Izindaba

Izinhlelo zokusebenza zokucubungula idatha yomhlaba wangempela zidinga amasistimu ekhompuyutha ahlangene, aphansi, anamandla aphansi.Ngamakhono ekhompuyutha aqhutshwa umcimbi, izakhiwo ezihambisanayo ze-metal-oxide-semiconductor hybrid memristive neuromorphic zinikeza isisekelo sehadiwe esifanelekile semisebenzi enjalo.Ukuze sibonise amandla aphelele amasistimu anjalo, siphakamisa futhi sibonise ngokuhlola isisombululo esibanzi sokucubungula inzwa sezinhlelo zokusebenza zokwenziwa kwasendaweni kwezinto zangempela.Ukudweba ugqozi oluvela ku-barn owl neuroanatomy, senze uhlelo lwendawo lwento oluphefumulelwe ephilayo, oluqhutshwa umcimbi oluhlanganisa i-transducer yesimanjemanje ye-piezoelectric micromechanical transducer enenkumbulo ye-computational graph-based neuromorphic resistive.Sibonisa izilinganiso zesistimu ekhiqiziwe ehlanganisa isitholi esisekelwe kwinkumbulo esiphikisayo, ukulibaziseka komugqa wokujikeleza, kanye ne-ultrasonic transducer engenziwa ngokwezifiso ngokugcwele.Sisebenzisa le miphumela yokuhlola ukuze silinganise izifaniso ezingeni lesistimu.Lezi zifaniso zibe sezisetshenziselwa ukuhlola ukulungiswa kwe-angular nokusebenza kahle kwamandla kwemodeli yokwenza kwasendaweni into.Imiphumela ibonisa ukuthi indlela yethu ingaba ama-oda amaningana wobukhulu bayonga amandla amaningi kunama-microcontroller enza umsebenzi ofanayo.
Singena enkathini yamakhompuyutha atholakala yonke indawo lapho inani lamadivayisi namasistimu asetshenzisiwe likhula kakhulu ukuze lisisize ezimpilweni zethu zansuku zonke.Lezi zinhlelo zilindeleke ukuthi zisebenze ngokuqhubekayo, zisebenzisa amandla amancane ngangokunokwenzeka ngenkathi zifunda ukuhumusha idatha eziyiqoqa kuzinzwa eziningi ngesikhathi sangempela futhi zikhiqize okukhiphayo kanambambili njengomphumela wokuhlukanisa ngezigaba noma imisebenzi yokuqashelwa.Esinye sezinyathelo ezibaluleke kakhulu ezidingekayo ukuze kuzuzwe lo mgomo ukukhipha ulwazi oluwusizo noluhlangene kudatha yezinzwa enomsindo futhi ngokuvamile engaphelele.Ubunjiniyela obujwayelekile busondela ngokujwayelekile amasignali enzwa yesampula ngezinga elingashintshi futhi eliphezulu, okukhiqiza inani elikhulu ledatha ngisho nalapho kungekho okokufaka okuwusizo.Ngaphezu kwalokho, lezi zindlela zisebenzisa amasu ayinkimbinkimbi okucubungula isignali yedijithali ukuze kucutshungulwe ngaphambili idatha yokokufaka (okuvame ukuba nomsindo).Esikhundleni salokho, i-biology inikeza ezinye izixazululo zokucubungula idatha yemizwa enomsindo kusetshenziswa izindlela ezonga amandla, ezingavumelaniyo, eziqhutshwa umcimbi (ama-spikes)2,3.I-Neuromorphic computing ithatha ugqozi kusuka ezinhlelweni zezinto eziphilayo ukuze kuncishiswe izindleko zokubala ngokwezidingo zamandla nenkumbulo uma kuqhathaniswa nezindlela zendabuko zokucubungula isignali4,5,6.Muva nje, amasistimu asekelwe ebuchosheni anenhloso evamile esebenzisa amanethiwekhi e-neural impulse (TrueNorth7, BrainScaleS8, DYNAP-SE9, Loihi10, Spinnaker11) abonisiwe.Lawa maphrosesa ahlinzeka ngamandla aphansi, izixazululo ze-latency ephansi yokufunda komshini kanye ne-cortical circuit modelling.Ukuze basebenzise ngokugcwele amandla abo okusebenza kahle, lawa ma-neuromorphic processors kufanele axhunywe ngokuqondile nezinzwa eziqhutshwa umcimbi12,13.Nokho, namuhla kunamadivayisi ambalwa wokuthinta ahlinzeka ngokuqondile ngedatha eqhutshwa umcimbi.Izibonelo ezivelele izinzwa eziguquguqukayo ezibukwayo (i-DVS) zezinhlelo zokusebenza zombono ezifana nokulandela umkhondo nokutholwa kokunyakaza14,15,16,17 i-silicon cochlea18 kanye nezinzwa ze-neuromorphic auditory (NAS)19 zokucubungula isignali yokuzwa, izinzwa zokunuka20 kanye nezibonelo eziningi21,22 zokuthinta..izinzwa zokuthungwa.
Kuleli phepha, sethula isistimu yokucubungula yokuzwa eqhutshwa umcimbi esanda kuthuthukiswa esetshenziswa ekwenziweni kwasendaweni kwento.Lapha, ngokokuqala ngqa, sichaza isistimu yokuphela ukuya-ekupheleni yokwenza kwasendaweni kwento etholwe ngokuxhuma i-piezoelectric micromachined ultrasonic transducer (pMUT) yesimanjemanje ngegrafu yekhompyutha esekelwe kumemori ye-neuromorphic resistive (RRAM).I-in-memory computing architectures esebenzisa i-RRAM iyisixazululo esithembisayo sokunciphisa ukusetshenziswa kwamandla23,24,25,26,27,28,29.Ukungaguquguquki kwazo okungokwemvelo—okungadingi ukusetshenziswa kwamandla okusebenzayo ukuze kugcinwe noma kuthuthukiswe ulwazi—kufanelana ngokuphelele nemvelo engavumelaniyo, eqhutshwa umcimbi yekhompyutha ye-neuromorphic, okuholela ekucisheni kungabikho ukusetshenziswa kwamandla lapho isistimu ingenzi lutho.I-Piezoelectric micromachined ultrasonic transducers (pMUTs) ayibizi, ama-miniaturized silicon-based ultrasonic transducers akwazi ukusebenza njengabathumeli nabamukeli30,31,32,33,34.Ukuze sicubungule amasiginali atholwe izinzwa ezakhelwe ngaphakathi, sithole ugqozi kusuka ku-barn owl neuroanatomy35,36,37.Isikhova esibizwa ngokuthi i-barn owl i-Tyto alba saziwa ngamakhono aso amangalisayo okuzingela ebusuku ngenxa yesistimu yokuzwa esebenza kahle kakhulu.Ukuze ubale indawo yesilwane esizingelayo, uhlelo lwendawo yesikhova esibayeni lufaka ikhodi isikhathi sokundiza (i-ToF) lapho amaza omsindo avela enyamazaneni efinyelela ezindlebeni zesikhova ngasinye noma izimukeli zomsindo.Uma kubhekwa ibanga eliphakathi kwezindlebe, umehluko phakathi kwezilinganiso ezimbili ze-ToF (I-Interural Time Difference, i-ITD) yenza kube nokwenzeka ukubala ngokuhlaziya indawo ye-azimuth yethagethi.Nakuba amasistimu ebhayoloji engafaneleki kahle ekuxazululeni izibalo ze-algebra, angakwazi ukuxazulula izinkinga zokwenziwa kwasendaweni ngempumelelo kakhulu.Uhlelo lwemizwa yesikhova esibayeni lusebenzisa isethi yomtshina wokuqondana (CD)35 neurons (okungukuthi, ama-neuron akwazi ukubona ukuhlobana kwesikhashana phakathi kwama-spikes asabalalisa phansi kuya eziphethweni ezijabulisayo eziguqukayo)38,39 ehlelwe yaba amagrafu ekhompyutha ukuze axazulule izinkinga zokuma.
Ucwaningo lwangaphambilini lubonise ukuthi i-complementary metal-oxide-semiconductor (CMOS) hardware kanye ne-RRAM-based neuromorphic hardware ephefumulelwe i-colliculus ephansi (“i-auditory cortex”) yesikhova esibayeni iyindlela ephumelelayo yokubala indawo usebenzisa i-ITD13, 40, 41, 42, 43, 44, 45, 46. Kodwa-ke, amandla ezinhlelo eziphelele ze-neuromorphic ezixhumanisa izinkomba zokuzwa kumagrafu we-neuromorphic computational ayikaboniswa.Inkinga enkulu ukuhlukahluka okungokwemvelo kwamasekhethi e-analog CMOS, okuthinta ukunemba kokutholwa komdlalo.Muva nje, okunye ukusetshenziswa kwezinombolo kwezilinganiso ze-ITD47 kubonisiwe.Kuleli phepha, siphakamisa ukusebenzisa ikhono le-RRAM ukushintsha inani lokuqhuba ngendlela engaguquguquki ukuze kuliwe nokuhlukahluka kumasekhethi e-analog.Senze isistimu yokuhlola ehlanganisa ulwelwesi olulodwa oludlulisa i-pMUT olusebenza ngemvamisa engu-111.9 kHz, i-pMUT emibili ethola ulwelwesi (izinzwa) ezilingisa izindlebe zezikhova ezibayeni, kanye neyodwa .Sihlole ngokuhlola isistimu yokuthola i-pMUT kanye negrafu yekhompyutha ye-ITD esekwe ku-RRAM ukuze sihlole isistimu yethu yokwenza okwasendaweni futhi sihlole ukulungiswa kwayo kwe-angular.
Siqhathanisa indlela yethu nokusetshenziswa kwedijithali ku-microcontroller eyenza umsebenzi ofanayo wokwenza indawo kusetshenziswa izindlela ezivamile ze-beamforming noma ze-neuromorphic, kanye nohlelo lwesango oluhlelekayo lwenkambu (FPGA) lokulinganisa kwe-ITD okuhlongoziwe kusithenjwa.47. Lokhu kuqhathanisa kugqamisa ukusebenza kahle kwamandla okuncintisana kwesistimu ehlongozwayo ye-analog neuromorphic system.
