《信息科学类专业英语》课件第7章

Lesson7ConfigurableComputing(第七课可重构计算)

Vocabulary(词汇)ImportantSentences(重点句)QuestionsandAnswers(问答)Problems(问题)ReadingMaterial(阅读材料)

Computersthatmodifytheirhardwarecircuitsastheyoperateareopeninganeweraincomputerdesign.Becausetheycanfilterdatarapidly,theyexcelatpatternrecognition,imageprocessingandencryption

Computerdesignersfaceaconstantstruggletofindtherightbalancebetweenspeedandgenerality.Theycanbuildversatilechipsthatperformmanydifferentfunctionsrelativelyslowly,ortheycandeviseapplication-specificchipsthatdoonlyalimitedsetoftasksbutdothemmuchmorequickly.Microprocessors(suchastheIntelPentiumorMotorolaPowerPCchipscommonlyfoundinpersonalcomputers)aregeneralpurpose:programminginstructionsencodedinbinaryformatcanleadamicroprocessorthroughvirtuallyanylogicalormathematicaloperationaprogrammercanconceive.TheIntelPentium,forexample,wasneverdesignedspecificallytoexecuteeitherMicrosoftWordorthecomputergameDOOM,butitcanrunboth.[1]1ConfigurableComputing

Incontrast,customhardwarecircuits,oftenknownasApplication-SpecificIntegratedCircuits(ASICs),providepreciselythefunctionalityneededforaspecifictask.BycarefullytuningeachASICtoagivenjob,thecomputerdesignercanproduceasmaller,cheaper,fasterchipthatconsumeslesspowerthanaprogrammableprocessor.AcustomgraphicschipforaPC,forinstance,candrawlinesorpaintpicturesonthescreen10or100timesasquicklyasageneral-purposecentralprocessingunitcan.[2]

Asdesignersmaketheirchoicesbetweenversatilityandspeed,theymustalsoconfronttheissueofcost.Awell-designedASICwillsolvethespecificproblemforwhichitwasdesigned,butprobablynotaslightlymodifiedproblemintroducedaftertheASICdesignisfinished.Furthermore,evenifamodifiedASICcanbedevelopedforthenewproblem,theoriginalhardwarecircuitsmaybetoohighlycustomizedtobereusedinsuccessivegenerations.Asaresult,theengineeringeffortrequiredtodesignandbuildanASICmustbeamortizedoverarelativelysmallnumberofunits.2ProgrammableCircuitry

Anewdevelopmentinintegratedcircuitsoffersathirdoption:large,fast,field-programmablegatearrays,orFPGAs—highlytunedhardwarecircuitsthatcanbemodifiedatalmostanypointduringuse.FPGAsconsistofarraysofconfigurablelogicblocksthatimplementthelogicalfunctionsofgates.Logicgatesarelikeswitcheswithmultipleinputsandasingleoutput.TheyareusedindigitalcircuitstoperformbasicbinaryoperationssuchasAND,NAND,OR,NORandXOR.Inmosthardwarethatisusedincomputingtoday,thelogicalfunctionsofgatesarefixedandcannotbemodified.

InFPGAs,however,boththelogicfunctionsperformedwithinthelogicblocksandtheconnectionsbetweentheblockscanbealteredbysendingsignalstothechip.TheseblocksarestructurallysimilartothegatearraysusedinsomeASICs,butwhereasstandardgatearraysareconfiguredduringmanufacture,theconfigurablelogicblocksinFPGAscanberewiredandreprogrammedrepeatedly,longaftertheintegratedcircuithasleftthefactory.[3]

ThekeythathasopenedthedoortoconfigurablecomputingisthedesignofnewFPGAsthatcanbeconfiguredextremelyquickly.Theearliestfield-programmablearraysrequiredseveralsecondsormoretochangetheirconnections—perfectlysuitableforengineerswhowantedtotestalternativecircuitdesignsorforcompaniesthatsolddevicesthatmightneedoccasionalupgrading.NewerFPGAscanbeconfiguredinonemillisecond,andweexpecttoseedeviceswithconfigurationtimesaslowas100microsecondswithintwoyears.Ultimately,computingdevicesmaybeabletoadapttheirhardwarealmostcontinuouslyinresponsetochangesintheinputdataorprocessingenvironment.

