Low Power CMOS Process Technology Scott Crowder IBM, SRDC, East Fishkill, NY Outline • • • • • What is Low Power? Process For Low Standby Power Process For Low Active Power Other Low Power Requirements Conclusions Process Technology/Scott Crowder 2 Power Components in Digital CMOS • Standby Power – – – – Power when no function is occurring Critical for battery driven Can be reduced through circuit optimization Temperature dependent leakage current dominates power • Active Power – Switching power plus passive power – Critical for higher performance applications • Other Sources of Power Consumption – Analog & I/O Power • Some circuit configurations have current flow even w/o switching – Dynamic Memory Refresh Power • eDRAM, 1T SRAM require periodic refresh of data Process Technology/Scott Crowder 3 Power Components (simple version) Power = Switching Power + Passive Power f*C*V*V Frequency Voltage Capacitance I (V,T) * V Current Process Technology/Scott Crowder Voltage 4 Power Trends Handheld Technology Desktop Processor Technology Base Devices, 10% Activity, 105C Gate Source Power for 10 x 10 mm chip (Watts) 200 Well without High-K Passive Power (picoWatts/Micron) High Vt Devices, 25C 150 100 50 0 180nm 130nm 90nm 65nm 45nm 100 Gate Sub Vt Active 80 60 40 20 0 180nm Process Technology/Scott Crowder 130nm 90nm 65nm 5 What is Low Power in CMOS? • Depends on product application • Examples #1: Handheld phone chip #2: Server microprocessor chip Process Technology/Scott Crowder 6 Example #1: Handheld Application • Key Attributes: – – – – – Battery Powered In off-state significant portion of time Standby condition at room temperature Lawsuit if chip gets too hot Standby dominates power requirements • Leakage current at room temperature dominates power • What is Low Power Optimized Design? – Want highest performance @ given leakage current Process Technology/Scott Crowder 7 Example #2: Server Application • Key Attributes: – – – – – Powered by electrical grid Almost always on Power condition at operating temperature Lower power reduces packaging & cooling expense Switching power dominated power requirements in past • Want lower capacitance, voltage at maximum frequency – Passive power at operating temperature now a major component • Want lower high temperature leakage current & lower voltage • What is Low Power Optimized Design? – Want switching power to dominate – Want to maximize performance & modulate voltage to meet power requirements Process Technology/Scott Crowder 8 One More Major Component: Cost • Critically important to technical solution • Example #2: Server microprocessor – Higher performance = higher end product price – Will add reasonable cost to enable performance • Higher cost process: strain engineering, added metal levels • Flip-chip packaging, heat sinks & fans • Example #1: Handheld application – Market will not allow higher cost solutions – Lower power specifications – Focus on lowest cost process that meets market demand Process Technology/Scott Crowder 9 Low Standby Power Process Technology/Scott Crowder 10 Components of Passive Power • Major components – Source leakage current • Sub threshold off-state current – Gate current VDD IG,off VDD IS, off IG,on GND IG,off • Tunneling current thru gate dielectric – Well current IW Process Technology/Scott Crowder IW IS, off IG,on • Band-to-band tunneling current to well VDD GND 11 IS: Source Sub threshold Leakage Sub threshold current Ioff (pA/um) 10000000 10000000 1000000 100000 100000 10000 1000 1000 100 10 10 1 0.1 0.1 0.01 0.001 0.001 1e-4 1E-5 1e-5 1e-6 1E-7 1e-7 0.1 0.2 0.2 0.3 0.3 0.4 0.4 0.5 0.5 0.6 