L16: 6.111 Spring 20071Introductory Digital Systems LaboratoryL16: Power Dissipation in Digital SystemsL16: Power Dissipation in Digital SystemsL16: 6.111 Spring 20072Introductory Digital Systems LaboratoryProblem #1: Power Dissipation/HeatProblem #1: Power Dissipation/Heat5KW 18KW 1.5KW 500W 40048008808080858086286386486Pentium® proc0.11101001000100001000001971 1974 1978 1985 1992 2000 2004 2008YearPower (Watts)40048008808080858086286386486Pentium® procP61101001000100001970 1980 1990 2000 2010YearPower Density (W/cm2)Hot PlateNuclearReactorRocketNozzleHow do you cool these chips??How do you cool these chips??chipheat sinkSun’sSurfaceCourtesy Intel (S. Borkar)L16: 6.111 Spring 20073Introductory Digital Systems LaboratoryProblem #2: Energy ConsumptionProblem #2: Energy Consumption(40+ lbs)BatteryYearNominalCapacity (Watt-hours/lb)Nickel-CadmiumNi-Metal Hydride65 70 75 80 85 90 95 0 10 20 30 40 50 Rechargable Lithium(from Jon Eager, Gates Inc. , S. Watanabe, Sony Inc.)No Moore’s law for batteries…Today: Understand where power goesand ways to manage itWhat can One Jouleof energy do?Send a 1 Megabyte file over 802.11bOperate a processor for ~ 7sThe Energy Problem7.5 cm3AA batteryAlkaline: ~10,000JMow your lawn for 1 msL16: 6.111 Spring 20074Introductory Digital Systems LaboratoryDynamic Energy DissipationDynamic Energy DissipationVDDCLE0→1= CLVDD2Ecap= 1/2CLVDD2iDDEdiss, RP= 1/2CLVDD2VDDCLIN =1Ediss,RN=1/2CLVDD2ChargingDischargingIN =0P = CLVDD2 fclkRNRPRNRPL16: 6.111 Spring 20075Introductory Digital Systems LaboratoryThe Transition Activity Factor The Transition Activity Factor αα00−−>>11Output TransitionNext InputCurrent Input0 −> 011110 −> 110110 −> 101110 −> 100111 −> 011101 −> 110101 −> 101101 −> 100101 −> 011011 −> 110011 −> 101011 −> 100011 −> 011001 −> 110001 −> 101001 −> 10000α0−>1= 3/16Assume inputs (A,B) arrive at f and are uniformly distributedWhat is the average power dissipation?P = α0−>1CLVDD2 fZABL16: 6.111 Spring 20076Introductory Digital Systems LaboratoryJunction (Silicon) TemperatureJunction (Silicon) TemperatureSimple ScenarioTj-Ta=RθJAPDSiliconRθJAis the thermal resistance between silicon and AmbientRθJAPDTj=Ta+ RθJAPDMake this as low as possibleRealistic ScenarioRθJCPDRθCA= RθCS +RθSA SinkCaseSiliconTJTATJTCTSTATJTCTSTARθCSRθSAis minimized by facilitating heat transfer (bolt case to extended metal surface – heat sink)L16: 6.111 Spring 20077Introductory Digital Systems LaboratoryIntel Pentium 4 Thermal GuidelinesIntel Pentium 4 Thermal Guidelines Pentium 4 @ 3.06 GHz dissipates 81.8W! Maximum TC= 69 °C RCA< 0.23 °C/W for 50 C ambient Typical chips dissipate 0.5-1W (cheap packages without forced air cooling)Execution core120oCCache70°CInteger & FP ALUsTemp(oC)Courtesy of Intel (Ram Krishnamurthy)L16: 6.111 Spring 20078Introductory Digital Systems LaboratoryPower Reduction StrategiesPower Reduction Strategies Reduce Transition Activity or Switching Events Reduce Capacitance (e.g., keep wires short) Reduce Power Supply Voltage Frequency is typically fixed by the application, though this can be adjusted to control powerP = α0−>1CLVDD2 fOptimize at all levels of design hierarchyOptimize at all levels of design hierarchyL16: 6.111 Spring 20079Introductory Digital Systems LaboratoryClock Gating is a Good Idea!Clock Gating is a Good Idea!