TherearemanyvariationsonFPGAdesign,butthebasicstructureconsistsofalargenumberofconfigurablelogicblocksandaprogrammablegridofconnectionsthatcanlinkthoseblocksinanypatternthedesignerchooses.ThoseFPGAsthatarecoarsegrainedhaveasmallnumberofpowerfulconfigurablelogicblocks;thosewithafiner-grainedstructurehavemanysimpleblocks.Asingleelementinacoarse-grainedFPGAmightbecapableofaddingorcomparingtwonumbers.Oneblockinafine-graineddevicemightbecapableonlyofcomparingtwobinarydigits—ineffect,itwouldbeasinglelogicgate.Adesignermightchoosetostartwitheitheracoarse-graintedorfine-grainedchipdependingontheapplicationathandandtheamountoftimeavailableforbuildingcomplexsubsystemsfromscratch.

Computingdevicescanmakeuseofconfigurableelementsinmanydifferentways.Theleastdemandingtechniqueistoswitchbetweenfunctionsoncommand—thehardwareequivalentofquittingoneprogramandthenrunninganother.Slowreconfiguration,ontheorderofseveralseconds,maywellbeacceptableinsuchanapplication.Fasterprogrammingtimespermitdynamicdesignswapping:asingleFPGAperformsaseriesoftasksinrapidsuccession,reconfiguringitselfbetweeneachone.

Suchdesignsoperatethechipinatime-sharingmodeandswapbetweensuccessiveconfigurationssorapidlythatitappearstheFPGAisperformingallitsfunctionsatonce.[4]Usingthisapproach,wehavebuiltasingle-chipvideotransmissionsystemthatreconfiguresitselffourtimespervideoframe.ItthusrequiresonlyaquarterofthehardwarethatwouldbeneededforafixedASIC.TheFPGAfirststoresanincomingvideosignalinmemory,thenappliestwodifferentimage-processingtransformationsandfinallytransformsitselfintoamodemtosendthesignalonward.

Themostchallengingandpotentiallymostpowerfulformofconfigurablecomputinginvolvesthehardwarereconfiguringitselfontheflyasitexecutesatask,refiningitsownprogrammingforimprovedperformance.Animage-recognitionchipmighttuneitselfinresponsetoatentativeidentificationoftheobjectitislookingat:ifanimagecontainedacaroratruck,partsofthecircuitryoriginallyintendedfortrackinghigh-speedaircraftorslow-movingpeoplecouldbereconfiguredtofocusinsteadonlandvehicles.Forsomeapplications,sucharadicaldeparturefromtraditionalcomputerdesign,inwhichthehardwareisspecifiedattheoutset,couldmakeformuchfasterandmoreversatilemachinesthanarepossiblewitheithergeneral-purposemicroprocessorsorcustomchips.3CuttingCriticalHardware

Oneofthemostpromisingapplicationsforconfigurablecomputinginvolvespatternmatching.Patternmatchingisusedintaskssuchashandwritingrecognition,faceidentification,databaseretrievalandautomatictargetrecognition.Afundamentaloperationofpatternmatchinginvolvescomparinganinputsetofbits(representinganimage,astringofcharactersorotherdata)withasetoftemplatescorrespondingtothepossiblepatternstoberecognized.Thesystemdeclaresrecognitionwhenthenumberofinputbitsthatmatchbitsinthetemplateexceedssomethreshold.[5]

Inthecaseoftargetrecognition—amilitaryapplicationthatdrovesomeofourinitialwork—thegreatestchallengeistherapidcomparisonofaninputimagetothousandsoftemplates.Atemplatecouldrepresent,forexample,afrontorsideviewofaspecifictypeofvehicle.Eachimagetypicallycontainsthousandsofpixels(pictureelements),andatargetcouldappearatanypositionwithinanimage.Torecognizetargetsfastenoughformilitaryapplications,asystemneedstoperformcomparisonsattherateofseveraltrillionoperationspersecond,becauseallthepixelsintheinputimagemustbecomparedwithallthepixelsinmanytemplates.