Esinye sezibonelo ezigqama kakhulu zesistimu yokwenziwa kwasendaweni kwento enembile nesebenza kahle ingatholakala ku-barn owl35,37,48.Ekuseni nasekuseni, isikhova esisesibayeni (Tyto Alba) sithembela kakhulu ekulaleleni, sifune ngenkuthalo inyamazane encane efana nama-voles noma amagundane.Lezi zingcweti zokuzwa zingakwazi ukuveza izimpawu zokuzwa ezivela enyamazaneni ngokunemba okumangalisayo (cishe u-2°)35, njengoba kuboniswe ku-Fig. 1a.Izikhova ze-Barn zibheka indawo yemithombo yomsindo endizeni ye-azimuth (evundlile) kusukela emehlukweni wesikhathi esingenayo sokundiza (ITD) ukusuka emthonjeni womsindo kuya ezindlebeni ezimbili.Indlela yokuhlanganisa ye-ITD yahlongozwa i-Jeffress49,50 ethembele ku-neural geometry futhi idinga izingxenye ezimbili ezibalulekile: i-axon, i-neuron's nerve fiber esebenza njengomugqa wokulibaziseka, kanye noxhaxha lwama-neurons omuthi wokuhlangana ahlelwe abe uhlelo lokubala.igrafu njengoba kukhonjisiwe kuMfanekiso 1b.Umsindo ufinyelela endlebeni nge-azimuth dependent time delay (ITD).Umsindo ube usuguqulwa ube iphethini ye-spike endlebeni ngayinye.Ama-axon ezindlebeni zesokunxele nakwesokudla asebenza njengemigqa yokulibaziseka futhi ahlangane kuma-CD neurons.Ngokombono, i-neuron eyodwa kuphela ohlwini lwama-neurons afanayo ezothola okokufaka ngesikhathi (lapho ukubambezeleka kukhanselwa khona ncamashi) futhi izodubula kakhulu (amaseli angomakhelwane nawo azodubula, kodwa ngefrikhwensi ephansi).Ukwenza kusebenze ama-neurons athile kufaka phakathi indawo okuqondiwe kuyo emkhathini ngaphandle kokuguqula i-ITD ibe ama-engeli.Lo mqondo ufinyezwe kuMfanekiso 1c: isibonelo, uma umsindo uvela ohlangothini lwesokudla lapho isignali yokufaka esuka endlebeni yesokudla ihamba indlela ende kunomzila osuka endlebeni yangakwesobunxele, inxephezela inani lama-ITD, isibonelo, lapho i-neuron 2 ifana.Ngamanye amazwi, i-CD ngayinye iphendula i-ITD ethile (eyaziwa nangokuthi ukubambezeleka okufanele) ngenxa yokubambezeleka kwe-axonal.Ngakho, ubuchopho buguqula ukwaziswa kwesikhashana kube ukwaziswa kwendawo.Ubufakazi be-anatomical balo mshini butholakele37,51.Ama-macronucleus neurons akhiywe esigabeni agcina ulwazi lwesikhashana mayelana nemisindo engenayo: njengoba negama lawo lisho, adubula ezigabeni ezithile zesignali.Ama-neurons okuthola ukuqondana kwemodeli ye-Jeffress angatholakala kumongo we-laminar.Bathola ulwazi kuma-macronuclear neurons, ama-axon awo asebenza njengemigqa yokulibaziseka.Inani lokulibaziseka elinikezwa umugqa wokulibaziseka lingachazwa ngobude be-axon, kanye nenye iphethini ye-myelination eshintsha isivinini sokuqhuba.Ngokugqugquzelwa uhlelo lokuzwa lwesikhova esibayeni, sisungule uhlelo lwe-biomimetic lokubeka izinto endaweni.Izindlebe ezimbili zimelelwe izamukeli ezimbili ze-pMUT.Umthombo womsindo umthumeli we-pMUT otholakala phakathi kwawo (Umfanekiso 1a), futhi igrafu yokubala yakhiwa igridi yamasekhethi e-CD asekelwe ku-RRAM (Umfanekiso 1b, oluhlaza), edlala indima yama-CD neurons okufakwa kwawo kulibaziseka.ngesekhethi, imigqa yokulibaziseka (eluhlaza okwesibhakabhaka) isebenza njengama-axon kumlingani webhayoloji.Isistimu yezinzwa ehlongozwayo iyahluka emazingeni okusebenza kunesikhova, uhlelo lwaso lokuzwa olusebenza ebangeni elingu-1–8 kHz, kodwa izinzwa ze-pMUT ezisebenza cishe ku-117 kHz zisetshenziswa kulo msebenzi.Ukukhethwa kwe-transducer ye-ultrasonic kubhekwa ngokuvumelana nezinqubo zobuchwepheshe kanye nokwenza kahle.Okokuqala, ukukhawulela umkhawulokudonsa owamukelayo kufrikhwensi eyodwa kuthuthukisa ukunemba kokulinganisa futhi kwenza isinyathelo sangemva kokucubungula sibe lula.Ngaphezu kwalokho, ukusebenza ku-ultrasound kunenzuzo yokuthi ama-pulses akhishiwe awazwakali, ngakho-ke ungaphazamisi abantu, ngoba uhla lwabo lokuzwa luyi-~20-20 kHz.
isikhova sasesibayeni sithola amaza omsindo avela endaweni eqondiwe, kulokhu sinyakazisa inyamazane.Isikhathi sokundiza (ToF) segagasi lomsindo sihlukile endlebeni ngayinye (ngaphandle uma inyamazane iphambi kwesikhova ngqo).Umugqa onamachashazi ubonisa indlela amaza omsindo ayihambayo ukuze afinyelele ezindlebeni zesikhova esisesibayeni.Inyamazane ingenziwa endaweni ngokunembile endizeni evundlile ngokusekelwe kumehluko wobude phakathi kwezindlela ezimbili ze-acoustic kanye nomehluko ohambisanayo wesikhathi se-interaural (ITD) (isithombe esingakwesokunxele sikhuthazwe yi-ref. 74, copyright 2002, Society for Neuroscience).Kusistimu yethu, isidluliseli se-pMUT (okuluhlaza okwesibhakabhaka) sikhiqiza amaza omsindo agxumayo asuka kulokho okuhlosiwe.Amagagasi e-ultrasound abonakalisiwe atholwa izamukeli ezimbili ze-pMUT (okuluhlaza okotshani) futhi acutshungulwe iphrosesa ye-neuromorphic (kwesokudla).b Imodeli yekhompyutha ye-ITD (Jeffress) echaza ukuthi imisindo engena ezindlebeni zesikhova esisesibayeni ifakwa kanjani ikhodi kuqala njengama-spikes akhiywe ngesigaba ku-nucleus enkulu (NM) bese kusetshenziswa igridi ehlelwe ngokwejometri yama-neurons omtshina afanayo kunucleus ye-lamellar.Iyacubungula (eNetherlands) (kwesokunxele).Umdwebo wegrafu yekhompyutha ye-neuroITD ehlanganisa imigqa yokulibaziseka kanye nama-neuron athola ukuqondana kwezinto, isistimu ye-owl biosensor ingamodelwa kusetshenziswa amasekhethi e-neuromorphic asekelwe ku-RRAM (kwesokudla).c I-Schematic yomshini omkhulu we-Jeffress, ngenxa yomehluko ku-ToF, izindlebe ezimbili zithola izisusa zomsindo ngezikhathi ezihlukene futhi zithumela ama-axon asuka kuzo zombili iziphetho kuya kumtshina.Ama-axon ayingxenye yochungechunge lwama-neurons okuthola ukuqondana kwezinto (CD), ngayinye esabela ngokukhetha okokufaka okuhlobene kakhulu nesikhathi.Ngenxa yalokho, ama-CD kuphela okokufaka kwawo okufika kunomehluko omncane wesikhathi ajabule kakhulu (i-ITD inxeshezelwa ngqo).I-CD izobe isifaka ikhodi indawo eyi-angular yalapho okuqondiwe khona.
Ama-transducers e-Piezoelectric micromechanical ultrasonic angama-scalable ultrasonic transducers angahlanganiswa nobuchwepheshe obuthuthukisiwe be-CMOS31,32,33,52 futhi abe ne-voltage yokuqala ephansi kanye nokusetshenziswa kwamandla kune-volumetric transducers yendabuko53.Emsebenzini wethu, i-diameter ye-membrane ingu-880 µm, futhi i-resonant frequency isatshalaliswa ebangeni lika-110–117 kHz (Fig. 2a, bona Izindlela ukuze uthole imininingwane).Eqoqweni lamadivayisi okuhlola ayishumi, isici sekhwalithi esimaphakathi sasicishe sibe ngama-50 (ref. 31).Ubuchwepheshe bufinyelele ekuvuthweni kwezimboni futhi abukho nge-bioinspired ngesekhondi ngalinye.Ukuhlanganisa ulwazi oluvela kumafilimu ahlukene e-pMUT kuyindlela eyaziwayo, futhi ulwazi lwama-engeli lungatholakala kuma-pMUT kusetshenziswa, isibonelo, amasu okwenza i-beamform31,54.Kodwa-ke, ukucutshungulwa kwesignali okudingekayo ukuze kukhishwe ulwazi lwe-engeli akufanelekile izilinganiso zamandla aphansi.Uhlelo oluhlongozwayo luhlanganisa idatha ye-neuromorphic preprocessing circuit i-pMUT negrafu yekhompyutha ye-neuromorphic esekelwe ku-RRAM egqugquzelwe imodeli ye-Jeffress (Umfanekiso 2c), ihlinzeka ngesinye isixazululo sezingxenyekazi zekhompuyutha ezonga amandla nezinsiza.Senze isilingo lapho izinzwa ezimbili ze-pMUT zabekwa ngokuqhelelana cishe ngo-10 cm ukuze kusetshenziswe imisindo ehlukene ye-ToF etholwe amalwelwesi amabili atholayo.I-pMUT eyodwa esebenza njengesidluliseli ihlala phakathi kwabemukeli.Okuhlosiwe kwakuyipuleti le-PVC elingu-12 cm ububanzi, elitholakala ebangeni D phambi kwedivayisi ye-pMUT (Fig. 2b).Umamukeli urekhoda umsindo ovela entweni futhi aphendule ngangokunokwenzeka phakathi nokudlula kwegagasi lomsindo.Phinda isilingo ngokushintsha indawo yento, enqunywa ibanga D kanye ne-engeli engu-θ.Ikhuthazwe isixhumanisi.55, siphakamisa ukucutshungulwa kwangaphambili kwe-neuromorphic kwamasiginali eluhlaza we-pMUT ukuguqula amagagasi abonisiwe abe iziqongo ukuze kufakwe igrafu yekhompyutha ye-neuromorphic.I-ToF ehambisana ne-amplitude ephakemeyo ikhishwa kusiteshi ngasinye kwamabili futhi ifakwe ikhodi njengesikhathi esiqondile seziqongo ngasinye.Emkhiwaneni.I-2c ibonisa ukujikeleza okudingekayo ukuze kuxhunywe inzwa ye-pMUT ngegrafu yekhompyutha esekelwe ku-RRAM: kowamukeli ngamunye we-pMUT, isiginali eluhlaza iyahlungwa ibhendi ukuze ibushelelezi, ilungiswe, bese idluliselwa kusihlanganisi esivuzayo kumodi yokunqoba.i-dynamic threshold (Fig. 2d) idala umcimbi wokuphumayo (spike) kanye ne-firing (LIF) neuron: isikhathi se-spike esiphumayo sihlanganisa isikhathi sendiza esitholiwe.Umkhawulo we-LIF ulinganiswa ngokuphikisana nempendulo ye-pMUT, ngaleyo ndlela yehlisa ukuhlukahluka kwe-pMUT kusuka kudivayisi kuye kwenye.Ngale ndlela, esikhundleni sokugcina lonke igagasi lomsindo enkumbulweni futhi silicubungule ngokuhamba kwesikhathi, simane sikhiqize inani eliphakeme elihambisana ne-ToF yegagasi lomsindo, okwenza okokufaka kugrafu yekhompyutha ephikisayo.Ama-spikes athunyelwa ngokuqondile emigqeni yokulibaziseka futhi afaniswa namamojula okuthola okufanayo kumagrafu wokubala we-neuromorphic.Ngenxa yokuthi zithunyelwa emasangweni ama-transistors, akukho umjikelezo wokukhulisa u-amplification odingekayo (bheka i-Supplementary Fig. 4 ukuze uthole imininingwane).Ukuze sihlole ukunemba kwe-angular yendawo okunikezwa yi-pMUT kanye nendlela yokucubungula isignali ehlongozwayo, silinganise i-ITD (okungukuthi, umehluko wesikhathi phakathi kwemicimbi ephakeme kakhulu ekhiqizwa abemukeli ababili) njengoba ibanga ne-engeli yento kwahlukahluka.Ukuhlaziywa kwe-ITD kwabe sekuguqulelwa kuma-engeli (bheka Izindlela) futhi kwahlelwa ngokumelene nendawo yento: ukungaqiniseki ku-ITD elinganisiwe kwanda ngebanga kanye ne-engeli entweni (Fig. 2e,f).Inkinga enkulu isilinganiso se-peak-to-noise (PNR) empendulweni ye-pMUT.Uma into ikude, isignali ye-acoustic iyancipha, ngaleyo ndlela inciphisa i-PNR (Fig. 2f, umugqa oluhlaza).Ukuncipha kwe-PNR kuholela ekwandeni kokungaqiniseki esilinganisweni se-ITD, okuholela ekwandeni kokunemba kwendawo (I-Fig. 2f, umugqa ohlaza okwesibhakabhaka).Entweni ebangeni elingu-50 cm ukusuka kusidlulisi, ukunemba kwe-angular kwesistimu cishe ku-10 °.Lo mkhawulo obekwe yizici zenzwa ungathuthukiswa.Isibonelo, ingcindezi ethunyelwe yi-emitter inganyuswa, ngaleyo ndlela ikhulise i-voltage eshayela ulwelwesi lwe-pMUT.Esinye isisombululo sokukhulisa isignali edlulisiwe ukuxhuma ama-transmitter amaningi 56. Lezi zixazululo zizokwandisa ububanzi bokuthola ngezindleko zezindleko zamandla ezikhuphukile.Ukuthuthukiswa okwengeziwe kungenziwa ohlangothini lokwamukela.Isitezi somsindo esamukeli se-pMUT singancishiswa kakhulu ngokwenza ngcono ukuxhumana phakathi kwe-pMUT nesiteji sokuqala sokukhulisa izwi, okwamanje kwenziwa ngoxhumo lwezintambo nezintambo ze-RJ45.
