News

Real-world data processing applications inoda compact, low-latency, low-power computing systems. Iine chiitiko-inotyairwa komputa kugona, inopindirana simbi-oxide-semiconductor hybrid memristive neuromorphic architecture inopa yakanakira hardware hwaro hwemabasa akadaro. Kuti tiratidze kugona kwakazara kweakadaro masisitimu, isu tinokurudzira uye nekuyedza kuratidza yakazara sensor yekugadzirisa mhinduro kune chaiyo-yepasirese chinhu mashandisirwo enzvimbo. Kudhirowa kufemerwa kubva kudura owl neuroanatomy, isu takagadzira bioinspired, inofambiswa nechiitiko chinhu localization system inosanganisa mamiriro-e-the-art piezoelectric micromechanical transducer transducer ine computational graph-based neuromorphic resistive memory. Isu tinoratidza zviyero zveyakagadzirwa sisitimu inosanganisira memory-based resistive coincidence detector, kunonoka mutsara wedunhu, uye inogoneka zvizere ultrasonic transducer. We use these experimental results to calibrate simulations at the system level. These simulations are then used to evaluate the angular resolution and energy efficiency of the object localization model. Mhedzisiro yacho inoratidza kuti maitiro edu anogona kuve akati wandei maodha ehukuru ane simba rakawanda pane mamicrocontrollers anoita basa rimwe chete.
Tiri kupinda munguva yeubiquitous computing uko huwandu hwemidziyo uye masisitimu akaiswa ari kukura zvakanyanya kutibatsira muhupenyu hwedu hwezuva nezuva. Aya masisitimu anotarisirwa kuti arambe achienderera mberi, achipedza simba shoma sezvinobvira achidzidzira kududzira data yavanounganidza kubva kune akawanda masensa munguva chaiyo uye kugadzira mabhinari akabuda semugumisiro wekuronga kana kuzivikanwa mabasa. Imwe yematanho akanyanya kukosha anodiwa kuti azadzise chinangwa ichi kutora ruzivo runobatsira uye rwakaomesesa kubva kune ruzha uye kazhinji rusina kukwana ruzivo rwe data. Injiniya yechinyakare inosvika kazhinji sampuli masaini masaini painogaro uye yakakwirira mwero, ichigadzira yakakura data kunyangwe pakashaikwa zvekushandisa zvinobatsira. Pamusoro pezvo, nzira idzi dzinoshandisa maitiro akaomarara edhijitari ekugadzirisa masaini kufanogadzirisa iyo (kazhinji ine ruzha) yekuisa data. Pane kudaro, biology inopa dzimwe mhinduro dzekugadzirisa zvine ruzha data data uchishandisa simba-rinoshanda, asynchronous, inofambiswa nezviitiko (spikes)2,3. Neuromorphic computing inotora kufemerwa kubva kubiological systems kuderedza computational cost maererano nesimba uye ndangariro zvinodiwa zvichienzaniswa nemaitiro echinyakare masignal process nzira4,5,6. Munguva pfupi yapfuura, hunyanzvi hwese chinangwa chehuropi-based masisitimu anoshandisa impulse neural network (TrueNorth7, BrainScaleS8, DYNAP-SE9, Loihi10, Spinnaker11) yakaratidzwa. Aya ma processor anopa yakaderera simba, yakaderera latency mhinduro dzemuchina kudzidza uye cortical wedunhu modelling. Kuti vashandise zvizere simba ravo resimba, aya neuromorphic processors anofanirwa kuve akabatana zvakananga kune chiitiko-inotyairwa sensors12,13. Nekudaro, nhasi kune mashoma ekubata maturusi anopa zvakananga data inofambiswa nechiitiko. Mienzaniso yakatanhamara ndeye dynamic visual sensors (DVS) yekuona maapplications senge tracking uye kufamba kwekuona14,15,16,17 the silicon cochlea18 uye neuromorphic auditory sensors (NAS)19 yeauditory sign process, olfactory sensors20 uye yakawanda mienzaniso21,22 yekubata. . texture sensors.
Mubepa rino, tinopa ichangoburwa chiitiko-inofambiswa neadhidhiyo system yekumisikidza inoshandiswa kune chinhu chinowanikwa munharaunda. Pano, kekutanga, tinotsanangura magumo-kusvika-kumagumo sisitimu yechinhu chinowanikwa munzvimbo inowanikwa nekubatanidza nyika-ye-the-art piezoelectric micromachined ultrasonic transducer (pMUT) ine computational graph yakavakirwa neuromorphic resistive memory (RRAM). In-memory computing architectures uchishandisa RRAM mhinduro inovimbisa yekudzikisa simba rekushandisa23,24,25,26,27,28,29. Kusagadzikana kwavo kwechisikigo-kusingade kushandiswa kwesimba rinoshanda kuchengetedza kana kugadzirisa ruzivo-yakanyatsoenderana neasynchronous, inofambiswa nechiitiko chimiro cheneuromorphic computing, zvichikonzera pedyo-pasina simba rekushandisa kana sisitimu isingaite. Piezoelectric micromachined ultrasonic transducers (pMUTs) isingadhuri, miniaturized silicon-based ultrasonic transducers inokwanisa kuita sevanotumira uye vanogamuchira30,31,32,33,34. Kuti tigadzirise masaini anogamuchirwa neakavakirwa-mukati masensa, isu takakwevera kurudziro kubva mudura rezizi neuroanatomy35,36,37. Zizi remudura reTyto alba rinozivikanwa nehunyanzvi hwaro hwekuvhima husiku nekuda kweiyo inoshanda kwazvo yekuongorora nzvimbo system. Kuti uverenge nzvimbo yemhuka, dura rezizi renzvimbo inoisa nguva yekubhururuka (ToF) apo masaisai eruzha kubva kune zvakapambwa anosvika kune imwe neimwe nzeve dzezizi kana manzwi ekugamuchira. Tichifunga chinhambwe chiri pakati penzeve, mutsauko uripo pakati pezviyero zviviri zveToF (Interaural Time Difference, ITD) inoita kuti zvikwanise kuverenga nekuongorora azimuth chinzvimbo chechinangwa. Kunyangwe masisitimu ebhayoloji asina kunyatsokodzera kugadzirisa algebraic equations, anogona kugadzirisa matambudziko enzvimbo nemazvo. Zizi rezizi retsinga rinoshandisa seti ye coincidence detector (CD) 35 neurons (kureva, neuroni dzinokwanisa kuona kuwirirana kwechinguvana pakati pezvipikisi zvinopararira zvichidzika kusvika kumagumo ekugadzirisa zvinofadza)
Tsvagiridzo yapfuura yakaratidza kuti complementary metal-oxide-semiconductor (CMOS) hardware uye RRAM-based neuromorphic hardware yakafuridzirwa neiyo yakaderera colliculus ("auditory cortex") yezizi remudura inzira inoshanda yekuverenga chinzvimbo uchishandisa ITD13, 40, 41, 42, 43 , 44, 45, 46. However, the potential of complete neuromorphic systems that link auditory cues to neuromorphic computational graphs has yet to be demonstrated. Dambudziko guru nderekusiyana kwakasiyana kweanalog CMOS maseketi, ayo anokanganisa kurongeka kwekuonekwa kwemutambo. Munguva pfupi yapfuura, kumwe kuita kwenhamba kweiyo ITD47 fungidziro kwakaratidzwa. Mubepa rino, isu tinokurudzira kushandisa kugona kweRRAM kushandura kukosha kwemaitiro nenzira isiri-inotenderera kurwisa kusiyanisa mumaseketi eanalog. Isu takaita hurongwa hwekuyedza hunosanganisira imwe pMUT yekutapurirana membrane inoshanda pafrequency ye111.9 kHz, maviri pMUT inogamuchira membrane (sensors) inotevedzera nzeve dzezizi, uye imwe. We experimentally characterized the pMUT detection system and RRAM-based ITD computational graph to test our localization system and evaluate its angular resolution.
