overview of emerging nonvolatile memory technologies by various authors.pdf

81

Upload: altin-guberi

Post on 16-Sep-2015

230 views

Category:

Documents


6 download

TRANSCRIPT

  • NanoscaleResearchLettersNanoscaleResLett.9(1):526-526

  • OverviewofemergingnonvolatilememorytechnologiesJaganSinghMeena1,SimonMinSze1,UmeshChand1,Tseung-YuenTseng1

    1.DepartmentofElectronicsEngineeringandInstituteofElectronics,NationalChiaoTungUniversity,Hsinchu30010,Taiwan

    Copyright2014Meenaetal.;licenseeSpringer.

    DOI:10.1186/1556-276X-9-526

    Publishedonline:25September2014

  • Abstract

    NonvolatilememorytechnologiesinSi-basedelectronicsdatebacktothe1990s.Ferroelectricfield-effecttransistor(FeFET)wasoneofthemostpromisingdevicesreplacingtheconventionalFlashmemoryfacingphysicalscalinglimitationsatthosetimes.AvariantofchargestoragememoryreferredtoasFlashmemoryiswidelyusedinconsumerelectronicproductssuchascellphonesandmusicplayerswhileNANDFlash-basedsolid-statedisks(SSDs)areincreasinglydisplacingharddiskdrivesastheprimarystoragedeviceinlaptops,desktops,andevendatacenters.TheintegrationlimitofFlashmemoriesisapproaching,andmanynewtypesofmemorytoreplaceconventionalFlashmemorieshavebeenproposed.Emergingmemorytechnologiespromisenewmemoriestostoremoredataatlesscostthantheexpensive-to-buildsiliconchipsusedbypopularconsumergadgetsincludingdigitalcameras,cellphonesandportablemusicplayers.Theyarebeinginvestigatedandleadtothefutureaspotentialalternativestoexistingmemoriesinfuturecomputingsystems.Emergingnonvolatilememorytechnologiessuchasmagneticrandom-accessmemory(MRAM),spin-transfertorquerandom-accessmemory(STT-RAM),ferroelectricrandom-accessmemory(FeRAM),phase-changememory(PCM),andresistiverandom-accessmemory(RRAM)combinethespeedofstaticrandom-accessmemory(SRAM),thedensityofdynamicrandom-accessmemory(DRAM),andthenonvolatilityofFlashmemoryandsobecomeveryattractiveasanotherpossibilityforfuturememoryhierarchies.Manyothernewclassesofemergingmemorytechnologiessuchastransparentandplastic,three-dimensional(3-D),andquantumdotmemorytechnologieshavealsogainedtremendouspopularityinrecentyears.Subsequently,notanexaggerationtosaythatcomputermemorycouldsoonearntheultimatecommercialvalidationforcommercialscale-upandproductionthecheapplasticknockoff.Therefore,thisreviewisdevotedtotherapidlydevelopingnewclassofmemorytechnologiesandscalingofscientificproceduresbasedonaninvestigationofrecentprogressinadvancedFlashmemorydevices.

  • Review

    BackgroundGeneraloverview

    Theideaofusingafloatinggate(FG)devicetoobtainanonvolatilememorydevicewassuggestedforthefirsttimein1967byKahngDandSzeSMatBellLabs[1].ThiswasalsothefirsttimethatthepossibilityofnonvolatileMOSmemorydevicewasrecognized.Fromthatday,semiconductormemoryhasmadetremendouscontributionstotherevolutionarygrowthofdigitalelectronicssincea64-bitbipolarRAMchiptobeusedinthecachememoryofanIBMcomputerwasreportedin1969[2].Semiconductormemoryhasalwaysbeenanindispensablecomponentandbackboneofmodernelectronicsystems.Allfamiliarcomputingplatformsrangingfromhandhelddevicestolargesupercomputersusestoragesystemsforstoringdatatemporarilyorpermanently[3].Beginningwithpunchcardwhichstoresafewbytesofdata,storagesystemshavereachedtomultiterabytesofcapacitiesincomparativelylessspaceandpowerconsumption.Regardingapplicationaspects,thespeedofstoragesystemsneedstobeasfastaspossible[4].SinceFlashmemoryhasbecomeacommoncomponentofsolid-statedisks(SSDs),thefallingpricesandincreaseddensitieshavemadeitmorecost-effectiveformanyotherapplications[5].MemorydevicesandmostSSDsthatuseFlashmemoryarelikelytoserveverydifferentmarketsandpurposes.Eachhasanumberofdifferentattributeswhichareoptimizedandadjustedtobestmeettheneedsofparticularusers.Becauseofnaturalinherentlimitations,thelong-establishedmemorydeviceshavebeenshortedoutaccordingtotheirinventionstomatchwithportableelectronicdatastoragesystems.Today,themostprominentoneisthelimitedcapacityforcontinuedscalingoftheelectronicdevicestructure.ResearchismovingalongthefollowingpathsforembeddedFlashdevices:(i)scalingdownthecellsizeofdevicememory,(ii)loweringvoltageoperation,and(iii)increasingthedensityofstatepermemorycellbyusingamultilevelcell.Tosustainthecontinuousscaling,conventionalFlashdevicesmayhavetoundergorevolutionarychanges.Basically,itisexpectedthatanentireDVDcollectionbeinthepalmofahand.Noveldeviceconceptswithnewphysicaloperationingprinciplesareneeded.Itisworthwhiletotakealookatsemiconductormemoriesagainstthebackgroundofdigitalsystems.Thewaysemiconductordevicesareusedinasystemsenvironmentdetermineswhatisrequiredofthemintermsofdensity,speed/power,andfunctions.Itisalsoworthwhiletolookintotheeconomicsignificanceofsemiconductormemoriesandtherelativeimportanceoftheirvarioustypes.Forthepastthreeandahalfdecadesinexistence,thefamilyofsemiconductormemorieshasexpandedgreatlyandachievedhigherdensities,higherspeeds,lowerpower,morefunctionality,andlowercosts[3,6,7].Atthesametime,someofthelimitationswithineachtypeofmemoryarealsobecomingmorerealized.Assuch,thereareseveralemergingtechnologiesaimingtogobeyondthoselimitationsandpotentiallyreplaceallormostoftheexistingsemiconductormemorytechnologiestobecomeauniversalsemiconductormemory(USM).Inaddition,therewardsforachievingsuchadevicewouldbetogaincontrolofanenormousmarket,whichhasexpandedfromcomputerapplicationstoallofconsumerelectronicproducts.Lookingforwardtothefuture,therearewiderangesofemergingmemoryapplicationsforautomationand

  • informationtechnologytohealthcare.Thespecificationofnonvolatilememory(NVM)isbasedonthefloatinggateconfiguration,whichisthefeatureofanerasedgateputintomanycellstofacilitateblockerasure.Amongthem,designedFlashmemoriessuchasNORandNANDFlashhavebeendevelopedandthenproposedascommercialproductsintobulkmarket.Theyhavebeenconsideredasthemostimportantproducts.NORhashighoperationspeedforbothcodeanddatastorageapplications;ontheotherhand,NANDhashighdensityforlargedatastorageapplications[8].SincetheinceptionofFlashmemory,therehasbeenanexponentialgrowthinitsmarketdrivenprimarilybycellphonesandothertypesofconsumerelectronicequipment.While,today,integrationofasiliconchipisnoteconomical,toys,cards,labels,badges,valuepaper,andmedicaldisposablescouldbeimaginedtobeequippedwithflexibleelectronicsandmemory.Withgrowingdemandsforhigh-densitydigitalinformationstorage,memorydensitywitharrivingtechnologyhasbeenincreaseddramaticallyfromthepastcoupleofyears.Themaindrivetodeveloporganicnonvolatilememoryiscurrentlyforapplicationsofthin-film,flexible,orevenprintedelectronics.Oneneedsatechnologytotageverythingtoelectronicfunctionalitywhichcanbeforeseeninaverylargequantityandataverylowcostonsubstratessuchasplasticandpaper.Accessiblepopularizationofroll-to-rollmemorycommercializationisawaytomakeanencounterinterestingandchallengingtohavechargestoragedevicesofchoiceforapplicationswithenormousflexibilityandstrength.Recently,polymer(plasticmemory)andorganicmemorydeviceshavesignificantconsiderationbecauseoftheirsimpleprocesses,fastoperatingspeed,andexcellentswitchingability[9,10].Onesignificantadvantagepolymermemoryhasoverconventionalmemorydesignsisthatitcanbestackedvertically,yieldingathree-dimensional(3-D)useofspace[11].Thismeansthatinterabytesolid-statedeviceswithextremelylowtransistorcountssuchasdrivesaboutthesizeofamatchbook,thedatapersistsevenafterpowerisremoved.TheNANDFlashmarketiscontinuallygrowingbythesuccessiveintroductionofinnovativedevicesandapplications.Tomeetthemarkettrend,3-DNVMsareexpectedtoreplacetheplanarones,especiallyfor10-nmnodesandbeyond.Moreover,simple-structureorganicbistablememoryexhibitingsuperiormemoryfeatureshasbeenrealizedbyemployingvariousnanoparticles(NPs)blendedintoasingle-layeredorganicmaterialsandwichedbetweentwometalelectrodes[12,13].TheNPsactastrapsthatcanbechargedanddischargedbysuitablevoltagepulses.NPblendsshowpromisingdataretentiontimes,switchingspeed,andcyclingendurance,buttheon-statecurrentistoolowtopermitscalingtonanometerdimensions[10,14].Alotofthesegreatideastendtodiebeforereachingthispointofdevelopment,butthatisnottosaythatwewillbeseeingplasticmemoryonstoreshelvesnextyear.Therearestillmanyhurdlestogetover;softwarealoneisabigtask,asisthemanufacturingprocess,butitdoesbringthistechnologyonestepclosertoreality[15].Itisnotanexaggerationtosaythattheequivalentof400,000CDs,60,000DVDs,or126yearsofMPGmusicmaybestoredonapolymermemorychipthesizeofacreditcard.

    Thevisionofthisreview

    Inthisreview,wefocusonelectricallyprogrammablenonvolatilememorychangesfromsiliconnanocrystalmemoryscalingtoorganicandmetallicNPmemorydevices.Further,thescalingtrendmovetowardstheemergingNVMtoflexibleandtransparentredox-basedresistiveswitchingmemorytechnologies.Thisreviewisintendedtogivean

  • overviewtothereaderofstoragesystemsandcomponentsfromconventionalmemorydevicesthathavebeenproposedinthepastyearsofrecentprogressincurrentNVMdevicesbasedonnanostructuredmaterialstoredox-basedresistiverandom-accessmemory(RRAM)to3-Dandtransparentmemorydevices.WedescribethebasicsofFlashmemoryandthenhighlightthepresentproblemswiththeissueofscalingtunneldielectricinthesedevices.Webrieflydescribeahistoricalchange,howtheconventionalFGnonvolatilememorysuffersfromachargelossproblemasthefeaturesizeofthedevicecontinuestoshrink.Adiscretepolysilicon-oxide-nitride-oxide-silicon(SONOS)memoryisthenproposedasareplacementoftheconventionalFGmemory.TheNCmemoryisexpectedtoefficientlypreservethetrappedchargeduetothediscretechargestoragenodewhilealsodemonstratingexcellentfeaturessuchasfastprogram/erasespeeds,lowprogrammingpotentials,andhighendurance.Wealsodiscusscurrentongoingresearchinthisfieldandthesolutionsproposedtosolvethescalingproblemsbydiscussingaspecificsolutionindetailwhichwouldbethecenterpieceinrecentmemoryworkprogress.Moreover,thisreviewmakesdistinctemergingmemoryconceptswithmorerecentmolecularandquantumdotprogrammablenonvolatilememoryconcepts,specificallyusingchargetrappinginconjugatedpolymersandmetalNPs.Weclassifyseveralpossibledevices,accordingtotheiroperatingprinciple,andcriticallyreviewtheroleof-conjugatedmaterialsinthedatastoragedeviceoperation.WedescribespecificationsforapplicationsofemergingNVMdevicesaswellasalreadyexistingNANDmemoryandreviewthestateoftheartwithrespecttothesetargetspecificationsinthefuture.Conclusionsaredrawnregardingfurtherworkonmaterialsandupcomingmemorydevicesandarchitectures.