Isithombe sekristalu ye-pMUT enamalwelwesi ayisithupha angu-880 µm ahlanganiswe ku-1.5 mm pitch.b Umdwebo wokusethwa kokulinganisa.Ithagethi itholakala endaweni ye-azimuth θ futhi ebangeni D. Isidluliseli se-pMUT sikhiqiza isignali engu-117.6 kHz egxuma isuke lapho kuqondiwe futhi ifinyelele izamukeli ezimbili ze-pMUT ezinesikhathi esihlukile sokundiza (ToF).Lo mehluko, ochazwa njengomehluko wesikhathi esiphakathi kokuzwakalayo (ITD), ubhala ngekhodi indawo yento futhi ungalinganiselwa ngokulinganisa impendulo ephezulu yezinzwa ezimbili ezamukelayo.c Ukuhlelwa kwezinyathelo zokucubungula ngaphambilini zokuguqula isignali eluhlaza ye-pMUT ibe ukulandelana kwe-spike (okungukuthi, okokufaka kugrafu yekhompyutha ye-neuromorphic).Izinzwa ze-pMUT namagrafu ekhompiyutha e-neuromorphic enziwe futhi ahlolwa, futhi ukucubungula kwangaphambili kwe-neuromorphic kusekelwe ekulingiseni isofthiwe.d Ukusabela kwe-membrane ye-pMUT lapho ithola isignali nokuguqulwa kwayo ibe isizinda se-spike.e Ukwenziwa kwasendaweni kokuhlola ukunemba kwe-angular njengomsebenzi we-engeli yento (Θ) nebanga (D) ukuya entweni eqondiwe.Indlela yokukhipha i-ITD idinga ukulungiswa kwe-angular okungenani okungaba ngu-4°C.f Ukunemba kwe-angular (umugqa oluhlaza okwesibhakabhaka) nesilinganiso esihambelanayo se-peak-to-noise (umugqa oluhlaza) uma kuqhathaniswa nebanga lento ye-Θ = 0.
Imemori ephikisayo igcina ulwazi esimweni sokuqhuba esingaguquguquki.Isimiso esiyisisekelo sendlela ukuthi ukuguqulwa kwezinto ezingeni le-athomu kubangela ushintsho ekuphatheni kwayo kagesi57.Lapha sisebenzisa inkumbulo ephikisayo esekwe ku-oxide ehlanganisa isendlalelo esingu-5nm se-hafnium dioxide ehlanganiswe phakathi kwe-titanium engaphezulu nangaphansi kanye nama-electrode e-titanium nitride.I-conductivity yamadivayisi e-RRAM ingashintshwa ngokusebenzisa i-waveform yamanje/i-voltage edala noma enqamule imicu yokuqhuba yezikhala ze-oxygen phakathi kwama-electrode.Sahlanganisa amadivaysi anjalo58 kunqubo evamile engu-130 nm ye-CMOS ukuze sakhe isifunda se-neuromorphic esilungiseka kabusha sisebenzisa umtshina wokuziqondana kanye nesekhethi yomugqa wokulibaziseka (Fig. 3a).Imvelo engaguquki ne-analog yedivayisi, kuhlanganiswe nemvelo eqhutshwa umcimbi yesekethe ye-neuromorphic, inciphisa ukusetshenziswa kwamandla.Isekhethi inomsebenzi wokuvula/ukuvala ngokushesha: isebenza ngokushesha ngemva kokuvulwa, ivumela amandla ukuthi acishwe ngokuphelele lapho isifunda singasebenzi.Amabhlogo wokwakha ayinhloko weskimu esihlongozwayo akhonjiswe emkhiwaneni.3b.Iqukethe izinhlaka ze-N parallel single-resistor single-transistor (1T1R) ezifaka izisindo ze-synaptic lapho kuthathwa khona imisinga enesisindo, ijovwe ku-synapse evamile ye-different pair integrator (DPI)59, futhi ekugcineni ijovwe ku-synapse ngokuhlanganisa kanye ukuvuza.icushiwe (LIF) neuron 60 (bona Izindlela ukuze uthole imininingwane).Ukufakwa kokufakwayo kusetshenziswa esangweni lesakhiwo se-1T1R ngendlela yokulandelana kwama-voltage pulses anobude besikhathi ngokulandelana kwamakhulu ama-nanoseconds.Inkumbulo ekwazi ukumelana nayo ingafakwa esimweni sokuqhuba kahle (i-HCS) ngokusebenzisa ireferensi yangaphandle evumayo ku-Vtop lapho i-Vbottom igxiliwe, futhi isethelwe esimweni sokuqhuba esiphansi (LCS) ngokusebenzisa i-voltage ephozithivu ku-Vbottom lapho i-Vtop imisiwe.Isilinganiso senani le-HCS lingalawulwa ngokukhawulela ukusebenza kwamanje (ukuthobela) kwe-SET (ICC) ngevolthi yesango yomthombo we-transistor yochungechunge (Fig. 3c).Imisebenzi ye-RRAM kusekethe ikabili: iqondisa futhi ikala ama-pulses okokufaka.
Isithombe se-electron microscope (SEM) yeskena sedivayisi ye-HfO2 1T1R RRAM eluhlaza okwesibhakabhaka ehlanganiswe kubuchwepheshe be-CMOS obungu-130 nm nama-transistors esikhethi (650 nm ububanzi) ngokuluhlaza.b Amabhlogo wokwakha ayisisekelo we-schema ye-neuromorphic ehlongozwayo.Ama-voltage okufakwayo ama-pulses (iziqongo) i-Vin0 ne-Vin1 zisebenzisa i-Iweight yamanje, elingana nezifunda zokuqhuba i-G0 ne-G1 yesakhiwo se-1T1R.Lokhu kwamanje kujovwa kuma-synapse e-DPI futhi kujabulise ama-neuron e-LIF.I-RRAM G0 ne-G1 ifakwe ku-HCS ne-LCS ngokulandelanayo.c Umsebenzi we-cumulative conductance density weqembu lamadivayisi we-16K RRAM njengomsebenzi wokufanisa wamanje we-ICC, olawula ngempumelelo izinga lokuqhuba.d Izilinganiso zesiyingi ku-(a) ezibonisa ukuthi i-G1 (ku-LCS) ivimba ngempumelelo okokufaka okuvela ku-Vin1 (okuluhlaza), futhi ngempela i-voltage ye-neuron ephumayo isabela kuphela kokokufaka okuluhlaza okwesibhakabhaka okuvela ku-Vin0.I-RRAM inquma ngempumelelo ukuxhumana kusekethe.e Ukulinganiswa kwesekhethi ku-(b) okubonisa umthelela wevelu ye-conductance G0 ku-voltage yolwelwesi lwe-Vmem ngemva kokufaka i-voltage pulse Vin0.Ukuziphatha okwengeziwe, impendulo inamandla: ngakho-ke, idivayisi ye-RRAM isebenzisa isisindo sokuxhuma se-I/O.Izilinganiso zenziwe kusekethe futhi zikhombisa ukusebenza okumbaxambili kwe-RRAM, umzila kanye nesisindo sama-pulses okokufaka.
Okokuqala, njengoba kunezimo ezimbili eziyisisekelo zokuqhuba (i-HCS ne-LCS), ama-RRAM angavimba noma aphuthe ama-pulses okokufaka uma esezifundeni ze-LCS noma ze-HCS, ngokulandelanayo.Njengomphumela, i-RRAM inquma ngempumelelo ukuxhumana kusekethe.Lesi isisekelo sokukwazi ukulungisa kabusha isakhiwo.Ukuze sibonise lokhu, sizochaza ukuqaliswa kwesekethe esunguliwe yebhlokhi yesifunda ku-Fig. 3b.I-RRAM ehambisana ne-G0 ihlelwe ku-HCS, futhi i-RRAM G1 yesibili ihlelwe yaba yi-LCS.Ama-pulses okokufaka asetshenziswa kukho kokubili i-Vin0 ne-Vin1.Imiphumela yokulandelana okubili kwama-pulses okokufaka yahlaziywa kuma-neurons okukhiphayo ngokuqoqa i-voltage ye-membrane ye-neuron kanye nesignali yokuphumayo kusetshenziswa i-oscilloscope.Ukuhlolwa kube yimpumelelo lapho kuphela idivayisi ye-HCS (G0) ixhunywe kukushaya kwemithambo ye-neuron ukuze kushukunyiswe ukungezwani kolwelwesi.Lokhu kuboniswa kuMfanekiso 3d, lapho isitimela sokushaya kwesibhakabhaka sibangela ukuthi i-voltage ye-membrane yakheke ku-membrane capacitor, kuyilapho isitimela esishaya ngamandla esiluhlaza sigcina i-voltage ye-membrane ingashintshi.
Umsebenzi wesibili obalulekile we-RRAM ukuqaliswa kwezisindo zokuxhuma.Ngokusebenzisa ukulungiswa kokuziphatha kwe-analog ye-RRAM, uxhumo lwe-I/O lungakalwa ngokufanele.Esivivinyweni sesibili, idivayisi ye-G0 yahlelwa yaba amazinga ahlukene e-HCS, futhi i-pulse yokufaka yasetshenziswa kokokufaka kwe-VIn0.I-pulse yokufaka idonsa i-current (Iweight) kusukela ocingweni, elingana nendlela yokuqhuba kanye ne-Vtop okungenzeka ibe khona ehambisanayo − Vbot.Lesi sikhathi esinesisindo sibe sesijovwa kuma-synapse e-DPI kanye nama-neuron aphumayo e-LIF.I-voltage ye-membrane yama-neurons okukhiphayo yarekhodwa kusetshenziswa i-oscilloscope futhi yaboniswa ku-Fig. 3d.Ukuphakama kwe-voltage yolwelwesi lwe-neuron ekuphenduleni ku-pulse okokufaka okukodwa kuhambisana nokuqhutshwa kwenkumbulo ephikisayo, okubonisa ukuthi i-RRAM ingasetshenziswa njengento ehlelekayo yesisindo se-synaptic.Lezi zivivinyo ezimbili zokuqala zibonisa ukuthi iplathifomu ehlongozwayo ye-neuromorphic esekelwe ku-RRAM iyakwazi ukusebenzisa izici eziyisisekelo zendlela ye-Jeffress eyisisekelo, okungukuthi umugqa wokulibaziseka kanye nesekhethi yomtshina wokuhlangana.Iplathifomu yesifunda yakhiwe ngokunqwabelanisa amabhlogo alandelanayo, njengamabhlogo akuMfanekiso 3b, nokuxhuma amasango awo emugqeni wokufaka ovamile.Sakhe, sakha, futhi sahlola inkundla ye-neuromorphic ehlanganisa ama-neuron amabili aphumayo athola okokufaka okubili (Fig. 4a).Umdwebo wesifunda ukhonjiswe kuMfanekiso 4b.I-matrix ye-2 × 2 RRAM ephezulu ivumela ama-pulses okokufaka ukuthi aqondiswe kuma-neurons amabili okukhiphayo, kuyilapho i-2 × 2 matrix ephansi ivumela ukuxhumeka okuphindaphindiwe kwama-neurons amabili (N0, N1).Sibonisa ukuthi le nkundla ingasetshenziswa ngokucushwa komugqa wokulibaziseka kanye nemisebenzi emibili ehlukene yesitholi sezinto eziqondane, njengoba kuboniswa izilinganiso zokuhlola ku-Fig. 4c-e.