WithsupportfromtheDefenseAdvancedResearchProjectsAgency(DARPA),wehavebuiltaprototyperecognitionsystemwithconfigurablehardwarethatachievessignificanthardwaresavingsbytuningitselftoeachtemplateinturn.Manyofthepixelsinatypicaltemplatedonotcontributetothematchingresults,andsotheconfigurablecomputingmachinecouldsimplyomitthemfromitscalculations.Aconventionalsystemcouldnoteasilypareitselfdowninasimilarway,becausethepixelstobeignoreddifferfromtemplatetotemplate.

Onecangofurtherinexploitingtheflexibilityofconfigurablemachinesbytuningthehardwaretotakeadvantageofsimilaritiesamongtemplates.Theconfigurablehardwarecanprocessasetoftemplatesinparallel,usingonlyonecomparisonunitforeachpixelwhosevalueisthesamefortemplatesinthatset.Forexample,ratherthanhavingeightseparatehardwarecircuitsconsideracertainpixelforeightdifferenttemplates,asinglecircuitcanconsiderthepixelandthenpropagateitsoutcometothesevenothertemplates.

Mostrecently,wehavebuiltaprototypeencryptionsystem(alsofundedbyDARPA)thattakesadvantageofconfigurablehardware.AnFPGAimplementstheDataEncryptionStandard(DES),whichuses56-bit-longkeystoencrypt64-bit-longblocksofdata.(Akeyinencryptionisanumberusedtoscrambleorunscrambleaconfidentialmessage.)DESencryptionusuallyproceedsintwosteps:subkeyschedulinganddataprocessing.

Inthefirststep,asetofrotationsandpermutationstranslatesthe56-bitencryptionkeyintoaseriesof16subkeys.Eachsubkeythenprocessesthedatainaseparateround;afullsetof16roundsencryptsordecryptseach64-bitblock.Whenthecomputerdealsconcurrentlywithmultipleusers,eachdialoguebetweenusersmusthaveadistinctkey,andtheencryptionhardwarewillchangekeysaspartsofmessagesarrivefordifferentusers.

InmanyapplicationsofDES,theencryptionkeyremainsconstantwhilealongblockofdatapassesthroughthedatapath.Forexample,iftwopeoplearecommunicatingoverasecurenetwork,theyexchangeasecureencryptionkeyonceandthenusethatkeythroughoutthedurationoftheirdialoguetogeneratethesubkeysforeachroundofencryptionordecryption.SomeASICsaredesignedtohandleonlyonekindofencryptionalgorithm,suchasDES;others—suchasprogrammabledigitalsignalprocessors—arecapableofimplementingmanyencryptionalgorithms.

Withaconfigurablechip,thesoftwarecancalculatethesubkeyvaluesonce,andthedata-processingcircuitrycanbeoptimizedforthosespecificsubkeys.Thisapproachallowsthesubkey-schedulinghardwaretobecompletelyremovedfromthesystem.ThesesavingshaveallowedustoimplementtheDESalgorithmina13,000-gateFPGA,insteadofthe25,000-gatecircuitpreviouslyrequired.Whentheencryptionkeymustbechanged,theusercanquicklyspecifyanewcircuit,customizedtothenewkey,anddownloadittotheFPGA.

