow’s cartoon, ca. 1966 about SiO2 instabilities


SiOSiO22 growth and instability, as-grown and during operation growth and instability, as-grown and during operation

CV-plot instabilities (VFB or VT shifts): Ions (mainly Na+ and K+, contamination in chambers, handling, gases, etc…) Interface states generation (stretch-out, Lai, Feigl, Sandia group, Technion, Siemens,…) Electron and hole traps (DiMaria, Young, Feigl):

• Neutral: H2O-related (mainly OH-) in wet oxides, radiation induced in processing, σ ≈ 10-15 to 10-17 cm2

• Charged-attractive: Ionic contamination, σ ≈ 10-13 cm2 , field-dependent• Charged-repulsive: Radiation-induced, σ ≈ 10-19 cm2


SiOSiO22 instability during operation instability during operation

Anomalous Positive Charge (APC): Caused by electron injection (Avalanche, Fowler-Nordheim) and also hole injection Related to Hydrogen: Boron deactivation in p-type substrates (Sah) Related to hole back-injection from anode? Dependent on gate-metal workfunction -

Au vs. Al vs. Mg (MVF&Weinberg, Chenming Hu) Occurring at Si-SiO2 interface even under negative bias: Neutral species such as

solitons, H2 diffusion…? (Weinberg). Connected to wear-out and breakdown (DiMaria, Stathis) Strongly correlated to interface traps (Pb-centers, Lenahan, Poindexter) Oxygen deficiency (Revesz)? Broken Si-H bonds (Si-D experiment, Lyding&Hess)?

Negative Bias Temperature Instability (NBTI): No time to discuss, but big issue in high-κ dielectrics


Understanding SiOUnderstanding SiO22 degradation: Two approaches degradation: Two approaches

MVF and DiMaria, INFOS 1989


SiOSiO22 growth and instability: Injection techniques and damage generation growth and instability: Injection techniques and damage generation


SiOSiO22 growth and instability: Electronic transport in SiO growth and instability: Electronic transport in SiO22

Electrons: Long-standing problems of high-field electron transport in polar insulators

(Karel Thornber’s 1970 PhD Thesis with Richard Feynman) LO-phonon scattering run-away connected to dielectric breakdown Experimental observations do not show predicted run-away at 2-3 MV/cm Umklapp scattering with acoustic phonons keeps electron energy under

control (MVF, DiMaria, Theis, Kirtley, Brorson, 1985) Holes: Small polaron (self-trapping) transport (Bob Hughes’ 1977 time-of-flight

experiments explained by David Emin’s 1975 theory).MVF et al., PR B (1985)


Hot electron effects in constant-voltage-scaled MOSFETsHot electron effects in constant-voltage-scaled MOSFETs

Two problems: Understand origin/spectrum of hot carrier Understand nature/process of damage generation

Practical problems: Unnecessary and expensive burn-in Wall Street “big glitch” in 1994

Theory: Shockley’s “lucky-electron model” widespread in EE community in the ’80s

(publicized by Chenming Hu, UCB): Even the Gods can be wrong at times… Full-band models (Sam Shichijo & Karl Hess, MVF&Laux, then others) Basic physics of electron scattering, injection into SiO2, etc. The mid-1990s “pseudo-full-band” frenzy (Bologna, UNC, Udine, Lille, TU-

Vienna, Aachen,..): Gain without pain… didn’t work…


Electron injection into SiOElectron injection into SiO22

MVF, Laux, and Crabbé, JAP (1996)


Timeline III once moreTimeline III once more

1975: 20 μm

1980: 10 μm

1985: 5 μm

1990: 1 μm

1995: 0.5 μm

2000: 0.23 μm

2005: 63 nm

2010: 32 nm

2015: 16 nm ?

Feature size Transport PhysicsFeature size Transport Physics

Drift-Diffusion

Hydrodynamic/Energy transport

Boltzmann

Quantum?

