Supercomputing ci gaban: ta amfani da neuromorphic na gani cibiyoyin sadarwa

A cikin ƴan shekarun da suka gabata, sanannen sananne kuma ingantaccen yanayin, Dokar Moore, wanda Gordon Moore na IBM ya annabta a cikin 1965, yanzu sannu a hankali ya zama ma'aunin ƙididdiga na aikin kwamfuta. Dokar Moore ta yi hasashen cewa kusan kowace shekara biyu adadin transistor a cikin na'ura mai haɗawa zai ninka, cewa za a sami ƙarin transistor a cikin adadin sararin samaniya, wanda zai haifar da haɓaka ƙididdiga kuma ta haka aikin kwamfuta. A cikin Afrilu 2005, a cikin wata hira, Gordon Moore da kansa ya bayyana cewa hasashensa ba zai dawwama ba: “Game da girma [na transistors] za ku iya ganin cewa muna gabatowa girman atom wanda shine babban shinge, amma ya za su kasance tsara biyu ko uku kafin mu kai ga wannan nisa-amma wannan ya yi nisa kamar yadda muka taɓa iya gani. Muna da wasu shekaru 10 zuwa 20 kafin mu kai ga iyaka. "   

Ko da yake dokar Moore za ta iya kaiwa ga mutuwa, wasu alamun ƙididdiga suna ganin haɓakar amfani. Tare da fasahar da muke amfani da ita a rayuwarmu ta yau da kullun, dukkanmu za mu iya ganin yadda kwamfutoci ke ƙara ƙanƙanta da ƙarami amma kuma batirin na'urar yana daɗe da tsayi. Halin na ƙarshe tare da batura ana kiransa Dokar Koomey, mai suna bayan farfesa na Jami'ar Stanford Jonathan Koomey. Dokar Koomey ta annabta cewa "... a ƙayyadaddun nauyin kwamfuta, adadin baturin da kuke buƙata zai faɗi da sau biyu a kowace shekara da rabi." Don haka, amfani da wutar lantarki ko ingancin makamashi na kwamfutoci yana ninka kusan kowane watanni 18. Don haka, abin da duk waɗannan canje-canje da canje-canje ke nunawa kuma suna bayyana shi ne makomar kwamfuta.

Makomar kwamfuta

Mun zo wani lokaci a tarihi inda za mu sake fayyace ƙididdiga kamar yadda abubuwa da dokokin da aka annabta shekaru da yawa da suka gabata ba su da amfani. Hakanan, yayin da ƙididdigewa ke matsawa zuwa ma'aunin nano da ma'aunin ƙididdiga, akwai bayyanannen gazawar jiki da ƙalubale masu zuwa. Wataƙila yunƙurin da ya fi fice a supercomputing, ƙididdige ƙididdigewa, yana da ƙalubalen ƙalubalen da ke tattare da haƙƙin haɗa ƙididdiga da gaske don ƙididdige ƙididdigewa, wato, yin ƙididdigewa kafin ƙayyadaddun ƙididdiga. Koyaya, duk da ƙalubalen ƙididdiga na ƙididdiga an sami ci gaba sosai a cikin ƴan shekarun da suka gabata. Mutum na iya samun samfura na gine-ginen kwamfuta na gargajiya na John von Neumann da aka yi amfani da su ga ƙididdigar ƙididdiga. Amma akwai wani sanannen daular (super) kwamfuta, mai suna neuromorphic computing wanda baya bin tsarin gine-ginen von Neumann na gargajiya. 

Kwamfuta na Neuromorphic ya hango ta Farfesa Caltech Carver Mead baya a cikin takardarsa ta seminal a cikin 1990. Ainihin, ka'idodin lissafin neuromorphic sun dogara ne akan ka'idodin nazarin halittu na aiki, kamar waɗanda ake tunanin za a yi amfani da su ta kwakwalwar ɗan adam a cikin lissafi. Wani ɗan taƙaitaccen bambanci tsakanin ka'idar lissafin neuromorphic da ka'idar lissafi ta von Neumann an taƙaita shi a cikin wata kasida ta Don Monroe a cikin Forungiyar Machineryungiyar Na'urar Na'ura jarida. Bayanin yana tafiya kamar haka: "A cikin tsarin gine-ginen von Neumann na gargajiya, babban mahimmancin dabaru (ko da yawa a layi daya) yana aiki akai-akai akan bayanan da aka samo daga ƙwaƙwalwar ajiya. Sabanin haka, lissafin 'neuromorphic' yana rarraba duka ƙididdigewa da ƙwaƙwalwar ajiya a tsakanin ɗimbin adadin 'neurons' masu yawa,' kowanne yana sadarwa tare da ɗaruruwa ko dubban sauran ƙwayoyin cuta ta hanyar 'synapses'.  