Isu tinofananidza nzira yedu nedhijitari yekumisikidza pane microcontroller ichiita yakafanana yenzvimbo basa uchishandisa yakajairwa beamforming kana neuromorphic nzira, pamwe nemunda programmable gedhi array (FPGA) yeITD fungidziro inotsanangurwa mureferensi. 47. Kuenzanisa uku kunoratidza kukwikwidzana kwemasimba ekugadzirisa kwakarongwa RRAM-based analog neuromorphic system.

the barn owl receives sound waves from a target, in this case moving prey. The time of flight (ToF) of the sound wave is different for each ear (unless the prey is directly in front of the owl). Mutsetse une madota unoratidza nzira iyo masaisai eruzha kuti asvike kunzeve dzezizi remudura. Prey can be accurately localized in the horizontal plane based on the length difference between the two acoustic paths and the corresponding interaural time difference (ITD) (left image inspired by ref. 74, copyright 2002, Society for Neuroscience). Mune yedu sisitimu, iyo pMUT transmitter (yakasviba bhuruu) inogadzira masaisai eruzha anobvongodza kubva pachinangwa. Reflected ultrasound mafungu anogamuchirwa nevaviri pMUT vanogamuchira (chiedza chegirini) uye chinogadziriswa neuromorphic processor (kurudyi). b An ITD (Jeffress) computational modhi inotsanangura manzwi anopinda munzeve dzezizi remudura anotanga kukodherwa semapikisi akakiyiwa munucleus hombe (NM) uyezve kushandisa geometrically yakarongwa gidhi yeanoenderana detector neurons mu lamellar nucleus. Kugadzira (Netherlands) (kuruboshwe). Illustration of a neuroITD computational graph combining delay lines and coincidence detector neurons, the owl biosensor system can be modeled using RRAM-based neuromorphic circuits (right). c Schematic of the main Jeffress mechanism, due to the difference in ToF, the two ears receive sound stimuli at different times and send axons from both ends to the detector. Iwo maaxoni chikamu chenhevedzano yema coincidence detector (CD) neurons, imwe neimwe inopindura inosarudza kune yakasimba-yakabatana-yakabatana mapindiro. Nekuda kweizvozvo, maCD chete ayo anopinza anosvika aine diki nguva musiyano anonakidzwa zvakanyanya (ITD inobhadharwa chaizvo). Iyo CD inozoisa encode yechinangwa cheangular chinzvimbo.

mufananidzo wekristaro yepMUT ine nhanhatu 880 µm membranes dzakabatanidzwa pa 1.5 mm pitch. b Dhiagiramu yekuyera kuseta. Chinangwa chiri panzvimbo yeazimuth θ uye nechekure D. Mutakuri we pMUT anogadzira chiratidzo che 117.6 kHz chinobvondoka kubva pachinangwa chosvika vaviri vanogamuchira pMUT vane nguva yakasiyana-yekubhururuka (ToF). Musiyano uyu, unotsanangurwa semusiyano wenguva yepakati-aural (ITD), inokodha chinzvimbo chechinhu uye inogona kufungidzirwa nekufungidzira mhinduro yepamusoro yemasensa maviri anogamuchira. c Schematic yepre-processing nhanho dzekushandura pMUT yakasvibira siginecha kuita spike sequences (kureva kupinza kune neuromorphic computation graph). Iyo pMUT sensors uye neuromorphic computational magirafu akagadzirwa uye akaedzwa, uye neuromorphic pre-processing yakavakirwa pasoftware simulation. d Mhinduro yepMUT membrane payakagamuchira chiratidzo uye shanduko yayo kuita spike domain. e Ongororo yenzvimbo yeangular kurongeka sebasa rechinhu kona (Θ) uye chinhambwe (D) kune chakanangana nechinhu. Iyo ITD yekubvisa nzira inoda kushoma angular kugadziriswa kweinenge 4°C. f Kurongeka kweAngular (mutsara webhuruu) uye inoenderana peak-to-ruzha reshiyo (yegirini mutsetse) maringe nechinhambwe chechinhu che Θ = 0.
Resistive memory inochengetedza ruzivo mune isingaite-volatile conductive state. Nheyo yakakosha yeiyo nzira ndeyokuti kushandurwa kwezvinhu padanho reatomu kunokonzera shanduko mune yayo magetsi conductivity57. Here we use an oxide-based resistive memory consisting of a 5nm layer of hafnium dioxide sandwiched between top and bottom titanium and titanium nitride electrodes. Iko conductivity yeRRAM madivayiri inogona kuchinjwa nekushandisa yazvino/voltage waveform iyo inogadzira kana kutyora conductive filaments yeokisijeni vacancies pakati pemagetsi. Isu takabatanidza midziyo yakadaro58 mune yakajairwa 130 nm CMOS maitiro kuti tigadzire yakagadzirwazve inogadziriswa neuromorphic reketi ichishandisa coincidence detector uye kunonoka mutsara wedunhu (Fig. 3a). Iyo isiri-inotenderera uye analog hunhu hwechishandiso, yakasanganiswa nechiitiko-inofambiswa chimiro cheiyo neuromorphic circuit, inoderedza kushandiswa kwesimba. Dunhu rine pakarepo on / off basa: rinoshanda pakarepo mushure mekushandura, richibvumira simba kuti ridzimwe zvachose kana dunhu risina basa. Zvivako zvikuru zvekuvaka zvechirongwa chakarongwa zvinoratidzwa mufig. 3b. Inoumbwa neN parallel single-resistor single-transistor (1T1R) zvimiro zvinoisa masinaptic huremu kubva painotorwa huremu hwemhepo, jekiseni mune yakajairwa synapse yeanosiyanisa pair integrator (DPI) 59, uye pakupedzisira inobaiwa mu synapse nekubatanidza uye. leakage. activated (LIF) neuron 60 (ona Nzira dzeruzivo). Iwo mabheji ekupinza anoiswa kugedhi reiyo 1T1R chimiro muchimiro chekutevedzana kwevoltage pulses nehuremu pahurongwa hwemazana ema nanoseconds. Resistive memory inogona kuiswa munzvimbo yakakwirira conductive state (HCS) nekushandisa yekunze yakanaka referensi kuVtop kana Vbottom yadzika, uye kudzoreredzwa kune yakaderera conductive state (LCS) nekuisa yakanaka voltage kuVbottom kana Vtop yadzika. Ivhareji kukosha kweHCS inogona kudzorwa nekudzikamisa purogiramu yezvino (kuteerera) yeSET (ICC) negedhi-source voltage ye transistor series (Fig. 3c). Mabasa eRRAM mudunhu akapetwa kaviri: anotungamira uye anoyera mapurusi ekuisa.