0.6 0.7 0.7 0.8 0.8 0.9 0.9 1.1 1.2 1.2 00 0.1 11 1.1 25C 125C Vt (V) • Strong Vth dependence • Strong Temperature dependence Process Technology/Scott Crowder 12 Under scaling Supply Voltage Ring Delay (psec) 30 1.0V 25 1.1V 20 1.0V 1.2V 15 1.1V 1.2V 10 1.0V 1.1V 1.2V 500mV 350mV 250mV Lower leakage power at same performance at higher Vth & Vdd 5 0 0.1 1 10 100 1000 Relative Leakage Power • • Higher performance at given leakage power by raising Vdd Vdd scaling has slowed dramatically for low standby power Process Technology/Scott Crowder 13 Methods to Modulate IS Leakages • Reverse Body Bias or Active Well NFET PFET Silicide N+ P+ PWell NWell N-band for NFET Active Well Added well implant to separate well and substrate P-type Substrate Process Technology/Scott Crowder 14 Reverse Body Bias LOW VT DEVICES HIGH VT DEVICES Total Leakage Total Leakage 1e-5 1e-3 Reverse bias 1e-6 1e-7 1e-4 125C ~5x 1e-8 Reverse bias 125C 1e-5 1e-6 1e-9 -0.4 25C 25C -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 1e-7 0.4 -0.4 -0.3 -0.2 Body Bias Voltage Body bias dependence limited by IG & IW leakage -0.1 0.0 0.1 0.2 0.3 0.4 Body Bias Voltage Leakage dominated by IS but low body effect Process Technology/Scott Crowder 15 IG: Gate Leakage 1.0 NFET Gate Leakage (µA) 0.9 ~10x per 0.2nm SiOxNy 0.8 0.7 0.6 0.5 0.4 0.3 0.2 ~12 Angstromgate oxide +1V 0.1 0.0 0 25 Q. Liang, IBM 50 75 100 Temperature (C) 125 150 Kraus, et al, IEEE TED Vol 52, No. 6 2005, page 1141 • Function of gate dielectric material and thickness • Low temperature dependence -> dominates at low T Process Technology/Scott Crowder 16 Methods to Reduce Gate Leakage: N fixed EOT contours Kraus, et al, IEEE TED Vol 52, No. 6 2005, page 1141 • Increased Nitrogen concentration reduces gate leakage Process Technology/Scott Crowder 17 Methods to Incorporate Nitrogen N plasma plasma nitridation thermal nitridation N-implant N SiO2 base oxide SiO2 base oxide N N 700-1000C NH3, NO N oxidation N N N N N N more N at top surface N N throughout N N Process Technology/Scott Crowder N more N at bottom 18 Methods to Reduce Gate Leakage: High-K PolySi or Metal Gate High-k Layer Bottom interface silicon High-k/Metal Gate Stack: Tinv = 14.5A, ToxGL = 16A Benefits • >1000x lower IG • Lower Tinv at same IG to enable device scaling Issues • Vth control (WF engineering) • Charge trapping • Mobility degradation • Increased complexity Process Technology/Scott Crowder 19 Mobility • Acceptable long channel mobility with Hf-based oxides • Requires optimization of HfSiO(N)/SiO(N) composition [M Frank et al., VLSi-TSA05 ] Process Technology/Scott Crowder 20 Contact S/D STI 2 Oxide/Nitride spacer HfSiO gate stack Counter doping interfacial layer ('cap') EXT EXT halos Capacitance density [µF/cm ] Gate Workfunction Tuning Contact S/D STI Bulk Si Fig.1 schematic to illustrate different components for vt adjustments for HfSIO gate stack 1.20 SiON 1.00 0.80 pFET nFET 0.60 High-k / SiON High-k / SiO2 0.40 0.20 0.00 -1.5 Conventional HfSiO -1.0 -0.5 0.0 0.5 Gate voltage [V] 1.0 1.5 Fig. 3 Split C-V curves demonstrating Optimized high-k dielectric stackthe reduces Vt shift by 200~300 mV Vt shiftpFET obtained by gate Selective use of ultra-thin interface/surface modification layers stack optimization onM.SiO and SiON 2 M. Frank et al VLSI-TSA 2005 interface layers Process Technology/Scott Crowder 21 IW: Tunneling & Trap-Assisted Current Gate Field Enhanced BTB * S/D De p le tio n * * BTB tunneling Trap Assisted Well 100 Current (pA/um) * * * * ** * 1000 10 IS IG IW 1 0.1 0.01 * * Schematic of IW sources 200 150 100 50 Technology Node (nm) Relative Contribution