ʻO nā holomua Supercomputing: me ka hoʻohana ʻana i nā ʻenehana optical neuromorphic

I nā makahiki i hala iho nei, ʻo ke ʻano i ʻike mua ʻia a pololei hoʻi, ʻo Moore's Law, i wānana ʻia e Gordon Moore o IBM i ka makahiki 1965, ke lilo lohi nei i kahi ana defunct o ka hana helu helu. Ua wānana ʻo Moore's Law ma kahi o kēlā me kēia ʻelua makahiki e pāpālua ʻia ka helu o nā transistors i loko o kahi kaapuni i hoʻohui ʻia, ʻoi aku ka nui o nā transistors i ka nui like o ka lewa, e alakaʻi ana i ka hoʻonui ʻana i ka helu a no laila ka hana kamepiula. Ma ʻApelila 2005, i loko o kahi ninaninau, ua ʻōlelo ʻo Gordon Moore ʻaʻole hiki ke kūpaʻa hou kāna kuhi: "Ma ke ʻano o ka nui [o nā transistors] hiki iā ʻoe ke ʻike ke hoʻokokoke nei mākou i ka nui o nā ʻātoma he mea pale kumu, akā. ʻelua a ʻekolu mau hanauna ma mua o ko mākou hiki ʻana i kēlā mamao - akā ʻo ia ka mamao e like me kā mākou i ʻike ai. Loaʻa iā mākou he 10 a 20 mau makahiki ma mua o ka hiki ʻana i kahi palena kumu. "   

ʻOiai ua hoʻopaʻi ʻia ke kānāwai o Moore e paʻi i kahi hopena make, ʻike ʻia nā hōʻailona ʻē aʻe o ka helu helu i ka piʻi ʻana o ka pono. Me ka ʻenehana a mākou e hoʻohana nei i ko mākou ola i kēlā me kēia lā, hiki iā mākou ke ʻike i nā ʻano o nā kamepiula e liʻiliʻi a liʻiliʻi akā ʻoi aku ka lōʻihi o nā ʻoihana. Ua kapa ʻia ke ʻano hope me nā pila ʻo Koomey's Law, i kapa ʻia ma muli o ke kumu aʻoaʻo o ke Kulanui ʻo Stanford ʻo Jonathan Koomey. Wahi a ke kānāwai o Koomey, "... ma kahi haʻahaʻa kamepiula paʻa, e hāʻule ka nui o ka pākaukau āu e pono ai ma ka helu ʻelua i kēlā me kēia makahiki a me ka hapa." No laila, pālua ka hoʻohana ʻana i ka mana uila a i ʻole ka ikehu o nā kamepiula i kēlā me kēia 18 mahina. No laila, ʻo ka mea a kēia mau ʻano a me nā hoʻololi e kuhikuhi nei a hōʻike ʻia ka wā e hiki mai ana o ka helu ʻana.

ʻO ka wā e hiki mai ana o ka helu helu

Ua hiki mai mākou i ka manawa o ka mōʻaukala kahi e pono ai mākou e wehewehe hou i ka helu helu ʻana no ka mea ʻaʻole pili nā ʻano a me nā kānāwai i wānana ʻia i nā makahiki i hala aku nei. Eia kekahi, ʻoiai ke neʻe nei ka helu helu i nā unahi nano a me quantum, aia nā palena kino a me nā pilikia e hiki mai ana. Malia paha ʻo ka hoʻāʻo ʻoi loa i ka supercomputing, quantum computing, loaʻa ka paʻakikī o ka hoʻohana ʻana i ka quantum entanglement no ka hoʻopili like ʻana, ʻo ia hoʻi, ka hana ʻana i nā helu ma mua o ka decoherence quantum. Eia naʻe, ʻoiai nā pilikia o ka computing quantum ua nui ka holomua i nā makahiki i hala iho nei. Hiki ke loaʻa i nā hiʻohiʻona o ka hana lolouila kahiko ʻo John von Neumann i hoʻohana ʻia i ka computing quantum. Akā aia kekahi aupuni kaulana ʻole o ka (super) computing, i kapa ʻia ʻo neuromorphic computing ʻaʻole i hahai i ka hana kuʻuna von Neumann. 

Ua ʻike ʻia ʻo Neuromorphic computing e Caltech professor Carver Mead i kāna pepa seminal i ka makahiki 1990.  ʻO ka mea nui, ua hoʻokumu ʻia nā loina o ka neuromorphic computing ma luna o nā kumu olaola i manaʻo ʻia e hoʻohana ʻia e ka lolo kanaka i ka helu ʻana. Ua hōʻuluʻulu ʻia kahi ʻokoʻa ʻokoʻa ma waena o ke kumumanaʻo computing neuromorphic me ke kumukānāwai computing classical von Neumann ma kahi ʻatikala na Don Monroe ma ka Association no ka mīkini hana uila puke pai. Penei ka ʻōlelo: “Ma ka hoʻolālā kuʻuna von Neumann, e hana maʻamau ka mana o ka logic core (a i ʻole kekahi mau mea like) ma ka ʻikepili i kiʻi ʻia mai ka hoʻomanaʻo. ʻO ka ʻokoʻa, ʻo ka 'neuromorphic' computing e puʻunaue ana i ka helu ʻana a me ka hoʻomanaʻo ʻana i waena o kahi heluna nui o nā 'neuron' maʻamau, e kamaʻilio ana kēlā me kēia me nā haneli a i ʻole mau tausani o nā neurons ma o nā 'synapses.'  