Kuongorora maelectronic microscope (SEM) mufananidzo weblue HfO2 1T1R RRAM mudziyo wakabatanidzwa mu130 nm CMOS tekinoroji ine selector transistors (650 nm paupamhi) mune yegirini. b Basic building blocks of the proposed neuromorphic schema. The input voltage pulses (peaks) Vin0 and Vin1 consume current Iweight, which is proportional to the conduction states G0 and G1 of the 1T1R structure. Iyi yazvino inobaiwa muDPI synapses uye inonakidza iyo LIF neurons. RRAM G0 uye G1 yakaiswa muHCS neLCS zvakateerana. c Function of cumulative conductance density for a group of 16K RRAM devices as a function of ICC current matching, which effectively controls the conduction level. d Circuit measurements in (a) showing that G1 (in the LCS) effectively blocks input from Vin1 (green), and indeed the output neuron's membrane voltage responds only to the blue input from Vin0. RRAM inonyatso sarudza zvinongedzo mudunhu. e Measurement of the circuit in (b) showing the effect of the conductance value G0 on the membrane voltage Vmem after applying a voltage pulse Vin0. Kunyanya kuita, kunowedzera kusimba kwemhinduro: saka, iyo RRAM mudziyo inoshandisa I / O yekubatanidza huremu. Zviyero zvakaitwa padunhu uye zvinoratidza iyo mbiri basa reRRAM, nzira uye kuyerwa kwemapurusi ekuisa.
First, since there are two basic conduction states (HCS and LCS), RRAMs can block or miss input pulses when they are in the LCS or HCS states, respectively. Nekuda kweizvozvo, RRAM inonyatso sarudza zvinongedzo mudunhu. Ichi ndicho chikonzero chekukwanisa kugadzirisa zvakare zvivakwa. Kuti tiratidze izvi, tichatsanangura kugadzirwa kwedunhu rakagadzirwa redunhu redunhu muFig. 3b. Iyo RRAM inoenderana neG0 yakarongerwa muHCS, uye yechipiri RRAM G1 yakarongerwa muLCS. Input pulses are applied to both Vin0 and Vin1. The effects of two sequences of input pulses were analyzed in the output neurons by collecting the neuron membrane voltage and the output signal using an oscilloscope. Kuedza kwacho kwakabudirira apo chete mudziyo weHCS (G0) wakabatanidzwa kune pulse yeNeuron kukurudzira kushushikana kwemembrane. This is demonstrated in Figure 3d, where the blue pulse train causes the membrane voltage to build up on the membrane capacitor, while the green pulse train keeps the membrane voltage constant.
Chechipiri chakakosha basa reRRAM ndeyekuitwa kwehuremu hwekubatanidza. Uchishandisa RRAM's analog conductance gadziriso, I / O yekubatanidza inogona kuyerwa zvinoenderana. Muchiyedzo chechipiri, chishandiso cheG0 chakarongedzerwa kune akasiyana mazinga eHCS, uye iyo yekupinza pulse yakaiswa kune iyo VIn0 yekuisa. Iyo yekupinza pulse inodhonza yazvino (Iweight) kubva pachishandiso, iyo yakaenzana neyekuitisa uye inoenderana inogona kudonha Vtop - Vbot. Iyi huremu yazvino inozoiswa muDPI synapses uye LIF inobuda neurons. Iyo membrane voltage yezvinobuda neuroni yakanyorwa uchishandisa oscilloscope uye yakaratidzwa muFig. 3d. Iyo voltage peak yeneuron membrane mukupindura kune imwechete yekupinza pulse inoenderana nemafambisirwo eresitive memory, zvichiratidza kuti RRAM inogona kushandiswa sechinhu chinorongwa chesynaptic huremu. Aya maviri ekutanga bvunzo anoratidza kuti yakatsanangurwa RRAM-yakavakirwa neuromorphic chikuva inokwanisa kuita zvinhu zvekutanga zveiyo Jeffress mechanic, kureva mutsara wekunonoka uye coincidence detector circuit. Chikuva chedunhu chinovakwa nekuturika zvidhinha zvakatevedzana padivi, sezvivharo zviri muMufananidzo 3b, uye kubatanidza magedhi avo kune yakajairika yekupinza mutsara. Isu takagadzira, takagadzira, uye takaedza neuromorphic papuratifomu inosanganisira maviri akabuda neurons anogamuchira maviri ekuisa (Fig. 4a). Dhiyagiramu yedunhu inoratidzwa muMufananidzo 4b. Iyo yepamusoro 2 × 2 RRAM matrix inobvumira pulses yekupinza kuti iendeswe kune maviri anobuda neuroni, nepo yakaderera 2 × 2 matrix inobvumira kudzokororwa kwekubatana kweviri neuroni (N0, N1). Isu tinoratidza kuti iyi puratifomu inogona kushandiswa nekunonoka mutsara kugadzirisa uye maviri akasiyana coincidence detector mabasa, sezvakaratidzwa nezviyero zvekuedza muFig. 4c-e.
Circuit diagram yakaumbwa ne two output neuron N0 neN1 inogamuchira two inputs 0 uye 1. Midziyo mina yepamusoro yearray inotsanangura synaptic connections kubva pakupinza kusvika pakubuda, uye maseru mana epasi anotsanangura hukama hunodzokororwa pakati peeuroni. Iwo mavara eRRAMs anomiririra zvishandiso zvakagadzirirwa muHCS kurudyi: zvishandiso zviri muHCS zvinobvumira kubatanidza uye zvinomiririra uremu, nepo michina iri muLCS inovhara mapurusi ekuisa uye kudzima kubatana kune zvinobuda. b Dhiagiramu yedunhu (a) ine sere RRAM modules yakasimbiswa mubhuruu. c Mitsetse yekunonoka inoumbwa nekungoshandisa masinapsesi eDPI neLIF neurons. Iyo RRAM yakasvibirira inotarwa kuitisa yakakwira zvakakwana kuti ikwanise kukonzeresa glitch pane zvakabuda mushure mekupinda kunonoka Δt. d Schematic mufananidzo wekutungamira-isinganzwe CD yekuona yenguva inoenderana nezviratidzo. Output neuron 1, N1, moto pane zvinopinda 0 uye 1 nekunonoka kupfupi. e Direction sensitive CD circuit, dunhu rinoona kana kupinza 1 kunosvika pakupinza 0 uye kunosvika mushure mekuisa 0. Kubuda kwedunhu kunomiririrwa neuron 1 (N1).