of IW • Band to band tunneling increases with doping levels & steeper gradients • Gate field enhanced tunneling increases with thinner gate dielectric • Reduction of trap assisted current varies with cause Process Technology/Scott Crowder 22 Methods to Reduce Junction Leakages 2.5 Capacitance (pF/um2) Current (pA/um) 2 1.5 1 1.5 6 1.4 5.5 1.3 5 1.2 4.5 1.1 4 1 3.5 0.9 0.5 25 30 35 40 45 50 25 30 35 40 3 45 Implant Energy (KeV) Halo implant angle (degrees)) NFET Junction Leakage vs. Halo Angle Current (pA/um) • Leakage modulated primarily by transistor doping profiles PFET Junction Leakage vs. S/D Energy Ref: Hook, ACEED 2005 Process Technology/Scott Crowder 23 Low Active Power Process Technology/Scott Crowder 24 Components of Active Power Power ~ f * C * V * V + I (V,T) * V Voltage Higher Transistor Performance at Same Leakage Same Fmax at Lower Voltage Lower Active Power Process Technology/Scott Crowder 25 Performance Scaling Traditional performance scaling: Shorter Gate Length Higher Gate Capacitance (IG) Shallower Junctions (IW) Lower Threshold Voltage (IS) Lower Supply Voltage All did not scale in 65nm technology Higher performance with: Higher Supply Voltage Higher Inversion Capacitance Higher Mobility Process Technology/Scott Crowder 26 Mobility Innovation: Stress Memory After anneal & Nit Removal n F E T Ioff (n A /u m ) Amorphous layer By implant 1000 1e-05 Control 100 1e-06 Tensile Nit stressor 10 1e-07 1 1e-08 0.006 0.007 0.008 0.009 0.01 0.011 0.012 0.013 0.014 600 700 800 900 1000 1100 1200 1300 1400 b (M 1 FN10 10 0 08 0 07 M 1 I ) nFET Ion (uA/um) Si is crystallized and stress is “memorized” Process Technology/Scott Crowder [Chan, CICC ’05] 27 Mobility Innovations: Strained Liner Compressive Tensile Nit +Nitride Ge Contact Tensile nit Compr nit N P ST 10-5 a) NMOS 10-6 10 -5 Ioff (A/µm) Ioff (A/µm) 10 -4 Un-stress 10 -6 tensile 10 -7 10 -8 700 900 1100 Ion (µA/µm) 1300 b) PMOS Un-stress Compressive 10-7 10-8 10-9 350 450 550 Ion (µA/µm) [Chan, CICC ’05] Process Technology/Scott Crowder 28 Mobility Innovations: Embedded SiGe 1700 [Luo, IEDM 2005] Rodlin (ohm) 1500 1300 1100 P+ 900 w/o eSiGe 700 eSiGe 500 300 eSiGe Compressive liner eSiGe eSiGe + CSL 0 [Luo, IEDM 2005] 0.01 0.02 0.03 0.04 0.05 0.06 0.07 PFET Lgate (um) PFET Liner Mobility Enhancement Embedded SiGe Cross-section • Introduces compressive stress • Lowers contact resistance Process Technology/Scott Crowder 29 Mobility Innovations: <100> PFET X (100) <011> channel Z (001) <011> channel X (100) 0o notch Y (010) <001> channel Z (001) <010> channel 45o notch Y (010) Process Technology/Scott Crowder 30 Mobility Innovations: <100> PFET p insensitive , , 10-6 100 05 Carrier direction <110> Carrier direction <100> 10 pMOS Ioff (A/um) pFET Ioff (nA/um) 1000 10-7 06 compressive 10-8 07 Neutral tensile 10-9 08 1 0.0002 200 0.0003 0.0004 0.0005 300 400 500 pFET Ion (uA/um) M 1 FPRVT1 x_08 M 1 Io n 0.0006 600 . 0 002 200 0 0025 0 003 300 0 0035 0 004 400 0 0045 0 005 500 0 0055 pMOS Ion (uA/um) 0 006 600 [Chan, CICC ’05] • Significant boost in PFET current without added process steps • Loss of ability to increase current with compressive stress Process Technology/Scott Crowder 31 Mobility Innovations: Hybrid Orientation • 110 surface: 2x higher PFET µ >2x lower NFET µ • Hybrid Orientation • 110 for PFET • 100 for NFET M. Yang, VLSI 2004 Process Challenges: - Low defect epitaxy - Planarization - Trench isolation PMOS Process Technology/Scott Crowder NMOS 32 Inversion Capacitance: Metal Gates Stability <700C Stability >700C 1 IA H IIA Li Be Na Mg 18 * = NFET; **=PFET 16 17 IIIA IVA VA VIA VIIA He B