    Classificationofsolid-statememorytechnologiesDatastoragedevicescanbeclassifiedbasedonmanyfunctionalcriteria.Ofthem,silicon-basedsemiconductormemoriesarecategorizedintotwo:volatileandnonvolatile[3,16].Involatilememories,theinformationeventuallyfadeswhilepowersupplyisturnedoffunlessthedevicesusedtostoredatawillbeperiodicallyrefreshed.Ontheotherhand,nonvolatilememoriesretainthestoredinformationevenwhenthepowersupplyisturnedoff.Volatilememories,suchasstaticrandom-accessmemory(SRAM)anddynamicrandom-accessmemory(DRAM),needvoltagesupplytoholdtheirinformationwhilenonvolatilememories,namelyFlashmemories,holdtheirinformationwithoutone.DRAM(dynamicstandsfortheperiodicalrefresh)isneededfordataintegrityincontrasttoSRAM.ThebasiccircuitstructuresofDRAM,SRAM,andFlashmemoriesareshowninFigure1.DRAM,SRAM,andFlasharetodaysdominantsolid-statememorytechnologies,whichhavebeenaroundforalongtime,withFlashtheyoungest,at25years.DRAMisbuiltusingonlyonetransistorandonecapacitorcomponent,andSRAMisusuallybuiltinCMOStechnologywithsixtransistors.Twocross-coupledinvertersareusedtostoretheinformationlikeinaflip-flop.Fortheaccesscontrol,twofurthertransistorsareneeded.Ifthewritelineisenabled,thendatacanbereadandsetwiththebitlines.TheFlashmemorycircuitworkswiththeFGcomponent.TheFGisbetweenthegateandthesource-drainareaandisolatedbyanoxidelayer.IftheFGisuncharged,thenthegatecancontrolthesource-draincurrent.TheFGgetsfilled(tunneleffect)withelectronswhenahighvoltageatthegateissupplied,andthenegativepotentialontheFGworksagainstthegateandnocurrentispossible.TheFGcanbe

  • erasedwithahighvoltageinreversedirectionofthegate.DRAMhasanadvantageoverSRAMandFlashofonlyneedingoneMOSFETwithacapacitor.ItalsohastheadvantageofcheapproductionaswellaslowerpowerconsumptionascomparedtoSRAMbutslowerthanSRAM.Ontheotherhand,SRAMisusuallybuiltinCMOStechnologywithsixtransistorsandtwocross-coupledinverters,andfortheaccesscontrol,twofurthertransistorsareneeded.SRAMhastheadvantageofbeingquick,easytocontrol,integratedinthechip,aswellasfastbecausenobusisneededlikeinDRAM.ButSRAMhasthedisadvantagesofneedingmanytransistorsandhenceexpensive,higherpowerconsumptionthanDRAM.IncomparisontoDRAMandSRAM,FlashmemoryhasFGbetweenthegateandthesource-drainareaandisolatedwithanoxidelayer.FlashmemorydoesnotrequirepowertostoreinformationbutisslowerthanSRAMandDRAM.

    Viewlargerversion

    Figure1.ThecircuitrystructuresofDRAM,SRAM,andFlashmemories.

    BothtypesofmemoriescanbefurtherclassifiedbasedonthememorytechnologythattheyuseandbasedondatavolatilityasshownintheclassificationflowchartdepictedinFigure2.VolatilememoriesconsistmostlyofDRAM[17],whichcanbefurtherclassifiedintoSDRAMandmobileRAMwhichonlyretaininformationwhencurrentisconstantlysuppliedtothedevice[18].AnothersmallbutveryimportantmemorydeviceisSRAM.ThemarketforDRAMdevicesfarexceedsthemarketforSRAMdevices,althoughasmallamountofSRAMdevicesisusedinalmostalllogicandmemorychips.However,DRAMusesonlyonetransistorandonecapacitorperbit,allowingittoreachmuchhigherdensitiesand,withmorebitsonamemorychip,bemuchcheaperperbit.SRAMisnotworthwhilefordesktopsystemmemory,whereDRAMdominates,butisusedforitscachememories.SRAMiscommonplaceinsmallembeddedsystems,whichmightonlyneedtensofkilobytesorless.ForthcomingvolatilememorytechnologiesthathopetoreplaceorcompetewithSRAMandDRAMincludeZ-RAM,TTRAM,A-RAM,andETARAM.Intheindustry,newuniversalandstablememorytechnologieswillappearasrealcontenderstodisplaceeitherorbothNANDFlashandDRAM.Flashmemoryispresentlythemostsuitablechoicefornonvolatileapplicationsforthefollowingreasons:Semiconductornonvolatilememoriesconsistmostlyoftheso-calledFlashdevicesandretaintheirinformationevenwhenthepoweristurnedoff.Othernonvolatilesemiconductormemoriesincludemaskread-onlymemory(MROM),antifuse-basedone-timeprogrammable(OTP)memory,andelectricallyerasableread-onlymemory(EEPROM).Flashisfurtherdividedintotwocategories:NOR,characterizedbyadirectwriteandalargecellsize,andNAND,characterizedbyapagewriteandsmallcellsize.Nonvolatilememoryisacomputermemorythatcanretainthestoredinformationevenwhennotpowered[3,19,20].Nonvolatilesemiconductormemoriesaregenerallyclassifiedaccordingtotheirfunctionalpropertieswithrespecttotheprogramminganderasingoperations,asshownintheflowchartdescribedinFigure2.Thesearefloatinggate,nitride,ROMandfuse,Flash,emerging,andothernewnext-generationmemorytechnologies.Today,thesenonvolatilememoriesarehighlyreliableandcanbeprogrammedusingasimplemicrocomputerandvirtuallyineverymodernelectronic

  • equipment,whichareexpectedtoreplaceexistingmemories.

    Viewlargerversion

    Figure2.Flowchartforthesemiconductormemoryclassificationaccordingtotheirfunctionalcriteria.

    Amongthem,emergingnonvolatilememoriesarenowverycaptivating.Thenext-generationmemorymarketwillcoveruptheseemergingmemorytechnologies[21].Therearemainlyfivetypesofnonvolatilememorytechnology:Flashmemory,ferroelectricrandom-accessmemory(FeRAM),magneticrandom-accessmemory(MRAM),phase-changememory(PCM),andRRAM.Nonvolatilememory,specificallyFlashmemory,whichischaracterizedbyalarge-block(orsector)erasingmechanism,hasbeenthefastestgrowingsegmentofthesemiconductorbusinessforthelast10years.SomeoftheseneweremergingtechnologiesincludeMRAM,FeRAM,PCM,spin-transfertorquerandom-accessmemory(STT-RAM),RRAMandmemristor.MRAMisanonvolatilememory[10,22].UnlikeDRAM,thedataisnotstoredinanelectricchargeflow,butbymagneticstorageelements.Thestorageelementsareformedbytwoferromagneticplates,eachofwhichcanholdamagneticfield,separatedbyathininsulatinglayer.Oneofthetwoplatesisapermanentmagnetsettoaparticularpolarity;theothersfieldcanbechangedtomatchthatofanexternalfieldtostorememory.STT-RAMisanMRAM(nonvolatile)butwithbetterscalabilityovertraditionalMRAM.TheSTTisaneffectinwhichtheorientationofamagneticlayerinamagnetictunneljunctionorspinvalvecanbemodifiedusingaspin-polarizedcurrent.Spin-transfertorquetechnologyhasthepotentialtomakeMRAMdevicescombininglowcurrentrequirementsandreducedcostpossible;however,theamountofcurrentneededtoreorientthemagnetizationisatpresenttoohighformostcommercialapplications.PCMisanonvolatilerandom-accessmemory,whichisalsocalledovonicunifiedmemory(OUM),basedonreversiblephaseconversionbetweentheamorphousandthecrystallinestateofachalcogenideglass,whichisaccomplishedbyheatingandcoolingoftheglass.Itutilizestheuniquebehaviorofchalcogenide(amaterialthathasbeenusedtomanufactureCDs),wherebytheheatproducedbythepassageofanelectriccurrentswitchesthismaterialbetweentwostates.Thedifferentstateshavedifferentelectricalresistancewhichcanbeusedtostoredata.Theidealmemorydeviceortheso-calledunifiedmemorywouldsatisfysimultaneouslythreerequirements:highspeed,highdensity,andnonvolatility(retention).Atthepresenttime,suchmemoryhasnotbeendeveloped.Thefloatinggatenonvolatilesemiconductormemory(NVSM)hashighdensityandretention,butitsprogram/erasespeedislow.DRAMhashighspeed(approximately10ns)andhighdensity,butitisvolatile.Ontheotherhand,SRAMhasveryhighspeed(approximately5ns)butlimitedfromverylowdensityandvolatility.ItisexpectedthatPCMwillhavebetterscalabilitythanotheremergingtechnologies.RRAMisanonvolatilememorythatissimilartoPCM.Thetechnologyconceptisthatadielectric,whichisnormallyinsulating,canbemadetoconductthroughafilamentorconductionpathformedafterapplicationofasufficientlyhighvoltage.Arguably,thisisamemristortechnologyandshouldbeconsideredaspotentiallyastrongcandidatetochallengeNANDFlash.Currently,FRAM,MRAM,andPCMareincommercialproductionbutstill,relativetoDRAMandNANDFlash,remain

  • limitedtonicheapplications.ThereisaviewthatMRAM,STT-RAM,andRRAMarethemostpromisingemergingtechnologies,buttheyarestillmanyyearsawayfromcompetingforindustryadoption[23].Anynewtechnologymustbeabletodelivermost,ifnotall,ofthefollowingattributesinordertodriveindustryadoptiononamassscale:scalabilityofthetechnology,speedofthedevice,andpowerconsumptiontobebetterthanexistingmemories.TheNVSMisininspiringsearchofnovelnonvolatilememories,whichwillsuccessfullyleadtotherealizationandcommercializationoftheunifiedmemory.

    Inprogress,anothernewclassofnonvolatilememorytechnologieswillofferalargeincreaseinflexibilitycomparedtodisks,particularlyintheirabilitytoperformfast,randomaccesses.UnlikeFlashmemory,thesenewtechnologieswillsupportin-placeupdates,avoidingtheextraoverheadofatranslationlayer.Further,thesenewnonvolatilememorydevicesbasedondeoxyribonucleicacid(DNA)biopolymerandorganicandpolymermaterialsareoneofthekeydevicesforthenext-generationmemorytechnologywithlowcost.NonvolatilememorybasedonmetallicNPsembeddedinapolymerhosthasbeensuggestedasoneofthesenewcross-pointmemorystructures.Inthissystem,traplevelssituatedwithinthebandgapofthepolymerareintroducedbytheNPs[24,25].Memorydevicesplayamassiveroleinallemergingtechnologies;assuch,effortstofabricateneworganicmemoriestobeutilizedinflexibleelectronicsareessential.Flexibilityisparticularlyimportantforfutureelectronicapplicationssuchasaffordableandwearableelectronics.Muchresearchhasbeendonetoapplytheflexibleelectronicstechnologytopracticaldeviceareassuchassolarcells,thin-filmtransistors,photodiodes,light-emittingdiodes,anddisplays[26-28].Researchonflexiblememorywasalsoinitiatedforthesefutureelectronicapplications.Inparticular,organic-basedflexiblememorieshavemeritssuchasasimple,low-temperature,andlow-costmanufacturingprocess.Severalfabricationresultsoforganicresistivememorydevicesonflexiblesubstrateshavebeenreported[29,30].Inaddition,withgrowingdemandforhigh-densitydigitalinformationstorage,NANDFlashmemorydensityhasbeenincreaseddramaticallyforthepastcoupleofdecades.Ontheotherhand,devicedimensionscalingtoincreasememorydensityisexpectedtobemoreandmoredifficultinabit-costscalablemannerduetovariousphysicalandelectricallimitations.Asasolutiontotheproblems,NANDFlashmemorieshavingstackedlayersareunderdevelopingextensions[31,32].In3-Dmemories,costcanbereducedbybuildingmultiplestackedcellsinverticaldirectionwithoutdevicesizescaling.Asabreakthroughforthescalinglimitations,various3-Dstackedmemoryarchitecturesareunderdevelopmentandexpectingthehugemarketof3-Dmemoriesinthenearfuture.Withlotsofexpectation,future-generationmemorieshavepotentialtoreplacemostoftheexistingmemorytechnologies.Thenewandemergingmemorytechnologiesarealsonamedtobeauniversalmemory;thismaygiverisetoahugemarketforcomputerapplicationstoalltheconsumerelectronicproducts.

    MarketmemorytechnologiesbyapplicationsThesemiconductorindustryhasexperiencedmanychangessinceFlashmemoryfirstappearedintheearly1980s.ThegrowthofconsumerelectronicsmarketurgesthedemandofFlashmemoryandhelpstomakeitaprominentsegmentwithinthesemiconductorindustry.TheFlashmemorieswerecommerciallyintroducedintheearly1990s,andsince

  • thattime,theyhavebeenabletofollowMooreslawandthescalingrulesimposedbythemarket.Thereareexpectedmassivechangesinthememorymarketoverthenextcoupleofyears,withmoredensityandreliabletechnologieschallengingthedominantNANDFlashmemorynowusedinSSDsandembeddedinmobileproducts.Server,storage,andapplicationvendorsarenowworkingonnewspecificationstooptimizethewaytheirproductsinteractwithNVM-movesthatcouldleadtothereplacementofDRAMandharddrivesalikeformanyapplications,accordingtoastoragenetworkingindustryassociation(SNIA)technicalworkinggroup[33,34].TheFlashmemorymarketplaceisoneofthemostvibrantandexcitinginthesemiconductorindustry,nottomentiononeofthemostcompetitive.Thecontinuousinventionofnewmemorytechnologiesandtheirapplicationsinthememorymarketalsoincreaseperformancedemands.Thesenewclassesofmemorieswiththelatesttechnologyincreasetheverticaldemandinthefuturememorymarket.Inthenextcomingyears,cumulativepricereductionsshouldbecomedisruptivetoDVDsandharddiskdrives(HDDs),stimulatehugedemand,andcreatenewFlashmarkets.