Mutsetse wekunonoka (Mufananidzo 4c) unongoshandisa maitiro ane simba eDPI synapses uye LIF neurons kubereka spike yekupinda kubva kuVin1 kusvika kuVout1 nekunonoka Tdel. Iyo chete G3 RRAM yakabatana neVin1 uye Vout1 yakarongwa muHCS, mamwe ese maRRAM akarongwa muLCS. Chishandiso cheG3 chakarongerwa 92.6 µs kuti ive nechokwadi chekuti pulse yega yega inopinza inowedzera membrane voltage yeinobuda neuron zvakakwana kuti isvike pachikumbaridzo uye kugadzira kunonoka kubuda pulse. Iko kunonoka Tdel kunotarirwa neiyo synaptic uye neural time constants. Ma coincidence detectors anoona kuitika kwemasaini anopindirana kwenguva pfupi asi akagovaniswa nzvimbo. Direction-insensitive CD inotsamira pane zvega zvega zvinoshandura kune zvakajairika kubuda neuron (Mufananidzo 4d). The two RRAMs connecting Vin0 and Vin1 to Vout1, G2 and G4 respectively are programmed for high conduction. Kusvika panguva imwe chete yezvipikisi paVin0 neVin1 kunowedzera voltage yeN1 neuron membrane pamusoro pechikumbaridzo chinodiwa kugadzira inobuda spike. Kana iwo maviri ekuisa ari kure kure nenguva, kubhadharisa pa membrane voltage yakaunganidzwa neyekutanga kupinza inogona kunge iine nguva yekuora, ichidzivirira iyo membrane inogona N1 kusvika pakukosha kwechikumbaridzo. G1 neG2 zvakarongerwa angangoita 65 µs, izvo zvinoita kuti kuvhiya kamwe chete kwekuisa hakuwedzeri membrane voltage zvakakwana kuti ikonzere kubuda. Kuona masanga pakati pezviitiko zvakagoverwa muchadenga nenguva ibasa rakakosha rinoshandiswa mumhando dzakasiyana siyana dzekunzwa senge optical kuyerera kwakavakirwa zvipingamupinyi kunzvenga uye ruzha kunobva kunzvimbo. Nekudaro, komputa kutungamira-inonzwa uye isinganzwisisike maCD chinhu chakakosha chivakwa chekuvaka anoona uye odhiyo enzvimbo masisitimu. Sezvinoratidzwa nemaitiro ezvimiro zvenguva (ona Supplementary Fig. 2), dunhu rinorongwa rinoshandisa mutsara wakakodzera wemirairo ina yezviyero zvenguva yakakura. Thus, it can simultaneously meet the requirements of visual and sound systems. Directional-sensitive CD is a circuit that is sensitive to the spatial order of arrival of pulses: from right to left and vice versa. Icho chivharo chakakosha chekuvaka mune yekutanga kufamba kwekuona network yeDrosophila yekuona system, inoshandiswa kuverenga mafambiro uye kuona kudhumhana62. Kuti uwane CD-inonzwisisa CD, mapeji maviri anofanirwa kuendeswa kune maviri neuroni akasiyana (N0, N1) uye hukama hwekutungamira hunofanira kusimbiswa pakati pavo (Fig. 4e). When the first input is received, NO reacts by increasing the voltage across its membrane above the threshold value and sending out a surge. This output event, in turn, fires N1 thanks to the directional connection highlighted in green. Kana chiitiko chekupinza Vin1 ikasvika uye inopa simba N1 uku membrane yayo ichiri yakakwira, N1 inogadzira chiitiko chinobuda chinoratidza kuti mutambo wawanikwa pakati pezviviri izvi. Directional connections inobvumira N1 kuti ibudise zvinobuda chete kana 1 ikauya mushure mekuisa 0. G0, G3, uye G7 zvakarongerwa ku73.5 µS, 67.3 µS, uye 40.2 µS, zvichiteerana, kuve nechokwadi chekuti spike imwe chete pakuisa Vin0 inonoke. output spike, while N1′s membrane potential only reaches threshold when both input bursts arrive in sync. .
Kusiyana ndiko kunokonzera kusakwana mumuenzaniso neuromorphic systems63,64,65. Izvi zvinotungamira kune heterogeneous maitiro eeurons uye synapses. Mienzaniso yezvakaipira zvakadaro inosanganisira 30% (kureva kutsauswa kwakajairwa) kusiyanisa mukuwana yekupinda, nguva isingachinji, uye nguva yekuramba, kungotaura zvishoma (ona Nzira). Dambudziko iri rinotonyanya kutaurwa kana akawanda neural circuits akabatana pamwechete, senge CD inorerekera-inonzwa inoumbwa nemaneuron maviri. Kuti ushande nemazvo, kuwana nekuora nguva dzemanyuroni maviri dzinofanira kunge dzakafanana sezvinobvira. For example, a large difference in input gain can cause one neuron to overreact to an input pulse while the other neuron is barely responsive. On fig. Figure 5a shows that randomly selected neurons respond differently to the same input pulse. Iyi neural musiyano yakakosha, semuenzaniso, kune basa rekutungamira-sensitive maCD. In the scheme shown in fig. 5b, c, the input gain of neuron 1 is much higher than that of neuron 0. Thus, neuron 0 requires three input pulses (instead of 1) to reach the threshold, and neuron 1, as expected, needs two input events. Kuita spike nguva-inotsamira biomimetic plasticity (STDP) inzira inogoneka yekudzikamisa kukanganiswa kweasina kujeka uye husimbe neural uye synaptic maseketi pane system performance43. Pano isu tinokurudzira kushandisa hunhu hwepurasitiki hwekuramba ndangariro senzira yekupesvedzera kukwidziridzwa kweneural input uye kuderedza mhedzisiro yekusiyana mumaseketi neuromorphic. As shown in fig. 4e, conductance levels associated with RRAM synaptic mass effectively modulated the corresponding neural membrane voltage response. We use an iterative RRAM programming strategy. Kune yakapihwa mapindiro, maitiro ehuremu eiyo synaptic huremu anorongwazve kudzamara hunhu hwekunangwa hwedunhu hwawanikwa (ona Nzira).
chiyero chekuedza chemhinduro yezvipfumbamwe zvakasarurwa neuron dzemunhu kune imwecheteyo yekupuruzira. Mhinduro yacho inosiyana pakati pehuwandu hwevanhu, inokanganisa kuwana kwekuisa uye nguva isingachinji. b Kuyedza kuyerwa kwesimba remaneuroni pakusiyana kwemaneuron kunokanganisa kutungamira-inonzwa CD. Iwo maviri madirection-sensitive CD inobuda neuroni inopindura zvakasiyana kune yekupinza stimuli nekuda kwekusiyana kweuron-to-neuron. Neuron 0 ine yakaderera yekupinda kuwana pane neuron 1, saka zvinotora matatu ekuisa pulses (panzvimbo pe1) kugadzira inobuda spike. Sezvinotarisirwa, neuron 1 inosvika pachikumbaridzo nezviitiko zviviri zvekupinza. Kana kupinza 1 kunosvika Δt = 50 µs mushure mekunge neuron 0 moto, CD inoramba yakanyarara nokuti Δt yakakura kudarika nguva inogara neuron 1 (inenge 22 µs). c inoderedzwa ne Δt = 20 µs, kuitira kuti kupinza 1 kunowedzera apo neuron 1's kupisa ichiri yakakwirira, zvichiita kuti pave nekuonekwa panguva imwe chete yezviitiko zviviri zvekupinza.