Thetarget-recognitionandencryptionprototypeswehavebuilthelpillustratetheenormousflexibilitythatariseswhenthehardwareinacomputercanbecustomizedtoadiverseandchangingsetofexternaldata.Therearemanyotherapplicationsthatcouldbenefitfromtheabilitytomodifythecomputationhardwareinthismanner,includingdigitalcommunications,designautomationanddigitalfilteringforradar.4TheFutureofConfigurableComputing

Configurablecomputingisstillanextremelyyoungfield.AlthoughGeraldEstrinoftheUniversityofCaliforniaatLosAngelesproposedconfigurablecomputinginthelate1960s,thefirstdemonstrationsdidnotoccuruntilafewyearsago,andcurrentFPGAs,withupto100,000logicelements,stilldonotcomeclosetoexploitingthefullpossibilitiesofthetechnique.

FutureFPGAswillbemuchlarger;aswithmanyotherintegratedcircuits,thenumberofelementsonasingleFPGAhasdoubledroughlyevery18months.Beforethedecadeisout,weexpecttoseeFPGAsthathaveamillionlogicelements.Suchchipswillhavemuchbroaderapplication,includinghighlycomplexcommunicationsandsignal-processingalgorithms.

Academicresearchersandmanufacturersareovercomingnumerousotherdesignlimitationsthathavehinderedtheadoptionofconfigurablecomputing.Notallcomputationscanbeimplementedefficientlywithtoday’sFPGAs:theyarewellsuitedtoalgorithmscomposedofbit-leveloperations,suchaspatternmatchingandintegerarithmetic,buttheyareillsuitedtonumericoperations,suchashigh-precisionmultiplicationorfloating-pointcalculations.

DedicatedmultipliercircuitssuchasthoseusedinmicroprocessoranddigitalsignalchipscanbeoptimizedtoperformmoreefficientlythanmultipliercircuitsconstructedfromconfigurablelogicblocksinanFPGA.Furthermore,FPGAscurrentlyprovideverylittleon-chipmemoryforstorageofintermediateresultsincomputations;thus,manyconfigurablecomputingapplicationsrequirelargeexternalmemories.ThetransferofdatatoandfromtheFPGAincreasespowerconsumptionandmayslowdownthecomputations.[6]

Fortunately,researchersaredevelopingadvancedFPGAsthatcontainmemory,arithmeticprocessingunitsandotherspecial-purposeblocksofcircuitry.AndréDeHonandThomasF.Knight,Jr.,oftheMassachusettsInstituteofTechnologyhaveproposedanFPGAthatstoresmultipleconfigurationsinaseriesofmemorybanks.Inasingleclockcycle,whichisontheorderoftensorhundredsofnanoseconds,thechipcouldswaponeconfigurationforanotherconfigurationwithouterasingpartiallyprocesseddata.

MeanwhileBradL.HutchingsofBrighamYoungUniversityhasusedconfigurablecomputingtobuildaDynamicInstructionSetComputer(DISC),whicheffectivelymarriesamicroprocessortoanFPGAanddemonstratesthepotentialofautomaticreconfigurationusingstoredconfigurations.Asaprogramruns,theFPGArequestsreconfigurationifthedesignatedcircuitisnotresident.DISCallowsadesignertocreateandstorealargenumberofcircuitconfigurationsandactivatethemmuchasaprogrammerwouldinitiateacalltoasoftwaresubroutineinamicroprocessor.

TheColtGroup,ledbyPeterM.AthanasofVirginiaPolytechnicInstituteandStateUniversity,isinvestigatingarun-timereconfigurationtechniquecalledWormholethatlendsitselftodistributedcomputing.Theunitofcomputingisastreamofdatathatcreatescustomlogicasitmovesthroughthereconfigurablehardware.

JohnWawrzynekandhiscolleaguesattheUniversityofCaliforniaatBerkeleyaredevelopingsystemsthatcombineamicroprocessorandanFPGA.SpecialcompilersoftwarewouldtakeordinaryprogramcodeandautomaticallygenerateacombinationofmachineinstructionsandFPGAconfigurationsforthefastestoverallperformance.ThisapproachtakesadvantageofopportunitiestointegrateaprocessorandanFPGAonasinglechip.