    Thenonvolatilememoriesofferthesystemadifferentopportunityandcoverawiderangeofapplications,fromconsumerandautomotivetocomputerandcommunication.Figure3showsNVSMmemoryconsumptionbyvariousapplicationsintheelectronicsindustrybymarketin2010extendingupwardsfromcomputersandcommunicationtoconsumerproducts[22].Itisnoticedthatthereisafastergrowthrateofthedigitalcellularphonesince1990;thevolumeofproductionhasincreasedby300times,e.g.,from5millionunitsperyeartoabout1.5billionunitsperyear.Nowadays,flexibilityandtransparencyareparticularlyofgreatsignificanceforfutureelectronicapplicationssuchasaffordableandwearableelectronics.Manyadvancedresearchtechnologiesareappliedtoflexibletechnologytobeusedinarealelectronicsarea[35].Althoughsilicon-basedsemiconductormemorieshaveplayedsignificantrolesinmemorystorageapplicationsandcommunicationinconsumerelectronics,now,therecentfocusisturningfromrigidsilicon-basedmemorytechnologyintoasoftnonvolatilememorytechnologyforlow-cost,large-area,andlow-powerflexibleelectronicapplications.Further,thememorymarketforthelongtermiscontinuouslygrowing,evenifwithsomeupsanddowns,andthisisexpectedtocontinueinthecomingyears[36].Sinceinnovationdrivesthesemiconductorindustry,anewtrendwithtransparencyaswellasflexibilityand3-Dtechnologieswillbeattractiveandmovetowardscontinuousgrowthinthenearfuture.

    Viewlargerversion

    Figure3.VariousNVSMapplicationsintheelectronicsindustrybymarketsizein2010.Reprintedfromref.[22].

    SuccessivecreationofnewmobiledevicesleadstothecontinualgrowthofNANDproductsasshowninFigure4.Tomeetthismarketdemand,earlythisyear,30-nmnodetechnologiesareinramping-upphase,20-nmnodetechnologiesareinthephaseoftransitiontomassproduction,anda10-nmnodetechnologyisunderdevelopment.Inaddition,thefuturemarketrequireshigh-speedoperationevenuptoapproximately1,500MB/sinordertosatisfyalargeamountofdatacorrespondence[37].However,high-speedoperationscausehighpowerconsumptionandchiptemperatureincrease,

  • whichcandeteriorateNANDreliability.Hence,reductionofoperatingvoltageisinevitabletoachievethefutureNAND.Opportunitiesfortheuseof3-Daswellaspolymermemorydesigninmodernelectroniccircuitsarerapidlyexpanding,basedontheveryhighperformanceanduniquefunctionality.However,theirpracticalimplementationinelectronicapplicationswillultimatelybedecidedbytheabilitytoproducedevicesandcircuitsatacostthatissignificantlybelowthatneededtomanufactureconventionalelectroniccircuitsbasedon,forexample,silicon.Ifsuccessful,theselow-costfabricationprocesseswillultimatelyresultintheprintingoflarge-areaorganicelectroniccircuitsonasheetofplasticpaperusingaroll-to-rollmethod,wherelow-temperaturedepositionoforganicsisfollowedbymetaldepositionandpatterninginacontinuous,high-speedprocessanalogous,perhaps,toprocessesusedintheprintingofdocumentsorfabrics.

    Viewlargerversion

    Figure4.GrowthofNANDFlashmarketupto2014(iSuppli)andtheinterfacespeedofvariousNANDapplications.Reproducedfromref.[37].

    Inrecentyears,IDTechExfindsthatthetotalmarketforprinted,flexible,andorganicelectronicswillgrowfrom$16.04billionin2013to$76.79billionin2023andthisgrowingtrendisexpectedtocontinueinthecomingyears(seeFigure5a).ThemajorityofthatisOLEDs(onlyorganic,notprinted)andconductiveinkusedforawiderangeofapplications.Ontheotherhand,stretchableelectronics,logicandmemory,andthin-filmsensorsaremuchsmalleringredientsbuthavinghugegrowthpotentialastheyemergefromR&D[38].Thereportspecificallyaddressesthebigpicturethatover3,000organizationsarepursuingprinted,organic,flexibleelectronics,includingprinting,electronics,materials,andpackagingcompanies.Whilesomeofthesetechnologiesareinusenow-indeedtherearemainsectorsofbusinesswhichhavecreatedbillion-dollarmarkets-othersarecommerciallyembryonic.

    Viewlargerversion

    Figure5.Marketvolume(a)andglobalflexibledisplaymarketshipmentforecast(b).Reproducedfromrefs.[38,39].

    Anotherkeypotentialmarketforprinted/flexibleelectronicsisnext-generationtransparentconductivefilmtoreplacebrittleandexpensiveindiumtinoxide(ITO)intouchscreensanddisplays,lighting,andphotovoltaics.Touchdisplayresearchsaysthatthemarketfornon-ITOtransparentconductorswillbeabout$206millionthisyearandgrowtosome$4billionby2020asshowninFigure5b.HighdemandfortouchscreensfornotebookandPCsizedisplayshascreatedashortageofITOtouchsensorssincetheendoflastyeartodrivemoreinterestinthesetechnologies,andthemoreflexibleandpotentiallycheaperreplacementtechnologiesaregettingmoremature,notesJenniferColegrove,presidentandanalyst,whowillspeakattheFlexTechworkshopontransparentconductors.ShenotesthatAtmel,Fujifilm,UnipixelandCambriosareallinsomephaseofproduction

  • [39].Alargeamountofthesemiconductormarket(approximately20%)isgivenbythesemiconductormemories;thus,themarketforchipswilldevelopinthenextfewyears.Thisstudyreportsthatthereisananalysisoftheproductionprocessandthesubsequentvaluechain,whichcomprisesabenchmarkanalysisofthemainsegmentsofthesemiconductorindustry.

    Recently,the3-DnonvolatilememorystructurehasalsoattractedconsiderableattentionduetoitspotentialtoreplaceconventionalFlashmemoryinnext-generationNVMapplications[37,40].3-Dmemoriesaregatheringincreasingattentionasfutureultra-high-densitymemorytechnologiestokeepatrendofincreasingbitdensityandreducingbitcost.TheNANDFlashmarketiscontinuouslygrowingbythesuccessiveintroductionofinnovativedevicesandapplications.Tomeetthemarkettrend,3-DNVMsareexpectedtoreplacetheplanarone,especiallyfor10-nmnodesandbeyond.Therefore,thefundamentalsandcurrentstatusofthe3-DNANDFlashmemoryarereviewedandfuturedirectionsarediscussed[41].3-DintegrationpromisestobeanexcellentreplacementofcurrenttechnologiesforthedevelopmentofNANDFlashmemory.TimeisrunningoutforplanarNANDtechnology.ItwillnotbelongthatplanarNANDwillbecompletelyreplacedby3-DNAND.3-DNANDpromisestosatisfythegrowingneedofNANDmemory[37].

    Finally,NVMtechnologieshaveabrightfuturesinceeveryend-useapplicationneedstostoresomeparametersorsomeamountofanapplicationprogramintheon-boardNVMtoenableittofunction.TheupcomingNVMsarethebighopeforasemiconductormemorymarket,whichprovidesmemoriesforsystemstorunwithflexibility,reliability,highperformance,andlowpowerconsumptioninatinyfootprintinnearlyeveryelectronicapplication.Recentmarkettrendshaveindicatedthatcommercializedornear-commercializedcircuitsareoptimizedacrossspeed,density,powerefficiency,andmanufacturability.Flashmemoryisnotsuitedtoallapplications,havingitsownproblemswithrandom-accesstime,bitalterability,andwritecycles.Withtheincreasingneedtolowerpowerconsumptionwithzero-powerstandbysystems,observersarepredictingthatthetimehascomeforalternativetechnologiestocaptureatleastsomeshareinspecificmarketssuchasautomotivesmartairbags,high-endmobilephones,andRFIDtags.AnembeddednonvolatilememorywithsuperiorperformancetoFlashcouldseewidespreadadoptioninsystem-on-chip(SoC)applicationssuchassmartcardsandmicrocontrollers.

    EmergingNVMtechnologiesforapplicationsThenewemergingnonvolatilerandom-accessmemoryproductsaddresstheurgentneedinsomespecificandsmall-formdevices.Therefore,iRAPfeltaneedtodoadetailedtechnologyupdateandmarketanalysisinthisindustry[42].Recently,YoleDveloppementreportsdescribethatemergingmemorytechnologieshavegreatpotentialtoimprovefuturememorydevicestobeincreasinglyusedinvariousmarketsofindustryandtransportation,enterprisestorage,mobilephones,massstorage,andsmartcards[43].EmergingNVMapplicationsinvariousmarketsareshowninFigure6.Buttherearenumerousopportunitiesexistingfornovelarchitecturesandapplicationsthattheseemergingmemorytechnologiescanenable.ThesenewemergingNVMproductsaddresstheurgentneedinsomespecificandsmall-formdevices.Therefore,emergingnonvolatilememoryproductsprovidemarketdataaboutthesizeofgrowthofthe

  • applicationsegmentsandthedevelopmentsofbusinessopportunities.Untilnow,onlyFeRAM,PCM,andMRAMwereindustriallyproducedandavailableinlow-densitychipstoonlyafewplayers.Thus,themarketwasquitelimitedandconsiderablysmallerthanthevolatileDRAM-andnonvolatileFlashNAND-dominantmarkets(whichenjoyedcombinedrevenuesof$50+billionin2012).However,inthenext5years,thescalabilityandchipdensityofthosememorieswillbegreatlyimprovedandwillsparkmanynewapplicationswithNVMmarketdriversexplainedinmoredetail.

    Viewlargerversion

    Figure6.EmergingNVMapplicationsinvariousmarkets.

    AccompaniedbytheadoptionofSTT-MRAMandPCMcachememory,enterprisestoragewillbethelargestemergingNVMmarket.NVMwillgreatlyimprovetheinput/outputperformanceofenterprisestoragesystemswhoserequirementswillintensifywiththegrowingneedforweb-baseddatasupportedbyfloatingmassservers.Inaddition,mobilephoneswillincreasetheiradoptionofPCMasasubstitutetoFlashNORmemoryinMCPpackagesto1-gigabyte(GB)chipsmadeavailablebyMicronin2012.Higher-densitychips,expectedin2015,willallowaccesstosmartphoneapplicationsthatarequicklyreplacingentry-levelphones.STT-MRAMisexpectedtoreplaceSRAMinSoCapplications,thankstolowerpowerconsumptionandbetterscalability.Smartcardsandmicrocontrollers(MCU)willlikelyadoptMRAM/STT-MRAMandPCMasasubstitutetoembedFlash.Indeed,Flashmemorycellsizereductionislimitedinthefuture.TheNVMcouldreducethecellsizeby50%andthusbemorecost-competitive.Additionalfeatureslikeincreasedsecurity,lowerpowerconsumption,andhigherendurancearealsoappealingNVMattributes.ThemassstoragemarketsservedbyFlashNANDcouldbeginusing3-DRRAMin2017to2018,when3-DNANDwillslowdownitsscalabilityaspredictedbyallofthemainmemoryplayers.Ifthishappens,thenamassiveRRAMramp-upwillcommenceinthenextdecadethatwillreplaceNAND;conditional3-DRRAMcost-competitivenessandchipdensityareavailable.ItisexpectedsurelythattheemergingNVMbusinesswillbeverydynamicoverthenext5years,thankstoimprovementsinscalability/costanddensityofemergingNVMchips[44].AccordingtoarecentlypublishedreportfromYoleDveloppement,EmergingNon-volatileMemoryTechnologies,IndustryTrendsandMarketAnalysis,theglobalmarketforemergingnonvolatilerandom-accessmemoryproductswasprojectedtohavereached$200millionin2012.Thismarketisexpectedtoincreaseto$2,500millionby2018atanaverageannualgrowthataCAGRof+46%throughtheforecastperiodwithmobilephones,smartcards,andenterprisestorageasmaingrowthdrivers(Figure7).Marketadoptionofmemoryisstronglydependentonitsscalability.ThisYoleDveloppementreportprovidesaprecisememoryroadmapintermsoftechnologicalnodes,cellsize,andchipdensityforeachemergingNVMsuchasFeRAM,MRAM/STT-MRAM,PCM,andRRAM.Amarketforecastisprovidedforeachtechnologybyapplication,units,revenues,andalsomarketgrowthasgivenadetailedaccountofemergingNVMmarketforecast(Figure7).PCMdevices,thedensestNVMin2012at1GB,willreach8GBby2018,whichareexpectedtoreplaceNORFlashmemoryinmobilephonesandwillalso

  • beusedasastorageclassmemoryinenterprisestorage.MRAM/STT-MRAMchipswillreach8to16GBin2018.TheywillbewidelysoldasastorageclassmemoryandpossiblyasaDRAMsuccessorinenterprisestorageafter2018.By2018,MRAM/STT-MRAMandPCMwillsurelybethetoptwoNVMonthemarket.Combined,theywillrepresenta$1.6billionbusinessby2018,andtheirsaleswillalmostdoubleeachyear,withdouble-densitychipslaunchedevery2years.FeRAMwillbemorestableintermsofscalability,with8-to16-MBchipsavailableby2018;thedevelopmentofanewFRAMmaterialcouldraisescalability,butwedonotexpectittobewidelyindustrializedandcommercializedbefore2018.FeRAMwillgrowatasteadygrowthrate(10%peryear)andwillfocusonindustrialandtransportationapplicationsbecauseofthelow-densityavailability,whereasRRAMrevenueswouldnotreallysurgeby2018,withtheavailabilityofhigh-densitychipsofseveraltensofgigabytesthatcouldreplaceNANDtechnology.Meanwhile,ithasalsobeenconsideredbymemorytechnologistexpertsthatforlarge-volumemarketslikemassstorageNAND,onlyonetechnologywillbeadoptedinordertoreduceproductioncostandRRAMseemstobethebestcandidate.ButtherealmassiveadoptionofemergingNVMasareplacementforNANDandDRAMwillhappenafter2020.

    Viewlargerversion

    Figure7.EmergingNVMmarketforecastbyapplicationsfrom2012to2018(inM$).Reproducedfromref.[43].