Izvo zvinhu zviviri zvinoshandiswa muITD calculation column ndiwo mutsara wekunonoka uye inotungamira isinganzwe CD. Maseketi ese ari maviri anoda kurongeka chaiko kuti ive nechokwadi chekuisa chinhu chakanaka kuita. The delay line must deliver a precisely delayed version of the input peak (Fig. 6a), and the CD must be activated only when the input falls within the target detection range. Nokuda kwemutsara wekunonoka, zviyero zve synaptic zvekubatanidzwa kwekubatanidza (G3 muFig. 4a) zvakarongwazve kusvikira chinangwa chinononoka chawanikwa. Isa kushivirira kwakatenderedza chinangwa kunonoka kumisa chirongwa: iyo idiki kushivirira, zvakanyanya kuoma kuti ubudirire kuseta mutsara wekunonoka. On fig. Mufananidzo 6b unoratidza mhedzisiro yekunonoka mutsara wekugadzirisa maitiro: zvinogona kuonekwa kuti chirongwa chinorongwa chinogona kunyatsopa kunonoka kwese kunodiwa muchirongwa chekugadzira (kubva pa10 kusvika ku300 μs). Huwandu hwehuwandu hwekugadzirisa iterations hunokanganisa kunaka kwemaitiro ekugadzirisa: 200 iterations inogona kuderedza kukanganisa kusvika pasi pe5%. One calibration iteration corresponds to a set/reset operation of an RRAM cell. The tuning process is also critical to improving the accuracy of CD module instant close event detection. It took ten calibration iterations to achieve a true positive rate (ie, the rate of events correctly identified as relevant) above 95% (blue line in Figure 6c). However, the tuning process did not affect false positive events (that is, the frequency of events that were erroneously identified as relevant). Imwe nzira inocherechedzwa mumasisitimu ezvipenyu ekukunda zvipingamupinyi zvenguva yekukurumidza kuita nzira nzira ndeyedundancy (kureva, makopi akawanda echinhu chimwe chete anoshandiswa kuita basa rakapihwa). Tichifemerwa nebiology66, takaisa akati wandei maCD maseketi mune yega CD module pakati pemitsetse miviri yekunonoka kuderedza kukanganiswa kwenhema. Sezvinoratidzwa mufig. 6c (green line), kuisa zvinhu zvitatu zveCD muCD module yega yega zvinogona kuderedza chiyero chenhema chearamu kusvika pasi pe10-2.
Mhedzisiro yekusiyana kweuronal pakunonoka mutsara maseketi. b Delay line circuits inogona kuyerwa kusvika kunonoka kukuru nekugadzirisa nguva dzenguva dzeLIF neurons dzinoenderana uye DPI synapses kune zvakakosha. Increasing the number of iterations of the RRAM calibration procedure made it possible to significantly improve the accuracy of the target delay: 200 iterations reduced the error to less than 5%. Imwe iteration inoenderana neSET/RESET oparesheni pane RRAM cell. Imwe neimwe CD modhi mu c Jeffress modhi inogona kuitwa uchishandisa N parallel CD zvinhu kuitira kuchinjika zvakanyanya maererano nekutadza kwehurongwa. d More RRAM calibration iterations increase the true positive rate (blue line), while the false positive rate is independent of the number of iterations (green line). Kuisa mamwe maCD zvinhu zvakafanana kunodzivirira kuonekwa kwenhema kwemaCD module.
Isu iye zvino tinoongorora kushanda uye kushandiswa kwesimba kwekupedzisira-ku-kuguma kwakabatanidzwa chinhu chekugadzirisa nzvimbo inoratidzwa muMufananidzo 2 tichishandisa zviyero zveacoustic properties yepMUT sensor, CD, uye kunonoka mutsara mitsara inoumba neuromorphic computing graph. Jeffress modhi (Fig. 1a). Kana iri neuromorphic computing graph, iyo yakawanda nhamba yeCD modules, zviri nani kugadzirisa angular, asiwo yakakwirira simba regadziriro (Fig. 7a). Chibvumirano chinogona kuwanikwa nekuenzanisa kurongeka kwezvikamu zvega (pMUT sensors, neurons, uye synaptic circuits) nekururama kwehurongwa hwose. Kugadziriswa kwemutsara wekunonoka kunogumira nenguva dzenguva dzema-synapses uye neurons, iyo muchirongwa chedu inodarika 10 µs, iyo inoenderana neangular resolution ye4 ° (ona Nzira). Mamwe manodhi epamberi ane tekinoroji yeCMOS anobvumira dhizaini yeneural uye synaptic maseketi ane yakaderera nguva isingachinjiki, zvichikonzera kurongeka kwakanyanya kwezvinhu zvekunonoka mutsara. Zvisinei, muhurongwa hwedu, kururamisa kunogumira nekukanganisa pMUT pakuenzanisa nzvimbo yeangular, kureva 10 ° (blue horizontal line muFig. 7a). Takagadzirisa nhamba yeCD modules pa40, iyo inofananidzwa nekugadzirisa kweangular inenge 4 °, kureva, kutendeseka kweangular ye graph computational (light blue horizontal line muFig. 7a). Padanho rehurongwa, izvi zvinopa sarudzo ye4 ° uye iko kurongeka kwe10 ° yezvinhu zviri 50 cm pamberi pe sensor system. Kukosha uku kunofananidzwa neuromorphic sound localization masisitimu anoshumwa mune ref. 67. Kuenzanisa kwegadziriro yakarongwa nehurumende yehutano inogona kuwanikwa muSupplementary Table 1. Kuwedzera mamwe pMUTs, kuwedzera acoustic signal level, uye kuderedza ruzha rwemagetsi inzira dzinobvira dzekuwedzera kuvandudza kururamisa kwenzvimbo. ) inofungidzirwa pa9.7. nz. 55. Yakapiwa 40 CD units pane computational graph, iyo SPICE simulation inofungidzira simba pakushanda (kureva, chinhu chekuisa simba) kuva 21.6 nJ. Iyo neuromorphic system inobatidzwa chete kana chiitiko chekupinza chasvika, kureva kana acoustic wave yasvika chero pMUT inogamuchira uye yakadarika chikumbaridzo chekuona, zvikasadaro inoramba isingaite. Izvi zvinodzivirira kushandiswa kwesimba kusingakoshi kana pasina chiratidzo chekuisa. Tichifunga nezve kuwanda kwekuita kwenzvimbo ye100 