Transistor Scaling: The Age of Innovation Kaizad Mistry VP, Technology and Manufacturing Group Intel Corporation April 2014 Outline Transistor Scaling – The Dennard Era – Moore’s Law – Dennard Scaling Transistor Scaling: The Age of Innovation – Strain – High-k/Metal Gate – Tri Gate – Dual/Quad Patterning Transistor Evolution: The Future 2 Moore’s Law - 1965 “Reduced cost is one of the big attractions of integrated electronics, and the cost advantage continues to increase as the technology evolves toward the production of larger and larger circuit functions on a single semiconductor substrate.” Electronics, Volume 38, Number 8, April 19, 1965 Integrated circuits = Lower cost per function 3 The Cost Reduction Engine that Drives the Industry 1010 10 As the number of transistors goes UP 100 10-1 $ 10-2 109 108 107 Price per transistor goes DOWN 10-3 10-4 10-5 106 105 10-6 104 10-7 103 ’70 ’80 ’90 ’00 Source: WSTS/Dataquest/Intel 4 And Enables Energy Efficiency As the number of transistors goes UP Energy per transistor 10 100 1010 109 10-1 108 10-2 107 10-3 10-4 Energy per transistor goes DOWN 106 105 10-5 10-6 104 10-7 103 ’70 ’80 ’90 ’00 Source: WSTS/Dataquest/Intel ~ 1 Million Factor Reduction In Energy/Transistor Over 30+ Years 5 Three Decades of Scaling Contact 1978 22nm SRAM Cell 2012 1 mm 2012 22nm SRAM cell is dwarfed by a 1978 SRAM cell CONTACT 29 6 Moore’s Law: circa 2008 Intel® Atom™ - dual-core Rice – single grain 47 Millions Transistors 45nm node Hi-k Metal Gate 193 dry Litho In 2014, on 14nm technology, the above chip would be 1/8 the size - Much smaller than the grain of rice! 7 Classical Transistor Scaling Id = W/L * m * Cox * (Vg-Vt)2 Dennard, Gaenssalen, Yu, Rideout, Bassous, LeBlanc, 1974 Classical Scaling drove transistor performance improvement for decades 8 Dimensional Scaling Source Wavelength scaling is slowing Id = W/L * m * Cox * (Vg-Vt)2 9 Mobility Scaling Mobility degrades with transistor scaling Id = W/L * m * Cox * (Vg-Vt)2 T. Ghani VLSI ‘00 10 Gate Oxide Scaling SiO2 scaling running out of atoms Does Cox stop scaling? Id = W/L * m * Cox * (Vg-Vt)2 1000 Poly SiON 100 Silicon 10 1 0.1 1 Gate Leakage (Rel.) Electrical (Inv) Tox (nm) 10 0.01 350nm 250nm 180nm 130nm 90nm 65nm 11 Voltage Scaling As VDD is scaled down below 1V, drive current reduces sigignificantly if VT cannot be reduced – Can VT be reduced? Id = W/L * m * Cox * (Vg-Vt)2 12 Classical Scaling Id = W/L * m * Cox * (Vg-Vt)2 Dennard, Gaenssalen, Yu, Rideout, Bassous, LeBlanc, 1974 Classical Scaling drove transistor performance improvement for decades But is it over? 13 Is the End of Scaling Near? © 2013 Jupiterimages Corp. 14 Outline Transistor Scaling – The Dennard Era – Moore’s Law – Dennard Scaling Transistor Scaling: The Age of Innovation – Strain – High-k/Metal Gate – Tri Gate – Dual/Quad Patterning Transistor Evolution: The Future 15 The Age of Material & Structure Innovation 2003 90 nm SiGe 2005 65 nm 2007 45 nm 2009 32 nm 2011 22 nm Gate-Last High-k Metal Gate 2nd Gen. Gate-Last High-k Metal Gate Tri-Gate SiGe SiGe 2nd Gen. SiGe Strained Silicon Strained Silicon 16 New Materials are key enablers 17 Uniaxial Strain Uniaxial strain structure has provided >2X improvement in carrier inversion mobility Id = W/L * m * Cox * (Vg-Vt)2 High Stress Film SiGe SiGe 18 Biaxial Tensile Strain Biaxial tensile strain studied extensively in 1990s – Much of the seminal work on strain with this structure – Large electron mobility gain, but... Two key problems: n+ Gate 1. Integration difficulties – Dislocations – Ge Up-diffusion – Fast diffusion of extensions – VT Shift – Cost 2. Poor hole mobility gain STI S Strained Si D Relaxed Si1-xGex Graded Si1-yGey Si Substrate 19 Biaxial vs. Uniaxial Strain: Electrons Electron mobility enhancement similar – Both cases result from splitting of 6-fold degenerate valleys <001> <010> <100> D2 D4 20 Biaxial vs. Uniaxial Strain: Holes Uniaxial stress more effective than biaxial for hole mobility improvement – Simple calculations using peizoresistance coefficients [C.S. Smith, Phys. Rev., 1954] 40 <110> channel orientation (001 surface) % Hole Dm / m 35 30 Longitudinal compressive 25 20 Tranverse tensile 15 10 Compressive stress <001> ~12.5 % Ge (Biaxial in plane tension) 5 0 0 200 400 / Stress MPa 600 21 High-k + Metal Gate Id = W/L * m * Cox * (Vg-Vt)2 High-k gate dielectrics enabled inversion charge to continue scaling 1.0 nm EOT Hi-K Dual workfunction metal gate electrodes Requires replacement metal gate flow so that metal workfunction materials come AFTER high temperature steps 22 Replacement Metal Gate Flow - I High temperature steps for S/D and strain formation completed with dummy polysilicon gate N+ N+ p-well STI SiGe SiGe N-well 23 Replacement Metal Gate Flow - II N+ N+ STI SiGe p-well SiGe N-well Dummy Poly removal 24 Replacement Metal Gate Flow - III N+ N+ STI SiGe p-well SiGe N-well PMOS WF Metal 25 Replacement Metal Gate Flow - IV N+ N+ p-well STI SiGe SiGe N-well NMOS WF Metal deposition 26 Replacement Metal Gate Flow - V Low Resistance Al Fill NMOS WF N+ N+ High-k p-well PMOS WF STI SiGe SiGe High-k N-well High-k + Metal gate transistor formation complete 27 Serendipity Strain formation and Hi-k Metal Gate flow interact NICELY – 50% INCREASE in strain during dummy poly removal 28 Tri-Gate Transistors Tri-Gate transistors provide improved electrostatics Enables voltage scaling Higher drive current per unit footprint Reduced Threshold Voltage Channel Current Id = W/L * m * Cox * (Vg-Vt)2 (normalized) Gate Voltage (V) Reduced Operating Voltage 29 Transistor Features Gate Tri-gate transistors Source/ Drain 25 nm gate length 0.9 nm EOT Hi-K strained silicon Dual workfunction metal gate electrodes NMOS Gate Source/ Drain PMOS 30 Excellent short channel effects 1.E-02 0.80V 0.80V 0.05V 0.05V IDSAT (A/mm) 1.E-03 1.E-04 1.E-05 PMOS 1.E-06 NMOS 1.E-07 1.E-08 1.E-09 SS ~72mV/dec DIBL ~50 mV/V -1.0 -0.6 -0.2 SS ~69mV/dec DIBL ~46 mV/V 0.2 VGS (V) 0.6 1.0 Low Subthreshold slope and DIBL 31 Short Channel Effects 0.5 NMOS 0.4 32nm Vds=0.05V 0.3 Vt (V) 0.2 0.1 22nm 0 -0.1 22nm -0.2 -0.3 -0.4 32nm PMOS -0.5 25 30 35 40 Lgate (nm) Excellent VT rolloff – Enables 100mV reduction in Vt 32 Tri-Gate Self-Aligned Double Patterning 1. Pattern Sacrificial Layer 2. Deposit & Etch spacer 3. Sacrificial layer removal / etch 120nm pitch sacrificial layer substrate Spacer 60nm pitch substrate substrate 4. Etch 5. Spacer removal 60nm pitch substrate substrate substrate substrate Id = W/L * m * Cox * (Vg-Vt)2 17 33 Self-Aligned Contact (SAC) 1. Std process Thru gate polish 2. Recess Gate 3. Fill/Polish Etch Stop Metal Gate Spacer 4. Etch Contacts Etch Stop Contact Allows Contact to land on gate without short – Scaling of alignment tolerances 34 Self-Aligned Contact (SAC) Misalignment % die passing 100 80 SAC Non-SAC 60 40 20 0 -25 -20 -15 -10 -5 0 5 10 15 20 25 Gate - Contact Shift (nm) “Infinite” CTG window demonstrated 35 Transistor Scaling Continues 130nm 90nm 65nm 45nm 32nm 22nm 36 The Golden Age of Transistor Innovation 2003 90 nm SiGe 2005 65 nm 2007 45 nm 2009 32 nm 2011 22 nm Gate-Last High-k Metal Gate 2nd Gen. Gate-Last High-k Metal Gate Tri-Gate SiGe SiGe 2nd Gen. SiGe Strained Silicon Strained Silicon Strained Silicon High-k Metal Gate Tri-Gate Can innovation driven scaling continue? 37 Outline Transistor Scaling – The Dennard Era – Moore’s Law – Dennard Scaling Transistor Scaling: The Age of Innovation – Strain – High-k/Metal Gate – Tri Gate – Dual/Quad Patterning Transistor Evolution: The Future 38 Enabling a Steady Technology Cadence TECHNOLOGY GENERATION 2009 22nm 2011 14nm 2013 MANUFACTURING 10nm 2015 7nm 2017 DEVELOPMENT 5nm 2019 Beyond 2020 RESEARCH To be defined 2007 32nm Defined 45nm What to do now to enable these future generations ? 