Transcript ppt - SEAS
ESE370: Circuit-Level Modeling, Design, and Optimization for Digital Systems Day 37: December 8, 2010 Adiabatic Amplification 1 Penn ESE370 Fall2010 -- DeHon Today • It is possible to switch without dissipating energy? – Dissipate less than CV2 driving load C to voltage V? • Energy dissipation can be proportional to speed – Slower switching reduces energy – even without reducing V 2 Penn ESE370 Fall2010 -- DeHon Adiabatic • Adiabatic – a thermodynamic process without heat transfer 3 Penn ESE370 Fall2010 -- DeHon Day 16 Look at Energy E E P(t)dt I(t)V dt dd 4 Penn ESE370 Fall2010 -- DeHon Day 16 Capacitor Charging Energy E Vdd I(t)dt Q CV I(t)dt 2 E CVdd 5 Penn ESE370 Fall2010 -- DeHon Energy Dissipation • When we switch node to zero – Dump charge to ground • Every 010 transition burns CV2 6 Penn ESE370 Fall2010 -- DeHon Energy Recycling? • Can we avoid discarding the charge? – Can we recycle the energy rather than throwing it away? – Slogan: “Cycling” rather than “Dumping” • Two sub-problems: 1.Pool of reusable charge 2.Moving to/from pool without loss 7 Penn ESE370 Fall2010 -- DeHon Energy Dissipation • Where does the dissipated energy go? Dissipated across transistor charging resistance E [V dd Vout ]I(t)dt dVout I(t) C dt 8 Penn ESE370 Fall2010 -- DeHon Dissipation in R dVout dt E [Vdd Vout ]C dt dVout dVout dt C Vout dt E Vdd C dt dt 2 2 1 C E Vdd C Vdd C Vdd 2 2 2 Penn ESE370 Fall2010 -- DeHon 9 Conventional CMOS • Spend CV2 in 010 cycle – 0.5CV2 dissipated in pullup transistor charging – 0.5CV2 dissipated in pulldown transistor discharging 10 Penn ESE370 Fall2010 -- DeHon Challenge 2: Reduce Dissipation • Can we charge capacitor without dissipation? – With less dissipation? • Two sub-problems: 1.Pool of reusable charge 2.Moving to/from pool without loss 11 Penn ESE370 Fall2010 -- DeHon Adiabatic Switching Described Two Ways (same idea) First Way 12 Penn ESE370 Fall2010 -- DeHon Constant Current Charging • Er = P×T =I2RT • Charge over time T – Want this T to be a control variable • I=(CVdd)/T Penn ESE370 Fall2010 -- DeHon CVdd E r RT T 2 RC 2 E r CVdd T 13 Slow Switching (chilling out?) • If can charge with constant current – Energy dissipated is inversely proportional to charging time – Slower we charge, the less energy we dissipate RC 2 E r CVdd T 14 Penn ESE370 Fall2010 -- DeHon How Make Constant? • Why normally constant not current? – Input changing (Vgs) changing Ids – I(t) = [Vdd-Vout(t)]/R • Vout changing • How make I(t) constant? – Input settle with no voltage across supply/output – Make DV constant – Ramp Vsupply with Vout Penn ESE370 Fall2010 -- DeHon 15 Adiabatic Switching Second Way 16 Penn ESE370 Fall2010 -- DeHon Charging with Small DV • Energy cost is due to large DV drop over R – P=IDV • Adiabatic discipline: – Never turn on a device with a large voltage drop across it • Spend 0.5C(DV)2 to charge DV – Charge in many small steps N=V/DV 17 Penn ESE370 Fall2010 -- DeHon Charging with Small DV • • • • • • Spend 0.5C(DV)2 to charge DV Charge in many small steps N=V/DV Etotal = N 0.5C(DV)2 Etotal = (V/DV) 0.5C(DV)2 = 0.5CV×DV Etotal = 0.5CV2/N Time ~ RC per step RC – Same ratio as before E r CVdd T 2 18 Penn ESE370 Fall2010 -- DeHon Visually • Charge from Vdd – N+N-1+N-2+….2+1=N2/2 • Charge from Ramp – 1+1+1+….+1 = N 19 Penn ESE370 Fall2010 -- DeHon Adiabatic Amplifier 20 Penn ESE370 Fall2010 -- DeHon Adiabatic Amplifier • Discipline: – Set input X before switching Vsupply • Y=/Y=Vsupply – Ramp Vsupply slowly to charge Y or /Y – Return Vsupply to zero before change X • Adiabatically – Move charge to Y, /Y 21 Penn ESE370 Fall2010 -- DeHon Power Supply • Want power supply looks like slow ramp • Not clear how to produce without energy cost 22 Penn ESE370 Fall2010 -- DeHon “Ramped” Supply • Can produce sine waves with LC circuit – LC circuit moves charge without loss 23 Penn ESE370 Fall2010 -- DeHon Challenge 1: Reusable Charge • Can we borrow and return charge? • Two sub-problems: 1. Pool of reusable charge 2. Moving to/from pool without loss 24 Penn ESE370 Fall2010 -- DeHon Pulsed Supply • Pulse enable FET to allow charge to slosh into circuit (or back) 25 Penn ESE370 Fall2010 -- DeHon Pulsed Supply and Load 26 Penn ESE370 Fall2010 -- DeHon Resonant Supply • Charge moves back and forth between circuit and supply like RLC circuit – Some loss based on circuit R – Small if LC slow (adiabatic switching) – Only that loss that needs to be replaced • Costs energy 27 Penn ESE370 Fall2010 -- DeHon Energy Adiabatic Amplifier E load 2 2 Kn 2 C Vdd T Cn Vdd 2Vth shape factor since sine instead of ramp >1for sine wave (~1.2) 28 Penn ESE370 Fall2010 -- DeHon Vdd Selection E load 2 2 Kn 2 C Vdd T Cn Vdd 2Vth • Minimize with Vdd=4Vth E load 2 16K n C Vth T Cn 29 Penn ESE370 Fall2010 -- DeHon Leakage and Vth • Concern with this solution – Runs slow, high leakage – Possibly compensate with large Vth • Need to run even slower • Traditional voltage scaling – Limited V scaling • Variation and leakage – Preventing us from scaling V down • Sets a lower bound on Energy/Operation • Saves energy without scaling down Vdd 30 Penn ESE370 Fall2010 -- DeHon Critical Questions • Can we make the supplies efficient enough? – Avoid just moving E loss to supplies • Can make sufficiently efficient resonator? • Can we get sufficiently good inductors? • Can contain leakage sufficiently? 31 Penn ESE370 Fall2010 -- DeHon Next Time • Asymptotically Zero Energy Computation? – Thermodynamically possible? – Connection between information and energy – Reversibility 32 Penn ESE370 Fall2010 -- DeHon Admin • Proj3b Friday • Review for final: Monday – Andrew 33 Penn ESE370 Fall2010 -- DeHon Idea • Asymptotically Zero Energy Switching – Energy proportional T-1 – Slower we switch, the more we save • Alternate to reducing Vdd • Two sub-problems: 1.Pool of reusable charge 2.Moving to/from pool without loss 34 Penn ESE370 Fall2010 -- DeHon