Hz uye activation nguva ye300 µs pakushanda (yakanyanya inogoneka ITD), kushandiswa kwesimba kweiyo neuromorphic computing graph ndeye 61.7 nW. Neuromorphic pre-processing inoshandiswa kune yega pMUT inogamuchira, simba rekushandisa rehurongwa hwose rinosvika 81.6 nW. Kuti tinzwisise kushanda kwesimba kweiyo yakarongwa neuromorphic maitiro kana ichienzaniswa neyakajairwa Hardware, takaenzanisa nhamba iyi nesimba rinodiwa kuita basa rimwe chete pane yazvino low power microcontroller uchishandisa ingave neuromorphic kana yakajairwa beamforming68 Skill. Iyo neuromorphic nzira inotarisa analog-to-digital converter (ADC) nhanho, inoteverwa nebhendi-pass sefa uye envelopu yekubvisa nhanho (Teeger-Kaiser nzira). Pakupedzisira, chikumbaridzo oparesheni inoitwa kubvisa iyo ToF. Isu takasiya kuverenga kweITD kwakavakirwa paToF uye kutendeuka kune inofungidzirwa nzvimbo yemakona sezvo izvi zvichiitika kamwe pachiyero chega chega (ona Nzira). Tichifunga sampling rate ye250 kHz pamatanho ese ari maviri (pMUT vanogamuchira), gumi nemasere bhendi pass sefa mashandiro, matatu ekubvisa hamvuropu mashandiro, uye 1 pachikumbaridzo oparesheni pamuenzaniso, simba rose rekushandisa rinofungidzirwa pa245 microwatts. Izvi zvinoshandisa iyo microcontroller's low-power mode69, iyo inobatidza kana algorithms isiri kuita, iyo inoderedza kushandiswa kwesimba kusvika 10.8 µW. Iko kushandiswa kwesimba kwebeamforming chiratidzo chekugadzirisa mhinduro yakakurudzirwa mureferensi. 31, ine 5 pMUT vanogamuchira uye 11 matanda zvakafanana akaparadzirwa mu azimuth ndege [-50 °, + 50 °], ari 11.71 mW (ona Methods chikamu ruzivo). Pamusoro pezvo, isu tinoshuma mashandisirwo emagetsi eFPGA47-based Time Difference Encoder (TDE) inofungidzirwa pa 1.5 mW sekutsiva iyo Jeffress modhi yenzvimbo yechinhu. Based on these estimates, the proposed neuromorphic approach reduces power consumption by five orders of magnitude compared to a microcontroller using classical beamforming techniques for object localization operations. Kutora nzira yeuromorphic yekusaina kusaina pane yakasarudzika microcontroller inoderedza mashandisiro emagetsi neanosvika maviri maodha ehukuru. The effectiveness of the proposed system can be explained by the combination of an asynchronous resistive-memory analog circuit capable of performing in-memory calculations and the lack of analog-to-digital conversion required to perceive signals.
Angular resolution (blue) uye simba rekushandisa (girini) yekushanda kwenzvimbo zvichienderana nehuwandu hwemaCD modules. Iyo yakasviba yebhuruu yakachinjika bhawa inomiririra kurongeka kweangular kwePMUT uye bhawa rebhuruu rakadzikama rinomiririra kurongeka kwemakona kweiyo neuromorphic computational graph. b Kushandiswa kwesimba kweiyo yakarongwa sisitimu uye kuenzanisa neaviri akakurukurwa ma microcontroller kuita uye kuita kwedhijitari kweTime Difference Encoder (TDE) 47 FPGA.
Kuti tideredze mashandisiro emagetsi echinangwa chenzvimbo yenzvimbo, takabata, takagadzira uye takaita inoshanda, inofambiswa nechiitiko RRAM-based neuromorphic circuit iyo inogadzirisa ruzivo rwechiratidzo chinogadzirwa neakavakirwa-mukati masensa kuti averenge nzvimbo yechinhu chakanangwa muchokwadi. nguva. . Nepo nzira dzechinyakare dzekugadzirisa dzichiramba dzichiita sampuli dzakaonekwa uye kuita masvomhu kuti ubvise ruzivo runobatsira, yakarongwa neuromorphic mhinduro inoita maverengero asynchronously sezvo ruzivo runobatsira runosvika, kuwedzera simba rehurongwa hwemagetsi nemaodha mashanu ehukuru. Mukuwedzera, isu tinosimbisa kushanduka kweRRAM-based neuromorphic circuits. Kugona kweRRAM kushandura mafambiro nenzira isingaite (plasticity) inotsiva kusiyanisa kwemukati kweyekupedzisira-yakaderera simba analog DPI's synaptic uye neural circuits. Izvi zvinoita kuti iyi RRAM-yakavakirwa dunhu riite zvakasiyana uye rine simba. Chinangwa chedu hachisi chekubvisa mabasa akaomarara kana mapatani kubva kumasaini, asi kuisa zvinhu munzvimbo munguva chaiyo. Yedu sisitimu inogona zvakare kunyatso kudzvanya chiratidzo uye pakupedzisira kuitumira kune mamwe matanho ekugadzirisa kuita sarudzo dzakaoma pazvinenge zvichidikanwa. Muchirevo chekushandiswa kwenzvimbo, yedu neuromorphic preprocessing nhanho inogona kupa ruzivo nezve nzvimbo yezvinhu. Ruzivo urwu runogona kushandiswa, semuenzaniso, pakuona mafambiro kana kuzivikanwa kwechimiro. Isu tinosimbisa kukosha kwekubatanidza ultra low power sensors senge pMUTs neultra low power electronics. Kune izvi, neuromorphic nzira dzave dzakakosha sezvo dzakatitungamira kugadzira kutsva kwedunhu maitirwo eiyo biologically yakafuridzirwa computational nzira senge Jeffress modhi. Muchirevo che sensor fusion application, sisitimu yedu inogona kusanganiswa neyakasiyana-siyana-based sensors kuti iwane ruzivo rwakanyanya. Kunyangwe mazizi ari shasha pakutsvaga mhuka murima, ane maziso akanaka kwazvo uye anoita mubatanidzwa wekunzwa uye kutsvaga kwekuona asati abata mhuka70. Kana imwe neuron inonzwika ikapfuta, zizi rinogashira ruzivo rwarinoda kuti rione kuti rinotangira kupi kutsvaga kwekuona, nokudaro richinangisa pfungwa dzaro pachikamu chidiki chenzvimbo inooneka. Mubatanidzwa wekuona ma sensors (DVS kamera) uye inokurudzirwa yekuteerera sensor (yakavakirwa paPMUT) inofanirwa kuongororwa kuitira kuvandudzwa kweamangwana vanozvimirira vamiririri.