Umdwebo wesiyingi owenziwe ama-neuron amabili okukhiphayo i-N0 kanye ne-N1 ethola okokufaka okubili okungu-0 no-1. Amadivayisi amane aphezulu ohlwini achaza ukuxhumana kwe-synaptic kusuka kokokufakayo kuye kokukhiphayo, futhi amaseli amane aphansi achaza ukuxhumana okuphindelelayo phakathi kwama-neurons.Ama-RRAM anemibala amelela amadivayisi alungiselelwe ku-HCS kwesokudla: amadivayisi ku-HCS avumela uxhumo futhi amele izisindo, kuyilapho amadivayisi aku-LCS avimba ama-pulses okokufaka futhi akhubaza ukuxhumeka kokuphumayo.b Umdwebo wesekethe (a) onamamojula ayisishiyagalombili e-RRAM agqanyiswe ngokuluhlaza okwesibhakabhaka.c Imigqa yokulibaziseka yenziwa ngokusebenzisa nje amandla e-DPI synapses nama-LIF neurons.I-RRAM eluhlaza isethelwe ekuqhubeni phezulu ngokwanele ukuze ikwazi ukubangela iphutha ekuphumeni ngemva kokubambezeleka kokokufaka Δt.d Umfanekiso wohlelo lokutholwa kwe-CD okungazweli kokuqondisa kwamasiginali ancike esikhathini.I-output neuron 1, N1, imililo kokokufaka 0 no-1 ngokulibaziseka okufushane.e Isekhethi ye-CD ezwelayo ekuqondeni, isekethe ethola lapho okokufaka 1 kusondela ku-0 futhi ifike ngemva kokufaka 0. Okukhiphayo kwesekhethi kumelelwe yi-neuron 1 (N1).
Ulayini wokulibaziseka (Umfanekiso 4c) umane usebenzise ukuziphatha okuguquguqukayo kwama-synapse e-DPI nama-neuron e-LIF ukuze akhiqize kabusha i-spike yokufaka ukusuka ku-Vin1 kuya ku-Vout1 ngokubambezela i-Tdel.I-G3 RRAM kuphela exhunywe ku-Vin1 ne-Vout1 ehlelwe nge-HCS, amanye ama-RRAM ahlelwe nge-LCS.Idivayisi ye-G3 yahlelelwa ama-92.6 µs ukuze kuqinisekiswe ukuthi i-pulse yokokufaka ngakunye inyusa i-voltage ye-membrane ye-neuron ephumayo ngokwanele ukuze ifinyelele umkhawulo futhi ikhiqize i-pulse yokukhiphayo ebambezelekile.Ukubambezeleka kwe-Tdel kunqunywa yi-synaptic kanye ne-neural time constants.Izitholi eziqondane zithola ukwenzeka kwezimpawu zokufaka ezihlotshaniswayo kodwa ezisatshalaliswa ngendawo.I-CD engazweli ekuqondeni incike kokokufaka ngakunye ukuguqulela ku-neuron ephumayo evamile (Umfanekiso 4d).Ama-RRAM amabili axhuma i-Vin0 ne-Vin1 ku-Vout1, G2 kanye ne-G4 ngokulandelanayo ahlelelwe ukuqhuba okuphezulu.Ukufika kanyekanye kwama-spikes ku-Vin0 naku-Vin1 kukhulisa i-voltage ye-N1 neuron membrane ngaphezu komkhawulo odingekayo ukuze kukhiqizwe i-spike ephumayo.Uma okokufaka okubili kuqhelelene kakhulu ngesikhathi, ukushaja ku-voltage ye-membrane eqoqwe okokufaka kokuqala kungase kube nesikhathi sokubola, kuvimbele ulwelwesi olunamandla lwe-N1 ukuthi lufinyelele inani lomkhawulo.I-G1 ne-G2 zihlelelwe cishe ama-65 µs, okuqinisekisa ukuthi okokufaka okukodwa akunyusi i-voltage ye-membrane ngokwanele ukuze kubangele ukukhuphuka kokuphumayo.Ukuthola ukuqondana phakathi kwezehlakalo ezisatshalaliswa emkhathini nesikhathi kuwumsebenzi obalulekile osetshenziswa ebangeni elibanzi lemisebenzi yezinzwa efana nokugwema izithiyo zokugeleza okubonakalayo kanye nokwenza kwasendaweni komthombo womsindo.Ngakho-ke, ama-CD azwela ukuqondisa futhi angenaluzwela kuyisakhiwo esiyisisekelo sokwakha amasistimu okwenziwa kwasendaweni abonwayo nenomsindo.Njengoba kuboniswa izici zezikhathi ezingaguquki (bona I-Supplementary Fig. 2), isifunda esihlongozwayo sisebenzisa uhla olufanele lwama-oda amane wezilinganiso zesikhathi sobukhulu.Ngakho-ke, ingakwazi ukuhlangabezana nezidingo zezinhlelo ezibukwayo nezomsindo kanyekanye.I-CD ye-Directional-sensitive isekethe ezwelayo kuhlelo lwendawo lokufika kwama-pulses: ukusuka kwesokudla kuye kwesokunxele nokuphambene.Kuyibhulokhi yokwakha eyisisekelo kunethiwekhi eyisisekelo yokuthola ukunyakaza yesistimu ebonakalayo ye-Drosophila, esetshenziselwa ukubala izikhombisi-ndlela ezinyakazayo nokubona ukungqubuzana62.Ukuze kuzuzwe i-CD ezwela isiqondiso, okokufaka okubili kufanele kuqondiswe kuma-neurons amabili ahlukene (N0, N1) futhi uxhumano oluqondisayo kufanele lusungulwe phakathi kwabo (Fig. 4e).Lapho okokufaka kokuqala kwamukelwa, CHA isabela ngokukhuphula i-voltage kulwelwesi lwaso ngaphezu kwevelu yomkhawulo nokuthumela ukuhlinza.Lo mcimbi ophumayo, wona, ushisa i-N1 ngenxa yokuxhumana okuqondile okugqanyiswe ngokuluhlaza.Uma umcimbi wokufaka i-Vin1 ifika futhi inika amandla i-N1 kuyilapho i-voltage ye-membrane isephezulu, i-N1 ikhiqiza umcimbi ophumayo obonisa ukuthi okufanayo kutholakele phakathi kwezinto ezimbili ezifakiwe.Uxhumano lwezikhombisi-ndlela luvumela i-N1 ukuthi ikhiphe okukhiphayo kuphela uma okokufaka 1 kuza ngemva kokufaka 0. I-G0, G3, ne-G7 zihlelwe zibe ngu-73.5 µS, 67.3 µS, no-40.2 µS, ngokulandelana, okuqinisekisa ukuthi isipike esisodwa kokokufaka kwe-Vin0 kubangela ukubambezeleka. i-spike yokuphumayo, kuyilapho amandla e-membrane ye-N1 efinyelela kuphela emngceleni lapho kokubili ukuqhuma okokufaka kufika ekuvumelaniseni..
Ukuhlukahluka kuwumthombo wokungapheleli kuma-modeled neuromorphic systems63,64,65.Lokhu kuholela ekuziphatheni okungafani kwama-neurons nama-synapses.Izibonelo zobubi obunjalo zifaka phakathi u-30% (okusho ukuchezuka okujwayelekile) ukuhlukahluka enzuzweni yokokufaka, ukungaguquguquki kwesikhathi, kanye nesikhathi sokuphikisa, ukusho okumbalwa (bona Izindlela).Le nkinga igqama nakakhulu lapho ama-neural circuits amaningi exhunywe ndawonye, ​​njenge-CD ezwelayo lapho ehlanganisa ama-neurons amabili.Ukuze kusebenze kahle, ukuzuza kanye nesikhathi sokubola kwama-neurons amabili kufanele kufane ngangokunokwenzeka.Isibonelo, umehluko omkhulu enzuzweni yokokufaka ungabangela i-neuron eyodwa ukuthi isabele ngokweqile ku-pulse yokokufaka kuyilapho enye i-neuron ingasabeli.Emkhiwaneni.Umfanekiso 5a ubonisa ukuthi ama-neuron akhethwe ngokungahleliwe aphendula ngokuhlukile ku-pulse yokufaka efanayo.Lokhu kuhlukahluka kwe-neural kuhlobene, isibonelo, emsebenzini wama-CD azwela ukuqondisa.Kuhlelo oluboniswe ku-fig.I-5b, c, inzuzo yokufaka ye-neuron 1 iphakeme kakhulu kune-neuron 0. Ngakho-ke, i-neuron 0 idinga ama-pulses okokufaka amathathu (esikhundleni se-1) ukuze ifinyelele embundwini, futhi i-neuron 1, njengoba kulindelekile, idinga imicimbi emibili yokufaka.Ukusebenzisa i-spike biomimetic plasticity (STDP) encike esikhathini se-spike kuyindlela engenzeka yokunciphisa umthelela wamasekhethi we-neural ne-synaptic angacacile futhi avilaphayo ekusebenzeni kwesistimu43.Lapha siphakamisa ukusebenzisa ukuziphatha kwepulasitiki kwenkumbulo ephikisayo njengendlela yokuba nomthelela ekuthuthukiseni okokufaka kwe-neural kanye nokunciphisa imiphumela yokuhlukahluka kumasekhethi we-neuromorphic.Njengoba kuboniswe emkhiwaneni.I-4e, amazinga okuqhuba ahlotshaniswa nesisindo se-synaptic ye-RRAM aguqule ngempumelelo impendulo ye-voltage ye-neural membrane ehambisanayo.Sisebenzisa isu lokuhlela le-RRAM eliphindaphindayo.Okokufaka okunikeziwe, amanani okuqhuba ezisindo ze-synaptic ahlelwa kabusha kuze kutholakale ukuziphatha okuqondiwe kwesekethe (bona Izindlela).
izilinganiso zokuhlola zempendulo yama-neurons angawodwana ayisishiyagalolunye akhethwe ngokungahleliwe kumbhobho wokufaka ofanayo.Impendulo iyahlukahluka kubo bonke abantu, okuthinta inzuzo yokufaka kanye nesikhathi esingaguquki.b Izilinganiso zokuhlola zethonya lama-neuron ekuguquguqukeni kwama-neuron athinta i-CD ezwela ukuqondiswa.Ama-neuron amabili azwelayo e-CD asabela ngendlela ehlukile kusisusa sokufakwayo ngenxa yokuhlukahluka kwe-neuron-to-neuron.I-Neuron 0 inenzuzo ephansi yokokufaka kune-neuron 1, ngakho-ke kuthatha ama-pulses okokufaka amathathu (esikhundleni sika-1) ukuze udale i-spike yokukhiphayo.Njengoba kulindelekile, i-neuron 1 ifinyelela emngceleni ngemicimbi emibili yokokufaka.Uma okokufaka 1 kufika Δt = 50 µs ngemva kokusha kwe-neuron 0, i-CD ihlala ithulile ngoba i-Δt inkulu kunesikhathi esingaguquki se-neuron 1 (cishe ama-22 µs).c incishiswa ngo-Δt = 20 µs, ukuze okokufaka okungu-1 kuphakame lapho ukudubula kwe-neuron 1 kusephezulu, okuholela ekutholweni ngesikhathi esisodwa kwemicimbi yokufakwayo emibili.