C N O F Ne P S Cl Ar 6 7 IIIB IVB VB VIB VIIB IB IIB Al* Si V* Cr Mn* Fe Co** Ni** Cu Zn Ga Ge As Se Br Kr Ru Rh** Pd** Ag Cd In Sn Sb Te I Xe W Re** Os Ir** Pt** Au Hg Tl Pb Bi Po At Rn Sc Ti* Rb Sr Y Zr* Nb* Mo Cs Ba * La Hf* Ta* Fr Ra ** Ac * ** Tc 11 15 5 VIII 10 14 4 Ca 9 13 3 K 8 VIIlA 12 Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb Lu Th Pa U Cf Es Fm Md No Lr Np Pu Am Cm Bk Metals 3-5 nm thick on 500 nm SiO2, anneal 3 oC/s up to 1000 oC in FG Binary/ternary alloys: W2N*, Ta2N*, TaN*, TaSiN*, RuO2**, CoSi2, NiSi • Metal gate incorporation to remove gate depletion • Holy Grail: Band edge workfunction & thermally stable Process Technology/Scott Crowder 33 Metal Gate Integration Options: FuSi “Two Steps” “One Step” Metal (Ni,Co) Polysilicon Silicide Metal Deposition Anneal, Strip, Anneal Metal (Ni,Co) Polysilicon Silicide NiSi Anneal, Strip, Anneal Silicide [Kedzierski, IEDM ’02] Process Technology/Scott Crowder 34 Challenges for FuSi O Al Ni Hf Si+Hf • Dual Workfunction creation Counts – Alloys – DopingAl spike – Phase change NiAlSi Al pile up HfO2 [ YH Kim IEDM’05] • Workfunction control & stability • Variation with gate density & size Process Technology/Scott Crowder 35 Metal Gate Integration Options: Replacement Gate Build Device Etch Gate Deposit High-k Deposit Gate Metal Deposit Fill Metal CMP Challenges: • Integration of different workfunction metals: high complexity • High-k deposition on small gate area • ALD probably required for gate metal Process Technology/Scott Crowder 36 Metal Gate Workfunction Stability 2 Capacitance Density (norm.) Workfunction of integrated metal gates has strong dependence on deposition & annealing conditions Deposition Annealing Capacitance Density µ(F/cm ) • 3.0 2.5 2.0 Method2 1.5 1.0 Method1 0.5 0.0 -0.5 0.0 0.5 1.0 1.5 2.0 As deposited 1.0 0.8 Reducing Ambient 0.6 Oxidizing Ambient 0.4 0.2 [E. Cartier, VLSI ’05] 0.0 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 Gate Voltage (V) • Potential root causes of workfunction shifts – – – – – Metal induced gap states (MIGS) Oxide charge (O vacancies) in HfO2 Oxidation of interfacial metal Hf-diffusion into metal (WF change) Dipole layers at SiO2/HfO2 interface Process Technology/Scott Crowder 37 Memory Limitation to Voltage Scaling • SRAM cell stability limits scaling of array Vdd – Vth mismatch due to dopant fluctuation in best case is constant with scaling • Lack of gate dielectric scaling has caused it to increase – Reduction of all other components of mismatch critical [VLSI 2005] 1.0 • Design Options 0.4 Cell area = 0.197µm2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Vleft [V] Vright [V] Vright [V] 0.8 0.6 0.2 8T 1.0 0.8 – Larger cells – Separate array Vdd – Alternative cells/memory • 8T SRAM • eDRAM 6T 0.6 0.4 0.2 Cell area = 0.262µm2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Vleft [V] Butterfly curve comparison for scaled cells Process Technology/Scott Crowder 38 Components of Switching Power Power ~ f * C * V * V Capacitance Transistor Ccg Metallization Cmg Cg Cd, area Cov Cvert Cd, per Cd, per Process Technology/Scott Crowder Chor 39 Capacitance Reduction: SOI Cd, per Cd, area Cd, per Cd, per Cd, area Cd, per Bulk Transistor SOI Transistor • Reduction of area diffusion capacitance to negligible level • Reduction of perimeter capacitance with film scaling • Additional AC drive current benefit from gate coupling Process Technology/Scott Crowder 40 Options to Gate Capacitance Scaling • Need better electrostatic control to scale LG – Higher gate inversion capacitance • High-K gate dielectric material to reduce EOT • Metal gate material