Iyo pMUT sensor iri paPCB ine maviri anogashira anenge 10 cm kureba, uye transmitter iri pakati pevanogamuchira. In this work, each membrane is a suspended bimorph structure consisting of two layers of piezoelectric aluminum nitride (AlN) 800 nm thick sandwiched between three layers of molybdenum (Mo) 200 nm thick and coated with a layer 200 nm thick. iyo yepamusoro passivating SiN layer sezvinotsanangurwa mureferensi. 71. Ma electrode emukati nekunze anoshandiswa kune pasi uye kumusoro kwepamusoro pe molybdenum, nepo pakati molybdenum electrode isingaenzaniswi uye inoshandiswa sepasi, zvichiita kuti membrane ine mapeji mana e electrode.
Ichi chivakwa chinobvumira kushandiswa kweyakajairwa membrane deformation, zvichikonzera kuvandudzwa kwekufambisa uye kugamuchira kunzwa. PMUT yakadaro inowanzo ratidza kunakidzwa kwekunzwa kwe700 nm/V seemitter, ichipa kumanikidza kwepamusoro kwe270 Pa/V. As a receiver, one pMUT film exhibits a short circuit sensitivity of 15 nA/Pa, which is directly related to the piezoelectric coefficient of AlN. The technical variability of the voltage in the AlN layer leads to a change in the resonant frequency, which can be compensated by applying a DC bias to the pMUT. DC senitivity yakayerwa pa 0.5 kHz/V. Kune acoustic characterization, maikorofoni inoshandiswa pamberi pePMUT.
To measure the echo pulse, we placed a rectangular plate with an area of ​​about 50 cm2 in front of the pMUT to reflect the emitted sound waves. Zvose zviri zviviri chinhambwe pakati pemahwendefa uye kona inoenderana nendege yepMUT inodzorwa nekushandisa zvakakosha vabati. A Tectronix CPX400DP voltage sosi inorerekera matatu pMUT membranes, kugadzirisa resonant frequency kusvika 111.9 kHz31, nepo ma transmitters achifambiswa neTectronix AFG 3102 pulse jenareta inotaridzwa kune resonant frequency (111.9 kHz) uye 0.0 basa kutenderera. The currents read from the four output ports of each pMUT receiver are converted to voltages using a special differential current and voltage architecture, and the resulting signals are digitized by the Spektrum data acquisition system. The limit of detection was characterized by pMUT signal acquisition under different conditions: we moved the reflector to different distances [30, 40, 50, 60, 80, 100] cm and changed the pMUT support angle ([0, 20, 40] o ) Mufananidzo 2b unoratidza yenguva ITD yekuona kugadzirisa zvichienderana nenzvimbo inowirirana yeakona mumadhigirii.



The function of RRAM in neuromorphic circuits is twofold: architecture definition (routing inputs to outputs) and implementation of synaptic weights. Iyo yekupedzisira pfuma inogona kushandiswa kugadzirisa dambudziko rekusiyana kweiyo modeled neuromorphic circuits. We have developed a simple calibration procedure that involves reprogramming the RRAM device until the circuit being analyzed meets certain requirements. Kune yakapihwa yekupinza, iyo inobuda inotariswa uye iyo RRAM inorongwazve kudzamara hunhu hwechinangwa hwawanikwa. A wait time of 5 s was introduced between programming operations to solve the problem of RRAM relaxation resulting in transient conductance fluctuations (Supplementary Information). Synaptic huremu hunogadziriswa kana kuenzaniswa maererano nezvinodiwa neuromorphic circuit iri kuenzanisirwa. The calibration procedure is summarized in additional algorithms [1, 2] that focus on two fundamental features of neuromorphic platforms, delay lines and direction insensitive CD. Kune dunhu rine mutsara wekunonoka, iyo inotariswa maitiro ndeyekupa yakabuda pulse nekunonoka Δt. Kana iyo chaiyo yedunhu kunonoka iri pasi peiyo inotarirwa kukosha, iyo synaptic uremu hweG3 hunofanira kuderedzwa (G3 inofanira kudzoreredzwa uye yozoiswa kune yakaderera inoenderana ikozvino Icc). Conversely, if the actual delay is greater than the target value, the conductance of G3 must be increased (G3 must first be reset and then set to a higher Icc value). This process is repeated until the delay generated by the circuit matches the target value and a tolerance is set to stop the calibration process. Kune maCD-asinganzwisisike maCD, maviri maRRAM zvishandiso, G1 uye G3, anobatanidzwa mukugadzirisa maitiro. Dunhu iri rine maviri ekuisa, Vin0 neVin1, inonokerwa nedt. Dunhu rinofanirwa kungopindura kunonoka kuri pazasi peiyo inofananidzira renji [0,dtCD]. Kana pasina peak yekubuda, asi iyo peak yekuisa iri padyo, ese maRRAM maturusi anofanirwa kukwidziridzwa kubatsira neuron kusvika pachikumbaridzo. Sezvineiwo, kana dunhu richipindura kunonoka kunopfuura dhairekitori reye dtCD, conductance inofanira kuderedzwa. Dzokorora maitiro kusvikira maitiro akakodzera awanikwa. Kuteerera ikozvino kunogona kugadziridzwa neyakavakwa-mukati analog wedunhu mune ref. 72.73. Neiyi yakavakirwa-mukati dunhu, maitiro akadaro anogona kuitwa nguva nenguva kuyera sisitimu kana kuishandisa zvakare kune imwe application.


Indiveri, G. & Sandamirskaya, Y. Kukosha kwenzvimbo uye nguva yekushandiswa kwechiratidzo muvamiririri veuromorphic: Dambudziko rekugadzira simba rakaderera, rinozvimiririra rinopindirana nenharaunda. Indiveri, G. & Sandamirskaya, Y. Kukosha kwenzvimbo uye nguva yekushandiswa kwechiratidzo muvamiririri veuromorphic: Dambudziko rekugadzira simba rakaderera, rinozvimiririra rinopindirana nenharaunda.Indiveri G. uye Sandamirskaya Y. Kukosha kwenzvimbo uye nguva yekugadzira zviratidzo mu neuromorphic agents: dambudziko rekugadzira pasi-simba rinozvimiririra rinosangana nenharaunda. Indiiver, G. & Sandamilskaya, Y. 空间和时间对于神经形态代理中信号处理的重要性: 开发与环境交互的低功耗, 自主代理的挑战. Indiveri, G. & Sandamirskaya, Y.Indiveri G. uye Sandamirskaya Y. Kukosha kwenzvimbo uye nguva yekugadzira zviratidzo mu neuromorphic agents: dambudziko rekugadzira pasi-simba rinozvimiririra rinosangana nenharaunda.IEEE Signal Processing. Journal 36, 16–28 (2019).
Thorpe, SJ Peak Nguva Yekusvika: Inoshanda Neural Network Coding Scheme. muEckmiller, R., Hartmann, G. & Hauske, G. (eds). muEckmiller, R., Hartmann, G. & Hauske, G. (eds).muEckmiller, R., Hartmann, G. uye Hauske, G. (eds.).MuEckmiller, R., Hartmann, G., uye Hauske, G. (eds.). Parallel processing mune neural system uye makomputa 91-94 (North-Holland Elsevier, 1990).