Izinto ezimbili ezisetshenziswe kukholomu yokubala ye-ITD umugqa wokulibaziseka kanye ne-CD enganaki isiqondisindlela.Womabili amasekhethi adinga ukulinganisa okunembile ukuze kuqinisekiswe ukusebenza kahle kokuma kwento.Ulayini wokulibaziseka kufanele ulethe inguqulo ebambezeleke ngokunembile yokuphakama kokufaka (Fig. 6a), futhi i-CD kufanele yenziwe isebenze kuphela uma okokufaka kuwela phakathi kwebanga lokuthola okuqondiwe.Ngomugqa wokulibaziseka, izisindo ze-synaptic zokuxhuma okokufaka (i-G3 ku-Fig. 4a) zahlelwa kabusha kuze kutholakale ukubambezeleka okuhlosiwe.Setha ukubekezelela eduze nokulibaziseka okuqondiwe ukuze umise uhlelo: uma kuncane ukubekezelelana, kuba nzima kakhulu ukusetha ngempumelelo umugqa wokulibaziseka.Emkhiwaneni.Umfanekiso 6b ubonisa imiphumela yenqubo yokulinganisa umugqa wokulibaziseka: kungabonakala ukuthi isikimu esihlongozwayo singanikeza ngokunembile konke ukubambezeleka okudingekayo esikimini sokuklama (kusuka ku-10 kuya ku-300 μs).Inombolo enkulu yokuphindaphinda kokulinganisa ithinta ikhwalithi yenqubo yokulinganisa: ukuphindaphinda okungu-200 kunganciphisa iphutha libe ngaphansi kuka-5%.Ukulinganisa okukodwa kuhambisana nokusebenza kokusetha/ukusetha kabusha kweseli ye-RRAM.Inqubo yokushuna ibalulekile futhi ekuthuthukiseni ukunemba kwemojuli ye-CD yokutholwa komcimbi oseduze ngokushesha.Kuthathe izikhathi eziyishumi zokulinganisa ukulinganisa ukuze kuzuzwe izinga leqiniso (okungukuthi, izinga lezehlakalo ezihlonzwe ngendlela efanele njengezifanele) ngaphezu kwama-95% (umugqa oluhlaza kuFigure 6c).Kodwa-ke, inqubo yokushuna ayizange ithinte izehlakalo ezinhle ezingamanga (okungukuthi, imvamisa yemicimbi ekhonjwe ngephutha njengezifanele).Enye indlela ebonwa ezinhlelweni zebhayoloji yokunqoba izithiyo zesikhathi sezindlela ezisebenza ngokushesha ukuphindaphinda (okungukuthi, amakhophi amaningi ento efanayo asetshenziselwa ukwenza umsebenzi othile).Sikhuthazwe yi-biology66, sibeke amasekhethi ama-CD amaningana kumojula ngayinye ye-CD phakathi kwemigqa yokulibaziseka emibili ukuze sinciphise umthelela wezinto ezingamanga.Njengoba kuboniswe emkhiwaneni.6c (umugqa oluhlaza), ukubeka izici ze-CD ezintathu kumojula ngayinye ye-CD kungehlisa izinga le-alamu elingamanga libe ngaphansi kuka-10–2.
Umthelela wokuhlukahluka kwe-neuronal kumasekhethi omugqa wokulibaziseka.b Amasekhethi omugqa wokulibaziseka angalinganiswa abe ukubambezeleka okukhulu ngokusetha izikhathi ezingaguquki zama-neurons e-LIF ahambisanayo nama-synapse e-DPI kube amanani amakhulu.Ukwenyusa inani lokuphindaphinda kwenqubo yokulinganisa ye-RRAM kwenze kwaba nokwenzeka ukuthuthukisa ngokuphawulekayo ukunemba kokulibaziseka okuqondiwe: ukuphindaphinda okungu-200 kwehlise iphutha laba ngaphansi kuka-5%.Ukuphindaphinda okukodwa kuhambisana nokusebenza kwe-SET/RESET kuseli ye-RRAM.Imojula ngayinye ye-CD kumodeli ye-c Jeffress ingasetshenziswa kusetshenziswa ama-elementi e-CD ahambisanayo angu-N ukuze kube nokuguquguquka okukhulu ngokuphathelene nokuhluleka kwesistimu.d Ukulinganisa okwengeziwe kwe-RRAM kokulinganisa kwenyusa izinga leqiniso eliphozithivu (umugqa oluhlaza), kuyilapho izinga elingelona iqiniso lizimele kwinani lokuphindaphinda (umugqa oluhlaza).Ukubeka izinto eziningi ze-CD ngokuhambisana kugwema ukutholwa okungamanga kokufana kwamamojula e-CD.
Manje sihlola ukusebenza namandla okusetshenziswa kwesistimu ehlanganisiwe yokwenziwa kwasendaweni kwento ephuma ekupheleni ukuya-ekupheleni eboniswe kuMfanekiso 2 sisebenzisa izilinganiso zezakhiwo ze-acoustic zenzwa ye-pMUT, i-CD, namasekhethi omugqa wokulibaziseka akha igrafu yekhompyutha ye-neuromorphic.Imodeli kaJeffress (Fig. 1a).Ngokuqondene negrafu ye-neuromorphic computing, inani elikhulu lamamojula e-CD, ngcono ukulungiswa kwe-angular, kodwa futhi amandla aphezulu esistimu (Fig. 7a).Ukuvumelana kungafinyelelwa ngokuqhathanisa ukunemba kwezingxenye ngazinye (izinzwa ze-pMUT, ama-neurons, nama-synaptic circuits) nokunemba kwalo lonke uhlelo.Ukulungiswa komugqa wokulibaziseka kukhawulelwe izikhathi ezingaguquki zama-synapses alingisa nama-neuron, okuthi ohlelweni lwethu lweqe u-10 µs, ohambisana nokulungiswa kwe-angular okungu-4° (bona Izindlela).Ama-node athuthuke kakhulu anobuchwepheshe be-CMOS azovumela ukuklanywa kwamasekhethi e-neural kanye ne-synaptic anezimo zesikhathi esiphansi, okuholela ekunembeni okuphezulu kwezinto zomugqa wokulibaziseka.Nokho, ohlelweni lwethu, ukunemba kunqunyelwe iphutha le-pMUT ekulinganiseni indawo eyi-angular, okungukuthi 10° (umugqa ovundlile oluhlaza okwesibhakabhaka ku-Fig. 7a).Silungise inani lamamojula e-CD ku-40, elihambisana nesinqumo se-angular esingaba ngu-4 °, okungukuthi, ukunemba kwe-angular kwegrafu yokubala (umugqa ovundlile okhanyayo okwesibhakabhaka ku-Fig. 7a).Ezingeni lesistimu, lokhu kunikeza isixazululo esingu-4° nokunemba kuka-10° ezintweni ezitholakala ngo-50 cm phambi kwesistimu yezinzwa.Leli nani liqhathaniswa nezinhlelo ze-neuromorphic sound localization ezibikwe kunkomba.67. Ukuqhathaniswa kwesistimu ehlongozwayo nesimo sobuciko kungatholakala kuThebula Lokwengeza 1. Ukwengeza ama-pMUT engeziwe, ukukhulisa izinga lesignali ye-acoustic, nokunciphisa umsindo we-elekthronikhi izindlela ezingenzeka zokuthuthukisa ukunemba kokwenziwa kwendawo.) ilinganiselwa ku-9.7.nz.55. Njengoba kunikezwe amayunithi angama-CD angama-40 kugrafu yokubala, ukulingisa kwe-SPICE kulinganisele amandla ngokusebenza ngakunye (okungukuthi, amandla okumisa into) abe ngu-21.6 nJ.Isistimu ye-neuromorphic yenziwa isebenze kuphela lapho umcimbi wokufakwayo ufika, okungukuthi lapho igagasi elizwakalayo lifinyelela kunoma yisiphi isamukeli se-pMUT futhi lidlula umkhawulo wokutholwa, ngaphandle kwalokho lihlala lingasebenzi.Lokhu kugwema ukusetshenziswa kwamandla okungadingekile uma ingekho isignali yokufaka.Uma kucatshangelwa imvamisa yokusebenza kwendawo ka-100 Hz nesikhathi sokwenza kusebenze esingu-300 µs ngokusebenza ngakunye (ubukhulu obungenzeka be-ITD), ukusetshenziswa kwamandla kwegrafu yekhompyutha ye-neuromorphic ngu-61.7 nW.Ngokucutshungulwa kwangaphambili kwe-neuromorphic kusetshenziswe kumamukeli ngamunye we-pMUT, ukusetshenziswa kwamandla kwalo lonke uhlelo kufinyelela ku-81.6 nW.Ukuze siqonde ukusebenza kahle kwamandla kwendlela ehlongozwayo ye-neuromorphic uma kuqhathaniswa nezingxenyekazi zekhompuyutha ezivamile, siqhathanise le nombolo namandla adingekayo ukwenza umsebenzi ofanayo kusilawuli esinamandla aphansi sesimanje sisebenzisa i-neuromorphic noma i-beamforming68 Skill evamile.Indlela ye-neuromorphic ibheka isigaba se-analog-to-digital converter (ADC), silandelwa isihlungi se-band-pass kanye nesiteji sokukhipha imvilophu (indlela kaTeeger-Kaiser).Ekugcineni, kwenziwa umsebenzi wokukhipha i-ToF.Sikweqile ukubala kwe-ITD okusekelwe ku-ToF kanye nokuguqulwa endaweni elinganiselwe ye-angular njengoba lokhu kwenzeka kanye ngesilinganiso ngasinye (bona Izindlela).Uma kubhekwa isilinganiso samasampula esingu-250 kHz kuzo zombili iziteshi (abamukeli be-pMUT), ukusebenza kwesihlungi sokudlula amabhendi angu-18, imisebenzi yokukhipha imvilophu emi-3, nokusebenza komkhawulo o-1 isampula ngayinye, ukusetshenziswa kwamandla okuphelele kulinganiselwa kuma-microwatts angu-245.Lokhu kusebenzisa i-microcontroller's low-power mode69, evula lapho ama-algorithms engasebenzi, okunciphisa ukusetshenziswa kwamandla ku-10.8 µW.Ukusetshenziswa kwamandla kwesixazululo sokucubungula isignali ekhazimulayo okuhlongoziwe kusithenjwa.31, enabamukeli abangu-5 pMUT kanye nemishayo engu-11 esatshalaliswe ngokufanayo endizeni ye-azimuth [-50°, +50°], ingu-11.71 mW (bona isigaba Sezindlela ukuze uthole imininingwane).Ngaphezu kwalokho, sibika ukusetshenziswa kwamandla kwe-FPGA47-based Time Difference Encoder (TDE) elinganiselwa ku-1.5 mW njengokungena esikhundleni semodeli ye-Jeffress yokwenza izinto zasendaweni.Ngokusekelwe kulezi zilinganiso, indlela ehlongozwayo ye-neuromorphic yehlisa ukusetshenziswa kwamandla ngama-oda amahlanu obukhulu uma kuqhathaniswa nesilawuli esincane sisebenzisa amasu akudala okwenza uhlaka okusebenza kokwenziwa kwasendaweni kwento.Ukwamukela indlela ye-neuromorphic yokucubungula isignali ku-microcontroller yakudala kunciphisa ukusetshenziswa kwamandla cishe ngama-oda amabili obukhulu.Ukusebenza kwesistimu ehlongozwayo kungachazwa ngenhlanganisela yesekethe ye-analog ye-resissive-memory ye-asynchronous ekwazi ukwenza izibalo ezisenkumbulweni kanye nokuntuleka kokuguqulwa kwe-analog-to-digital okudingekayo ukuze uthole amasignali.
ukulungiswa kwe-Angular (okuluhlaza okwesibhakabhaka) kanye nokusetshenziswa kwamandla (okuluhlaza) kokusebenza kokwenziwa kwasendaweni kuye ngenani lamamojula e-CD.Ibha evundlile eluhlaza okwesibhakabhaka okumnyama imelela ukunemba kwe-angular kwe-PMUT kanye nebha evundlile eluhlaza okwesibhakabhaka ngokukhanyayo imelela ukunemba kwe-angular kwegrafu yekhompyutha ye-neuromorphic.b Ukusetshenziswa kwamandla kwesistimu ehlongozwayo kanye nokuqhathanisa nokusetshenziswa kokubili okuxoxiwe kwe-microcontroller kanye nokuqaliswa kwedijithali kwe-Time Difference Encoder (TDE)47 FPGA.