ʻO nā hiʻohiʻona nui ʻē aʻe o ka neuromorphic computing ʻo ia ka hewa intolerance, kahi e manaʻo ai e hoʻohālike i ka hiki o ka lolo o ke kanaka ke nalowale i nā neurons a hiki ke hana. Hoʻohālikelike, ma ka helu kuʻuna, ʻo ka nalowale o hoʻokahi transistor e pili ana i ka hana kūpono. ʻO kekahi pono i manaʻo ʻia a i manaʻo ʻia o ka neuromorphic computing ʻaʻole pono e hoʻolālā ʻia; ʻO kēia pahuhopu hope ke hoʻohālike hou i ka hiki o ka lolo kanaka ke aʻo, pane a hoʻololi i nā hōʻailona. No laila, ʻo ka neuromorphic computing ka moho maikaʻi loa no ke aʻo ʻana i ka mīkini a me nā hana naʻauao. 

Nā holomua o ka neuromorphic supercomputing

ʻO ke koena o kēia ʻatikala e komo i ka holomua o ka neuromorphic supercomputing. ʻO ka mea kikoʻī, ua paʻi hou ʻia ka noiʻi ma ka Arxiv mai Alexander Tait et. al. mai ke Kulanui ʻo Princeton e hōʻike ana ua ʻoi aku ka hoʻohālikelike ʻana o kahi ʻōnaehana neural photonic i hoʻokumu ʻia i ke kiʻi kiʻekiʻe ma mua o ke ʻano hoʻohālikelike maʻamau ma kahi kokoke i 2000-fold. Hiki i kēia neuromorphic photonic platform of computing ke alakaʻi i ka hoʻoili ʻike ultrafast. 

ʻO ka Tait et. al. pepa i kapaia ʻO Neuromorphic Silicon Photonics Hoʻomaka ka wehewehe ʻana i nā pono a me nā pōʻino o ka hoʻohana ʻana i ke ʻano kukui photonic o ka pāhawewe electromagnetic no ka helu ʻana. ʻO nā kumu nui mua o ka pepa, ʻo ia ka hoʻohana nui ʻana o ke kukui no ka hoʻouna ʻana i ka ʻike akā ʻaʻole no ka hoʻololi ʻana i ka ʻike, ʻo ia hoʻi, ka hoʻopili ʻana i ka ʻike. Pēlā nō, i ka computing quantum, aia nā pilikia kino koʻikoʻi i ka computing optical digital. A laila hele ka pepa i nā kikoʻī o kahi papa hoʻonohonoho kiʻi kiʻi neuromorphic i manaʻo mua ʻia ʻo Tait et. al. hui i paʻi ʻia i ka makahiki 2014, nona ke poʻo Hoʻolaha a me ke kaumaha: He pūnaewele i hoʻohui ʻia no ka hoʻoili ʻana i ka spike photonic scalable. Hōʻike kā lākou pepa hou i nā hopena o ka hōʻike hoʻokolohua mua o kahi pūnaewele neural photonic integrated. 

Ma ka hoʻolālā hoʻonohonoho "hoʻolaha a me ke kaumaha", ua hāʻawi ʻia nā "nodes" i kahi "wavelength carrier" kūikawā ʻo ia ka "wavelength division multiplexed (WDM)" a laila hoʻolaha ʻia i nā "nodes" ʻē aʻe. ʻO nā "nodes" i loko o kēia hoʻolālā i manaʻo ʻia e hoʻohālikelike i ka ʻano neuron i loko o ka lolo kanaka. A laila, hoʻoponopono ʻia nā hōʻailona "WDM" ma o nā kānana waiwai mau i kapa ʻia ʻo "microring (MRR) weight banks" a laila hōʻuluʻulu ʻia me ka uila i loko o kahi waiwai ʻike mana. ʻO ka laina ʻole o kēia hoʻololi / helu helu electro-optic hope loa ʻo ia ka non-linearity i koi ʻia e hoʻohālike i ka hana neuron, pono no ka helu ʻana ma lalo o nā kumu neuromorphic. 

I loko o ka pepa, kūkākūkā lākou i kēia mau hoʻokolohua hoʻololi electro-optic hoʻololi dynamics i ka makemakika me ka "2-node hoomau-manawa recurrent neural network" (CTRNN) model. Hōʻike kēia mau hopena paionia e hiki ke hoʻohana ʻia nā mea hana hoʻolālā i hoʻohana ʻia no nā hiʻohiʻona CTRNN i nā kahua neuromorphic pilikika. Wehe kēia ʻike i ke ala e hoʻololi ai i ke ʻano CTRNN i nā photonics silicon neuromorphic. Ma kā lākou pepa, hana lākou i kahi hoʻohālikelike hoʻohālike i kā lākou "hoʻolaha a me ke kaumaha". Hōʻike nā hualoaʻa i ka hoʻohālike CTRNN i hoʻohālikelike ʻia i kā lākou 49-node architecture e hāʻawi i ka neuromorphic computing architecture e ʻoi aku ka maikaʻi o nā kumu hoʻohālike maʻamau e 3 mau kauoha o ka nui.