Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Sushant

1,071 views

Published on

THIS IZ ABOUT FIN FETS ....

  • Be the first to comment

Sushant

  1. 1. 1 Presented by :- SUSHANT MISHRA EC-2 Under the guidance of :- FinFETs: From Circuit to Architecture
  2. 2. Talk Outline <ul><li>Background </li></ul><ul><li>Low Power FinFET Circuits </li></ul><ul><ul><li>Unusual Logic Styles </li></ul></ul><ul><ul><li>Unusual Dual-V dd /Dual-V th Circuits </li></ul></ul><ul><li>Architectural Impact </li></ul><ul><li>Other Ongoing Work </li></ul><ul><li>Conclusions </li></ul>2
  3. 3. Why Double-gate Transistors ? <ul><li>DG-FETs can be used to fill this gap </li></ul><ul><li>DG-FETs are extensions of CMOS </li></ul><ul><ul><li>Manufacturing processes similar to CMOS </li></ul></ul><ul><li>Key limitations of CMOS scaling addressed through </li></ul><ul><ul><li>Better control of channel from transistor gates </li></ul></ul><ul><ul><li>Reduced short-channel effects </li></ul></ul><ul><ul><li>Better Ion/Ioff </li></ul></ul><ul><ul><li>Improved sub-threshold slope </li></ul></ul><ul><ul><li>No discrete dopant fluctuations </li></ul></ul>3 Non-Si nano devices Bulk CMOS Feature size 32 nm 10 nm DG-FETs Gap
  4. 4. What are FinFETs? <ul><li>Fin-type DG-FET </li></ul><ul><ul><li>A FinFET is like a FET, but the channel has been “turned on its edge” and made to stand up </li></ul></ul>4 Si Fin
  5. 5. Independent-gate FinFETs <ul><li>Both the gates of a FET can be independently controlled </li></ul><ul><li>Independent control </li></ul><ul><ul><li>Requires an extra process step </li></ul></ul><ul><ul><li>Leads to a number of interesting analog and digital circuit structures </li></ul></ul>5 Back Gate Oxide insulation
  6. 6. FinFET Width Quantization <ul><li>Electrical width of a FinFET with n fins: W = 2* n * h </li></ul><ul><li>Channel width in a FinFET is quantized </li></ul><ul><li>Width quantization is a design challenge if fine control of transistor drive strength is needed </li></ul><ul><ul><ul><li>E.g., in ensuring stability of memory cells </li></ul></ul></ul>6 FinFET structure Ananthan, ISQED’05
  7. 7. Talk Outline <ul><li>Background </li></ul><ul><li>Low Power FinFET Circuits </li></ul><ul><ul><li>Unusual Logic Styles </li></ul></ul><ul><ul><li>Unusual Dual-V dd /Dual-V th Circuits </li></ul></ul><ul><li>Architectural Impact </li></ul><ul><li>Other Ongoing Work </li></ul><ul><li>Conclusions </li></ul>
  8. 8. Motivation: Power Consumption <ul><li>Traditional view of CMOS power consumption </li></ul><ul><ul><li>Active mode: Dynamic power (switching + short circuit + glitching) </li></ul></ul><ul><ul><li>Standby mode: Leakage power </li></ul></ul><ul><li>Problem: rising active leakage </li></ul><ul><ul><li>40% of total active mode power consumption (70nm bulk CMOS) † </li></ul></ul>† J. Kao, S. Narendra and A. Chandrakasan, “Subthreshold leakage modeling and reduction techniques,” in Proc. ICCAD, 2002.
  9. 9. Logic Styles: NAND Gates SG-mode NAND IG-mode NAND LP-mode NAND IG/LP-mode NAND pull up bias voltage pull down bias voltage IG-mode pull up LP-mode pull down
  10. 10. Comparing Logic Styles † Average leakage current for two-input NAND gate (V dd = 1.0V) Design Mode Advantages Disadvantages SG Fastest under all load conditions High leakage † (1 μ A) LP Very low leakage (85nA), low switched capacitance Slowest, especially under load. Area overhead (routing) IG Low area and switched capacitance Unmatched pull-up and pull-down delays. High leakage (772nA) IG/LP Low leakage (337nA), area and switched capacitance Almost as slow as LP mode