Ukuze kuncishiswe ukusetshenziswa kwamandla kwesistimu yendawo eqondiwe, sicabange, saklama futhi sasebenzisa isekethe ye-neuromorphic esekelwe ku-RRAM eqhutshwa umcimbi esebenza kahle ecubungula ulwazi lwesiginali olukhiqizwe izinzwa ezakhelwe ngaphakathi ukuze sibale indawo yento eqondiwe ngokoqobo. isikhathi..Nakuba izindlela zokucubungula zendabuko ziqhubeka nokusampula amasignali atholiwe futhi zenza izibalo ukuze kukhishwe ulwazi oluwusizo, isixazululo esihlongozwayo se-neuromorphic senza izibalo ngendlela efanayo njengoba ulwazi oluwusizo lufika, lukhulise ukusebenza kahle kwamandla esistimu ngama-oda amahlanu obukhulu.Ngaphezu kwalokho, sigqamisa ukuguquguquka kwamasekethe e-neuromorphic asuselwa ku-RRAM.Ikhono le-RRAM lokushintsha ukusebenza ngendlela engaguquguquki (i-plasticity) linxephezela ukuhlukahluka okungokwemvelo kwe-ultra-low power analog DPI's synaptic and neural circuits.Lokhu kwenza le sekethi esekwe ku-RRAM ibe nezinto eziningi futhi ibe namandla.Umgomo wethu awukona ukukhipha imisebenzi eyinkimbinkimbi noma amaphethini kumasiginali, kodwa ukwenza izinto zibe ezasendaweni ngesikhathi sangempela.Isistimu yethu ingacindezela kahle isiginali futhi ekugcineni iyithumele ezinyathelweni eziqhubekayo zokucubungula ukwenza izinqumo eziyinkimbinkimbi uma kudingeka.Kumongo wezinhlelo zokusebenza zasendaweni, isinyathelo sethu sokucubungula kusengaphambili se-neuromorphic singanikeza ulwazi mayelana nendawo yezinto.Lolu lwazi lungasetshenziswa, isibonelo, ukuthola ukunyakaza noma ukubonwa kokuthinta.Sigcizelela ukubaluleka kokuhlanganisa izinzwa zamandla aphansi kakhulu njengama-pMUT nama-electronics anamandla aphansi.Kulokhu, izindlela ze-neuromorphic zibe ukhiye njengoba zisiholele ekuthuthukiseni ukusetshenziswa kwesekethe okusha kwezindlela zekhompiyutha eziphefumulelwe ngokwebhayoloji njengemodeli ye-Jeffress.Ngomongo wezinhlelo zokusebenza zokuhlanganisa izinzwa, isistimu yethu ingahlanganiswa nezinzwa eziningana ezisekelwe kumcimbi ukuze kutholwe ulwazi olunembe kakhudlwana.Nakuba izikhova zizinhle kakhulu ekutholeni inyamazane ebumnyameni, zikwazi ukubona kahle kakhulu futhi zenza ukusesha okuhlangene kokuzwa nokubukwayo ngaphambi kokubamba inyamazane70.Lapho i-neuron ethile ezwayo ivutha umlilo, isikhova sithola ulwazi esiludingayo ukuze sinqume ukuthi singaqala ngakuphi ukusesha kwaso okubukwayo, ngaleyo ndlela sigxilise ukunaka kwaso engxenyeni encane yendawo yokubuka.Inhlanganisela yezinzwa ezibukwayo (ikhamera ye-DVS) kanye nenzwa yokulalela ehlongozwayo (ngokusekelwe ku-pMUT) kufanele ihlolwe ukuze kuthuthukiswe ama-agent azayo azimele.
Inzwa ye-pMUT itholakala ku-PCB enazamukeli ezimbili eziqhelelene cishe ngo-10 cm, futhi isidlulisi sitholakala phakathi kwabamukeli.Kulo msebenzi, ulwelwesi ngalunye luyisakhiwo se-bimorph esimisiwe esihlanganisa izingqimba ezimbili ze-piezoelectric aluminium nitride (AlN) engu-800 nm obukhulu ehlanganiswe phakathi kwezingqimba ezintathu ze-molybdenum (Mo) engu-200 nm enogqinsi futhi embozwe ngengqimba engu-200 nm obukhulu.isendlalelo se-SiN esiphezulu esidlulayo njengoba kuchazwe kusithenjwa.71. Ama-electrode angaphakathi nangaphandle asetshenziselwa izingqimba ezingezansi nangaphezulu ze-molybdenum, kuyilapho i-electrode ephakathi ye-molybdenum ingenakulinganiswa futhi isetshenziswe njengomhlabathi, okuholela kulwelwesi olunamapheya amane ama-electrode.
Lesi sakhiwo sivumela ukusetshenziswa kwe-membrane deformation evamile, okuholela ekudluliseni okuthuthukisiwe nokuthola ukuzwela.I-pMUT enjalo ngokuvamile ibonisa ukuzwela kwenjabulo okungu-700 nm/V njengesithumeli, ihlinzeka ngomfutho ongaphezulu ongu-270 Pa/V.Njengomamukeli, ifilimu eyodwa ye-pMUT ibonisa ukuzwela kwesekethe emfushane engu-15 nA/Pa, ehlobene ngokuqondile ne-piezoelectric coefficient ye-AlN.Ukuhlukahluka kobuchwepheshe be-voltage kungqimba lwe-AlN kuholela ekushintsheni kwemvamisa ye-resonant, enganxeshezelwa ngokusebenzisa ukuchema kwe-DC ku-pMUT.Ukuzwela kwe-DC kukalwe ku-0.5 kHz/V.Ukuze kufakwe abalingisi be-acoustic, kusetshenziswa imakrofoni ngaphambi kwe-pMUT.
Ukuze silinganise i-echo pulse, sibeke ipuleti elingunxande elinendawo engaba ngu-50 cm2 ngaphambi kwe-pMUT ukukhombisa amaza omsindo akhishiwe.Kokubili ibanga phakathi kwamapuleti kanye ne-engeli ehlobene nendiza ye-pMUT kulawulwa kusetshenziswa izibambi ezikhethekile.Umthombo wamandla kagesi we-Tectronix CPX400DP uchemile ulwelwesi lwe-pMUT oluthathu, lulungisa imvamisa ye-resonant iye ku-111.9 kHz31, kuyilapho ama-transmitters ashayelwa i-Tectronix AFG 3102 pulse generator eshunswe imvamisa ezwakalayo (111.9 kHz) kanye nomjikelezo womsebenzi ongu-0.01.Ama-currents afundwa kumachweba amane okukhiphayo kwesamukeli ngasinye se-pMUT aguqulelwa kuma-voltage kusetshenziswa ukwakheka okukhethekile okuhlukile kwamanje kanye ne-voltage, futhi amasiginali aphumayo afakwa kudijithali uhlelo lokutholwa kwedatha ye-Spektrum.Umkhawulo wokutholwa wawubonakala ngokutholwa kwesignali ye-pMUT ngaphansi kwezimo ezihlukene: sihambise isibonisi kumabanga ahlukene [30, 40, 50, 60, 80, 100] cm futhi sashintsha i-engeli yokusekela ye-pMUT ([0, 20, 40] o ) Umfanekiso 2b ubonisa ukulungiswa kokutholwa kwe-ITD yesikhashana kuye ngendawo ye-angular ehambisanayo ngamadigri.
Lesi sihloko sisebenzisa amasekhethi e-RRAM amabili ahlukene angekho eshalofini.Eyokuqala iwuxhaxha lwamadivayisi angu-16,384 (16,000) (amadivaysi angu-128 × 128) ekucushweni kwe-1T1R nge-transistor eyodwa kanye ne-resistor eyodwa.I-chip yesibili iyipulatifomu ye-neuromorphic eboniswe ku-Fig. 4a.Iseli ye-RRAM iqukethe ifilimu ye-HfO2 engu-5 nm obukhulu eshumekwe kusitaki se-TiN/HfO2/Ti/TiN.Isitaki se-RRAM sihlanganiswe ne-back-of-line (BEOL) yenqubo evamile ye-130nm CMOS.Amasekhethi e-neuromorphic asekelwe ku-RRAM aveza inselele yokuklama yawo wonke amasistimu kagesi we-analog lapho amadivayisi e-RRAM ahlangana khona nobuchwepheshe bendabuko be-CMOS.Ikakhulukazi, isimo sokuqhuba sedivayisi ye-RRAM kufanele sifundwe futhi sisetshenziswe njengokuguquguquka komsebenzi wesistimu.Kuze kube manje, isekethe yaklanywa, yaqanjwa futhi yahlolwa efunda amandla amanje kusukela kudivayisi lapho i-pulse yokufaka itholwa futhi isebenzisa le yamanje ukuze kukale impendulo ye-different pair integrator (DPI) synapse.Lo mjikelezo uboniswa kuMfanekiso 3a, omele amabhlogo wokwakha ayisisekelo weplatifomu ye-neuromorphic ku-Figure 4a.Ukushaya kwenhliziyo okokufaka kwenza isango ledivayisi ye-1T1R lisebenze, lenze i-current nge-RRAM ngokulingana nendlela yedivayisi engu-G (Iweight = G(Vtop – Vx)).Okokufaka okuguqulayo kwesekethe ye-amplifier (op-amp) kune-DC bias voltage Vtop engashintshi.Impendulo engalungile ye-op-amp izohlinzeka nge-Vx = Vtop ngokuhlinzeka ngokulinganayo kwamanje kusuka ku-M1.I-Iweight yamanje ebuyisiwe kusukela kudivayisi ijovwe ku-synapse ye-DPI.Amandla amanje anamandla azoholela ekulahlekeni okwengeziwe, ngakho ukuqhuba kwe-RRAM kusebenzisa izisindo ze-synaptic ngempumelelo.Lo mshini wamanje we-exponential synaptic ujovwa nge-membrane capacitor ye-Leaky Integration and Excitation (LIF) neurons, lapho ihlanganiswa khona njenge-voltage.Uma i-threshold voltage ye-membrane (i-voltage yokushintsha ye-inverter) ihlulwa, ingxenye ephumayo ye-neuron iyasebenza, ikhiqize i-spike ephumayo.Lokhu kushaya kwenhliziyo kuyabuya futhi kuvimbe i-membrane capacitor ye-neuron ukuba ishone phansi, ibangele ukuthi iphume.Lo mjikelezo ube usulekelelwa ngesandisi se-pulse (esingabonisiwe ku-Fig. 3a), esilolonga i-pulse ephumayo ye-LIF neuron ukuya kububanzi be-pulse eqondiwe.Ama-Multiplexer nawo akhelwe emugqeni ngamunye, okuvumela i-voltage ukuthi isetshenziswe kuma-electrode aphezulu naphansi wedivayisi ye-RRAM.
Ukuhlolwa kukagesi kufaka phakathi ukuhlaziya nokurekhoda ukuziphatha okuguquguqukayo kwamasekhethi e-analog, kanye nokuhlela nokufunda amadivayisi e-RRAM.Zombili izinyathelo zidinga amathuluzi akhethekile, wonke axhunywe ebhodini lezinzwa ngesikhathi esisodwa.Ukufinyelela kumadivayisi e-RRAM kumasekethe e-neuromorphic kwenziwa kusuka kumathuluzi angaphandle nge-multiplexer (MUX).I-MUX ihlukanisa iseli elingu-1T1R kuwo wonke amasekhethi eyingxenye yawo, okuvumela idivayisi ukuthi ifundwe futhi/noma ihlelwe.Ukuze uhlele futhi ufunde amadivaysi e-RRAM, umshini we-Keithley 4200 SCS usetshenziswa ngokubambisana ne-microcontroller ye-Arduino: eyokuqala yokukhiqiza ukushaya kwenhliziyo enembile nokufunda kwamanje, kanti okwesibili ukufinyelela okusheshayo kuzakhi zomuntu ngamunye ze-1T1R kuhlu lwenkumbulo.Umsebenzi wokuqala ukwakha idivayisi ye-RRAM.Amaseli akhethwa ngamunye ngamunye futhi i-voltage eqondile isetshenziswa phakathi kwama-electrode aphezulu naphansi.Kulesi simo, okwamanje kunqunyelwe ku-oda lamashumi ama-microamperes ngenxa yokunikezwa kwe-voltage yesango elihambisanayo ku-transistor yokukhetha.Iseli le-RRAM lingakwazi ukujikeleza phakathi kwe-low conductive state (LCS) kanye ne-high conductive state (HCS) lisebenzisa imisebenzi ye-RESET ne-SET, ngokulandelanayo.Ukusebenza kwe-SET kwenziwa ngokusebenzisa i-voltage engunxande enesikhathi esingu-1 μs kanye ne-voltage ephakeme engu-2.0-2.5 V ukuya ku-electrode ephezulu, kanye ne-sync pulse yomumo ofanayo ne-voltage ephakeme engu-0.9-1.3 V ukuya isango lesikhethi transistor.Lawa manani avumela ukuqondisa kwe-RRAM ngezikhawu ezingama-20-150 µs.EKUSETHWENI KABUSHA, i-1 µs ububanzi, 3 V ishayela eliphakeme isetshenziswa ku-electrode engezansi (umugqa webhithi) weseli lapho voltage yesango ikububanzi obungu-2.5-3.0 V. Okokufaka nokuphumayo kwamasekhethi e-analog amasignali ashukumisayo. .Okokufaka, sihlukanise amajeneretha amabili e-HP 8110 namajeneretha esignali e-Tektronix AFG3011.I-pulse yokufaka inobubanzi obungu-1 µs kanye nomkhawulo wokukhuphuka/wehla ongu-50 ns.Lolu hlobo lwe-pulse kucatshangwa ukuthi luyi-glitch evamile kumasekhethi asekelwe ku-analog glitch.Ngokuqondene nesignali yokuphumayo, isignali yokuphumayo yaqoshwa kusetshenziswa i-Teledyne LeCroy 1 GHz oscilloscope.Isivinini sokutholwa kwe-oscilloscope sifakazelwe ukuthi asisona isici esikhawulelayo ekuhlaziyweni nasekutholweni kwedatha yesekethe.
