I-Los Ama-chip e-2nm asephenduke umgomo omusha omkhulu kusukela embonini ye-semiconductor. Asisakhulumi ngomqondo okude, kodwa ubuchwepheshe obuqala ukungena ekukhiqizweni kwangempela futhi obuzobeka inkambo yezizukulwane ezilandelayo zama-CPU, ama-GPU nama-SoC kubathengi, izikhungo zedatha kanye nobuhlakani bokwenziwa.
Ngesikhathi esifanayo, Ukushintshela ku-2 nm kubonakala sengathi akulula neze.Izindleko ezikhuphukayo, izithiyo ekupakisheni okuthuthukisiwe, ukungabaza ngenzuzo nge-wafer ngayinye kanye nokuncintisana okukhulu phakathi kwe-TSMC, i-Intel ne-Samsung, kuyilapho abadlali abanjengo-Apple, i-NVIDIA, i-AMD, i-Qualcomm, i-Fujitsu noma i-Broadcom bezibeka esimweni sokusebenzisa le lithography entsha ngendlela yabo.
Isho ukuthini ngempela i-chip ye-2nm namuhla?
Into yokuqala okufanele icaciswe ukuthi, okwamanje, Isibalo esithi “2 nm” siwuphawu lokumaketha kakhulu kunesilinganiso esiqondile somzimba.Ayisachazi ngokunembile ubude besango lama-transistors, ibanga eliphakathi kwawo, noma ezinye izilinganiso ezithile zejiyometri. I-foundry ngayinye (i-TSMC, i-Intel, i-Samsung, njll.) ichaza ama-node ayo njengoba ibona kufanele, ihlanganisa izilinganiso zangaphakathi ezahlukahlukene, okwenza kube nzima kakhulu ukuqhathanisa okuqondile phakathi kwezinqubo.
Noma kunjalo, leli gama liyaqhubeka lisetshenziswa ngoba liphawula isigaba sokuhlanganiswa esithuthuke kakhuluLokhu kuhlotshaniswa nobuningi be-transistor obuphezulu, ukusebenza okungcono, kanye nokusetshenziswa kwamandla okuphansi ngokusebenza ngakunye uma kuqhathaniswa nama-node angaphambilini njenge-5 nm, 4 nm, noma i-3 nm. Ngamanye amazwi, nakuba "ama-nanometer" engasasebenzi ngokoqobo, asebenza ukukhombisa ukuthi i-node ngayinye ikuphi esitebhisini sobuchwepheshe.
Empeleni, ukusuka ezinqubweni ezifana ne-3 nm noma i-4 nm kuya ku-2 nm kuhilela thola ama-chip asheshayo noma asebenza kahle kakhulu, noma ibhalansi engcono phakathi kokubiliLokhu kuholela kuma-core amaningi, i-cache eyengeziwe, amayunithi e-AI amaningi, kanye nama-frequency aphezulu ngobukhulu obufanayo be-chip, noma ukugcina ukusebenza okufanayo ngenkathi kunciphisa kakhulu ukusetshenziswa kwamandla.
Ngakho-ke, i-lithography ye-2 nm akuyona nje indaba yokumaketha: Kuyisisekelo segagasi elilandelayo lemikhiqizo esebenza kahle kakhuluKusukela kuma-processor edeskithophu nama-laptop kuya kuma-accelerator e-AI nama-supercomputer, kufaka phakathi ama-SoC eselula asezingeni eliphezulu.
Kusukela ku-FinFET kuya ku-GAAFET/Nanosheet: ushintsho lwesizukulwane kuma-transistors
Enye yezinguquko ezinkulu ezixhumene ne-2 nm akuyona nje ubukhulu "obujwayelekile" be-node, kodwa futhi nohlobo lwe-transistor. Imboni iyasuka kuma-FinFET endabuko, okube yisisekelo kusukela ku-22 nm/16 nm, ukuze kuthuthelwe ku-GAAFET (Gate-All-Around) architectures, eyaziwa nangokuthi i-nanosheet noma i-nanolaminate.