FPGAswillneverreplacemicroprocessorsforgeneral-purposecomputingtasks,buttheconceptofconfigurablecomputingislikelytoplayagrowingroleinthedevelopmentofhigh-performancecomputingsystems.ThecomputingpowerthatFPGAsofferwillmakethemthedevicesofchoiceforapplicationsinvolvingalgorithmsinwhichrapidadaptationtotheinputisrequired.

Inaddition,thelinebetweenprogrammableprocessorsandFPGAswillbecomelessdistinct:futuregenerationsofFPGAswillincludefunctionssuchasincreasedlocalmemoryanddedicatedmultipliersthatarestandardfeaturesoftoday’smicroprocessors;therearealsonext-generationmicroprocessorsunderdevelopmentwhosehardwaresupportslimitedamountsofFPGA-likereconfiguration.Indeed,justascomputersconnectedtotheInternetcannowautomaticallydownloadspecial-purposesoftwarecomponentstoperformparticulartasks,futuremachinesmightdownloadnewhardwareconfigurationsastheyareneeded.Computingdevices10yearsfromnowwillincludeastrongmixofsoftware-programmablehardwareandhardware-configurablelogic.[7]

1. generalityn.普遍性，普通性，一般性。.

2. versatile*adj.多才多艺的；多用途的或者多功能的；反复无常的；易变的。

3. operation*n.操作运算；动作；运算[操作]指令。

4. configurableadj.可配置的，结构的。Vocabulary

5. tune*n.【音乐】曲调，尤指简单的、易记的调子；准确的音高；和谐；融洽；【电子学】调谐调整接收器或电路以获得对于给定的信号或频率的最强反应v.tr.【音乐】调音使音高正确；使……和谐，调整或调节，尤指为了达到和谐一致【电子学】调整(接收器的频率)、调谐(接收器)以达到希望的频率；调整(电路)、调谐(电路)以使其与输入的信号共振。

6. ASICabbr.[电]专用集成电路。

7. FPGAabbr.[电]现场可编程门阵列。

8. coarseadj.粗糙的,粗鄙的。

9. multipliern.乘法器，倍增器；乘式[MLPR]。

10. swapvt.n.交换。

11. framen.帧，画面。

12. templaten.标准框，样板，模板；【软】属性单元，同templet[TEM]。

13. thresholdn.阈，阈值，门限，门槛;地点,开端。

14. pixeln.像素，像元，图素。

15. paralleladj.并行的，并联的，平行[P]。

16. scramblev.intr.爬，尤指用手和膝盖迅速移动或攀爬；争夺，为了得到某物而疯狂地争夺或者争斗。v.tr.随便地促在一起，偶然地混合或丢在一起；杂乱的收集，匆忙或无秩序地聚集在一起【电子学】扰频打乱或扰乱无线电信号的频率而使在无特殊仪器的情况下无法接收n.攀爬的动作或事例；争夺，粗暴地争夺或斗争。

17. confidential*adj.秘密的,机密的。

18. schedule*n.图表，目录，清单；进度表；【动】调度。

19. bank*n.(数据)库，集；簇；排，列，组，堆，区；存储体；银行。

[1]Computerdesignersfaceaconstantstruggletofindtherightbalancebetweenspeedandgenerality.Theycanbuildversatilechipsthatperformmanydifferentfunctionsrelativelyslowly,ortheycandeviseapplication-specificchipsthatdoonlyalimitedsetoftasksbutdothemmuchmorequickly.ImportantSentencesMicroprocessors(suchastheIntelPentiumorMotorolaPowerPCchipscommonlyfoundinpersonalcomputers)aregeneralpurpose:programminginstructionsencodedinbinaryformatcanleadamicroprocessorthroughvirtuallyanylogicalormathematicaloperationaprogrammercanconceive.TheIntelPentium,forexample,wasneverdesignedspecificallytoexecuteeitherMicrosoftWordorthecomputergameDOOM,butitcanrunboth.计算机设计者面临着不断地在速度和通用性之间找到合适的平衡点的问题。他们可以制造能完成多种不同的功能但速度比较慢的芯片，或设计出能完成有限任务但速度很快的专用芯片。微处理器(如个人电脑中常见的Intel的Pentium处理器和MotorolaPowerPC的芯片)都是通用的：经编码成二进制格式的程序指令实际上可以指导微处理器完成程序员想得到的所有逻辑或数学运算。例如Intel的Pentium处理器决不是专门设计来运行Word的，也不是专门为DOOM游戏而设计的，但是它可以运行这两个程序。