Ukusebenzisa amandla e-analog electronics ukulingisa ukuziphatha kwama-neurons nama-synapses kuyisixazululo esihle nesisebenzayo sokuthuthukisa ukusebenza kahle kwekhompyutha.Ububi balokhu okungaphansi kokubala ukuthi kuzohluka kuye ngohlelo.Silinganise ukuhlukahluka kwama-neurons nama-synaptic circuits (I-Supplementary Fig. 2a,b).Kukho konke ukubonakaliswa kokuhlukahluka, lokho okuhlotshaniswa nokungaguquguquki kwesikhathi kanye nenzuzo yokokufaka kunomthelela omkhulu kakhulu ezingeni lesistimu.Isikhathi esingaguquguquki se-LIF neuron kanye ne-synapse ye-DPI kunqunywa isekethe ye-RC, lapho inani lika-R lilawulwa amandla kagesi okuchema asetshenziswa esangweni le-transistor (i-Vlk ye-neuron ne-Vtau ye-synapse), inquma ukuvuza izinga.Ukuzuza kokokufaka kuchazwa ngokuthi i-voltage ephakeme kakhulu efinyelelwe yi-synaptic kanye ne-neuronal membrane capacitor evuselelwe ukushaya kwenhliziyo okokufaka.Inzuzo yokokufaka ilawulwa enye i-bias transistor elungisa okokufaka kwamanje.Ukulingisa kwe-Monte Carlo okulinganiselwe kunqubo ye-ST Microelectronics' 130nm kwenziwe ukuze kuqoqwe inzuzo ethile yokokufaka kanye nezibalo ezingaguquki zesikhathi.Imiphumela yethulwa Kumfanekiso Owengeziwe wesi-2, lapho inzuzo yokufaka nesikhathi esingaguquki kubalwa njengomsebenzi we-voltage echema elawula izinga lokuvuza.Omaka abaluhlaza balinganisa ukuchezuka okujwayelekile kokungaguquki kwesikhathi ukusuka kuncazo.Kokubili ama-neurons nama-synaptic circuits akwazi ukuveza ububanzi besikhathi esingaguquki ebangeni le-10-5-10-2 s, njengoba kuboniswe kuhlelo lwe-Supplementary Fig.Ukukhulisa okokufaka (I-Supplementary Fig. 2e,d) yokuhlukahluka kwe-neuronal kanye ne-synapse cishe bekuyi-8% ne-3%, ngokulandelana.Ukuntuleka okunjalo kubhalwe kahle ezincwadini: izilinganiso ezihlukahlukene zenziwa ohlwini lwama-chip e-DYNAP ukuze kuhlolwe ukungafani phakathi kwezibalo zama-neurons e-LIF63.Ama-synapse ku-chip yesignali exubile ye-BrainScale akalwa futhi ukungahambisani kwawo kwahlaziywa, futhi inqubo yokulinganisa yahlongozwa ukuze kwehliswe umphumela wokuhlukahluka kwezinga lesistimu64.
Umsebenzi we-RRAM kumasekethe we-neuromorphic ukabili: incazelo yezakhiwo (okufakwayo komzila kokuphumayo) kanye nokuqaliswa kwezisindo ze-synaptic.Impahla yokugcina ingasetshenziselwa ukuxazulula inkinga yokuhlukahluka kwamasekhethi enziwe nge-neuromorphic.Senze inqubo yokulinganisa elula ebandakanya ukuhlela kabusha idivayisi ye-RRAM kuze kube yilapho umjikelezo ohlaziywayo uhlangabezana nezimfuneko ezithile.Okokufaka okunikeziwe, okukhiphayo kuyagadwa futhi i-RRAM ihlelwa kabusha kuze kube kufinyelelwa ukuziphatha okuqondiwe.Isikhathi sokulinda samasekhondi angu-5 safakwa phakathi kwemisebenzi yokuhlela ukuze kuxazululwe inkinga yokuphumula kwe-RRAM okuholela ekuguquguqukeni kokuziphatha kwesikhashana (Ulwazi Olwengeziwe).Izisindo ze-Synaptic ziyalungiswa noma zilinganiswe ngokuya ngezidingo zesekethe ye-neuromorphic emodelayo.Inqubo yokulinganisa ifinyezwa kuma-algorithms engeziwe [1, 2] agxile ezicini ezimbili ezibalulekile zamapulatifomu e-neuromorphic, imigqa yokulibaziseka kanye ne-CD engazweli.Kumjikelezo onomugqa wokulibaziseka, ukuziphatha okuqondiwe ukuhlinzeka nge-pulse ephumayo ngokulibaziseka Δt.Uma ukubambezeleka kwesekethe kwangempela kungaphansi kwevelu eqondiwe, isisindo se-synaptic se-G3 kufanele sehliswe (i-G3 kufanele isethwe kabusha bese isethelwa ku-Icc yamanje efanayo).Ngokuphambene, uma ukubambezeleka kwangempela kukukhulu kunevelu eqondiwe, ukuqhuba kwe-G3 kufanele kukhushulwe (i-G3 kufanele iqale isethwe kabusha bese isethelwa kunani eliphezulu le-Icc).Le nqubo iphindaphindiwe kuze kube yilapho ukubambezeleka okukhiqizwa yisekethe kufana nenani eliqondiwe futhi ukubekezelelana kusethelwe ukumisa inqubo yokulinganisa.Kuma-CD angezwani ne-orientation, amadivayisi amabili e-RRAM, i-G1 ne-G3, ayabandakanyeka enqubweni yokulinganisa.Lo mjikelezo unokufaka okubili, i-Vin0 ne-Vin1, ibambezeleke yi-dt.Isekhethi kufanele iphendule kuphela ekubambezelekeni okungaphansi kobubanzi obufanayo [0,dtCD].Uma lingekho inani eliphakeme lokukhiphayo, kodwa inani eliphakeme lokokufaka liseduze, womabili amadivayisi e-RRAM kufanele akhuliswe ukuze kusizwe i-neuron ifinyelele embundwini.Ngokuphambene, uma isifunda siphendula ekubambezelekeni okudlula ububanzi obuhlosiwe be-dtCD, ukuqhutshwa kufanele kwehliswe.Phinda inqubo kuze kutholakale ukuziphatha okufanele.Ukuthobela kwamanje kungashintshwa yisekethe ye-analog eyakhelwe ngaphakathi ku-ref.72.73.Ngalesi sekhethi eyakhelwe ngaphakathi, izinqubo ezinjalo zingenziwa ngezikhathi ezithile ukuze kulinganiswe isistimu noma kuphinde kusetshenziswe olunye uhlelo lokusebenza.
Sihlola ukusetshenziswa kwamandla kwendlela yethu yokucubungula isignali ye-neuromorphic ku-32-bit microcontroller68 ejwayelekile.Kulokhu kuhlola, sithatha ukusebenza ngokusetha okufanayo nakuleli phepha, ngesidlulisi esisodwa se-pMUT kanye nabamukeli ababili be-pMUT.Le ndlela isebenzisa isihlungi se-bandpass, silandelwa yisinyathelo sokukhipha imvilophu (Teeger-Kaiser), futhi ekugcineni kusetshenziswa umsebenzi wokuvala isignali ukuze kukhishwe isikhathi sendiza.Ukubalwa kwe-ITD nokuguqulwa kwayo kuma-engeli okuthola kushiywe ekuhlolweni.Sicabangela ukusetshenziswa kwesihlungi se-band pass sisebenzisa isihlungi sempendulo ye-impulse ye-oda yesi-4 esidinga ukusebenza kwamaphoyinti antantayo angu-18.Ukukhipha imvilophu kusebenzisa eminye imisebenzi emithathu yamaphuzu antantayo, futhi umsebenzi wokugcina usetshenziselwa ukusetha umkhawulo.Kudingeka isamba sokusebenza samaphuzu angu-22 ukuze kucutshungulwe isiginali.Isignali edlulisiwe ingukuqhuma okufushane kwe-111.9 kHz sine waveform ekhiqizwa njalo ngo-10 ms okuholela ekumisweni kwemvamisa yokusebenza engu-100 Hz.Sisebenzise isilinganiso samasampula esingu-250 kHz ukuze sithobelane ne-Nyquist kanye newindi elingu-6 ms lesilinganiso ngasinye ukuze sithwebule ibanga eliyimitha elingu-1.Qaphela ukuthi ama-millisecond angu-6 isikhathi sokundiza sento eqhele ngemitha elingu-1.Lokhu kunikeza ukusetshenziswa kwamandla okungu-180 µW ekuguquleni kwe-A/D ku-0.5 MSPS.Ukucutshungulwa kwangaphambili kwesiginali kungu-6.60 MIPS (imiyalo ngomzuzwana), ikhiqiza u-0.75 mW.Nokho, isilawuli esincane singashintshela kumodi yamandla aphansi engu-69 lapho i-algorithm ingasebenzi.Le modi inikeza ukusetshenziswa kwamandla okumile okungu-10.8 μW kanye nesikhathi sokuvuka esingu-113 μs.Njengoba kunikezwe imvamisa yewashi engu-84 MHz, i-microcontroller iqeda yonke imisebenzi ye-algorithm ye-neuromorphic phakathi kuka-10 ms, futhi i-algorithm ibala umjikelezo wemisebenzi engu-6.3%, ngaleyo ndlela isebenzisa imodi yamandla aphansi.Umphumela wokuchithwa kwamandla ngu-244.7 μW.Qaphela ukuthi siyakushiya okukhiphayo kwe-ITD kusuka ku-ToF kanye nokuguqulelwa ku-engeli yokuthola, ngaleyo ndlela sibukeza ukusetshenziswa kwamandla kwesilawuli esincane.Lokhu kunikeza inani elengeziwe ekusebenzeni kahle kwamandla kwesistimu ehlongozwayo.Njengesimo esingeziwe sokuqhathanisa, sihlola ukusetshenziswa kwamandla kwezindlela zakudala ze-beamforming ezihlongozwayo kusithenjwa.31.54 uma ishumekwe ku-microcontroller68 ku-voltage yokunikezela engu-1.8V.Ama-membrane e-pMUT amahlanu ahlukaniswe ngokulinganayo asetshenziselwa ukuthola idatha yokwenza i-beamform.Ngokuqondene nokucubungula ngokwako, indlela yokwenza i-beamforming esetshenzisiwe iwukubambezeleka kokuhlanganisa.Imane ihlanganisa ukusebenzisa ukubambezeleka emigwaqweni ehambisana nomehluko olindelekile ezikhathini zokufika phakathi komzila owodwa nolayini wereferensi.Uma amasignali esesigabeni, isamba lalawa masiginali sizoba namandla aphezulu ngemuva kokuhamba kwesikhathi.Uma ziphumile esigabeni, ukuphazamiseka okulimazayo kuzokhawulela amandla esamba sazo.nginothize engithandananaye.Emkhiwaneni.31, isilinganiso samasampula esingu-2 MHz sikhethwa ukuze kushintshwe isikhathi idatha ngenombolo ephelele yamasampuli.Indlela enesizotha kakhulu iwukugcina isilinganiso sesampula esimaholoholo esingu-250 kHz nokusebenzisa isihlungi se-Finite Impulse Response (FIR) ukuze kuhlanganiswe ukubambezeleka kwengxenye.Sizothatha ngokuthi inkimbinkimbi ye-algorithm yokwenza i-beamforming inqunywa ngokuyinhloko ngokuhamba kwesikhathi, njengoba isiteshi ngasinye sihlanganiswa nesihlungi se-FIR esinompompi abangu-16 ohlangothini ngalunye.Ukuze ubale inombolo ye-MIPS edingekayo kulo msebenzi, sicabangela iwindi elingu-6ms ngesilinganiso ukuze sithwebule ububanzi bemitha elingu-1, iziteshi ezingu-5, izikhombisi-ndlela ezingu-11 ezikhazimulayo (ububanzi +/- 50° ngezinyathelo ezingu-10°).Izilinganiso ezingama-75 ngomzuzwana ziphushe isilawuli-micro esifinyelela ku-100 MPS.Isixhumanisi.68, okuholela ekushabalaleni kwamandla ka-11.26 mW kumandla aphelele angu-11.71 mW ngemva kokwengeza umnikelo we-ADC ogibele.
