资源描述:
《Analog Neuromorphic Module Based on carbon nanotube synapses.pdf》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、ARTICLEAnalogNeuromorphicModuleBasedonCarbonNanotubeSynapsesAlexMingShen,Chia-LingChen,KyunghyunKim,ByungjinCho,AndrewTudor,andYongChen*DepartmentofMechanicalandAerospaceEngineering,andCaliforniaNanoSystemsInstitute,UniversityofCalifornia,LosAngeles,California90095,UnitedStatesABS
2、TRACTWereportananalogneuromorphicmodulecomposedofp-typecarbonnanotube(CNT)synapsesandanintegrate-and-fire(I&F)circuit.TheCNTsynapsehasafield-effecttransistorstructurewitharandomCNTnetworkasitschannelandanaluminumoxidedielectriclayerimplantedwithindiumionsasitsgate.Apositivevoltagepul
3、se(spike)appliedonthegateattractselectronsintothedefectsitesofthegatedielectriclayer,andthetrappedelectronsaregraduallyreleasedafterthepulseisremoved.TheelectronsmodifytheholeconcentrationandinduceadynamicpostsynapticcurrentintheCNTchannel.Multipleinputspikesinduceexcitatoryorinhi
4、bitorypostsynapticcurrentsviaexcitatoryorinhibitoryCNTsynapses,whichflowtowardanI&Fcircuittotriggeroutputspikes.Thedynamictransferfunctionbetweentheinputandoutputspikesoftheneuromorphicmoduleisanalyzed.Themodulecouldpotentiallybescaleduptoemulatebiologicalneuralnetworksandtheirfunc
5、tions.KEYWORDS:carbonnanotube.neuromorphicmodule.synapse.analog.spikeequiringlesspowerthana100Wlightdimensionalinformation.Supercomputersbulb,thehumanbraincanprocesshavebeenusedtoimitatethecomplexRlargedimensionalinformationsuchnonlineardynamicfunctionsofaneuralasimagesandspeechmu
6、chfasterthananetwork.However,asupercomputercon-1,2supercomputer.Thehumanbraincon-sumesmuchmoreenergyandtimethan11tains∼10neuronsworkinginparalleltoaneuralnetworkwhenthenonlineardy-processspikesignals,whicharemillise-namicalgorithmsoftheneuralnetworkcond-longpotentialpulses.Atypica
7、lneuronareimplementedinthesupercomputer13receivesspikesfromafewthousandfellowbysoftware.Neuralnetworkshavealsoneuronsviasynapses,whicharethecon-beenemulatedbylarge-scaleSineuro-57nectionsbetweentwoneurons.Aspikemorphiccircuits.However,thesecircuitstriggersadynamicanalogpostsynap
8、ticcur-consumedsubstantiallymoree