Kuma-FinFET, isango lizungeza ingxenye yesiteshi se-transistor, kanti kuma-GAAFET Isango lizungeze ngokuphelele umseleokunikeza ukulawula okunembe kakhulu kokugeleza kwamanje. Lokhu kuvumela ukuvuza okuncishisiwe, ukushabalaliswa kokushisa okungcono, kanye nokukhuliswa okuqhubekayo kwamandla kagesi kanye nemvamisa ngisho noma ubuningi be-transistor bukhula kakhulu.
Isibonelo, i-TSMC ibize indlela yayo ngokuthi I-N2 Nanosheet enobuchwepheshe be-NanoflexLeli cebo livumela ukuhlanganisa ukuphakama okuhlukene kwe-nanosheet kanye nobuningi bamaseli e-logic endaweni encane kakhulu, ukulungisa umklamo ngokwengxenye ye-chip: ezinye izindawo ezilungiselelwe ukusebenza okuphezulu kanti ezinye zenzelwe ukusebenza kahle kwamandla.
I-Samsung ibilokhu icwaninga izinhlobo ze-GAAFET isikhathi esithile, futhi Uze wacabanga ngezinto ezingasetshenziswa njenge-molybdenum. ukuthuthukisa ukuhamba kwama-electron nokuxazulula kangcono izinkinga zokushisa ezivelayo ngokuncishiswa okwengeziwe. I-Intel, yona, ikhuluma ngama-transistors ayo e-GAA asekelwe kuma-nanosheet ku ama-node afana no-20A no-18Alapho lolu lwakhiwo luphinde luhlanganiswe nezinhlelo zokunikezwa kwamandla ngemuva kwe-chip.
Yiziphi izithuthukisi ezithenjiswa ama-node e-2nm?
Abakhiqizi bavame ukunikeza izibalo ngokuqhathanisa i-node neyandulelayo. Endabeni ye-2 nm, Izibalo ezixoxwa ngazo zinenzuzo enkulu.Noma kunjalo, zihlala zincike ekwakhiweni kwe-chip ethile kanye nohlobo lomsebenzi:
Ngokusho kuka-Abhijeet Chakraborty, umakhi wesofthiwe kwa-Synopsys, ikhasimende elifuduka lisuka ku-3 nm liya ku-2 nm lingalindela, cishe:
- Ukusebenza okungaphezulu phakathi kuka-10% no-15% ngamandla afanayo.
- Ukusetshenziswa kwamandla okungaphansi ngo-20% no-30% ngokusebenza okulinganayo.
- Cishe ubuningi be-transistor obungu-15% ngaphezulu endaweni efanayo.
I-TSMC inikeza izibalo ezifanayo ze-node yayo ye-N2 uma kuqhathaniswa ne-N3E: ukusebenza okwengeziwe kufikela ku-15% noma ukusetshenziswa okungaphansi ngo-30%Njengoba i-transistor density iphindwe ngo-1,15 ngenxa yama-Nanoflex nanosheet transistors, le nhlanganisela yokusebenza kanye nokusebenza kahle yikona kanye okuqhuba iziqhwaga ezifana ne-Apple ne-NVIDIA ukuthi zigcine umthamo kule node kusenesikhathi.
Egameni le-Intel, uBen Sell, iphini likamongameli wentuthuko yezobuchwepheshe, ugcizelele ukuthi kuma-node alinganayo angu-2nm, njenge-18A, Okubalulekile akukhona nje ukufinyelela imvamisa ephezulukodwa kunalokho ukuthuthukisa ukusebenza nge-watt ngayinye nokunciphisa indawo edingekayo ezingeni elithile lamandla okusebenzisa ikhompyutha. Ngokombono wemakethe, lokhu kuholela emandleni amaningi kuphakheji efanayo noma ekusebenzeni kahle okukhulu ngenkathi kugcinwa ukusebenza.