  11. 11. FinFET Circuit Power Optimization <ul><li>Construct FinFET-based Synopsys technology libraries </li></ul><ul><li>Extend linear programming based cell selection † for FinFETs </li></ul><ul><li>Use optimized netlists to compare logic styles at a range of delay constraints </li></ul>† D. Chinnery and K. Keutzer, “Linear programming for sizing, V dd and V t assignment,” in Proc. ISLPED , 2005. Benchmark Minimum-delay synthesis in Design Compiler SG-mode netlist Power-optimized mixed-mode netlists SG+ IG/LP SG+IG SG+LP Linear programming based cell selection 32 nm PTM FinFET models Delay/power characterization in SPICE LP IG/LP IG SG Synopsys libraries 32 nm PTM inFET models FinFET models (UFDG, PTM) Logic gate designs Logic gate designs
  12. 12. Power Consumption of Optimized Circuits <ul><li>Leakage power savings </li></ul><ul><li>110% a.t. (68.5%) </li></ul><ul><li>120% a.t. (80.3%) </li></ul><ul><li>Total power savings </li></ul><ul><li>110% arrival time (a.t.) (34%) </li></ul><ul><li>120% a.t. ( 47.5%) </li></ul>Estimated total power consumption for ISCAS’85 benchmarks V dd = 1.0V, α = 0.1, 32nm FinFETs Available modes
  13. 13. Talk Outline <ul><li>Background </li></ul><ul><li>Low Power FinFET Circuits </li></ul><ul><ul><li>Unusual Logic Styles </li></ul></ul><ul><ul><li>Unusual Dual-V dd /Dual-V th Circuits </li></ul></ul><ul><li>Architectural Impact </li></ul><ul><li>Other Ongoing Work </li></ul><ul><li>Conclusions </li></ul>
  14. 14. Dual-V dd FinFET Circuits <ul><li>Conventional low- </li></ul><ul><li>power principle: </li></ul><ul><ul><li>1.0V V dd for critical logic, 0.7V for off-critical paths </li></ul></ul><ul><li>Our proposal: overdriven gates </li></ul><ul><ul><li>Overdriven FinFET gates leak a lot less! </li></ul></ul>1.08V 1V Leakage current Vin Reverse bias V gs =+0.08V Overdriven inverter Higher V th
  15. 15. <ul><li>Using only two V dd ’s saves leakage only in P-type FinFETs, but not in N-type FinFETs </li></ul><ul><li>Solution </li></ul><ul><ul><li>Use a negative ground voltage (V H ss ) to symmetrically save leakage in N-type FinFETs </li></ul></ul>V th Control with Multiple V dd ’s (TCMS) V dd H V ss H V dd L V ss L TCMS buffer Symmetric threshold control for P and N V dd H 1.08V V dd L 1.0V V ss H -0.08V V ss L 0.0V
  16. 16. Exploratory Buffer Design <ul><li>Size of high-V dd inverters kept small to minimize leakage in them </li></ul><ul><li>Wire capacitances not driven by high-V dd inverters </li></ul><ul><li>Output inverter in each buffer overdriven and its size (and switched capacitance) can be reduced </li></ul>l opt S 1 S 2 V H dd V H ss V L ss V L dd S 1 S 2 V H dd V H ss V L ss V L dd i i’
  17. 17. Power Savings <ul><li>Benchmarks are nets extracted from real layouts and scaled to 32nm </li></ul><ul><ul><li>http://dropzone.tamu.edu/~zhouli/GSRC/fast_buffer_insertion.html </li></ul></ul>Power component Savings Dynamic power -29.8% Leakage power 57.9% Total power 50.4%
  18. 18. Fin-count Savings <ul><li>Transistor area is measured as the total number of fins required by all buffers </li></ul><ul><li>TCMS can save 9% in transistor area </li></ul>
  19. 19. TCMS Extension Delay-minimized netlist Power : 283.6uW Area: 538 fins Power-optimized netlist Power : 149.9uW Area: 216 fins
  20. 20. Power Reduction (ISCAS’85 Benchmarks)
  21. 21. Power-minimized vs Delay-minimized Netlists at 130% ATC TCMS TCMS (Single-Vth Dual-Vdd % reduction in dynamic power 53.3 49.8 51.4 % reduction in leakage power 95.8 95.7 95.8 % reduction in total power 67.6 65.3 66.3 % reduction in Fin-count 65.2 59.5 61.6
  22. 22. Talk Outline <ul><li>Background </li></ul><ul><li>Low Power FinFET Circuits </li></ul><ul><ul><li>Unusual Logic Styles </li></ul></ul><ul><ul><li>Unusual Dual-V dd /Dual-V th Circuits </li></ul></ul><ul><li>Architectural Impact </li></ul><ul><li>Other Ongoing Work </li></ul><ul><li>Conclusions </li></ul>