Sauran mahimman fasalulluka na lissafin neuromorphic sun haɗa da rashin haƙuri, wanda ke da nufin yin ƙima ga ikon kwakwalwar ɗan adam na rasa ƙwayoyin jijiya kuma har yanzu yana iya aiki. Hakazalika, a cikin kwamfuta na gargajiya, asarar transistor ɗaya yana shafar aikin da ya dace. Wani fa'idar hangen nesa da niyya na ƙididdigar neuromorphic shine babu buƙatar yin shiri; Wannan manufa ta ƙarshe ita ce sake yin ƙira ga ikon kwakwalwar ɗan adam don koyo, amsawa da daidaitawa da sigina. Don haka, lissafin neuromorphic a halin yanzu shine mafi kyawun ɗan takara don koyan na'ura da ayyukan fasaha na wucin gadi. 

Ci gaban neuromorphic supercomputing

Sauran wannan labarin zai shiga cikin ci gaban neuromorphic supercomputing. Musamman, kwanan nan da aka buga bincike akan Arxiv daga Alexander Tait et. al. daga Jami'ar Princeton ya nuna cewa samfurin hanyar sadarwa na photonic na tushen siliki ya fi dacewa da tsarin lissafin al'ada ta kusan 2000. Wannan dandali na photonic neuromorphic na kwamfuta na iya haifar da sarrafa bayanai na ultrafast. 

Tait et. al. takarda mai suna Neuromorphic Silicon Photonics ya fara bayyana fa'idodi da rashin lahani na amfani da nau'in haske na photonic radiation na lantarki don ƙididdigewa. Babban mahimman abubuwan farko na takarda shine cewa an yi amfani da haske sosai don watsa bayanai amma ba don canza bayanai ba, watau dijital na gani kwamfuta. Hakazalika, zuwa ƙididdige ƙididdigewa, akwai ƙalubalen ƙalubale na zahiri ga ƙididdigar gani na dijital. Takardar ta shiga cikin cikakkun bayanai na wani dandamalin kwamfuta na zamani na zamani wanda aka gabatar a baya mai suna Tait et. al. tawagar da aka buga a 2014, mai suna Watsa shirye-shirye da nauyi: Haɗe-haɗen hanyar sadarwa don sarrafa karu na photonic. Sabuwar takarda tasu ta bayyana sakamakon gwajin gwaji na farko na cibiyar sadarwa ta jijiyar hoto. 

A cikin gine-ginen kwamfuta na "watsawa da nauyi", an sanya "nodes" na musamman "mai ɗaukar nauyi" wanda shine "Rashin tsayin tsayin raƙuman ruwa (WDM)" sannan kuma watsa shi zuwa wasu "nodes". “Masu nodes” a cikin wannan gine-gine ana nufin su kwaikwayi halayen neuron a cikin kwakwalwar ɗan adam. Sannan ana sarrafa siginar “WDM” ta hanyar ci gaba da ƙima mai ƙima da ake kira “microring (MRR) banks weight” sa’an nan kuma a tattara ta hanyar lantarki zuwa jimlar ƙimar gano wuta. Rashin layin layi na wannan canji na electro-optic na ƙarshe / ƙididdigewa shine ainihin rashin layin da ake buƙata don kwaikwayon aikin neuron, mai mahimmanci ga ƙididdiga a ƙarƙashin ka'idodin neuromorphic. 

A cikin takarda, sun tattauna cewa waɗannan sauye-sauyen canjin lantarki da aka tabbatar da gwaji sun kasance daidai da lissafi da tsarin "2-node ci gaba da ci gaba da ci gaba da ci gaba da ci gaba na jijiyoyi" (CTRNN). Waɗannan sakamakon majagaba suna ba da shawarar cewa kayan aikin shirye-shirye waɗanda aka yi amfani da su don samfuran CTRNN za a iya amfani da su zuwa dandamalin neuromorphic na tushen silicon. Wannan binciken yana buɗe hanya don daidaita hanyoyin CTRNN zuwa ƙirar silicon photonics na neuromorphic. A cikin takardar su, suna yin irin wannan ƙirar ƙirar kawai akan tsarin gine-ginen "watsawa da nauyi". Sakamakon ya nuna cewa ƙirar CTRNN da aka kwaikwayi akan gine-ginen su na 49-node yana haifar da ƙirar ƙirar ƙirar neuromorphic don haɓaka ƙirar ƙira ta gargajiya ta umarni 3 na girma.