Ngo-1965, uGordon Moore, owayengumseki we-Intel, wachaza oko kwaba “nguMthetho kaMoore.” Kangangeminyaka engaphezu kwesiqingatha, yayixhasa ukuphumelela okuqhubekayo ekusebenzeni kwesekethe edibeneyo (IC) kunye nokwehla kweendleko—isiseko setekhnoloji yedijithali yanamhlanje. Ngamafutshane: inani lee-transistors kwi-chip liphindaphindeka phantse kabini rhoqo emva kweminyaka emibini.
Kangangeminyaka, inkqubela phambili ilandele loo santya. Ngoku umfanekiso uyatshintsha. Ukuncipha ngakumbi kuye kwaba nzima; ubukhulu beempawu buphantsi kakhulu kunee-nanometers ezimbalwa. Iinjineli zifikelela kwimida ebonakalayo, amanyathelo enkqubo anzima ngakumbi, kunye neendleko ezikhulayo. Iijometri ezincinci nazo ziyayithomalalisa isivuno, nto leyo eyenza imveliso ephezulu ibe nzima. Ukwakha nokusebenza ngendlela egqwesileyo kufuna imali eninzi kunye nobuchule obukhulu. Ngenxa yoko abaninzi bathi uMthetho kaMoore uyaphelelwa ngumdla.
Olo tshintsho luvule ithuba lokusebenzisa indlela entsha: ii-chiplets.
I-chiplet yidayi encinci eyenza umsebenzi othile—ngokuyintloko isilayi sento eyayisakuba yitshiphu enye ye-monolithic. Ngokudibanisa ii-chiplets ezininzi kwiphakheji enye, abavelisi banokuhlanganisa inkqubo epheleleyo.
Ngexesha le-monolithic, yonke imisebenzi yayiphila kwidayi enye enkulu, ngoko ke isiphene naphi na sasinokususa yonke itship. Ngee-chiplets, iinkqubo zakhiwe nge-"known-good die" (KGD), nto leyo ephucula kakhulu ukusebenza kakuhle kwemveliso kunye nemveliso.
Ukuhlanganiswa okwahlukeneyo—ukudibanisa ii-dies ezakhelwe kwii-node zenkqubo ezahlukeneyo nakwimisebenzi eyahlukeneyo—kwenza ii-chiplets zibe namandla kakhulu. Iibhloko zekhompyutha ezisebenzayo kakhulu zinokusebenzisa ii-node zamva nje, ngelixa iisekethe zememori kunye nee-analog zihlala kwiiteknoloji ezivuthiweyo nezingabizi kakhulu. Isiphumo: ukusebenza okuphezulu ngexabiso eliphantsi.
Ishishini leemoto linomdla ngokukodwa. Abenzi beemoto abakhulu basebenzisa ezi ndlela ukuphuhlisa iiSoCs ezizayo ezikwiimoto, apho ukwamkelwa ngobuninzi kujoliswe khona emva kowama-2030. IiChiplets zibavumela ukuba bandise i-AI kunye nemizobo ngokufanelekileyo ngelixa bephucula isivuno—benyusa ukusebenza kunye nokusebenza kwii-semiconductors zeemoto.
Ezinye iindawo zeemoto kufuneka zihlangabezane nemigangatho engqongqo yokusebenza nokhuseleko kwaye ngaloo ndlela zixhomekeke kwiindawo ezindala neziqinisekisiweyo. Okwangoku, iinkqubo zanamhlanje ezifana noncedo lomqhubi oluphambili (i-ADAS) kunye nezithuthi ezichazwe kwisoftware (ii-SDV) zifuna ukubala okungakumbi. IiChiplets zivala loo msantsa: ngokudibanisa ii-microcontrollers zeklasi yokhuseleko, imemori enkulu, kunye nee-accelerators ze-AI ezinamandla, abavelisi banokulungelelanisa ii-SoCs neemfuno zomenzi ngamnye wemoto—ngokukhawuleza.
Ezi nzuzo zidlulela ngaphaya kweemoto. Uyilo lweChiplet lusasazeka luye kwi-AI, kwi-telecom, nakwezinye iindawo, lukhawulezisa uphuhliso kumashishini kwaye luba yintsika yesicwangciso se-semiconductor.
Ukuhlanganiswa kweChiplet kuxhomekeke kuqhagamshelo oluncinci nolukhawulezayo lwe-die-to-die. Isixhobo esiphambili esivumelayo yi-interposer—umaleko ophakathi, odla ngokuba yi-silicon, phantsi kwee-dies ezihambisa imiqondiso efana kakhulu nebhodi yesekethe encinci. Ii-interposer ezingcono zithetha ukudibana okuqinileyo kunye nokutshintshiselana kwesignali ngokukhawuleza.
Ukupakisha okuphucukileyo kuphucula ukuhanjiswa kwamandla. Iindidi ezixineneyo zonxibelelwano lwesinyithi esincinci phakathi kweedayi zibonelela ngeendlela ezaneleyo zamandla kunye nedatha nakwiindawo ezixineneyo, okuvumela ukudluliselwa kwe-bandwidth ephezulu ngelixa kusetyenziswa ngokufanelekileyo indawo yephakheji encinci.
Indlela ephambili yanamhlanje kukudityaniswa kwe-2.5D: ukubeka ii-dies ezininzi ecaleni kwenye i-interposer. Inqanaba elilandelayo kukudityaniswa kwe-3D, apho ii-dies ezi-stacks zithe nkqo kusetyenziswa ii-vias ze-through-silicon (ii-TSV) ukuze kubekho uxinano oluphezulu nangakumbi.
Ukudibanisa uyilo lweetshiphusi eziziimodyuli (ukwahlula imisebenzi kunye neentlobo zeesekethe) kunye ne-3D stacking ivelisa ii-semiconductors ezikhawulezayo, ezincinci, nezisebenzisa amandla amaninzi. Ukudibanisa imemori kunye ne-computing kubonelela nge-bandwidth enkulu kwiiseti zedatha ezinkulu—ilungele i-AI kunye neminye imisebenzi yokusebenza ephezulu.
Nangona kunjalo, ukuqokelelana ngokuthe nkqo kuzisa imingeni. Ubushushu buqokelelana lula, nto leyo eyenza kube nzima ukulawula ubushushu kunye nokuvelisa. Ukujongana noku, abaphandi baphuhlisa iindlela ezintsha zokupakisha ukuze bajongane ngcono nemida yobushushu. Nangona kunjalo, amandla okusebenza aqinile: ukudibana kwee-chiplets kunye nokuhlanganiswa kwe-3D kujongwa ngokubanzi njengepharadaym ephazamisayo—elungele ukuthwala itotshi apho uMthetho kaMoore ushiye khona.
Ixesha leposi: Oktobha-15-2025