Transcript ppt - SEAS
ESE534 Computer Organization Day 6: September 21, 2016 Energy, Power, Reliability 1 Penn ESE534 Fall 2016 -- Mehta & DeHon Today • Energy tradeoffs • Voltage limits and leakage • Variations • [more of the gory transistor equations…] 2 Penn ESE534 Fall 2016 -- Mehta & DeHon At Issue • Many now argue energy will be the ultimate scaling limit – (not lithography, costs, …) • Proliferation of portable and handheld devices – …battery size and life biggest issues • Cooling, energy costs may dominate cost of electronics – Even server room applications 3 Penn ESE534 Fall 2016 -- Mehta & DeHon Preclass 1 • 1GHz case – Voltage? – Energy per Operation? – Power required for 2 processors? • 2GHz case – Voltage? – Energy per Operation? – Power required for 1 processor? 4 Penn ESE534 Fall 2016 -- Mehta & DeHon Preclass 1 Lesson • What does Preclass 1 result tell us? 5 Penn ESE534 Fall 2016 -- Mehta & DeHon Energy and Delay 1 2 E CV 2 tgd=Q/I=(CV)/I Id,sat=(mCOX/2)(W/L)(Vgs-VTH 2 ) 6 Penn ESE534 Fall 2016 -- Mehta & DeHon Energy/Delay Tradeoff • EV2 • tgd1/V 1 E CV 2 2 tgd=(CV)/I Id,sat (Vgs-VTH)2 • We can trade speed (latency) for energy • E×(tgd)2 constant Martin et al. Power-Aware Computing, Kluwer 2001 http://caltechcstr.library.caltech.edu/308/ 7 Penn ESE534 Fall 2016 -- Mehta & DeHon Area/Time Tradeoff • Also have Area-Time tradeoffs – HW2 spatial vs temporal multipliers – See more next week • Compensate slowdown with additional parallelism – For throughput problems • …trade Area for Energy Architectural Option 8 Penn ESE534 Fall 2016 -- Mehta & DeHon Question • By how much can we reduce energy? • What limits us? 9 Penn ESE534 Fall 2016 -- Mehta & DeHon Challenge: Power 10 Penn ESE534 Fall 2016 -- Mehta & DeHon Origin of Power Challenge • Limited capacity to remove heat – ~100W/cm2 force air – 1-10W/cm2 ambient • Transistors per chip grow at Moore’s Law rate = (1/F)2 • Energy/transistor must decrease at this rate to keep constant power density • P/tr CV2f • E/tr CV2 – …but V scaling more slowly than F 11 Penn ESE534 Fall 2016 -- Mehta & DeHon ITRS Vdd Scaling: More slowly than F 12 Penn ESE534 Fall 2016 -- Mehta & DeHon ITRS CV2 Scaling: More slowly than (1/F)2 13 Penn ESE534 Fall 2016 -- Mehta & DeHon Origin of Power Challenge • Transistors per chip grow at Moore’s Law rate = (1/F)2 • Energy/transistor must decrease at this rate to keep constant • E/tr CV2 14 Penn ESE534 Fall 2016 -- Mehta & DeHon 15 Penn ESE534 Fall 2016 -- Mehta & DeHon Intel Power Density 16 Penn ESE534 Fall 2016 -- Mehta & DeHon Impact Power Limits Integration Density Limit Constant Power Limit 45nm 32nm 22nm 16nm 11nm Source: Carter/Intel 17 Penn ESE534 Fall 2016 -- Mehta & DeHon 17 Impact • Power density is limiting scaling – Can already place more transistors on a chip than we can afford to turn on! • Power is potential challenge/limiter for all future chips. – Only turn on small percentage of transistors? – Operate those transistors as much slower frequency? – Find a way to drop Vdd? 18 Penn ESE534 Fall 2016 -- Mehta & DeHon How far can we reduce Vdd? 19 Penn ESE534 Fall 2016 -- Mehta & DeHon Limits • Ability to turn off the transistor • Parameter Variations • Noise (not covered today) 20 Penn ESE534 Fall 2016 -- Mehta & DeHon MOSFET Conduction From: http://en.wikipedia.org/wiki/File:IvsV_mosfet.png Penn ESE534 Fall 2016 -- Mehta & DeHon 21 Transistor Conduction gate drain src channel • Three regions – Subthreshold (Vgs<VTH) – Linear (Vgs>VTH) and (Vds < (Vgs-VTH)) – Saturation (Vgs>VTH) and (Vds > (Vgs-VTH)) 22 Penn ESE534 Fall 2016 -- Mehta & DeHon Saturation Region • (Vgs>VTH) • (Vds > (Vgs-VTH)) Ids,sat=(mCOX/2)(W/L)(Vgs-VTH 2 ) 23 Penn ESE534 Fall 2016 -- Mehta & DeHon Linear Region • (Vgs>VTH) • (Vds < (Vgs-VTH)) Ids,lin=(mCOX)(W/L)((Vgs-VTH)Vds-(Vds)2/2) 24 Penn ESE534 Fall 2016 -- Mehta & DeHon Subthreshold Region • (Vgs<VTH) V Isub IVT 10 gs VTH / S S (ln( 10))kT / q [Frank, IBM J. R&D v46n2/3p235] 25 Penn ESE534 Fall 2016 -- Mehta & DeHon IDS vs. VGS 26 Penn ESE370 Fall2014 -- DeHon Operating a