[2]Incontrast,customhardwarecircuits,oftenknownasApplication-SpecificIntegratedCircuits(ASICs),providepreciselythefunctionalityneededforaspecifictask.BycarefullytuningeachASICtoagivenjob,thecomputerdesignercanproduceasmaller,cheaper,fasterchipthatconsumeslesspowerthanaprogrammableprocessor.AcustomgraphicschipforaPC,forinstance,candrawlinesorpaintpicturesonthescreen10or100timesasquicklyasageneral-purposecentralprocessingunitcan.

相反，定制的硬件电路，也经常称为专用集成电路，恰恰可以提供为完成某些特定任务所需的功能。通过针对给定的工作来细心调整专用集成电路，计算机设计者可以制造出更小、更便宜、更快且比可编程处理器消耗更少能量的芯片。一个专用的PC图形(处理)芯片，可以用通用中央处理单元10倍甚至100倍的速度在屏幕上画线和作图。

[3]Anewdevelopmentinintegratedcircuitsoffersathirdoption:large,fast,field-programmablegatearrays,orFPGAs—highlytunedhardwarecircuitsthatcanbemodifiedatalmostanypointduringuse.FPGAsconsistofarraysofconfigurablelogicblocksthatimplementthelogicalfunctionsofgates.Logicgatesarelikeswitcheswithmultipleinputsandasingleoutput.TheyareusedindigitalcircuitstoperformbasicbinaryoperationssuchasAND,NAND,OR,NORandXOR.Inmosthardwarethatisusedincomputingtoday,thelogicalfunctionsofgatesarefixedandcannotbemodified.InFPGAs,however,boththelogicfunctionsperformedwithinthelogicblocksandtheconnectionsbetweentheblockscanbealteredbysendingsignalstothechip.TheseblocksarestructurallysimilartothegatearraysusedinsomeASICs,butwhereasstandardgatearraysareconfiguredduringmanufacture,theconfigurablelogicblocksinFPGAscanberewiredandreprogrammedrepeatedly,longaftertheintegratedcircuithasleftthefactory.集成电路的新发展提供了第三种选择：大规模、高速度和现场可编程的门阵列，或者叫FPGA——经过高度调整的，可以在使用过程中的任何时刻进行修改的硬件电路。现场可编程门阵列由可以实现逻辑门功能的可配置逻辑块阵列组成。逻辑门就像多输入单输出的开关一样，它们在数字电路中用来完成基本的二进制运算(如与、与非、或、或非和异或)。在当今大多数用于计算的硬件中，逻辑门的功能是固定不变的。但在现场可编程的门阵列中，不但逻辑块可完成逻辑功能，而且逻辑块之间的连接都可以通过向芯片送信号来改变。这些逻辑块的结构和某些专用集成电路中的门阵列类似，但标准门阵列在生产的时候就被配置好了，而现场可编程的门阵列中的逻辑块可以在集成电路出厂后很长时间里反复重新布线和重新编程。