    AdvancesinFlashmemorytechnologiesFlashmemoryisbasicallyaMOSFETnonvolatiledevicethatcanbeelectricallyerasedandreprogrammed[3,45].ItisatechnologythatisprimarilyusedinmemorycardsandFlashdrivesforgeneralstorageandtransferofdatabetweencomputersandotherdigitalproducts.Sincetheinventionofthetransistor,NVSMhadbeenthemostimportantinventionintheelectrondevicefield.Thefloatinggatememorywasusedtostoretheinformationandatunnelingcurrentforprogramminganderasingoperations.Thechargeisinjectedintoorremovedfromthefloatinggateandthefloatinggateremainsinthatstate,evenafterpowerisremoved,whichmeansthatFlashmemoryisnonvolatile.TheinventionofNVSMfurthergaverisetoanewclassofmemorydevicesandhencebroadeneditsapplicationstobecomeubiquitous.TherearealargenumberofproductsinthemarketnowwhichuseFlashdevicesexclusivelyassecondarystorage.Fewexamplesoftheirapplicationsincludemedicaldiagnosticsystems,notebookcomputers,digitalaudioplayers,digitalcameras,mobilephones,personaldigitalassistants,digitaltelevisions,universalserialbus(USB)Flashpersonaldisks,GlobalPositioningSystems,andmanymore.Semiconductorstoragedevicesstoredataintinymemorycellsmadeofverysmalltransistorsandcapacitorsmadeofsemiconductormaterialssuchassilicon.Eachcellcanhold1bitofinformationandanarrayofcellsstoresalargechunkofinformation.Flashdevicesaregainingpopularityoverconventionalsecondarystoragedeviceslikeharddisks.TheFlashmemoryfabricationprocessiscompatiblewiththecurrentCMOSprocessandisasuitablesolutionforembeddedmemoryapplications.A

  • FlashmemorycellissimplyaMOSFETcell,exceptthatapolysiliconfloatinggate[46](orasiliconnitridechargetraplayer)issandwichedbetweenatunneloxideandaninter-polyoxidetoformachargestoragelayer[47].AlthoughFlashmemoryislikelythestandardchargestoragedeviceforthenextgeneration,scalingmayeventuallybelimitedbythetunneloxidelimit[8].Intermsoftheoperationspeedofprogramanderase,Flashmemoryrequiresathintunneloxidetoenhancethecarriertransportbetweenthefloatinggateandthesiliconsubstrate.However,theverythintunneloxidesuffersfrommanyreliabilityissueslikereductioninoperationvoltage,andafteraconsiderablenumberofprogramanderasecycles,thetunneloxideundergoesdeteriorationloss[48].Thus,researchershavefocusedonpossiblesolutionsandproposedalternatetechnologies,includingnitride-basedmemory,nanocrystalmemory,andswitchingmemory.AllothernonvolatilememoriesrequireintegrationofnewmaterialsthatarenotascompatibleastheconventionalCMOSprocess.

    NORandNANDFlashmemorytechnologies

    NORandNANDFlash,twomajorFlashtypes,aredominantinthememorymarket.NORFlashhaslowerdensitybutarandom-accessinterface,whileNANDFlashhashigherdensityandinterfaceaccessthroughacommandsequence[49].TheircorrespondingstructuresareshowninFigure8.NORandNANDFlashcomefromthestructureusedfortheinterconnectionsbetweenmemorycells.Intelisthefirstcompanytointroduceacommercial(NORtype)Flashchipin1988,andToshibareleasedtheworldsfirstNANDFlashin1989[50].Dependingonhowthecellsareorganizedinthematrix,itispossibletodistinguishbetweenNANDFlashmemoriesandNORFlashmemories.InNORFlash,cellsareconnectedinparalleltothebitlines,whichnotablyallowthecellstobereadandprogrammedindividually.TheparallelconnectionofNORFlashcellsresembletheparallelconnectionoftransistorsinaCMOSNORgatearchitecture.Ontheotherhand,inNANDFlash,thecellsareconnectedinseries,resemblingaNANDgate.Theseriesconnectionsconsumelessspacethantheparallelones,reducingthecostofNANDFlash.Itdoesnot,byitself,preventNANDcellsfrombeingreadandprogrammedindividually.Mostoftheengineersandscientistsarenotsofamiliarwiththedifferencesbetweenthesetwotechnologies.Generally,theyusuallyrefertotheNORarchitectureasFlashandareunawareoftheNANDFlashtechnologyanditsmanybenefitsoverNOR[51].ThiscouldbeduetothefactthatmostFlashdevicesareusedtostoreandruncodes(usuallysmall),forwhichNORFlashisthedefaultchoice,althoughweareprovidingsomemajordifferencesbetweenNORandNANDFlashtechnologiesbytheirarchitectureandtheinternalcharacteristicfeaturesoftheindividualFlash.

    Viewlargerversion

    Figure8.ComparisonofNORFlasharrayandNANDFlasharrayarchitectures.

    NORFlashisslowerineraseoperationandwriteoperationcomparedtoNANDFlash[52].ThismeansthatNANDFlashhasfastereraseandwritetimes.Moreover,NANDFlashhassmallereraseunits,sofewererasesareneeded.NORFlashcanreaddataslightlyfasterthanNANDFlash.NORFlashofferscompleteaddressanddatabusesto

  • randomlyaccessanyofitsmemorylocations(addressabletoeverybyte).ThismakesitasuitablereplacementforolderROMBIOS/firmwarechips,whichrarelyneedstobeupdated.Itsenduranceis10,000to1,000,000erasecycles.NORFlashishighlysuitableforstoringcodesinembeddedsystems.Mostoftodaysmicrocontrollerscomewithbuilt-inFlashmemory[53].

    NANDFlashoccupiesasmallerchipareapercell.ThismakesNANDFlashavailableingreaterstoragedensitiesandatlowercostsperbitthanNORFlash.ItalsohasuptotentimestheenduranceofNORFlash.NANDismorefitasstoragemediaforlargefilesincludingvideoandaudio.USBthumbdrives,SDcards,andMMCcardsareofNANDtype[54].NANDsadvantagesarefastwrite(program)anderaseoperations,whileNORsadvantagesarerandomaccessandbytewritecapability.NORsrandomaccessabilityallowsforexecuteinplace(XiP)capability,whichisoftenarequirementinembeddedapplications.NANDisslowrandomaccessible,whileNORishamperedbyhavingslowwriteanderaseperformance.NANDisbettersuitedforfilingapplications.However,moreprocessorsincludeadirectNANDinterfaceandcanbootdirectlyfromNAND(withoutNOR).However,NANDcannotperformreadandwriteoperationssimultaneously;itcanaccomplishtheseatasystemlevelusingamethodcalledshadowing,whichhasbeenusedonPCsforyearsbyloadingtheBIOSfromtheslowerROMintothehigh-speedRAM.

    Table1highlightsthemajordifferencesbetweenNORandNAND.ItshowsthatNANDisidealforhigh-capacitydatastoragewhileNORisbestusedforcodestorageandexecution,usuallyinsmallcapacities.Therearemanyotherdifferencesbetweenthesetwotechnologieswhichwillbefurtherdiscussedindividually.However,thoselistedinthetableareenoughtostronglydifferentiatethetypesofapplicationsusingthem:NORistypicallyusedforcodestorageandexecution.This,mainlyincapacitiesupto4MB,iscommoninapplicationssuchassimpleconsumerappliances,low-endcellphones,andembeddedapplications,whilerawNANDisusedfordatastorageinapplicationssuchasMP3players,digitalcameras,andmemorycards[55-57].ThecodesforrawNAND-basedapplicationsarestoredinNORdevices.

    Seefulltable

    Table1.ComparisonbetweenNORandNANDFlashmemories[55-57]

    ScalingandchallengesofFlashmemorytechnologies

    Currently,therehavebeenincreasingdemandsonreducingthefeaturesizeinmicroelectronicproductsandmoreinterestinthedevelopmentofFlashmemorydevicestomeetthegrowingworldwidedemand.AconventionalFGmemorydevicemusthaveatunneloxidelayerthicknessof8nmtopreventchargelossandtomake10yearsdataretentioncertain.ThisnecessitywilllimitscalabilityforFlashmemorydevices[8,58].Thus,inordertomeettechnologyscalinginthefieldofmemoryanddatastoragedevices,

  • mainstreamtransistor-basedFlashtechnologieswillbedevelopedgraduallytoincorporatematerialandstructuralinnovations[59].Dielectricscalinginnonvolatilememorieshasbeenreachedneartothepointwherenewapproacheswillberequiredtomeetthescalingrequirementswhilesimultaneouslymeetingthereliabilityandperformancerequirementsforfutureproducts.High-dielectric-constantmaterialsarebeingexploredaspossiblecandidatestoreplaceboththetraditionalSiO2andoxide/nitride/oxide(ONO)filmsusedinFlashmemorycells.FlashcellscalinghasbeendemonstratedtobereallypossibleandtobeabletofollowMooreslawdowntothe90-nmtechnologygenerations.Thetechnologydevelopmentandtheconsolidatedknow-howareexpectedtosustainthescalingtrenddowntothe50-nmtechnologynodeandbelowasforecastedbytheInternationalTechnologyRoadmapforSemiconductors(ITRS)inFigure9,whichindicatesthatthesiliconMOSFETwasalreadyinthenanoscale.TheminimumfeaturesizeofanindividualCMOSFEThasshrunkto15nmwithanequivalentgateoxidethickness(EOT)of0.8nmin2001[13].However,semiconductorFlashmemoryscalingisfarbehindCMOSlogicdevicescaling.Forexample,theEOTofthegatestackinsemiconductorFlashmemoryisstillmorethan10nm.Moreover,semiconductorFlashmemorystillrequiresoperationvoltagesofmorethan10V,whichisstillfarfromtheoperationvoltageofCMOSlogicdevices.ItisimportanttoscaletheEOTofthegatestacktoachieveasmallmemorycellsizeandalsoprolongbatterylife.

    Viewlargerversion

    Figure9.ThetrendofMOSFETscalingfromITRS.ReproducedfromITRSCorp.

    AnotherlimitationofFGtechnologyisthattunneloxidescalingislimitedbystress-inducedleakagecurrent(SILC)relatedtochargetransferproblemasindicatedinFigure10[60,61].TheSILKincreaseswithdecreasingoxidethickness.Thiscanbeattributedtotunnelingassistedbythetrapsinthebulkofthedielectric.Trap-assistedtunnelingcantakeplaceatverylowelectricfields.Ifthedensityoftrapsisincreased,theleakagewillalsoincrease.Electricalstresscanincreasethenumberofthesetraps.Soitbecomesanimportantlimitationofscalingdownthememorydevice[62].ForEOT

  • comparedtoSiO2withasimilarEOT[64].Sucharelentlessreductionofdevicedimensionshasmanychallengeslikeretention,endurance,reductioninthenumberofelectronsintheFG,dielectricleakage,cell-to-cellcrosstalk,thresholdvoltageshift,andreductioninmemorywindowmargins[65,66].ThekeyconceptofrealscalingissuessuchasmaterialandstructuralchangesinFlashmemorytechnologiesisprovidedindetailinthenextdistinctpart.

    Viewlargerversion

    Figure10.SchematicplotsofaFlashmemorycellandthedegradationofitstunneloxide.ThedegradationleadstotheformationofpercolationpathsresponsiblefortheFGchargeloss,hencethelossofthestoredinformation.Thepresenceoftrapsintheenergybarrieryieldsthetrap-assistedtunneling

    mechanismandoriginatesthestress-inducedleakagecurrent(SILC).

    FGFlashmemorytechnology

    TheFGNVmemoryisabasicbuildingblockofFlashmemory,whichisbasedonFGthin-filmstorage(TFS)memoriesthathavebeendevelopedwiththeadditionofanerasegateconfiguration.TheconventionalFGmemory(Figure11a)consistsofaMOSFETconfigurationthatismodifiedtoincludepolysiliconasachargestoragelayersurroundedbyaninsulatedinnergate(floatinggate)andanexternalgate(controlgate).ThiswhatmakesFlashmemorynonvolatileandallfloatinggatememoriestohavethesamegenericcellstructure.Chargeistransferredtoorfromthefloatinggatethroughathin(8to10nm)oxide[1,67].Becausethefloatinggateiselectricallyisolatedbytheoxidelayer,anyelectronsplacedonitaretrappedthere.Flashmemoryworksbyadding(charging)orremoving(discharging)electronstoandfromafloatinggate.Abits0or1statedependsuponwhetherornotthefloatinggateischargedordischarged.Whenelectronsarepresentonthefloatinggate,currentcannotflowthroughthetransistorandthebitstateis0.Thisisthenormalstateforafloatinggate.Whenelectronsareremovedfromthefloatinggate,currentisallowedtoflowandthebitstateis1.TheFGmemoryhasachievedhighdensity,goodprogram/erasespeed,goodreliability,andlowoperatingvoltageandpromotesenduranceforFlashmemoryapplication.

    Viewlargerversion

    Figure11.SchematicsoftheconventionalFGmemoryandSONOS.Schematicsof(a)floatinggateandthin-filmstorage-basedembeddednonvolatilememorybitcells,dependingonthechargestoredinsidethegatedielectricofaMOSFET,and(b)thenitride

    traps(SONOS),embeddedintothegateoxideofaMOSFET.