Levy, WB & Calvert, VG Kukurukurirana inoshandisa simba rakapetwa ka35 pane kuverenga mucortex yevanhu, asi zvose zvinodhura zvinodiwa kufanotaura nhamba ye synapse. Levy, WB & Calvert, VG Kukurukurirana inoshandisa simba rakapetwa ka35 pane kuverenga mucortex yevanhu, asi zvose zvinodhura zvinodiwa kufanotaura nhamba ye synapse.Levy, WB neCalvert, WG Kukurukurirana inoshandisa simba rakapetwa ka35 pane kuverenga mucortex yemunhu, asi zvose zvinodhura zvinodiwa kufanotaura nhamba ye 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 KukurukuriranaLevy, WB neCalvert, WG Kukurukurirana inoshandisa simba rakapetwa ka35 pane kuverenga mucortex yemunhu, asi zvose zvinodhura zvinoda kufanotaura nhamba ye synapses.process. National Academy yeSainzi. sainzi. US 118, https://doi.org/10.1073/pnas.2008173118 (2021).
Dalgaty, T., Vianello, E., De Salvo, B. & Casas, J. Insect-inspired neuromorphic computing. Dalgaty, T., Vianello, E., De Salvo, B. & Casas, J. Insect-inspired neuromorphic computing.Dalgati, T., Vianello, E., DeSalvo, B. uye Casas, J. Insect-inspired neuromorphic computing.Dalgati T., Vianello E., DeSalvo B. uye Casas J. Insect-inspired neuromorphic computing. Current. Maonero. Sayenzi yezvipembenene. 30, 59–66 (2018).
Roy, K., Jaiswal, A. & Panda, P. Towards spike-based machine intelligence ine neuromorphic computing. Roy, K., Jaiswal, A. & Panda, P. Towards spike-based machine intelligence ine neuromorphic computing. Roy, K., Jaiswal, A. & Panda, P. Towards Spike-Based Machine Intelligence neNeuromorphic Computing.Roy K, Jaiswal A, uye Panda P. Pulse-based artificial intelligence vachishandisa neuromorphic computing. Zvisikwa 575, 607–617 (2019).
Indiveri, G. & Liu, S.-C. Indiveri, G. & Liu, S.-C.Indiveri, G. naLiu, S.-K. Indiveri, G. & Liu, S.-C. Indiveri, G. & Liu, S.-C.Indiveri, G. naLiu, S.-K.Memory uye kugadzirisa ruzivo mune neuromorphic systems. process. IEEE 103, 1379–1397 (2015).
Akopyan F. et al. Truenorth: Dhizaini uye kit yemidziyo ye65 mW 1 miriyoni neuron inorongwa synaptic chip. IEEE kutengeserana. Kugadzirwa kwekombuta yeakabatanidzwa edunhu masisitimu. 34, 1537–1557 (2015).
Schemmel, J. nevamwe. Live demo: yakadzikira pasi vhezheni yeBrainScaleS neuromorphic system pachiyero cheplate. 2012 IEEE International Symposium pamusoro peMasekete neSystems (ISCAS), (IEEE ed.) 702–702 (2012).
Moradi, S., Qiao, N., Stefanini, F. & Indiveri, G. A scalable multicore architecture ane heterogeneous memory structures for dynamic neuromorphic asynchronous processors (DYNAPs). Moradi, S., Qiao, N., Stefanini, F. & Indiveri, G. A scalable multicore architecture ane heterogeneous memory structures for dynamic neuromorphic asynchronous processors (DYNAPs).Moradi S., Qiao N., Stefanini F. uye Indiviri G. A scalable multicore architecture ine heterogeneous memory structures for 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的廆理器(DYNAP) Moradi, S., Qiao, N., Stefanini, F. & Indiveri, G. Rudzi rwekuwedzeredzwa kweakawanda-musimboti architecture, ine yakasarudzika ndangariro chimiro che dynamic neural processing (DYNAP).Moradi S., Qiao N., Stefanini F. uye Indiviri G. A scalable multicore architecture ine heterogeneous memory structures for dynamic neuromorphic asynchronous processors (DYNAP).IEEE Shanduko pane Biomedical sainzi. hurongwa hwemagetsi. 12, 106–122 (2018).
Davis, M. et al. Loihi: A neuromorphic multi-core processor ine yakadzikwa kudzidza. IEEE Micro 38, 82–99 (2018).
Furber, SB, Galluppi, F., Tembere, S. & Plana, LA Iyo SpiNNaker chirongwa. Furber, SB, Galluppi, F., Tembere, S. & Plana, LA Iyo SpiNNaker chirongwa.Ferber SB, Galluppi F., Temple S. uye Plana LA SpiNNaker chirongwa.Ferber SB, Galluppi F., Temple S. uye Plana LA SpiNNaker chirongwa. process. IEEE 102, 652–665 (2014).
Liu, S.-K. & Delbruck, T. Neuromorphic sensory systems. & Delbruck, T. Neuromorphic sensory systems.uye Delbrück T. Neuromorphic sensory systems. & Delbruck, T. 神经形态感觉系统. & Delbruck, T.uye Delbrück T. Neuromorphic sensory system.Current. Maonero. Neurobiology. 20, 288–295 (2010).
Chope, T. et al. Neuromorphic sensory integration for combined sound source localization and collision avoidance. Muna 2019 paIEEE Musangano weBiomedical Circuits uye Systems (BioCAS), (IEEE Ed.) 1-4 (2019).
Risi, N., Aimar, A., Donati, E., Solinas, S. & Indiveri, G. A spike-based neuromorphic architecture ye stereo vision. Risi, N., Aimar, A., Donati, E., Solinas, S. & Indiveri, G. A spike-based neuromorphic architecture ye stereo vision.Risi N, Aymar A, Donati E, Solinas S, uye Indiveri G. A 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, uye Indiveri G. Spike-based neuromorphic architecture ye stereo vision.
Osswald, M., Ieng, S.-H., Benosman, R. & Indiveri, G. A spiking neural network model of 3Dperception for event-based neuromorphic stereo vision systems. Osswald, M., Ieng, S.-H., Benosman, R. & Indiveri, G. A spiking neural network model of 3Dperception for event-based neuromorphic stereo vision systems.Oswald, M., Ieng, S.-H., Benosman, R., uye 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.Oswald, M., Ieng, S.-H., Benosman, R., uye Indiveri, G. Spiked 3Dperception Neural Network Model for an Event-Based Neuromorphic Stereo Vision System.sainzi. Chirevo 7, 1-11 (2017).
Dalgaty, T. et al. Zvipembenene zvakafemerwa nekutanga kuona mafambiro anosanganisira resistive memory uye bursty neural network. Bionic biohybrid system. 10928, 115–128 (2018).
D'Angelo, G. et al. Chiitiko-chakavakirwa eccentric kufamba kwekuona uchishandisa temporal musiyano wekodha. pamberi. Neurology. 14, 451 (2020).