[4]Computingdevicescanmakeuseofconfigurableelementsinmanydifferentways.Theleastdemandingtechniqueistoswitchbetweenfunctionsoncommand—thehardwareequivalentofquittingoneprogramandthenrunninganother.Slowreconfiguration,ontheorderofseveralseconds,maywellbeacceptableinsuchanapplication.Fasterprogrammingtimespermitdynamicdesignswapping:asingleFPGAperformsaseriesoftasksinrapidsuccession,reconfiguringitselfbetweeneachone.Suchdesignsoperatethechipinatime-sharingmodeandswapbetweensuccessiveconfigurationssorapidlythatitappearstheFPGAisperformingallitsfunctionsatonce.计算装置可从多方面利用可配置元件。最基本的技术就是在不同命令下实现功能的转换，即退出一个程序后运行另一个程序对应的硬件动作。几秒钟时间的慢速重配置在这种应用中是完全可以接受的。更快的编程使动态设计交换技术成为可能：单个现场可编程的门阵列完成一系列相继的任务，在任务之间重新配置自己。这种设计为分时式操作芯片，在相继的配置间的转换速度很快，给人的感觉好像它同时完成了多个功能。

[5]Oneofthemostpromisingapplicationsforconfigurablecomputinginvolvespatternmatching.Patternmatchingisusedintaskssuchashandwritingrecognition,faceidentification,databaseretrievalandautomatictargetrecognition.Afundamentaloperationofpatternmatchinginvolvescomparinganinputsetofbits(representinganimage,astringofcharactersorotherdata)withasetoftemplatescorrespondingtothepossiblepatternstoberecognized.Thesystemdeclaresrecognitionwhenthenumberofinputbitsthatmatchbitsinthetemplateexceedssomethreshold.可配置计算最有前途的应用之一包括模式匹配。模式匹配可用于手写输入识别、面容识别、数据库检索和自动目标识别等。模式匹配的基本运算包括把输入的一组比特值(代表一幅图、一串字符或其他数据)与一组对应于可能被识别的模式的模板相比较。当与模板中的比特匹配的输入比特的数量超过一个门限值时，系统将认为识别成功。

[6]Academicresearchersandmanufacturersareovercomingnumerousotherdesignlimitationsthathavehinderedtheadoptionofconfigurablecomputing.Notallcomputationscanbeimplementedefficientlywithtoday’sFPGAs:theyarewellsuitedtoalgorithmscomposedofbit-leveloperations,suchaspatternmatchingandintegerarithmetic,buttheyareillsuitedtonumericoperations,suchashigh-precisionmultiplicationorfloating-pointcalculations.DedicatedmultipliercircuitssuchasthoseusedinmicroprocessoranddigitalsignalchipscanbeoptimizedtoperformmoreefficientlythanmultipliercircuitsconstructedfromconfigurablelogicblocksinanFPGA.Furthermore,FPGAscurrentlyprovideverylittleon-chipmemoryforstorageofintermediateresultsincomputations;thus,manyconfigurablecomputingapplicationsrequirelargeexternalmemories.ThetransferofdatatoandfromtheFPGAincreasespowerconsumptionandmayslowdownthecomputations.理论研究者和生产厂商正在试图克服大量其他阻碍采用可配置计算的设计限制。并不是所有的计算都可以在与今日的现场可编程门阵列相结合的情况下能有效地实现：它们可以很好地适用于包含位运算的运算法则，如模式匹配和整数运算，但并不适用于数字的运算，如高精度的乘法或浮点运算。例如用于微处理器和数字信号处理芯片的专用乘法器电路，在优化后能够比现场可编程门阵列中可配置逻辑块构成的乘法器电路更加高效地运行。此外，当前的现场可编程门阵列几乎不能提供集成在芯片上的内存来存储计算的中间结果；因此，很多可配置计算的应用要求大量的外存。对现场可编程门阵列数据的来回传输增加了功耗，同时会降低计算速度。