    SONOSmemorytechnology

    InordertosolvethescalingissueoftheFGmemory,theSONOSmemoryhasbeenproposedasaFlashtechnologysincethe1980s[68,69].TheacronymSONOSisderivedfromthestructureofthedeviceasshowninFigure11b.TheSONOSdeviceisbasicallyaMOSFET,wherethegatehasbeenreplacedbyanONOdielectric.TheSONOSmemoryhasabetterchargeretentionthantheFGmemorywhentheFGbitcellstunnelingoxidelayerisbelow10nm[70].Moreover,theSONOSmemoryexhibitsmanyadvantages,e.g.,easytofabricate,highprogram/erase(P/E)speed,lowprogrammingvoltageand

  • powerconsumption,andbetterpotentialforscalabilitybelowthe70-nmnode,accordingtotheITRS[71].Thecharge,holesorelectrons,areinjectedintothenitridelayerusingdirecttunnelingthroughthetunneloxidelayer.Thenitridelayeriselectricallyisolatedfromthesurroundingtransistor,althoughchargesstoredonthenitridedirectlyaffecttheconductivityoftheunderlyingtransistorchannel.SincetheSONOSmemorypossessesspatiallyisolateddeep-leveltraps,asingledefectinthetunnelingoxidewillnotcausedischargeofthememorycell.ThethicknessofthetopoxideisimportanttopreventtheFowler-Nordheimtunnelingofelectronsfromthegateduringerase.Whenthepolysiliconcontrolgateisbiasedpositively,electronsfromthetransistorsourceanddrainregionstunnelthroughtheoxidelayerandgettrappedinthesiliconnitride.Thisresultsinanenergybarrierbetweenthedrainandthesource,raisingthethresholdvoltageVth(thegate-sourcevoltagenecessaryforcurrenttoflowthroughthetransistor).Moreover,thenitridelayeriselectricallyisolatedfromthesurroundingtransistor,althoughchargesstoredonthenitridedirectlyaffecttheconductivityoftheunderlyingtransistorchannel.Theoxide/nitridesandwichtypicallyconsistsofa2-nm-thickoxidelowerlayer,a5-nm-thicksiliconnitridemiddlelayer,anda5-to10-nm-thickoxideupperlayer[72,73].However,SONOS-typeFlashmemorieshaveseveraldrawbackssuchasshallowtrapenergylevel,erasesaturation,andverticalstoredchargemigration[74].Theprogrammingspeedandoperatingvoltageproblemscanbesolvedbyreducingthetunneloxidethickness.Atlowtunneloxidethickness,theissuesthatimpactSONOS-typememoriesincludeerasesaturationandverticalchargemigration,whichseriouslydegradetheretentioncapabilityofthememory[75].Thus,manyconcernsstillremainfortheSONOStypeofmemories,whichwillbediscussedinthenextsection.

    LimitationsofFGandSONOSmemorytechnologies

    Scalingdemandsverythingateinsulatorsinordertokeepshortchanneleffectsandcontroltheshrinkageofthedevicesizeandmaximizetheperformance.Whenthetunnelingoxidethicknessisbelow10nm,thestoragedchargeintheFGiseasytoleakduetoadefectinthetunnelingoxideformedbyrepeatedwrite/erasecyclesordirecttunnelingcurrent.

    ThetunnelinggateoxidethicknessinaconventionalFlashmemorycannotbescaleddowntosub-7nmbecauseofchargeretention[76].TheSONOSFlashmemorycanrelievetheproblembutstillhasarelativelythickgatedielectricthicknessofabout7nm.Therefore,conventionalSONOSFlashmemoryalsohasascaling-downproblem.ManystudieshaveshownthatthechargeretentioncharacteristicsinscaledSONOSnonvolatilememorydeviceswithalowgateoxidethicknessandathightemperatureareproblematicwithshallow-leveltraps[48,77,78].FortheconventionalSONOSmemory,erasesaturationandverticalstoredchargemigration[79,80]arethetwomajordrawbacks;themostchallengingtasksarehowtomaintainanacceptablechargecapabilityofthediscretestoragenodesandhowtofabricatenanocrystalswithconstantsize,highdensity,anduniformdistributions[81].Whenthetrapenergylevelisshallow,erasesaturationandverticalmigrationoccurandtheelectronchargedecayrateincreasesduetolowtunneloxidethickness,issuesthatimpactSONOS-typememoriesasshowninFigure12.ThiserasesaturationmakesSONOSeraselessastheerasevoltageorthetunneloxidethicknessisincreased.SincetheSONOSmemoryusessiliconnitrideasachargetrappinglayer,theelectronsintheSisub-conductionbandwilltunnelthroughthetunnelingoxide

  • andaportionofthenitride,andthisconsequentlydegradestheprogramspeed.Besidesthis,theconductionbandoffsetofnitrideisonly1.05eVandback-tunnelingofthetrappedelectronmayalsooccur.Althoughapplyingaveryhighelectricfieldmayacceleratethede-trappingrate,thegateelectroninjectioncurrentexceedsthede-trappingbutresultinginpracticallyanincreaseinchargeandnoerasing.Usinganultra-thin(
  • theidealcharacteristicsofnewemergingmemorytechnologieshavetobemeetingtheperformanceofSRAMandthedensityofNANDFlashintermsofstability,scalability,andswitchingspeed.Thus,goingbeyondthetraditionalbistablememory,thepossibilitiesofmultilevel,high-performancememorydevicessuitableformarketmustbeexplored.Currently,thereareseveraltechnologiesthatshowsomepromise;someofthesenewemergingtechnologiesareMRAM,FeRAM,PCM,STT-RAM,nano-random-accessmemory(NRAM),racetrackmemory,RRAMandmemristor,molecularmemory,andmanyothers[10,83].Eachofthesememorytechnologieswillbebrieflyoutlinedanddiscussedinthefollowingsections.Inviewofthecommercialproduction,currently,MRAM,FeRAM,andPCMareincommercialproductionbutstillremainlimitedtonicheapplicationsrelativetoDRAMandNANDFlash.Thereisaprospectthatamongtheemergingmemorytechnologies,MRAM,STT-RAM,andRRAMarethemostpromisingones,buttheyarestillmanyyearsawayfromcompetingforindustryadoption[84].Itisnecessaryforanynewtechnologytobeabletodelivermostforindustryadoption.Forindustryadoptiononamassscale,someparametersmustbematchedwithexistingmemorytechnologies.Inconsiderationofnewtechnologyforindustryapplication,thescalabilityofthetechnology,speedofthedevice,powerconsumptiontobebetterthanexistingmemories,endurance,densities,betterthanexistingtechnologiesandmostimportantlythecost;iftheemergingtechnologycanonlyrunoneortwooftheseattributes,then,atmostdesirable,itislikelytoberesignedtonicheapplications.

    MRAM

    MRAMormagneticRAMisanonvolatileRAMtechnologyunderdevelopmentsincethe1990s.RRAMmethodsofstoringdatabitsusemagneticchargesinsteadoftheelectricalchargesusedbyDRAMandSRAMtechnologies.MRAM,firstdevelopedbyIBMinthe1970s[85],isexpectedtoreplaceDRAMasthememorystandardinelectronics.MRAMisbasicallybasedonmemorycellshavingtwomagneticstorageelements,onewithafixedmagneticpolarityandanotherwithaswitchablepolarity.ThesemagneticelementsarepositionedontopofeachotherbutseparatedbyathininsulatingtunnelbarrierasshowninthecellstructureinFigure13.Moreover,scientistsdefineametalasmagnetoresistiveifitshowsaslightchangeinelectricalresistancewhenplacedinamagneticfield.BycombiningthehighspeedofstaticRAMandthehighdensityofDRAM,proponentssaythatMRAMcouldbeusedtosignificantlyimproveelectronicproductsbystoringgreateramountsofdata,enablingittobeaccessedfasterwhileconsuminglessbatterypowerthanexistingelectronicmemories.Technically,itworkswiththestateofthecell,whichissensedbymeasuringtheelectricalresistancewhilepassingacurrentthroughthecell.Becauseofthemagnetictunneleffect[86],ifbothmagneticmomentsareparalleltoeachother,thentheelectronswillbeabletotunnelandthecellisinthelowresistanceONstate.However,ifthemagneticmomentsareantiparallel,thecellresistancewillbehigh.ThememorycharacteristicsofMRAMofwritinganderasingarefulfilledbypassingacurrentthroughthewritelinetoinduceamagneticfieldacrossthecell.MRAMhasbeenslowlygettingoffthegroundbuthasnowenteredthemarketandwillbecomeincreasinglyavailableformassproductioninthecoupleofyearsandbeyond.Currently,ithasreachedsomelevelofcommercialsuccessinnicheapplications[87].VariouscompaniessuchasSamsung,IBM,HitachiandToshiba,andTSMCareactivelydevelopingvarianttechnologiesofMRAMchips.Inviewof

  • powerconsumptionandspeed,MRAMcompetesfavorablythanotherexistingmemoriessuchasDRAMandFlash,withanaccesstimeofafewnanoseconds[88-90].Althoughithassomelimitationduringthewriteoperation,thesmallercellsizecouldbelimitedbythespreadofthemagneticfieldintoneighboringcellsandneedanamendmenttocompetecompletelyasauniversalmemory.ThepriceofMRAMisalsoanotherissueandconsideredalimitingfactor,withpricesfarinexcessofallthecurrentlyestablishedmemoriesatapproximately2to3($3to$5)permegabyte[91].Accordingtothispricelevel,MRAMisinexcessof1,000timesthepriceofFlashmemoryandover10,000timesthepriceofharddiskdrives.Itisexpectedthatofthenext-generationmemorytechnologies,MRAM,inthefuture,willhavethebiggestmarket,followedbyFeRAM,PCRAM,andmemristors.

    Viewlargerversion

    Figure13.BasicMRAMcellstructure.

    STT-MRAM

    STT-MRAMisamagneticmemorytechnologythatexertsthebaseplatformestablishedbyanexistingmemorycalledMRAMtoenableascalablenonvolatilememorysolutionforadvancedprocessnodes[92,93].ItisanewkindofmagneticRAMwiththefollowingfeatures:fastreadandwritetimes,smallcellsizes,potentiallyevensmaller,andcompatibilitywithexistingDRAMandSRAM.Aswehavediscussedintheprevioussection,MRAMstoresdataaccordingtothemagnetizationdirectionofeachbitandthenanoscopicmagneticfieldssetthebitsinconventionalMRAM.Ontheotherhand,STT-MRAMusesspin-polarizedcurrents,enablingsmallerandlessenergy-consumingbits.ThebasiccellstructureofSTT-RAMisdepictedinFigure14.Inaddition,STT-RAMwritingisatechnologyinwhichanelectriccurrentispolarizedbyaligningthespindirectionoftheelectronsflowingthroughamagnetictunneljunction(MTJ)element.Datawritingisperformedbyusingthespin-polarizedcurrenttochangethemagneticorientationoftheinformationstoragelayerintheMTJelement[94].TheresultantresistancedifferenceoftheMTJelementisusedforinformationreadout.STT-RAMisamoreappropriatetechnologyforfutureMRAMproducedusingultra-fineprocessesandcanbeefficientlyembeddedinsubsequentgenerationsofsuchsemiconductordevicesasFPGAs,microprocessors,microcontrollers,andSoC.AspecialbonusforembeddeddesignersisthefactthattheinternalvoltageSTT-RAMrequiresisonly1.2V.ThedifferencebetweenSTT-MRAMandaconventionalMRAMisonlyinthewritingoperationmechanism;thereadsystemisthesame.ThememorycellofSTT-MRAMiscomposedofatransistor,anMTJ,awordline(WL),abitline(BL),andasourceline(SL)[95].Currently,STT-RAMisbeingdevelopedincompaniesincludingEverspin,Grandis,Hynix,IBM,Samsung,TDK,andToshiba.However,forSTT-RAMtobeadoptedasauniversalmainstreamsemiconductormemory,somekeychallengesshouldberesolved:thesimultaneousachievementoflowswitchingcurrentandhighthermalstability.Itmustbedense(approximately10F2),fast(below10nsofreadandwritespeeds),andoperatingatlowpower[96].

  • Viewlargerversion

    Figure14.BasicSTT-RAMcellstructure.

    FeRAM

    FeRAMisanonvolatileRAMthatcombinesthefastreadandwriteaccessofDRAMcells,consistingofacapacitorandtransistorstructureasshowninFigure15.Thecellisthenaccessedviathetransistor,whichenablestheferroelectricstateofthecapacitordielectrictobesensed.Inspiteofitsname,FeRAMdoesnotcontainiron.Thepolarizationpropertiesofaferroelectricsubstanceareusedasamemorydevice.TodaysFeRAMusesleadzirconatetitanate(PZT);othermaterialsarebeingconsidered.ThemaindeveloperofFeRAMisRamtronInternational.FeRAMisthemostcommonkindofpersonalcomputermemorywiththeabilitytoretaindatawhenpoweristurnedoffasdoothernonvolatilememorydevicessuchasROMandFlashmemory[97].InaDRAMcell,thedataperiodicallyneedrefreshingduetothedischargingofthecapacitor,whereasFeRAMmaintainsthedatawithoutanyexternalpowersupply.Itachievesthisbyusingaferroelectricmaterialintheplaceofaconventionaldielectricmaterialbetweentheplatesofthecapacitor.Whenanelectricfieldisappliedacrossdielectricorferroelectricmaterials,itwillpolarize,andwhilethatfieldisremoved,itwilldepolarize.Buttheferroelectricmaterialexhibitshysteresisinaplotofpolarizationversuselectricfield,anditwillretainitspolarization.OnedisadvantageofFeRAMisthathasadestructivereadcycle.Thereadmethodinvolveswritingabittoeachcell;ifthestateofthecellchanges,thenasmallcurrentpulseisdetectedbyindicatingthatthecellwasintheOFFstate.However,itisafastmemorythatcanendureahighnumberofcycles(e.g.,1014)[98],meaningthattherequirementforawritecycleforeveryreadcyclewillnotresultinshortproductliveswithaverylowpowerrequirement.Itisexpectedtohavemanyapplicationsinsmallconsumerdevicessuchaspersonaldigitalassistants(PDAs),handheldphones,powermeters,andsmartcards,andinsecuritysystems.FeRAMisfasterthanFlashmemory.ItisalsoexpectedtoreplaceEEPROMandSRAMforsomeapplicationsandtobecomeakeycomponentinfuturewirelessproducts.EvenafterFeRAMhasachievedalevelofcommercialsuccess,withthefirstdevicesreleasedin1993[99,100],currentFeRAMchipsofferperformancethatiseithercomparabletoorexceedingcurrentFlashmemories[98,101],butstillslowerthanDRAM.