+XGlobal ClockAdder ClockMultiplier ClockAdder OffEnable_AdderEnable_MultiplierMultiplier On100’s of different clocks in a microprocessorClock Gating Reduces Energy, does it reduce Power?Clock Gating Reduces Energy, does it reduce Power?Clock gating reduces activityand is the most common low-powertechnique used todayL16: 6.111 Spring 200710Introductory Digital Systems LaboratoryDoes your GHz Processor run at a GHz? Does your GHz Processor run at a GHz? ProcessorThermalSensor Note that there is a difference between average and peak power On-chip thermal sensor (diode based), measures the silicon temperature If the silicon junction gets too hot (say 125 °C), then the activity is reduced (e.g., reduce clock rate or use clock gating)ChipActivity ControlUse of Thermal FeedbackUse of Thermal FeedbackL16: 6.111 Spring 200711Introductory Digital Systems LaboratoryPower Supply ResonancePower Supply ResonanceLboardLpackageRgridSwitchingcurrentsBoard decapOn-diedecapCourtesy of Motorola(David Blaauw)Courtesy of MotorolaCourtesy of Motorola(David Blaauw)(David Blaauw)200MhzDesignCan write a Virus to Activate Can write a Virus to Activate Power Supply Resonance!Power Supply Resonance!L16: 6.111 Spring 200712Introductory Digital Systems LaboratoryNumber Representation:Number Representation:TwoTwo’’s Complement vs. Sign Magnitudes Complement vs. Sign MagnitudeTwo’s complement0000011100111011111111101101110010101001100001100101010000100001+0+1+2+3+4+5+6+7-0-1-2-3-4-5-6-7Sign-MagnitudeConsider a 16 bit bus where inputs togglesbetween +1 and –1 (i.e., a small noise input)Which representation is more energy efficient?L16: 6.111 Spring 200713Introductory Digital Systems LaboratoryBus Coding to Reduce ActivityBus Coding to Reduce ActivityMajorityFunctioninvertDQInputData BusNOutput[Stan94]Extra bit to indicated if thebus is invertedL16: 6.111 Spring 200714Introductory Digital Systems LaboratoryTime Sharing is a Bad IdeaTime Sharing is a Bad IdeaTime Sharing Increases Switching ActivityTime Sharing Increases Switching Activity2L16: 6.111 Spring 200715Introductory Digital Systems LaboratoryNot just a 6Not just a 6--1 Issue: 1 Issue: ““CoolCool””Software ???Software ???CPU01111111000000000111111100000001011111110000001001111111000000111000000000000000100000000000000110000000000000101000000000000011float a [256], b[256];float pi= 3.14;for (i = 0; i < 255; i++) {a[i] = sin(pi * i /256);}for (i = 0; i < 255; i++) {b[i] = cos(pi * i /256);}float a [256], b[256];float pi= 3.14;for (i = 0; i < 255; i++) {a[i] = sin(pi * i /256);b[i] = cos(pi * i /256);}a[0]a[1]a[2]a[3]b[0]b[1]b[2]b[3]addressMEMORYaddress16512(8)+2+4+8+16+32+64+128+256= 4607 bit transitions2(8)+2(2+4+8+16+32+64+128+256)= 1030 transitionsL16: 6.111 Spring 200716Introductory Digital Systems LaboratoryGlitchingGlitchingTransitionsTransitions Balancing paths reduces glitching transitions Structures such as multipliers have lot of glitching transitions Keeping logic depths short (e.g., pipelining) reduces glitching+++ABCD(A+B) + (C+D)+++ABCD(((A+B) + C)+D)Chain
View Full Document