[7]Inaddition,thelinebetweenprogrammableprocessorsandFPGAswillbecomelessdistinct:futuregenerationsofFPGAswillincludefunctionssuchasincreasedlocalmemoryanddedicatedmultipliersthatarestandardfeaturesoftoday’smicroprocessors;therearealsonext-generationmicroprocessorsunderdevelopmentwhosehardwaresupportslimitedamountsofFPGA-likereconfiguration.Indeed,justascomputersconnectedtotheInternetcannowautomaticallydownloadspecial-purposesoftwarecomponentstoperformparticulartasks,futuremachinesmightdownloadnewhardwareconfigurationsastheyareneeded.Computingdevices10yearsfromnowwillincludeastrongmixofsoftware-programmablehardwareandhardware-configurablelogic.除此之外，可编程处理器和现场可编程门阵列的界限变得更加不清晰：未来几代现场可编程的门阵列将包含更多的功能，诸如越来越大的内部存储器和当今微处理器才有的专用乘法器等；同时将会出现正在开发中的下一代微处理器，其硬件在一定范围内支持像现场可编程门阵列一样的重配置。事实上，正如网络上的计算机可以自动下载专用软件组件来完成特定的任务一样，未来的机器也可以在需要时下载新的硬件配置。10年后的计算装置将会把如同软件一样可编程的硬件和硬件可配置逻辑有力地结合在一起。

1.Accordingtothecontentbefore“ProgrammableCircuitry”,answerthefollowingquestions.

(1) Whatproblemhasbotheredforcomputerdesignersforquitealongtime?()

A. Balancebetweencostandfunctionality.

B. Balancebetweenstabilityandpracticality.

C. Balancebetweenspeedandgenerality.

D. Allofthem.QuestionsandAnswers

(2) Whichofthefollowingsisakindofmicroprocessorinthepassage?()

A. DOOM.

B. MotorolaPowerPCchips.

C. MicrosoftWord.

D. ASIC.

(3) WhichofthefollowingsayingsisTrue?()

A. IntelPentiumcanrunMicrosoftWordandthecomputergameDOOM.

B. IntelPentiumcanrunMicrosoftWordbutnotDOOM.

C. IntelPentiumcanrunDOOMbutnotMicrosoftWord.

D. IntelPentiumcanrunneitherMicrosoftWordnorDOOM.

(4) WhichofthefollowingsisnotafeatureofASIC?()

A. Fast.

B. Energy-saving.

C. Small.

D. Slow.

(5) WhatproblemofASICshouldbesolved?()

A. Toospecialized.

B. Tooslow.

C. Tooexpensive.

D. Allofthem.

2.Accordingtothecontentbetween“ProgrammableCircuitry”and“CuttingCriticalHardware”,answerthefollowingquestions.

(1) Whatisthethirdoptionofferedbythenewdevelopmentinintegratedcircuits?()

A. Versatilemicroprocessor.

B. ASIC.

C. FPGA.

D. Specializedmicroprocessor.

(2) WhatisthemoststrikingadvantageofFPGAoverASIC?()

A. Itcanberewiredandreprogrammed.

B. Itcanrunmuchfaster.

C. Itismuchsmaller.

D. Itismuchcheaper.

(3) Whatmakesconfigurablecomputingpossible?()

A. Versatilemicroprocessor.

B. ASIC.

C. FPGA.

D. Specializedmicroprocessor.

(4) HowfastcanFPGAsbeconfigurednow?()

A. Seconds.

B. Milliseconds.

C. Microseconds.

D. Nanoseconds.

(5) WhatisthebasicstructureofaFPGAdesign?()

A. Configurablelogicblocks.

B. Aprogrammablegridofconnections.

C. Programmableprocessor.

D. AandB.

(6) WhichofthefollowingstatementsisWrong?()

A. Acoarse-grainedFPGAmightbecapableofaddingorcomparingtwonumbers.

B. Afine-graineddevicemightbecapableonlyofcomparingtwobinarydigits.

C. Oneblockinafine-graineddevicemightonlybeasinglelogicgate.

D. Coarse-grainedFPGAsarelesswidelyusedthanfine-graineddevices.

(7) WhichofthefollowingsayingsisWrong?()

A. Dynamicdesignswappingoperatesthechipinatime-sharingmode.

B. Dynamicdesignswappingswapsbetweensuccessiveconfigurationsrapidly.

C. Dynamicdesignswappin