    Viewlargerversion

    Figure15.BasicstructureofaFeRAMcell.Thecrystalstructureofaferroelectricandanelectricpolarization-electricfieldhysteresiscurvearealsoshown.

    PCRAM

    PCRAM,alsoknownasPCM,perfectRAM(PRAM),OUM,andchalcogenideRAM(CRAM),isatypeofnonvolatileRAMbasedonaclassofmaterialcalledchalcogenideglassesthatcanexistintwodifferentphasestates(e.g.,crystallineandamorphous)[102,103].ThebasicPCRAMcellstructureisdepictedinFigure16.Mostphase-change

  • materialscontainatleastoneelementfromgroup6oftheperiodictable,andthechoiceofavailablematerialscanbefurtherwidenedbydopingthesematerials[104-107].Inparticular,themostpromisingaretheGeSbTealloyswhichfollowapseudobinarycomposition(betweenGeTeandSb2Te3),referredtoasGST.Thesematerialsareinfactcommonlyusedasthedatalayerinrewritablecompactdisksanddigitalversatiledisks(CD-RWandDVD-RW)wherethechangeinopticalpropertiesisexploitedtostoredata.Thestructureofthematerialcanchangerapidlybackandforthbetweenamorphousandcrystallineonamicroscopicscale.Thematerialhaslowelectricalresistanceinthecrystallineororderedphaseandhighelectricalresistanceintheamorphousordisorderedphase.ThisallowselectricalcurrentstobeswitchedONandOFF,representingdigitalhighandlowstates.Thisprocesshasbeendemonstratedtobeontheorderofafewtensofnanoseconds[108],whichpotentiallymakesitcompatiblewithFlashforthereadoperation,butseveralordersofmagnitudefasterforthewritecycle.ThismakesitpossibleforPCMtofunctionmanytimesfasterthanconventionalFlashmemorywhileusinglesspower.Inaddition,PCMtechnologyhasthepotentialtoprovideinexpensive,high-speed,high-density,high-volumenonvolatilestorageonanunprecedentedscale.Thephysicalstructureisthree-dimensional,maximizingthenumberoftransistorsthatcanexistinachipoffixedsize.PCMissometimescalledperfectRAMbecausedatacanbeoverwrittenwithouthavingtoeraseitfirst.PossibleproblemsfacingPCRAMconcernthehighcurrentdensityneededtoerasethememory;however,ascellsizesdecrease,thecurrentneededwillalsodecrease.PCMchipsareexpectedtolastseveraltimesaslongascurrentlyavailableFlashmemorychipsandmayprovecheaperformassproduction.WorkingprototypesofPCMchipshavebeentestedbyIBM,Infineon,Samsung,Macronix,andothers.Also,theproductionofPCMhasbeenannouncedrecentlybybothcollaborationsbetweenIntelandSTMicroelectronicsaswellaswithSamsung[109,110].

    Viewlargerversion

    Figure16.BasicPCRAMcellstructure.ReproducedfromIBM-Macronix-Qimonda.

    ComparisonofprimarycontendersforMRAM,STT-RAM,FeRAM,andPCMtechnologies

    Beforegoingtootheremergingmemories,wehereinprovideacomparisonamongMRAM,FeRAM,andPCM.ThespecificfeaturesofthesememorydevicesareprovidedinTable2.Relativelymature,new-materialmemoriessuchasMRAM,STT-RAM,FeRAM,andPCMcanofferavarietyoffeaturesthathavepotentialtobethecandidatesfornext-generationnonvolatilememorydevices.Brand-newconceptssuchasRRAM,molecular,organic/polymer,andothernanowire-basedmemorytechnologieshavealsobeenproposed.Thesearediscussedindetailinthefollowingsection.

  • Seefulltable

    Table2.SummaryofprimarycontendersforMRAM,FeRAM,STT-RAM,andPCMtechnologies

    RRAM

    RRAMisadisruptivetechnologythatcanrevolutionizetheperformanceofproductsinmanyareas,fromconsumerelectronicsandpersonalcomputerstoautomotive,medical,military,andspace.Amongallthecurrentmemorytechnologies,RRAMisattractingmuchattentionsinceitiscompatiblewiththeconventionalsemiconductorprocesses.Memristor-basedRRAMisoneofthemostpromisingemergingmemorytechnologiesandhasthepotentialofbeingauniversalmemorytechnology[111].Itoffersthepotentialforacheap,simplememorythatcouldcompeteacrossthewholespectrumofdigitalmemories,fromlow-cost,low-performanceapplicationsuptouniversalmemoriescapableofreplacingallcurrentmarket-leadingtechnologies,suchasharddiskdrives,random-accessmemories,andFlashmemories[112].RRAMisasimple,two-terminalmetal-insulator-metal(MIM)bistabledeviceasshowninthebasicconfigurationinFigure17.Itcanexistintwodistinctconductivitystates,witheachstatebeinginducedbyapplyingdifferentvoltagesacrossthedeviceterminals.RRAMusesmaterialsthatcanbeswitchedbetweentwoormoredistinctresistancestates.Manycompaniesareinvestingmetaloxidenanolayersswitchedbyvoltagepulses.Researchersgenerallythinkthatthepulseselectricfieldsproduceconductingfilamentsthroughtheinsulatingoxide.HPLabsplanstoreleaseprototypechipsthisyearbasedonmemristorsinwhichmigratingoxygenatomschangeresistance[113].Xuetal.havealsodefinedthatamongallthetechnologycandidates,RRAMisconsideredtobethemostpromisingasitoperatesfasterthanPCRAMandithasasimplerandsmallercellstructurethanmagneticmemories(e.g.,MRAMorSTT-RAM)[114].IncontrasttoaconventionalMOS-accessedmemorycell,amemristor-basedRRAMhasthepotentialofformingacross-pointstructurewithoutusingaccessdevices,achievinganultra-highdensity.Thisdeviceisbasedonthebistableresistancestatefoundforalmostanyoxidematerial,includingNiO,ZrO2,HfO2,SrZrO3,andBaTiO3[115-119].Currently,SamsungandIBMareactivelyinvestigatingRRAM.

    Viewlargerversion

    Figure17.BasicRRAMcellstructure.Aschematicdiagramofthemechanismoftheresistiveswitchinginametal/oxide/metal-structuredmemorycellisalsoshown.Reproducedfromref.[123].

    Kamiyaetal.haverevealedbyatheoreticalmechanismthatRRAMshowsfilamentary-typeresistiveswitching,wheretheoxygenvacancyisconsideredtoformconductivefilamentsintheresistivematerialasshowninFigure17[120].TheformationanddisruptionofthesefilamentsarethusthemechanismsresponsiblefortheON-OFFswitchinginRRAMdevices.Thekeyissueis,therefore,torevealelectronicrolesinthe

  • formationanddisruptionofthevacancyfilaments.RRAMcanbeswitchedbetweenthelowresistancestate(LRS)andthehighresistancestate(HRS)oftheresistivematerialbyapplyingvoltagestotheelectrodes.LeehasexplainedthatduringtheSETprocess,thecurrentlevelincreasesfromHRStoLRSasthevoltageincreasesfrom0Vtothecriticalpointwhichiscalledthesetvoltage(Vset),whilethecurrentlevelabruptlydecreasesfromLRStoHRSattheresetvoltage(Vreset)undertheRESETprocess.TheSETandRESETprocessesarerepeatedlycarriedoutbysweepingthegatevoltagewiththebinarystatesLRSandHRS[121].WangandTsengandLinetal.haveindicatedthattheinterfaceplaysanimportantroleinenhancingtheperformancesofRRAM[122,123].Recently,Gouxetal.haveexplainedthatusingastackedRRAMstructurehasbeenshowntobeoneofthemostpromisingmethodstoimprovethememorycharacteristics[124].Althoughbeingamostpromisingmemoryelement,criticalissuesforthefuturedevelopmentofRRAMdevicesarereliable,suchasdataretentionandmemoryendurance[125].Adataretentiontimeofover10yearscanbeextrapolatedfromretentioncharacteristicsmeasuredathightemperaturesandamemoryenduranceofover106cycles[126].Therefore,astatisticalstudyofreliability,availability,andmaintainabilityisessentialforthefuturedevelopmentofRRAM.

    Polymermemory

    Throughoutthelastfewyears,polymershavefoundgrowinginterestasaresultoftheriseofanewclassofnonvolatilememories.Inapolymermemory,alayerconsistsofmoleculesand/ornanoparticlesinanorganicpolymermatrixissandwichedbetweenanarrayoftopandbottomelectrodesasillustratedinFigure18.Moreover,polymermemoryhastheadvantageofasimplefabricationprocessandgoodcontrollabilityofmaterials[127].Polymermemorycouldbecalleddigitalmemorywiththelatesttechnology.Itisnotpossibleforasilicon-basedmemorytobeestablishedinlessspace,butitispossibleforpolymermemory.Lingetal.explainedthatpolymermaterialshavesimplicityinstructure,freereadandwritecapability,betterscalability,3-Dstackingability,low-costpotential,andhugecapacityofdatastorage[128].Theyrevealedthatapolymermemorystoresinformationinamannerthatisentirelydifferentfromthatofsilicon-basedmemorydevices.Ratherthanencoding0and1fromthenumberofchargesstoredinacell,apolymermemorystoresdataonthebasisofhighandlowconductivitywhilerespondingtoanappliedvoltage.Amongthelargenumberofemergingmemorytechnologies,polymermemoryistheleadingtechnology.Itismainlybecauseofitsexpansioncapabilityin3-Dspace[129]sincemostpolymersareorganicmaterialsconsistingoflongchainsofsinglemolecules.Priortopolymermemoryfabrication,depositionofanorganiclayerisusuallydonebythesol-gelspincoatingtechnique.Alltheothernecessaryconstituentmaterialsaredissolvedinasolventwhichisthenspin-coatedoverasubstrate.Whenthesolventisevaporated,athinfilmofmaterialwith10-to100-nmthicknessissuccessfullydepositedatbottomelectrodes.Topelectrodesaredepositedasthefinalstep.Theconductivityoftheorganiclayeristhenchangedbyapplyingavoltageacrossthememorycell,allowingbitsofdatatobestoredinthepolymermemorycell.Whenthepolymermemorycellbecomeselectricallyconductive,theelectronsareintroducedandremoved.Eventhepolymerisconsideredasasmartmaterialtotheextentthatfunctionalityisbuiltintothematerialitselfofswitchabilityandchargestore.Thiswillopenuptremendousopportunitiesinthe

  • electronicsworld,wheretailor-madememorymaterialsrepresentanunknownterritory.Thenonvolatilenessandotherfeaturesareinbuiltatthemolecularlevelandoffersveryhighadvantagesintermsofcost.Butturningpolymermemoryintoacommercialproductwouldnotbeeasy.Memorytechnologiescompetenotonlyonstoragecapacitybutonspeed,energyconsumption,andreliability.Thedifficultyisinmeetingalltherequirementsofcurrentsiliconmemorychips,saysThomas,theDirectorofPhysicalSciencesatIBMsWatsonResearchCenterinYorktownHeights,NY.Theyarelikelytobelimitedtonicheapplications[130].

    Viewlargerversion

    Figure18.Structureofapolymermemorydevice.

    Racetrackmemory

    Inaracetrackmemory,informationisstoredonaU-shapednanowireasapatternofmagneticregionswithdifferentpolarities.TheU-shapedmagneticnanowireisanarrayofkeys,whicharearrangedverticallyliketreesinaforestasshowninFigure19.AchievingcapacitiescomparabletoverticalRMorharddriveswouldrequirestacksofthesearrays.Thenanowireshaveregionswithdifferentmagneticpolarities,andtheboundariesbetweentheregionsrepresent1or0s,dependingonthepolaritiesoftheregionsoneitherside[131,132].Themagneticinformationitselfisthenpushedalongthewire,pastthewriteandreadheadsbyapplyingvoltagepulsestothewireends.Themagneticpatterntospeedalongthenanowire,whileapplyingaspin-polarizedcurrent,causesthedatatobemovedineitherdirection,dependingonthedirectionofthecurrent.AseparatenanowireperpendiculartotheU-shapedracetrackwritesdatabychangingthepolarityofthemagneticregions.Aseconddeviceatthebaseofthetrackreadsthedata.Datacanbewrittenandreadinlessthanananosecond.Aracetrackmemoryusinghundredsofmillionsofnanowireswouldhavethepotentialtostorevastamountsofdata[133,134].Inthisway,thememoryrequiresnomechanicalmovingofpartsandithasagreaterreliabilityandhigherperformancethanHDDs,withtheoreticalnanosecondoperatingspeeds.Foradeviceconfigurationwheredatastoragewiresarefabricatedinrowsonthesubstrate,conventionalmanufacturingtechniquesareadequate.However,forthemaximumpossiblememorydensity,thestoragewiresareproposedtobeconfiguredrisingfromthesubstrateinaUshape,givingrisetoa3-Dforestofnanowires.Whilethislayoutdoesallowhighdatastoragedensities,italsohasthedisadvantageofcomplexfabricationmethods,withsofar,only3-bitoperationofthedevicesdemonstrated[133].Astheaccesstimeofthedataisalsodependentonthepositionofthedataonthewire,thesewouldalsobeperformancelossesiflongwiresareusedtoincreasethestoragedensityfurther.Thespeedofoperationofthedeviceshasalsobeenanissueduringdevelopment,withmuchslowermovementofthemagneticdomainsthanoriginallypredicted.Thishasbeenattributedtocrystalimperfectionsinthepermalloywire,whichinhibitthemovementofthemagneticdomains.Byeliminatingtheseimperfections,adatamovementspeedof110m/shasbeendemonstrated[133].