Kunoma ikuphi, ukugxuma kwe-"nanometers of marketing" okungu-2-3 kuyabonakala ekusebenzeni: Umehluko phakathi kuka-5 nm no-3 nm ube mkhulu kakhulu Kubakhiqizi abanjengo-AMD, NVIDIA, Qualcomm noma i-Apple, ukuthuthela ku-2 nm kulandela indlela efanayo, yize kunezindleko zokukhiqiza eziphakeme kakhulu kanye nezinselele zobuchwepheshe ezikhulayo.
I-TSMC: umholi obeka ijubane le-2nm
I-TSMC ibilokhu iyi- "umenzi wesithunzi" omkhulu wengxenye yomkhakha wezobuchwephesheIzimboni zabo zikhiqiza ama-chip omakhalekhukhwini, amakhadi ehluzo, ama-CPU edeskithophu nama-laptop, ama-processor amaseva, ama-console, nokunye okuningi. Isabelo sabo semakethe embonini yokwenziwa kwezinto ezithuthukisiwe sicishe sibe ngu-60%, okubanika isikhundla esicacile nesibusayo.
Inkampani isiqalile kakade ukukhiqizwa kwe-wafer endaweni yayo ye-N2 Ngokusekelwe kubuchwepheshe be-nanosheet, i-node edumile ethi “2 nm”, lobu buchwepheshe bumelela ushintsho lwesizukulwane oluvela ku-N3E, hhayi nje ngenxa ye-lithography uqobo, kodwa futhi nangenxa yokugxuma kusuka kuma-transistors e-FinFET kuya kuma-transistors e-GAA anobuchwepheshe be-Nanoflex. Omunye wamakhasimende okuqala amakhulu ukusebenzisa lobu buchwepheshe kuzoba yi-AMD, okubikwa ukuthi isiqede ukukhipha ama-CCD ayo ezakhiwo ze-Zen 6 esikhathini esizayo ku-node ye-N2, ngokubikezelwa ukuthi izokwethulwa cishe ngo-2026.
Ukuze ihambisane ne-N2, i-TSMC iklame i- umndeni ophelele wezinqubo ezisuselwe kokunye:
- I-N2P, ehloselwe ukusebenza kahle kwamandla okukhulu.
- I-N2X, igxile ku ukusebenza ngokwedlulele kwezicelo ze-TDP eziphezulu.
- I-A16 SPR, i-node ethuthuke kakhulu enesiphehli sikagesi sangemuva (i-Super Power Rail).
Endabeni ye-A16 SPR, idatha yokuqala iphakamisa ukuthi kufike ku- 8-10% ngokushesha kune-N2P ku-voltage efanayo, noma ukwehla okungu-15-20% kokusetshenziswa ngenkathi kugcinwa ukusebenza. Uma sibheka phambili, i-TSMC iphinde yachaza imephu yayo yendlela eya ku-A14, iveza ukusebenza okungaphezu kwezikhathi ezingu-1,8 kumandla afanayo kanye nokusebenza kahle kwamandla okungcono okuphindwe izikhathi ezingu-4,2 ezimweni ezithile.
Ukuze kuqhubeke lokhu kuhlasela, i-TSMC ibeke eceleni okungenani izimboni ezimbili ezibalulekile (i-Fab 20 ne-Fab 22) ekukhiqizweni kwe-2nmAkukhona nje ngokuthuthukisa imishini: izigidigidi zamaRandi zidinga ukutshalwa emishinini ye-EUV ethuthuke kakhulu, izinhlelo ze-metrology, imigqa emisha yokupakisha nokuhlola, kanye nayo yonke ingqalasizinda ehlotshaniswa ne-node eyinkimbinkimbi kangaka.
Inkinga yokupakisha okuthuthukisiwe: i-CoWoS-L kanye nenkampani
Olunye uhlangothi lwemali ukuthi amakhasimende e-TSMC awasafuni nje "i-wafer engu-2nm" futhi akukho okunye. Isidingo sigxile kuma-chips ayinkimbinkimbi aneziphako ezithuthukisiwe., njenge-CoWoS-L, ekwazi ukuhlanganisa ama-dies amaningana ku-interposer enamandla kakhulu futhi, ezimweni eziningi, nememori ehlanganisiwe.
