Ngokubhekiselele kwisithintelo esisondelayo kwimveliso ye-chip, into engenakwenzeka ukuhla ngakumbi kweenkqubo zobugcisa, ukupakishwa kwe-multi-chip ye-crystals kuza phambili. Ukusebenza kweeprosesa zexesha elizayo kuya kulinganiswa ngobunzima, okanye kungcono kodwa, ubunzima bezisombululo. Imisebenzi emininzi inikezelwe kwi-chip encinci yeprosesa, amandla ngakumbi kunye nokusebenza kakuhle iqonga lilonke liya kuba. Kule meko, iprosesa ngokwayo iya kuba yiplatifomu yobuninzi beekristale ezingafaniyo ezidityaniswe yibhasi yesantya esiphezulu, engayi kuba yimbi ngakumbi (ngokwesantya kunye nokusetyenziswa) kunokuba ibe yikristale enye ye-monolithic. Ngamanye amazwi, iprosesa iya kuba zombini ibhodi yomama kunye neseti yamakhadi okwandisa, kuquka imemori, i-peripherals, njalo njalo.
I-Intel sele ibonisile ukuphunyezwa kweetekhnoloji ezimbini zobunini bokupakishwa kwendawo yeekristale ezingafaniyo kwiphakheji enye. Ezi EMIB kunye . Eyokuqala yibhulorho-ujongano olwakhelwe kwi-substrate "enyukayo" yokulungelelaniswa okuthe tye kweekristale, kwaye okwesibini lulungiselelo olunamacala amathathu okanye oluqokelelweyo lweekristale kusetyenziswa, phakathi kwezinye izinto, ngokusebenzisa imijelo ethe nkqo yeentsimbi ze-TSV. Ukusebenzisa iteknoloji ye-EMIB, inkampani ivelisa i-Stratix X isizukulwana se-FPGAs kunye ne-Kaby Lake G ye-hybrid processors, kwaye iteknoloji yeFoveros iya kuphunyezwa kwiimveliso zorhwebo kwisiqingatha sesibini salo nyaka. Umzekelo, iya kusetyenziselwa ukuvelisa iLakefield laptop processors.
Ewe kunjalo, i-Intel ayiyi kumisa apho kwaye iya kuqhubeka iphuhlisa ngokusebenzayo itekhnoloji yokupakishwa kwe-chip eqhubekayo. Abakhuphisana nabo benza into efanayo. Njani , kwaye i-Samsung iphuhlisa itekhnoloji yokulungelelanisa indawo yeekristale (i-chiplets) kwaye inenjongo yokuqhubeka nokutsala ingubo yamathuba amatsha kubo.
Kungekudala, kwinkomfa ye-SEMICON West, i-Intel kwakhona ukuba ubugcisa bayo bokupakishwa kweetshiphu ezininzi buphuhla ngesantya esilungileyo. Lo msitho ubonise itekhnoloji ezintathu, ukuphunyezwa kwazo okuza kwenzeka kungekudala. Kufuneka kuthiwe zonke itekhnoloji ezintathu aziyi kuba yimigangatho yoshishino. I-Intel igcina lonke uphuhliso ngokwalo kwaye iya kubonelela kuphela kubathengi kwimveliso yekhontrakthi.
Eyokuqala kobuchwephesha obuthathu obutsha bokupakishwa kwendawo yee-chiplets yi-Co-EMIB. Olu ludibaniso lwetekhnoloji yojongano lwebhulorho ye-EMIB enexabiso eliphantsi kunye neechiplets zeFoveros. Uyilo lwe-Foveros multi-chip stack lunokudityaniswa noqhagamshelo oluthe tyaba lwe-EMIB kwiinkqubo ezintsonkothileyo ngaphandle kokuncama ugqithiso okanye ukusebenza. Ibanga le-Intel ukuba i-latency kunye ne-output yazo zonke ii-interfaces ze-multi-layer aziyi kuba mbi ngakumbi kune-monolithic chip. Enyanisweni, ngenxa yokuxinana okugqithisileyo kweekristale ezingafaniyo, ukusebenza ngokubanzi kunye nokusebenza kwamandla kwesisombululo kunye ne-interfaces kuya kuba phezulu ngakumbi kunemeko yesisombululo se-monolithic.
Okokuqala, itekhnoloji ye-Co-EMIB ingasetyenziselwa ukuvelisa i-Intel hybrid processors ye-Aurora supercomputer, ekulindeleke ukuba ithunyelwe ngasekupheleni kuka-2021 (iprojekthi edibeneyo phakathi kwe-Intel kunye neCray). Iprosesa yeprototype yaboniswa kwi-SEMICON West njenge-stack ye-18 encinci ifa kwi-difa enye enkulu (i-Foveros), isibini esidityaniswe ngokuthe tye ngoqhagamshelwano lwe-EMIB.
Eyesibini kwiitekhnoloji ezintathu ze-Intel zokupakisha itshiphu entsha ibizwa ngokuba yi-Omni-Directional Interconnect (ODI). Le teknoloji ayikho enye into ngaphandle kokusetyenziswa kwe-EMIB kunye ne-Foveros ujongano loqhagamshelwano oluthe tye kunye noluthe nkqo lombane lweekristale. Yintoni eyenza i-ODI ibe yinto eyahlukileyo yinto yokuba inkampani iphumeze unikezelo lwamandla kwii-chiplets kwi-stack isebenzisa uqhagamshelwano oluthe nkqo lwe-TSVs. Le ndlela iya kwenza ukuba kube lula ukusasaza ukutya ngokufanelekileyo. Kwangaxeshanye, ukuxhathisa 70-ΞΌm amajelo TSVs unikezelo lwamandla kuncitshiswe kakhulu, nto leyo eya kunciphisa inani imijelo efunekayo ukuze unikezelo amandla kwaye ukhulule indawo kwi chip for transistors (umzekelo).
Ekugqibeleni, i-Intel yabiza i-chip-to-chip interface ye-MDIO iteknoloji yesithathu yokupakishwa kwendawo. Le yiBhasi yoNxibelelwano oluPhambili (i-AIB) ngendlela yomaleko obonakalayo wotshintshiselwano lwe-inter-chip yesignali. Ukuthetha ngokuthe ngqo, esi sisizukulwana sesibini sebhasi ye-AIB, leyo i-Intel iphuhlisa i-DARPA. Isizukulwana sokuqala se-AIB saziswa kwi-2017 ngokukwazi ukudlulisa idatha phezu koqhagamshelwano ngalunye ngesantya se-2 Gbit / s. Ibhasi ye-MDIO iya kubonelela ngokutshintshiselana ngesantya se-5,4 Gbit / s. Eli khonkco liya kuba ngumntu okhuphisana nebhasi ye-TSMC LIPINCON. Isantya sokudlulisa i-LIPINCON siphezulu - i-8 Gbit / s, kodwa i-Intel MDIO inobuninzi be-GB / s kwi-millimeter nganye: i-200 xa iphikisana ne-67, ngoko i-Intel ibango lophuhliso olungekho mbi ngakumbi kunolo olukhuphisana nalo.
umthombo: 3dnews.ru