  • Viewlargerversion

    Figure19.RacetrackmemorydiagramshowinganarrayofU-shapedmagneticnanowires.Thenanowiresarearrangedverticallyliketreesinaforestandapairoftinydevicesthatreadandwritethedata.AdoptedfromIBM.

    OthernewmemorytechnologiesResearchersarealreadyworkinghardonseveralemergingtechnologies,asdiscussedinprevioussections,topursuestorage-classmemorieswithamoretraditionaldesignthanthatoftheracetrackmemory,whichplacesthebitsinhorizontalarrays.

    Molecularmemory

    Amolecularmemoryisanonvolatiledatastoragememorytechnologythatusesmolecularspeciesasthedatastorageelement,ratherthan,e.g.,circuits,magnetics,inorganicmaterials,orphysicalshapes[135].Inamolecularmemory,amonolayerofmoleculesissandwichedbetweenacross-pointarrayoftopandbottomelectrodesasshowninFigure20.Themoleculesarepackedinahighlyorderedway,withoneendofthemoleculeelectricallyconnectedtothebottomelectrodeandtheotherendofthemoleculeconnectedtothetopelectrode,andthismolecularcomponentisdescribedasamolecularswitch[136].Langmuir-Blodgett(LB)depositionisideallysuitedfordepositingthemolecularlayerforthefabricationofmolecularmemorydevices[137,138].Then,regardingthemolecularmemoryoperation,byapplyingavoltagebetweentheelectrodes,theconductivityofthemoleculesisaltered,enablingdatatobestoredinanonvolatileway.Thisprocesscanthenbereversed,andthedatacanbeerasedbyapplyingavoltagetotheoppositepolarityofthememorycell.Theincreasingdemandfornonvolatileelectronicmemorieswillgrowrapidlyinordertokeeppacewiththerequirementsforsubsystemsinvolvedinflightdemonstrationprojectsanddeepspaceoperations.Atthesametime,mass,volume,andpowermustbeminimizedformissionaffordabilityconcerningtheserequirements;molecularmemorycouldbeaverypromisingcandidatetofillthisneed.

    Viewlargerversion

    Figure20.Cellstructureofamolecularmemorydevice.

    Recently,Plafkehasrevealedclearlyviaanarticlethatlikemostexperimentaltechnologythatsoundssoamazingthatwewantitrightnow,themolecularmemorycelldoesnotprovideenoughpowerforacommercialdevice[139].Thisiscurrentlyonlyabletoproducea20%jumpinconductivity.However,theareaofmolecularswitchingmemoryispromising,havingeliminatedtheneedfornear-absolutezerotemperaturesandremovedsomeoftheconstraintsoftheshapeandnumberoflayersofthemoleculesheetswhichintendtoconveythattwoofthebiggestbarriersaretakenaway.Thus,molecularmemoryrequiresstrongattentiontoworkoversuchissuesandneedsimmediateamendmenttoseethepossibilityofauniversalmemoryinthefuture.

    MNW

  • Inthelasttwodecades,anincreasinginterestisobservedforelectronics-relateddevicesandthesearchforauniversalmemorydatastoragedevicethatcombinesrapidreadandwritespeeds,highstoragedensity,andnonvolatilityisdrivingtheinvestigationofnewmaterialsinthenanostructuredform[140].AsanalternativetothecurrentFlashmemorytechnology,anoveltransistorarchitectureusingmolecular-scalenanowirememorycellsholdsthepromiseofunprecedentlycompactdatastorage.Themolecularnanowirearray(MNW)memoryisfundamentallydifferentfromothersemiconductormemories;informationstorageisachievedthroughthechannelofananowiretransistorthatisfunctionalizedwithredox-activemoleculesratherthanthroughmanipulationofsmallamountsofcharge.Itisrelativelyslowandlackstherandomaccesscapability,whereindatathatcanberandomlyreadandwrittenateverybytearebeingactivelypursued.Figure21showstheschematicdesignofaMNWmemorycell.Lieber,andAgarwalandLieberhaverevealedthatthenanowire-basedmemorytechnologyisapowerfulapproachtoassembleelectronic/photonicdevicesatultra-smallscalesowingtotheirsub-lithographicsize,defect-freesingle-crystallinestructure,anduniquegeometry[141,142].Nanowiressynthesizedbychemicalorphysicalprocessesarenearlyperfectsingle-crystalstructureswithasmallgeometryandperfectsurface.Thechannelofananowiretransistorisfunctionalizedwithredox-activemolecules.Duringprogramming,controlofthevoltageactingonthesubstrateispossibletochangetheoxidationandreductionstatesoftheactivemolecules.Finally,bymeasurementoftheconductanceofthenanowirewiththegatebiasfixedat0Vorasmallvoltageandfromthehysteresis,thetwostatescanbedefinedasahigh-conductanceONstateandalow-conductanceOFFstate.TheMNWmemoryhasadvantagesoflowpowerdissipation,ultra-highdensity,simplefabricationprocess,3-Dstructure,andmultilevelstorage,anditfunctionsatthenanoscalewithafewelectronsbutlimitedbylowretentiontimeparameter[143,144].Moreover,thedepositionofmetalsontoamonolayerofmolecularwirescanleadtolowdeviceyield,andthisproblemremainsamajorchallenge[145].However,mentioningthetermemergingclassmemory,itcouldbeexpectedthattheMNWmemoryrepresentsanimportantsteptowardsthecreationofmolecularcomputersthataremuchsmallerandcouldbemorepowerfulthantodayssilicon-basedcomputers.

    Viewlargerversion

    Figure21.AMNWmemorycellstructure.

    SNW

    Semiconductormemoryisessentialforinformationprocessingasakeypartofsilicontechnology;semiconductormemoryhasbeencontinuouslyscaledtoachieveahigherdensityandbetterperformanceinaccordancewithMooreslaw[146].Flashmemorymayreachfundamentalscalinglimits,however,becauseathicktunnelingoxideisrequiredtopreventchargeleakageandachieve10yearsretention.AsFlashmemoryapproachesitsscalinglimit,severalalternativestrategieshavebeenproposedtoextendorreplacethecurrentFlashmemorytechnology[147].Theseapproachesarerevolutionary,butmajorchallengesmustbeovercometoachievesmallmemorysizeandaggressivetechnologydesignarchitecture.Inadditiontotheengineeringoftrappinglayers,thedevice

  • performancecanalsobeimprovedbyusinginnovativenonplanarchannelgeometries.Amongthevariousnanostructurematerials,semiconductornanowirememory(SNW)hasinducedgreatscientificinterestaspossiblebuildingblocksforfuturenanoelectroniccircuitry.InaSNWmemorydevice,nanowiresareintegratedwithSONOStechnology.ThebasicschematicdesignofSNWisdepictedinFigure22.TheSNWmemoryshowshighmobility,lesspowerdissipation,andhighperformance.Moreover,being3-D-stacked,theSNWmemoryenhancescelldensityanddatacapacitywithoutrelyingonadvancesinprocesstechnology.Thenanowire-basedmemorydevicecanstoredataelectricallyandisnonvolatile,meaningitretainsdatawhenthepoweristurnedoff,likethesilicon-basedFlashmemoryfoundinsmartphonesandmemorycards[148],withminimalincreaseinchipsize.Inaddition,theSNWdeviceexhibitsreliablewrite/read/eraseoperationswithalargememorywindowandhighon-to-offcurrentratio,whicharehighlyadvantageousforapplicationsinnonvolatilememory[149].TheSNWmemorycannotholddataaslongastheexistingFlash,butitisslowerandhasfewerrewritecyclesanditcouldpotentiallybemadesmallerandpackedtogethermoredensely.Anditsmainadvantageisthatitcanbemadeusingsimpleprocessesatroomtemperature,whichmeansthatitcanbedepositedevenontopofflexibleplasticsubstrates[150].TheSNWcould,forinstance,bebuiltintoaflexibledisplayandcouldbepackedintosmallerspacesinsidecellphones,MP3players,plasticRFIDtags,andcreditcards.

    Viewlargerversion

    Figure22.Abottom-gateFET-basednonvolatileSNWmemorydevice.

    NRAM

    NRAMisacarbonnanotube(CNT)-basedmemory,whichworksonananomechanicalprinciple,ratherthanachangeinmaterialproperties[151].NRAMusescarbonnanotubesforthebitcells,andthe0or1isdeterminedbythetubesphysicalstate:upwithhighresistance,ordownandgrounded.NRAMisexpectedtobefasteranddenserthanDRAMandalsoveryscalable,abletohandle5-nmbitcellswheneverCMOSfabricationadvancestothatlevel.Itisalsoverystableinits0or1state.ProducedbyNantero,thesememoriesconsistofthestructureshowninFigure23awithanarrayofbottomelectrodescoveredbyathininsulatingspacerlayer[152].CNTsarethendepositedonthespacerlayer,leavingthemfreestandingabovethebottomelectrodes.UnwantedCNTsareremovedfromtheareasaroundtheelectrode,withtopcontactsandinterconnectsdepositedontopofthepatternedCNTlayer.DuringthetimethattheCNTsarefreestanding,thereisnoconductionpathbetweenthebottomandtopelectrodesandhencethememorycellisintheOFFstate.However,ifalargeenoughvoltageisappliedoverthecell,thenanotubesareattractedtothebottomelectrodewheretheyareheldinplacebyvanderWaalsforces[153].DuetotheconductivenatureoftheCNTs,theelectrodesarenowconnectedandthecellreadsthelowconductivityONstateasshowninFigure23b.TheOFFstatecanbereturnedbyrepellingthenanotubeswiththeoppositeelectrodepolarity.NonvolatilityisachievedduetothestrengthofthevanderWaalsforcesovercomingthemechanicalstrainofthebentnanotubes,henceholdingthecellintheONstate.NRAMoffersthepossibilityofasimplecellarchitecture,whichcouldoperateat

  • muchhigherspeedsthantheconventionalFlashandwithlowpoweruse.Cuietal.reportedCNTmemorydevicesexhibitinganextraordinarilyhighchargestoragestabilityofmorethan12daysatroomtemperature[154].However,asNRAMisbasedonCNTs,itsuffersfromfabricationproblemsthatareinherentincarbonnanotube-baseddevices.TheissuesincludethecostandfabricationcomplexityofproducingtheCNTs,ensuringuniformdispersionsofnanotubes,anddifficultiesinremovingnanotubesfromtheunwantedpositionsonthesubstrate.

    Viewlargerversion

    Figure23.NRAMstructurewith(a)OFFstateand(b)ONstate.

    Millipedememory

    In2002,IBMdevelopedapunchcardsystemknownasMillipede,whichisanonvolatilecomputermemorystoredinathinpolymersheetwithnanoscopicholestoprovideasimplewaytostorebinarydata[155].Itcanstorehundredsofgigabytesofdatapersquarecentimeter.However,thepolymerrevertstoitspre-punchedformovertime,losingdataintheprocess.Millipedestoragetechnologyisbeingpursuedasapotentialreplacementformagneticrecordinginharddrives,atthesametimereducingtheformfactortothatofFlashmedia.Theprototypescapacitywouldenablethestorageof25DVDsor25millionpagesoftextonapostagestamp-sizedsurfaceandcouldenable10GBofstoragecapacityonacellphone.Millipedeusesthousandsoftinysharppoints(hencethename)topunchholesinathinplasticfilm.Eachofthe10-nmholesrepresentsasinglebit.Thepatternofindentationsisadigitizedversionofthedata.Thelayoutofthemillipedecantilever/tipincontactwiththedatastoragemediumisshowninFigure24.AccordingtoIBM,Millipedecanbethoughtofasananotechnologyversionofthepunchcarddataprocessingtechnologydevelopedinthelatenineteenthcentury[156].However,therearesignificantdifferences:Millipedeisrewritable,anditmayeventuallyenablestorageofover1.5GBofdatainaspacenolargerthanasingleholeinthepunchcard.StoragedevicesbasedonIBMstechnologycanbemadewithexistingmanufacturingtechniques,sotheywillnotbeexpensivetomake.AccordingtoP.Vettiger,headoftheMillipedeproject,thereisnotasinglestepinfabricationthatneedstobeinvented.VettigerpredictsthatananostoragedevicebasedonIBMstechnologycouldbeavailableasearlyas2005[155].Now,researchersatIBMsZurichResearchLaboratoryinSwitzerlandhaveclockedtherateofdataloss.Theyhavecalculatedthatat85C,atemperatureoftenusedtoassessdataretention,itwouldlosejust10%to20%ofinformationoveradecade,comparabletoFlashmemory[157].