to eliminate gate depletion – Fully depleted double gate structures • Better channel control through structural change • LG must decrease more than Cinv increases Process Technology/Scott Crowder 41 Capacitance Scaling with Metal Gate • LG must decrease to prevent capacitance increase Si Band edge DIBL at constant LG Towards mid-gap Gate Workfunction • Short channel effect degradation when metal gate workfunction deviates too far from band edge – Lose portion of performance benefit and most of power benefit Process Technology/Scott Crowder 42 One Double Gate Structure: FinFET NiSi Gate NiSi Si Fin Tox = 1.6nm Tsi = 25nm Si [D. Fried] BOX Kedzierski, IEDM 03 Process Challenges: • Fin creation and control • Source/drain resistance engineering • Contact formation and control Process Technology/Scott Crowder 43 BEOL Capacitance Scaling Performance ~ R * C Cvert h Chor Capacitance ~ keff * h / space Resistance ~ ρ / (h * w) Reduce inter-level dielectric constant => lower capacitance Reduce resistance => lower height => lower capacitance Process Technology/Scott Crowder 44 Capacitance Scaling: Reducing BEOL k A. Grill et al., Adv. Metalliz. Conf. 2001 Cohesive Strength, Γ (J/m2) 12.0 X-H. Liu et al., Adv. Metal. Conf. (2004); 8.0 4.0 0.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 Nominal Dielectric Constant, k • Multiple porous dielectric options with k < 2.4 – Spin-on and CVD deposition options • All materials have lower material strength as k decreases – Optimization of BEOL stack & packaging required Process Technology/Scott Crowder 45 Capacitance Scaling: Scaling h Line Resistivity (µΩ-cm) 14 Assumptions Grain size = 2x line width Smooth surfaces Mean free paths: Al=15 nm; Cu=39 nm; Au =40 nm; Ag=45 nm Cu 12 10 Ag 8 Au 6 Note: Pure Al has poor E-M; Al(Cu) alloy has poor R. 4 2 0 0.01 Al (pure) 45 nm 0.1 Linewidth (µm) 1 S. Rossnagel, Semicon West (2004). • Copper will continue to be material for low ρ metallization Process Technology/Scott Crowder 46 Capacitance Scaling: Scaling h Ta/TaN Liner Cu [Edelstein, 2005] ITRS 45nm PVD liners demonstrated • Cu BEOL using Ru liner Scaling requires thinner liner & seed films – 45nm dimensions demonstrated – Porous dielectrics must be compatible with thin liner to have benefit • ALD liner & Cu seed reduction provide opportunity for additional scaling Process Technology/Scott Crowder 47 2X pitch wires Min. pitch wires Native BEOL capacitor Capacitance/Area (fF/um 2) Passives Scaling: BEOL Capacitors 3 2.5 Native BEOL 2 1.5 1 0.5 0 0 50 100 150 Technology (nm) • Benefit of scaling: better “free” BEOL capacitors • Good tolerance due to averaging over many levels Process Technology/Scott Crowder 48 Passives Scaling: Decoupling Capacitors Capacitance (fF/um2) 200 150 Trench Capacitor 100 Gate Oxide Caps 50 0 0 0.5 1 1.5 Leakage (nA/um2) 2 2.5 Trench Capacitors • Gate capacitor scaling not possible with low passive power • >10x reduction in area & >100x reduction in leakage with trench capacitors Process Technology/Scott Crowder 49 Conclusions: Handheld Low Power • Standby power dominated • Choices: – Add strain engineering without gate scaling – Add high-K dielectrics to enable gate scaling – Add design complexity to modulate leakage • Decision will be driven by lowest cost solution Process Technology/Scott Crowder 50 Conclusions: Plugged In Low Power • Active Power Dominated • Performance elements added as developed • Choices: – Lower supply voltage • May require new memory choice – Add design complexity to reduce both switching and passive power • Voltage islands, clock gating, etc… Process Technology/Scott Crowder 51