  23. 23. Orion-FinFET <ul><li>Extends ORION for FinFET-based power simulation for interconnection networks </li></ul><ul><li>FinFET power libraries for various temperatures and technologies nodes </li></ul><ul><li>Power breakdown of interconnection networks for different FinFET modes </li></ul><ul><li>Power comparison for different FinFET modes under different traffic patterns </li></ul>
  24. 24. Router Microarchitecture & Pipeline Stages
  25. 25. Power Simulation Flow
  26. 26. Power Breakdown for SG/LP Modes <ul><li>4X4 mesh network: 5 ports/router, 48-flit buffer/port </li></ul><ul><li>Flit width = 128 bits </li></ul><ul><li>Clock frequency = 1GHz </li></ul>Router power breakdown Network power breakdown
  27. 27. Bulk CMOS vs. LP-mode FinFETs <ul><li>Bulk CMOS simulation: 32nm predictive technology model </li></ul><ul><li>Leakage power of bulk CMOS network 2.68X as compared to an LP-mode FinFET network </li></ul>
  28. 28. Router Leakage Power vs. Temp. <ul><li>Leakage power of SG-mode router grows much faster with temp. than for LP-mode </li></ul><ul><li>Leakage power ratio at 105 o C: 7:1 </li></ul>
  29. 29. Talk Outline <ul><li>Background </li></ul><ul><li>Low Power FinFET Circuits </li></ul><ul><ul><li>Unusual Logic Styles </li></ul></ul><ul><ul><li>Unusual Dual-V dd /Dual-V th Circuits </li></ul></ul><ul><li>Architectural Impact </li></ul><ul><li>Other ongoing work </li></ul><ul><li>Conclusions </li></ul>
  30. 30. FinFET SRAM and Embedded DRAM Design <ul><li>FinE: Two-tier FinFET simulation framework for FinFET circuit design space exploration: </li></ul><ul><ul><li>Sentaurus TCAD+UFDG SPICE model </li></ul></ul><ul><ul><li>Quasi Monte-Carlo simulation for process variation analysis </li></ul></ul><ul><ul><li>Thermal analysis using ThermalScope </li></ul></ul><ul><ul><li>Yield estimation </li></ul></ul><ul><li>Variation-tolerant ultra low-leakage FinFET SRAMs at lower technology nodes </li></ul><ul><li>Gated-diode FinFET embedded DRAMs </li></ul>
  31. 31. Extension of CACTI for FinFETs <ul><li>Selection of any of the FinFET SRAM and embedded DRAM cells </li></ul><ul><li>Use of any of the FinFET operating modes </li></ul><ul><li>Scaling of FinFET designs from 32nm to 22nm, 16nm and 10nm technology nodes </li></ul><ul><li>Accurately modeling the behavior of a wide range of cache configurations </li></ul>
  32. 32. FPGA vs. ASICs NATURE CMOS fabrication compatible Nano RAM on-chip storage Run-time reconfiguration Temporal logic folding Design flexibility Logic density <ul><li>Distributed non-volatile nano RAMs: main storage for reconfiguration bits </li></ul><ul><li>Fine-grain reconfiguration (even cycle-by-cycle) and logic folding </li></ul><ul><ul><li>More than an order of magnitude increase in logic density and area-delay product </li></ul></ul><ul><ul><li>Competitive performance and moderate power consumption </li></ul></ul><ul><ul><li>Non-volatility: useful in low power & secure processing </li></ul></ul><ul><li>NanoMap to map application to NATURE </li></ul><ul><ul><li>Significant area-delay trade-off flexibility </li></ul></ul>
  33. 33. Conclusions <ul><li>FinFETs a necessary semiconductor evolution step because of bulk CMOS scaling problems beyond 32nm </li></ul><ul><li>Use of the FinFET back gate leads to very interesting design opportunities </li></ul><ul><li>Rich diversity of design styles, made possible by independent control of FinFET gates, can be used effectively to reduce total active power consumption </li></ul><ul><li>TCMS able to reduce both delay and subthreshold leakage current in a logic circuit simultaneously </li></ul><ul><li>Time has arrived to start exploring the architectural trade-offs made possible by switch to FinFETs </li></ul>

×