    Viewlargerversion

    Figure24.Schematiclayoutofthemillipedecantilever/tipincontactwiththedatastoragemedium.Adoptedfromref.[157].

    WORMmemorybasedonDNAbiopolymernanocomposite

  • TheuseofDNAiswellknownasagoodmodelformetalNPsynthesisduetoitsaffinitytothemetalions[158].Inrecentyears,DNAhasalsobeenshowntobeapromisingopticalmaterialwiththematerialprocessingfullycompatiblewithconventionalpolymerforthin-filmoptoelectronicapplications[159,160].ResearchersfromNationalTsingHuaUniversityinTaiwanandtheKarlsruheInstituteofTechnologyinGermanyhavecreatedaDNA-basedmemorydevice,thatis,write-once-read-many-times(WORM),thatusesultraviolet(UV)lighttoencodeinformation[161].Thedeviceconsistsofasinglebiopolymerlayersandwichedbetweenelectrodes,inwhichelectricalbistabilityisactivatedbyinsituformationofsilvernanoparticlesembeddedinabiopolymeruponlightirradiation(Figure25).ThedevicefunctionallyworkswhenshiningUVlightonthesystem,whichenablesalight-triggeredsynthesisprocessthatcausesthesilveratomstoclusterintonanosizedparticlesandreadiesthesystemfordataencoding.Forsomeparticularinstance,theteamhasfoundthatusingDNAmaybelessexpensivetoprocessintostoragedevicesthanusingtraditional,inorganicmaterialslikesilicon,theresearcherssay[161,162].TheysaidthatwhennovoltageorlowvoltageisappliedthroughtheelectrodestotheUV-irradiatedDNA,onlyalowcurrentisabletopassthroughthecomposite;thiscorrespondstotheOFFstateofthedevice.ButtheUVirradiationmakesthecompositeunabletoholdachargeunderahighelectricfield,sowhentheappliedvoltageexceedsacertainthreshold,anincreasedamountofchargeisabletopassthrough.ThishigherstateofconductivitycorrespondstotheONstateofthedevice.Theteamfoundthatthischangefromlowconductivity(OFF)tohighconductivity(ON)wasirreversible:oncethesystemhadbeenturnedon,itstayedon,nomatterwhatvoltagetheteamappliedtothesystem.Onceinformationiswritten,thedeviceappearstoretainthatinformationindefinitely.Theresearchershopethatthetechniquewillbeusefulinthedesignofopticalstoragedevicesandsuggestthatitmayhaveplasmonicapplicationsaswell.Consequently,WORMmemoriesbasedonDNAabiopolymernanocompositehaveemergedasanexcellentcandidatefornext-generationinformationstoragemediabecauseoftheirpotentialapplicationinflexiblememorydevices.ThisworkcombinesnewadvancesinDNAnanotechnologywithaconventionalpolymerfabricationplatformtorealizeanewemergingclassofDNA-basedmemory.

    Viewlargerversion

    Figure25.SchematicdesignofamemorydeviceconsistingofathinDNAbiopolymerfilmsandwichedbetweenelectrodes.Thememoryswitchingeffectisactivateduponlightirradiation.Adoptedfromref.[161].

    QDmemory

    Memorymadefromtinyislandsofsemiconductors-knownasquantumdots-couldfillagapleftbytodayscomputermemory,allowingstoragethatisfastaswellaslonglasting.Researchershaveshownthattheycanwriteinformationintoquantumdotmemoryinjustnanoseconds.Memoryisdividedintotwoforms:DRRAMandFlash[163,164].ComputersuseDRAM,forshort-termmemory,butdatadoesnotpersistforlongandmustberefreshedover100timespersecondtomaintainitsmemory.Ontheotherhand,Flashmemory,likethatusedinmemorycards,canstoredataforyearswithoutrefreshingbutwritesinformationabout1,000timesslowerthanDRAM.Newresearchshowsthatmemorybasedonquantumdotscanprovidethebestofboth:long-termstoragewithwrite

  • speedsnearlyasfastasDRAM.Atightlypackedarrayoftinyislands,eacharound15nmacross,couldstore1terabyte(1,000GB)ofdatapersquareinch,theresearcherssay.DieterBimbergandcolleaguesattheTechnicalUniversityofBerlin,Germany,withcollaboratorsatIstanbulUniversity,Turkey,demonstratedthatitispossibletowriteinformationtothequantumdotsinjust6ns[165,166].Thekeyadvantagesofquantumdot(QD)NVMsarethehighread/writespeed,smallsize,lowoperatingvoltage,and,mostimportantly,multibitstorageperdevice.However,thesefeatureshavenotbeenrealizedduetovariationsindotsizeandlackofuniforminsulatorcladdinglayersonthedots[167].IncorporatingQDsintothefloatinggateresultsinareductioninchargeleakageandpowerdissipationwithenhancedprogrammingspeed.ResearchersinIndiaandGermanyhavenowunveiledthememorycharacteristicsofsiliconandsilicon-germaniumQDsembeddedinepitaxialrare-earthoxidegadoliniumoxide(Gd2O3)grownonSi(111)substratesasshownintheDQMstructureinFigure26.MultilayerSiaswellassingle-layerSi1xGex(wherex=0.6)QDsshowexcellentmemorycharacteristics,makingthemattractivefornext-generationFlash-floating-gatememorydevices[168,169].

    Viewlargerversion

    Figure26.Structureofquantumdotmemory.Adoptedfromref.[168].

    3-Dcross-pointmemory

    Memoryproducersarealsotryingtodevelopalternativetechnologiesthatmaybescalablebeyond20-nmlithography.Fortruescalabilitybeyond20-nmtechnologynodes,itisnecessarytodesignacross-pointmemoryarraywhichdoesnotrequirediodesforaccesselements[170].Thecross-pointmemoryarchitecturecouldbedesignedsuchthatitcanbeeasilyfabricatedinmultiplelayerstoformastacked3-Dmemory[171].The3-DtechnologyhasbroughttohighvolumeanNVMwherearraysofmemorycellsarestackedabovecontrollogiccircuitryinthethirddimension,andstacking3-DmemorydirectlyoverCMOSallowsforhigharrayefficiencyandverysmalldiesize[172].The3-Dtechnologyusesnonewmaterials,processes,orfabricationequipment,whichcontrollogiccircuitrycomposedoftypicalCMOS.Thememoryconstructionusestypicalback-endprocessingtools,andeachmemorylayerisarepeatofthelayersbelowit.Thebasicdesignofthe3-DcellconsistsofaverticaldiodeinserieswithamemoryelementasshowninFigure27.Buildingintegratedcircuitsverticallyallowsforareducedchipfootprintwhencomparedtoatraditional2-Ddesign,byanapproximatefactorofthenumberoflayersused.Thisofferssignificantadvantagesintermsofreducedinterconnectdelaywhenroutingtoblocksthatotherwisewouldhavebeenplacedlaterally.Theprocessforthe2-Dcross-pointarraycanbebuiltintoamultilayer3-Darchitecture.Traditionally,a3-Dintegratedcircuit(3-D-IC)hasusedmorethanoneactivedevicelayer.Whileresistance-changememorycellsarenotactivedevices,theyfunctionasrectifyingdevicesindesign.Furthercharacterizationoftheresistance-changematerialisalsonecessaryin

  • ordertoguaranteethatthe3-Dcross-pointmemorywillbepracticalfordatastorage.Also,thescalabilityofmetal-oxideresistance-changematerialsbeyond20-nmtechnologynodesstillneedstobestudied.Moreover,theprogrammingoperationisexpectedtobecompetitivewithbothNANDandNORFlashintermsofspeedbecauseoftherelativelylowvoltagerequirementsofresistance-changematerials.Iftheperipheralcircuitryforaccommodatingthewriteoperationcanbemadesufficientlycompact,thenthe3-Dcross-pointmemorywillindeedbeaviablereplacementforNANDandNORFlashinfutureprocessgenerations.

    Viewlargerversion

    Figure27.Thebasicdesignofa3-Dcellthatconsistsofaverticaldiodeinseries.(top)Sideview,(bottomright)topview,and(bottomleft)3-Dview.

    TFM

    Transparentandflexibleelectronics(TFE)is,today,oneofthemostadvancedtopicsforawiderangeofdeviceapplications,wherethekeycomponentistransparentconductingoxides(TCOs),whichareuniquematerialsthatoxidesofdifferentoriginplayanimportantrole,notonlyasapassivecomponentbutalsoasanactivecomponent[173].TFEisanemergingtechnologythatemploysmaterials(includingoxides,nitrides,andcarbides)andadevicefortherealizationofinvisiblecircuitsforimplementingnext-generationtransparentconductingoxidesinaninvisiblememorygeneration[174].Ingeneral,theTF-RRAMdeviceisbasedonacapacitor-likestructure(e.g.,ITO/transparentresistivematerial/ITO/transparentandflexiblesubstrate),whichprovidestransmittanceinthevisibleregion[175].Forsuchnewclassofmemorytechnology,dataretentionisexpectedtobeabout10years.Thebasicstructuraldesignofthenewmemorychipsisconfigured,namelywithtwoterminalsperbitofinformationonatransparentandflexiblesubstrateratherthanthestandardthreeterminalsperbitonarigidandopaquesubstrate(Figure28).Theyaremuchbettersuitedforthenextrevolutioninelectronic3-DmemorythanFlashmemory.Thesenewmemorychipsthataretransparent,areflexibleenoughtobefoldedlikeasheetofpaper,shrugoff1,000Ftemperaturestwiceashotasthemaxinakitchenoven,andsurviveotherhostileconditionscouldusherinthedevelopmentofnext-generationFlash-competitivememoryfortomorrowskeychaindrives,cellphones,andcomputers,ascientistreportedtoday.Speakingatthe243rdNationalMeetingandExpositionoftheAmericanChemicalSociety,theworldslargestscientificsociety,hesaidthatdeviceswiththesechipscouldretaindatadespiteanaccidentaltripthroughthedrierorevenavoyagetoMars.Andwithaunique3-Dinternalarchitecture,thenewchipscouldpackextragigabytesofdatawhiletakinguplessspace[176].DespitetherecentprogressinTF-RRAM,itneedslotsofworktosatisfythedualrequirementsofresistancetorepeatedbendingstressandtransparentproperties.Thus,itissupposedthatanachievementofsuchTF-RRAMdevicewillbethenextsteptowardstherealizationoftransparentandflexibleelectronicsystems.WehopethatFT-RRAMdeviceswillmarkamilestoneinthecurrentprogressofsuchuniqueandinvisibleelectronicsystemsinthenearfuture.

  • Viewlargerversion

    Figure28.AschematicdesignofFT-RRAMandaflexible,transparentmemorychipimagecreatedbyresearchersatRiceUniversity.ReproducedfromTourLab/RiceUniversity.

    1T1R-RRAM

    One-transistorone-resistor(1T1R)-RRAMisalsooneclassofemergingmemorytechnologywithimpressivecharacteristics.Itmeetsthedemandsfornext-generationmemorysystems.Itisexpectedthat1T1R-RRAMcouldbeabletomeetthedemandofhigh-speed(e.g.,performance)memorytechnology.The1T1Rstructureischosenbecausethetransistorisolatescurrenttocells,whichareselectedfromcellswhichdonot.Thebasiccellstructureof1T1RisdepictedinFigure29.1T1R-RRAMconsistsofanaccesstransistorandaresistorasastorageelement.ZangenehandJoshihavealsomentionedthatthe1T1RcellstructureissimilartothatofaDRAMcellinthatthedataisstoredastheresistanceoftheresistorandthetransistorservesasanaccessswitchforreadingandwritingdata[177,178].Inreferencetothis,theyrevealedthe1T1RcellasthebasicbuildingblockofaNVRRAMarrayasitavoidssneakpathproblemtoensurereliableoperation.Moreover,the1T1Rstructureismorecompactandmayenableverticallystackingmemorylayers,ideallysuitedformassstoragedevices.But,intheabsenceofanytransistor,theisolationmustbeprovidedbyaselectordevice,suchasadiode,inserieswiththememoryelement,orbythememoryelementitself.Suchkindsofisolationcapabilitieshavebeeninferiortotheuseoftransistors,limitingtheabilitytooperateverylargeRRAMarraysin1T1Rarchitecture.1T1Rmemorypolaritycanbeeitherbinaryorunary.Bipolareffectscausepolaritytoreverseresetoperationtosetoperation.Unipolarswitchingleavespolarityunaffectedbutusesdifferentvoltages.

    Viewlargerversion

    Figure29.Thebasiccellstructureof1T1R-RRAM.

    MTM,PFRAM,SPBMM,andCMORRAM-futurealternateNVMs

    Otherpotentialemergingclassesofmemorytechnologies,wearedescribinginshort,aremoleculartunnelmemory(MTM),polymericferroelectricRAM(PFRAM),spin-polarizedbeammagneticmemory(SPBMM),lightmemory,andcomplexmetal-oxideRRAM(CMORRAM).Wecansaythatthesearesistermemorytechnologiesofmolecularmemory,ferroelectric/polymermemory,magneticmemory,andmetal-oxideRRAM,respectively.Althoughthesenewtechnologieswillalmostcertainlyresultinmorecomplexmemoryhierarchiesthantheirfamilymemories,theyarelikelytoallowtheconstructionofmemorychipsthatarenonvolatile,havelowenergy,andhavedensityanddevelopmentclosetoorbetterthanthoseofDRAMchips,withimprovedperformanceandallowingmemor