39 How Far Can We Go? Innovation-Enabled Technology Pipeline is Full 65nm 2005 45nm 2007 32nm 2009 22nm 2011 14nm 2013 MANUFACTURING 10nm 7nm 2015 Beyond … 2017 2019 DEVELOPMENT III-V 5nm RESEARCH 3-D EUV Interconnects 2021+ Graphene Dense Memory Photonics Materials Synthesis Nanowires *projected Our limit to visibility goes out ~10 years 41 Optimizing Choices for Transistors on Multiple Fronts Energy Band Diagram Increasing MOBILITY Si d SEM Micrograph Gate n-Ge Source Strain Drain Ge III-V InP CNT QW Graphene InAlAs Barriers Increasing COUPLING Planar With High K UTB SOI (or QW) Fins Wires/Ribbons 42 Production 22nm Tri-Gate Transistors ~400,000 atoms 1.E-02 IDSAT (A/mm) 1.E-03 1.E-04 Lg=26nm 0.80V 0.05V 0.05V 1.E-05 PMOS 1.E-06 NMOS 1.E-07 1.E-08 SS ~72mV/dec DIBL ~50 mV/V 1.E-09 -1.0 -0.6 -0.2 SS ~69mV/dec DIBL ~46 mV/V 0.2 VGS (V) 36 43 0.80V C. Auth, VLSI 2012 0.6 1.0 Research Nanowire Transistors ~40,000 atoms 1.E-03 Ids (A/mm) 1.E-05 1.E-07 SS ~ 61 mV/dec DIBL ~ 10 mV/V 1.E-09 SS ~ 67 mV/dec DIBL ~ 23 mV/V 1.E-11 Lg=30nm 1.E-13 -1.5 -1 -0.5 0 0.5 1 1.5 Vgs (V) K. Kuhn, IEDM 2012 44 III-V Progress Scorecard Integration of III-V on Si – Feasibility demonstrated using MBE Intel paper @ IEDM 2007 Enhancement-mode operation – Feasibility demonstrated – – MIT papers @ IEDM 2006 and 2007 Intel paper @ IEDM 2007 Research on surface prep, novel materials Scalability compared to Si devices unknown – In progress – – n+cap Work started on self-alignment, alternative geometries Modeling efforts underway at universities and internal Manufacturing tool feasibility – Research tool selected Source HiK S/D n+cap InAlAs Intel paper @ IEDM 2008 Ge PMOS QW devices may be alternatives Gate dielectric on III-V layers of interest – Demonstrated – HiK InP III-V hole mobility (P-type) not high enough – Strain demo – – S/D Gate – III-V Barrier In0.7Ga0.3As QW InAlAs Barrier In0.53Ga0.47As cap layer InP etch stop In0.52Al0.48As top barrier Drain 40nm gate length P-channel III-V In0.7Ga0.3As QW Gate P-channel QW with strain In0.52Al0.48As bottom barrier 45 Future Visibility: Interconnects Current Status Bottoms-up fill okay to about 14nm, barrier is the limiter No “better than Cu” option <14nm L/S might exceed dielectric breakdown limit 14nm filled trench Needed Focus Thin conformal plateable barrier … or self forming barrier Tall vias might use non-Cu Non-SiO2 dielectrics Exotic long interconnects: CNT (10’s um), optical (>mm) 5nm conformal Cu On-chip optical interconnect ~15nm Cu nanowire CNT 46 3-D Chip Stacking & Other ways to integrate + High density chip-chip connections Top Chip + Small form factor TSV Bottom Chip + Combine dissimilar technologies Package ? Added cost ? Degraded power delivery, heat sinking CPU TSV Memory Package ? Area impact on lower chip 3-D chip stacking using through-silicon vias 47 Beyond 2020 and possible futures Conventional fabrication architectures continue – Individual steps continue as 2D layers – More and more layers stacked to give increasing function High resolution TEM of graphene Graphene layers can couple together and create a quantum condensate Bilayer graphene structure Theoretically >10000x less power Source: M. Gilbert et.al J Comput Electron (2009) 48 Our limit to visibility goes out ~10 years TECHNOLOGY GENERATION 45nm 2007 32nm 2009 MANUFACTURING 22nm 2011 14nm 2013 10nm 2015 DEVELOPMENT 7nm 2017 Beyond 2020 RESEARCH Carbon Nanotube ~1nm diameter Graphene 1 atom thick QW III-V Device 5nm Not to scale Nanowire 10 atoms across 5 nm Silicon lattice is ~ 0.5nm, hard to imagine good devices smaller than 10 lattices across – reached in 2020 49 Is the End of Scaling Near? © 2013 Jupiterimages Corp. Rumors of my death are greatly exaggerated - Mark Twain 50 Discussion 51