Transistor • Concerned about Ion and Ioff • Ion drive (saturation) current for charging – Determines speed: Tgd = CV/I • Ioff leakage current – Determines leakage power/energy: • Pleak = V×Ileak • Eleak = V×Ileak×Tcycle 27 Penn ESE534 Fall 2016 -- Mehta & DeHon Leakage • To avoid leakage want Ioff very small • Switch V from Vdd to 0 • Vgs in off state is 0 (Vgs<VTH) V Isub IVT 10 gs VTH Ioff IVT 10 / S VTH / S 28 Penn ESE534 Fall 2016 -- Mehta & DeHon Leakage Ioff IVT 10 • • • • VTH / S S90mV for single gate S70mV for double gate For lowest leakage, want S small, VTH large 4 orders of magnitude IVT/IoffVTH>280mV Leakage limits VTH in turn limits Vdd 29 Penn ESE534 Fall 2016 -- Mehta & DeHon How maximize Ion/Ioff ? • Maximize Ion/Ioff – for given Vdd ? EswCV2 • Get to pick VTH Id,sat=(mCOX/2)(W/L)(Vgs-VTH 2 ) Id,lin=(mCOX)(W/L)(Vgs-VTH)Vds-(Vds)2/2 V Isub IVT 10 gs VTH Penn ESE534 Fall 2016 -- Mehta & DeHon / S 30 IDS vs. VGS 31 Penn ESE370 Fall2014 -- DeHon Preclass 2 • E = Esw + Eleak • Eleak = V×Ileak×Tcycle • EswCV2 V Isub IVT 10 gs VTH • Ichip-leak = Ndevices ×Itr-leak 32 Penn ESE534 Fall 2016 -- Mehta & DeHon / S Preclass 2 • Eleak(V) ? • Tcycle(V)? 33 Penn ESE534 Fall 2016 -- Mehta & DeHon In Class • Assign calculations – SIMD – each student computes for a different Voltage • Collect results on board • Group: Identify minimum energy point and discuss 34 Penn ESE534 Fall 2016 -- Mehta & DeHon Graph for In Class 36 Penn ESE534 Fall 2016 -- Mehta & DeHon Impact • Subthreshold slope prevents us from scaling voltage down arbitrarily. • Induces a minimum operating energy. 37 Penn ESE534 Fall 2016 -- Mehta & DeHon Challenge: Variation 38 Penn ESE534 Fall 2016 -- Mehta & DeHon Statistical Dopant Count and Placement 39 Penn ESE534 Fall 2016 -- Mehta & DeHon [Bernstein et al, IBM JRD 2006] Vth Variability @ 65nm 40 Penn ESE534 Fall 2016 -- Mehta & DeHon [Bernstein et al, IBM JRD 2006] Variation • Fewer dopants, atoms increasing Variation • How do we deal with variation? % variation in VTH (From ITRS prediction) 41 Penn ESE534 Fall 2016 -- Mehta & DeHon 41 Impact of Variation? • Higher VTH? – Not drive as strongly slower – Id,sat (Vgs-VTH)2 • Lower VTH? – Not turn off as well leaks more Ioff IVT 10 VTH / S 42 Penn ESE534 Fall 2016 -- Mehta & DeHon Variation • Margin for expected variation • Must assume VTH can be any value in range Ion,min=Ion(Vth,max) Probability Distribution Id,sat (Vgs-VTH)2 Vgs = Vdd VTH 43 Penn ESE534 Fall 2016 -- Mehta & DeHon Margining • Must raise Vdd to increase drive strength • Increase energy Ion,min=Ion(Vth,max) Probability Distribution Id,sat (Vgs-VTH)2 Vgs = Vdd VTH 44 Penn ESE534 Fall 2016 -- Mehta & DeHon Variation • Increasing variation forces higher voltages – On top of our leakage limits 45 Penn ESE534 Fall 2016 -- Mehta & DeHon 45 • Margins growing due to increasing variation Probability Distribution Variations Old New Delay • Margined value may be worse than older technology? 46 Penn ESE534 Fall 2016 -- Mehta & DeHon End of Energy Scaling? Black nominal Grey with variation [Bol et al., IEEE TR VLSI Sys 17(10):1508—1519]47 Penn ESE534 Fall 2016 -- Mehta & DeHon Minimum Energy vs Technology • Similar experiment for FPGA logic 6 5 4 3 Energy Increase Energy Increase 2 • …and a hint at how to mitigate Energy Decrease 1 [Mehta PhD thesis 2012 (refined from FPGA 2012)] Penn ESE534 Fall 2016 -- Mehta & DeHon 48 Chips Growing • Larger chips (billions of transistors) sample further out on distribution curve From: http://en.wikipedia.org/wiki/File:Standard_deviation_diagram.svg 49 Penn ESE534 Fall 2016 -- Mehta & DeHon Big Ideas • Can trade time for energy – … area for energy • Variation and leakage limit voltage scaling • Power major limiter going forward – Can put more transistors on a chip than can switch • Continued scaling demands – Deal with noisier components • High variation • … other noise sources 50 Penn ESE534 Fall 2016 -- Mehta & DeHon Admin • Homework 2 due Tonight • Reading for Monday on web • Homework 3 due next Wednesday 51 Penn ESE534 Fall 2016 -- Mehta & DeHon