↓

↕

↓


Electron transport in Si at 3 eV: A big headacheElectron transport in Si at 3 eV: A big headache

Effective-mass approximation valid only for E ≈ a few kBT Scattering rates at E ≥ {a few kBT} totally unknown Moments of the BTE (DDE, Hydrodynamic) not sufficiently accurate


Electron transport in Si at 3 eV ca. 1992: A depressing picture…Electron transport in Si at 3 eV ca. 1992: A depressing picture…

The state-of-the art circa 1992


A good example of experiments-theory feedbackA good example of experiments-theory feedback

XPS (McFeely, Cartier, Eklund at the Brookhaven IBM synchrotron line, 1993) Carrier separation (DiMaria, 1992)

Cartier et al. APL (1993)


Electron transport in Si at 3 eV ca. 1994: Much better…Electron transport in Si at 3 eV ca. 1994: Much better…

The state-of-the art circa 1994

MVF et al., JAP (1996)


1980 → 1985 → 1988 → 1991 → 1994 → 1999 → 2003 …LGATE = 10 μm 5 μm 2.5 μm 1.3 μm 0.63 μm 0.25 μm 130 nm …. 1980 → 1985 → 1988 → 1991 → 1994 → 1999 → 2003 …LGATE = 10 μm 5 μm 2.5 μm 1.3 μm 0.63 μm 0.25 μm 130 nm …. 1980 → 1985 → 1988 → 1991 → 1994 → 1999 → 2003 …LGATE = 10 μm 5 μm 2.5 μm 1.3 μm 0.63 μm 0.25 μm 130 nm …. 1980 → 1985 → 1988 → 1991 → 1994 → 1999 → 2003 …LGATE = 10 μm 5 μm 2.5 μm 1.3 μm 0.63 μm 0.25 μm 130 nm ….

1980 → 1985 → 1988 → 1991 → 1994 → 1999 → 2003 …LGATE = 10 μm 5 μm 2.5 μm 1.3 μm 0.63 μm 0.25 μm 130 nm …. 1980 → 1985 → 1988 → 1991 → 1994 → 1999 → 2003 …LGATE = 10 μm 5 μm 2.5 μm 1.3 μm 0.63 μm 0.25 μm 130 nm …. 1980 → 1985 → 1988 → 1991 → 1994 → 1999 → 2003 …LGATE = 10 μm 5 μm 2.5 μm 1.3 μm 0.63 μm 0.25 μm 130 nm ….

ScalingScaling

Shrink dimensions maintaining aspect-ratio Must shrink electrostatic features as well (depletion regions→ doping level and profiles)

↔1 μm


ScalingScaling

Electrostatic integrity (Well-tempered MOSFET, Antoniadis): SOI, DGFETs, FinFETs, NW-FETs Reduce power, an example: The tunnel FET n (tFET) Reduced leakage: High-κ gate-insulators Improve (or, at least, maintain) performance: Alternative channel materials?


Scaling – Electrostatic integrity: SOIScaling – Electrostatic integrity: SOI

22 nm strained-Si nFET: SOI to prevent punch- through, strained Si to improve performance (B. Doris, IBM, 2006)


Scaling – Electrostatic integrity: Double gate FETScaling – Electrostatic integrity: Double gate FET

AIST (2003)


Scaling – Electrostatic integrity: Multibridge FETs (TEM, SEM)Scaling – Electrostatic integrity: Multibridge FETs (TEM, SEM)

Samsung Electronics Ltd. (2005)


Multibridge FETs: Process flowMultibridge FETs: Process flow

Samsung Electronics Ltd. (2005)


Scaling – Electrostatic integrity: FinFETsScaling – Electrostatic integrity: FinFETs

Freescale Semiconductors


Scaling – Electrostatic integrity: Si Nanowire TransistorsScaling – Electrostatic integrity: Si Nanowire Transistors

KAIST (2007)


Scaling – Reduce power: The tunnel-FET (tFET)Scaling – Reduce power: The tunnel-FET (tFET)

InAs Tunnel-FET: structure (M. Haines, UMass 2009)

InAs Tunnel-FET: pair generation rate (M. Haines, UMass 2009)

Stand-by power dissipation approaching “on” power dissipation… Cannot continue like this!

60 mV/dec → ΔVG ≈ 250 mV for Ioff/Ion ≈ 10-4

VT + ΔVG ≥ 0.45 V at 300 K (nFETs) Must increase slope (i.e., go below 60 mV/dec) if

we want the `Green’ FET (term coined by C. Hu) Problem: Ion too low in all attempts (DARPA to

IBM, UCB, Stanford,…) so far


Scaling – Reduce leakageScaling – Reduce leakage

Off-leakage: Accepted value increasing: Ioff/Ion ≈ 10-4 for the 32 nm node (used to be 10-6 or lower!) Connected to electrostatic integrity (punch-through, junction leakage, gate leakage)

Gate leakage: C = εox/tox, so if tox has reached its limit (≈ 1nm, too aggressive so far), scale εox:

High-κ insulators such as HfO2, ZrO2, Al2O3, etc. Problem: Low mobility in high-κ MOS systems (scattering with interfacial optical phonons) Metals with different workfunction needed!