Idatha esekela imiphumela yalolu cwaningo iyatholakala kumbhali ofanele, i-FM, uma kunesicelo esifanele.
I-Indiveri, G. & Sandamirskaya, Y. Ukubaluleka kwesikhala nesikhathi sokucubungula isignali kuma-ejenti we-neuromorphic: Inselele yokuthuthukisa amandla aphansi, ama-agent azimele asebenzisana nemvelo. I-Indiveri, G. & Sandamirskaya, Y. Ukubaluleka kwesikhala nesikhathi sokucubungula isignali kuma-ejenti we-neuromorphic: Inselele yokuthuthukisa amandla aphansi, ama-agent azimele asebenzisana nemvelo.I-Indiveri G. kanye ne-Sandamirskaya Y. Ukubaluleka kwesikhala nesikhathi sokucubungula isignali kuma-agent we-neuromorphic: inselele yokuthuthukisa ama-agent anamandla aphansi azimele asebenzisana nemvelo. I-Indiveri, G. & Sandamirskaya, Y. I-Indiveri, G. & Sandamirskaya, Y.I-Indiveri G. kanye ne-Sandamirskaya Y. Ukubaluleka kwesikhala nesikhathi sokucubungula isignali kuma-agent we-neuromorphic: inselele yokuthuthukisa ama-agent anamandla aphansi azimele asebenzisana nemvelo.IEEE Isignali Iyacubungula.Ijenali 36, 16–28 (2019).
I-Thorpe, SJ Peak Isikhathi Sokufika: Uhlelo Olusebenzayo Lwekhodi Yenethiwekhi Ye-Neural. e-Eckmiller, R., Hartmann, G. & Hauske, G. (eds). e-Eckmiller, R., Hartmann, G. & Hauske, G. (eds).e-Eckmiller, R., Hartmann, G. and Hauske, G. (eds.).Ku-Eckmiller, R., Hartmann, G., and Hauske, G. (eds.).Ukucubungula okufanayo kumasistimu we-neural namakhompyutha 91-94 (North-Holland Elsevier, 1990).
I-Levy, i-WB & Calvert, i-VG Communication idla amandla aphindwe izikhathi ezingu-35 kune-computing ku-cortex yomuntu, kodwa zombili izindleko ziyadingeka ukubikezela inombolo ye-synapse. I-Levy, i-WB & Calvert, i-VG Communication idla amandla aphindwe izikhathi ezingu-35 kune-computing ku-cortex yomuntu, kodwa zombili izindleko ziyadingeka ukubikezela inombolo ye-synapse.I-Levy, i-WB ne-Calvert, i-WG Communication idla amandla aphindwe izikhathi ezingu-35 kune-computing ku-cortex yomuntu, kodwa zombili izindleko ziyadingeka ukubikezela inani lama-synapses. Levy, WB & Calvert, VG Communication 消耗的能量是人类皮层计算的35 倍，但這两种成本都需要预测突触数量. Levy, WB & Calvert, VG CommunicationI-Levy, i-WB ne-Calvert, i-WG Communication idla amandla aphindwe izikhathi ezingu-35 kune-computing ku-cortex yomuntu, kodwa zombili izindleko zidinga ukubikezela inani lama-synapses.inqubo.I-National Academy of Science.isayensi.I-US 118, https://doi.org/10.1073/pnas.2008173118 (2021).
I-Dalgaty, T., Vianello, E., De Salvo, B. & Casas, J. I-insect-inspired neuromorphic computing. I-Dalgaty, T., Vianello, E., De Salvo, B. & Casas, J. I-insect-inspired neuromorphic computing.I-Dalgati, T., Vianello, E., DeSalvo, B. kanye ne-Casas, i-J. I-insect-inspired neuromorphic computing.Dalgati T., Vianello E., DeSalvo B. kanye ne-Casas J. I-insect-inspired neuromorphic computing.Okwamanje.Umbono.Isayensi yezinambuzane.30, 59–66 (2018).
Roy, K., Jaiswal, A. & Panda, P. Towards spike-based machine intelligence ene-neuromorphic computing. Roy, K., Jaiswal, A. & Panda, P. Towards spike-based machine intelligence ene-neuromorphic computing. Roy, K., Jaiswal, A. & Panda, P. Towards Spike-based Machine Intelligence ene-Neuromorphic Computing.U-Roy K, u-Jaiswal A, kanye no-Panda P. Ubuhlakani bokwenziwa obususelwa kuPulse kusetshenziswa i-neuromorphic computing.Imvelo 575, 607–617 (2019).
I-Indiveri, G. & Liu, S.-C. I-Indiveri, G. & Liu, S.-C.I-Indiveri, G. kanye no-Liu, S.-K. I-Indiveri, G. & Liu, S.-C. I-Indiveri, G. & Liu, S.-C.I-Indiveri, G. kanye no-Liu, S.-K.Ukucubungula inkumbulo nolwazi kumasistimu we-neuromorphic.inqubo.IEEE 103, 1379–1397 (2015).
U-Akopyan F. et al.I-Truenorth: Idizayini kanye nekhithi yamathuluzi ye-65 mW 1 million neuron ehlelekayo ye-synaptic chip.Ukushintsha kwe-IEEE.Ukwakhiwa kwekhompyutha yezinhlelo zesekethe ezididiyelwe.34, 1537–1557 (2015).
Schemmel, J. et al.Idemo ebukhoma: inguqulo eyehlisiwe yesistimu ye-BrainScaleS neuromorphic esikalini sepuleti.2012 IEEE International Symposium on Circuits and Systems (ISCAS), (IEEE ed.) 702–702 (2012).
I-Moradi, S., Qiao, N., Stefanini, F. & Indiveri, G. I-architecture ye-multicore scalable enezakhiwo zememori ezihlukile zama-dynamic neuromorphic asynchronous processors (DYNAPs). I-Moradi, S., Qiao, N., Stefanini, F. & Indiveri, G. I-architecture ye-multicore scalable enezakhiwo zememori ezihlukile zama-dynamic neuromorphic asynchronous processors (DYNAPs).I-Moradi S., Qiao N., Stefanini F. kanye ne-Indiviri G. I-architecture ye-multicore ekalakayo enezakhiwo zememori ezihlukile zama-dynamic neuromorphic asynchronous processors (DYNAP). Moradi, S., Qiao, N., Stefanini, F. & Indiveri, G. 一种可扩展的多核架构，具有用于动态神经形态异步处理器(DYNAP) I-Moradi, S., Qiao, N., iStefanini, F. & Indiveri, G. Uhlobo olunwebekayo lwezakhiwo eziningi eziyisisekelo, ezinesakhiwo sememori esiyingqayizivele sokucubungula kwe-neural (DYNAP).I-Moradi S., Qiao N., Stefanini F. kanye ne-Indiviri G. I-architecture ye-multicore ekalakayo enezakhiwo zememori ezihlukile zama-dynamic neuromorphic asynchronous processors (DYNAP).IEEE Transactions on Biomedical science.uhlelo lukagesi.12, 106–122 (2018).
UDavis, M. et al.I-Loihi: I-neuromorphic multi-core processor enokufunda okushunyekiwe.IEEE Micro 38, 82–99 (2018).
I-Furber, SB, Galluppi, F., Temple, S. & Plana, LA Iphrojekthi ye-SpiNNaker. I-Furber, SB, Galluppi, F., Temple, S. & Plana, LA Iphrojekthi ye-SpiNNaker.Ferber SB, Galluppi F., Temple S. kanye nephrojekthi ye-Plana LA SpiNNaker.Ferber SB, Galluppi F., Temple S. kanye nephrojekthi ye-Plana LA SpiNNaker.inqubo.IEEE 102, 652–665 (2014).
U-Liu, S.-K. & Delbruck, T. Neuromorphic sensory systems. & Delbruck, T. Neuromorphic sensory systems.kanye ne-Delbrück T. Neuromorphic sensory systems. & Delbruck, T. 神经形态感觉系统. & Delbruck, T.kanye ne-Delbrück T. Neuromorphic sensory system.Okwamanje.Umbono.I-Neurobiology.20, 288–295 (2010).
Chope, T. et al.Ukuhlanganiswa kwezinzwa ze-Neuromorphic zokwenziwa kwasendaweni komthombo womsindo ohlanganisiwe nokugwema ukungqubuzana.Ngo-2019 eNgqungqutheleni ye-IEEE ye-Biomedical Circuits and Systems (BioCAS), (IEEE Ed.) 1–4 (2019).
Risi, N., Aimar, A., Donati, E., Solinas, S. & Indiveri, G. A spike-based neuromorphic architecture of stereo vision. Risi, N., Aimar, A., Donati, E., Solinas, S. & Indiveri, G. A spike-based neuromorphic architecture of stereo vision.Risi N, Aymar A, Donati E, Solinas S, kanye ne-Indiveri G. I-spike-based neuromorphic stereovision architecture. Risi, N., Aimar, A., Donati, E., Solinas, S. & Indiveri, G. 一种基于脉冲的立体视觉神经形态结构。 Risi, N., Aimar, A., Donati, E., Solinas, S. & Indiveri, G.Risi N, Aimar A, Donati E, Solinas S, kanye ne-Indiveri G. Spike-based neuromorphic architecture yombono we-stereo.ngaphambili.I-Neurorobotics 14, 93 (2020).
Osswald, M., Ieng, S.-H., Benosman, R. & Indiveri, G. Imodeli yenethiwekhi ye-spiking neural ye-3Dperception yezinhlelo zombono we-neuromorphic stereo. Osswald, M., Ieng, S.-H., Benosman, R. & Indiveri, G. Imodeli yenethiwekhi ye-spiking neural ye-3Dperception yezinhlelo zombono we-neuromorphic stereo.Oswald, M., Ieng, S.-H., Benosman, R., kanye ne-Indiveri, G. A 3D Pulsed Neural Network Perception Model for Event-Based Neuromorphic Stereo Vision Systems. Osswald, M., Ieng, S.-H., Benosman, R. & Indiveri, G. 基于事件的神经形态立体视觉系统的3Dperception 脉冲神经网统模。 Osswald, M., Ieng, S.-H., Benosman, R. & Indiveri, G. 3Dperception 脉冲神经网络模型.U-Oswald, M., Ieng, S.-H., Benosman, R., kanye no-Indiveri, G. Spiked 3Dperception Neural Network Model ye-Event-based Neuromorphic Stereo Vision System.isayensi.Umbiko 7, 1–11 (2017).
Dalgaty, T. et al.Ukutholwa kokunyakaza okuyisisekelo okugqugquzelwe yizinambuzane kubandakanya inkumbulo ephikisayo kanye namanethiwekhi emizwa aqhumayo.Isistimu ye-Bionic biohybrid.10928, 115–128 (2018).
D'Angelo, G. et al.Ukutholwa kokunyakaza okusekelwe kumcimbi kusetshenziswa ikhodi yokuhlukanisa yesikhashana.ngaphambili.I-Neurology.14, 451 (2020).