Hi-res TEM from Susanne Stemmer, UCSB MVF et al., JAP (2001)


Scaling – Reduce leakage: Gate oxide scaling at IntelScaling – Reduce leakage: Gate oxide scaling at Intel

C. Hu et al., IEDM (1996)


Scaling – Improve performanceScaling – Improve performance

Taken for granted early on (ca. 1986) Slow realization that early optimism was unjustified

MVF and S. Laux, EDL (1987)


Scaling – Improve performanceScaling – Improve performance

Look for high-velocity, low-effective mass semiconductors… or should we? Problems:

High-energy (≈ 0.5 eV ≈ 20 kBT) DOS and rates identical in most fcc semiconductors Low DOS → loss of transconductance Low DOS → smaller density in quasi-ballistic conditions → lower Ion

Low DOS → less scattering in source → source starvation


Scaling – Improve performance: DOS bottleneckScaling – Improve performance: DOS bottleneck

MVF and S. Laux, TED (1991)


Scaling – Improve performance: Strained SiScaling – Improve performance: Strained Si

MVF and S. Laux, JAP (1996)


Scaling – Improve performance: Strained SiScaling – Improve performance: Strained Si

IBM 32 nm strained (tensile) Si nFET on SiGe virtual substrate

Intel 45 nm strained (compressive) Si pFET with regrown SiGe S/D


Why are sub-40 nm devices getting slower?Why are sub-40 nm devices getting slower?

Power dissipation → reduce frequency or fry! Parasitics play a bigger role (Antoniadis, MIT) Higher oxide fields squeeze carriers against interface → increased scattering

(Antoniadis, MIT) Intrinsic Coulomb effects!

MVF and S. Laux, JAP (2001)


Why are sub-40 nm devices getting slower? The effect of e-e interactionsWhy are sub-40 nm devices getting slower? The effect of e-e interactions


Sub-32 nm Si CMOS devices: Where do we stand?Sub-32 nm Si CMOS devices: Where do we stand?

22 nm: Planar (Intel), SOIs (IBM), FinFETs doable but too expensive. 16 nm: Possibly FinFETs, still Si Below 16 nm:

Ge pFETs and III-V nFETs (IMEC)? A pipedream… Ge nFETs still lousy, improvements promised at Dec 2009 IEDM, we’ll see III-Vs in the works:

•MIT (del Alamo): Great HEMTs, but huge S/D-gate gap not easily scalable•SRC/UCSB MOSFETs: Wait and see…


The future and “post Si CMOS” devices: What do we need?The future and “post Si CMOS” devices: What do we need?

Three terminal devices (Josephson computers taught us something…!) At least some gain (preserving signal over billions of devices, beating kBT) At least a few devices must have high Ion to charge external loads On/off behavior (Landauer’s water faucet analogy) Low power, possibly non-charge-switching (spins, QCA,…). BUT: If we use ≈ kBT to

switch, the heat bath will switch for us even if we do not want to… Notable historic failures:

Josephson: Excessively strict tolerances (on insulators), complicated 2-terminal logic SETs: No output current (`a slight impedance matching issue’, as someone kindly put

it….) Optical computers: Photons are huge! Clumsy 3-terminal devices Resonant tunneling diodes and multi-state logic: Non off-off switches, impossible to

control manufacturing tolerances High hopes:

Nanowires: They are just thin and narrow FinFETs Long shots:

Spins and QCA: Low power but no gain CNT: No current in single tube, must use many in parallel III-Vs: Battle already lost in 1991 (DOS bottleneck),,, why bother again?


And by popular demand… The future of “post-Si CMOS” logic technologyAnd by popular demand… The future of “post-Si CMOS” logic technology


More slides about the lunatic fringe….More slides about the lunatic fringe….


The lunatic fringe: Exploratory devicesThe lunatic fringe: Exploratory devices





Carbon NanoTube (CNT) FETIFF-Jülich, Germany (2004)


CNT TransistorsCNT Transistors

IFF-Jülich, Germany (2004)


CNT FET inverterCNT FET inverter

J. Appenzeller, IBM

Documents

Department of Electrical and Computer Engineering (i.e., A brief history of sand) M. Fischetti October 2, 2009 A (partial, biased?) history of the MOSFET