I-CoWoS-L ivumela ama-chip amaningi (i-CPU, i-GPU, ama-chiplet e-cache, i-HBM, njll.) ukuthi abekwe ku-silicon interposer noma into efanayo nayo ukuxhumana okuminyene kakhulu, okune-bandwidth ephezuluInkinga iwukuthi lolu hlobo lokupakisha luletha izinselele ezengeziwe, njengokwanda kokushisa: izinto ezahlukene zanda ngokungalingani ngokushisa, okungabangela ukuguquguquka, ukucindezeleka komshini, imifantu emincane, noma ukwehluleka kokuxhumana.
Kulinganiselwa ukuthi Kubikwa ukuthi i-NVIDIA igcine cishe u-70% wamandla okukhiqiza e-TSMC e-CoWoS-LLokhu kuholele ekushodeni okukhulu kwala maphakheji athuthukile kwamanye amakhasimende. Imikhiqizo efana nama-NVIDIA's AI GPU, ama-processor e-AMD's EPYC ane-3D V-Cache, kanye nama-processor e-Xeon esikhathi esizayo anama-cache block amakhulu athembele ngqo kulolu hlobo lokuhlanganiswa ukuze afinyelele izibalo zawo zokusebenza.
Lokhu kuphazamiseka sekuvele kube nemiphumela ebonakalayo: I-TSMC iphoqelekile ukuba ilibazise ukulethwa okuhlobene nokwakhiwa kwe-NVIDIA's Blackwellecaphuna izinkinga zokusebenza nge-CoWoS-L kanye nokulinganiselwa kwamandla. Ngesikhathi esifanayo, inkampani ilungiselela inqubo yayo ye-A16 (cishe i-1,6 nm) nge-Super Power Rail, lapho i-NVIDIA isivele ibhalwe njengekhasimende eliza kuqala.
Ngamafuphi, nakuba i-lithography engu-2 nm iyisihloko esidonsa amehlo, Inani elikhulu (kanye nezinkinga) livela ekupakishweni kwe-2.5D kanye ne-3DImiklamo esekelwe kuma-chiplets, i-stacked cache, kanye nememori ecishe ibe yikhompyutha yikona okwandisa izinzuzo zala ma-node, futhi yilapho i-TSMC idinga khona ukukhulisa amandla ayo ngokushesha okukhulu.
I-Intel ne-Samsung: abanye abancintisanayo abazobusa imakethe ye-2nm
Njengoba i-TSMC iqinisa ubuholi bayo, I-Intel ne-Samsung babona i-2nm njengethuba elihle ukuze bathole amandla futhi bakholise amakhasimende angaphandle ukuthi ubuchwepheshe babo buyancintisana, hhayi ngemikhiqizo yabo kuphela.
I-Samsung ibhekene nesikhathi esinzima: Imali engenayo yayo ye-semiconductor yehle cishe ngo-37,5% ngo-2023 Ngokusho kukaGartner, inkampani iphoqelekile ukulungisa izinhlelo zayo zokwandisa kanye nabasebenzi uma kuqhathaniswa no-2022. Noma kunjalo, igcina umgomo wayo wokudlula i-TSMC kanye nezinye izimbangi ezifana ne-Intel zingakapheli iminyaka emihlanu, njengoba kushiwo ngumphathi wayo jikelele wesigaba se-semiconductor, uKye Hyun Kyung.
Isu le-Samsung lihilela ukwethula ama-node ayo e-GAAFET futhi alungele ukukhiqizwa ngobuningi be-2nm I-Samsung izoqala isu layo lokukhiqiza uma isidingo semakethe sivuma, kokubili kuma-SoC eselula kanye nezixazululo zesikhungo sedatha esisebenza kahle kakhulu. Le nkampani ibilokhu isebenza kubuchwepheshe bokupakisha obuthuthukisiwe kanye nezinto ezihlukile zokunciphisa izinkinga zokushisa ezihlobene nokukhula okukhulu. Imininingwane eyengeziwe ngesu lokukhiqiza le-Samsung ingatholakala ku-[isixhumanisi sewebhusayithi/isisetshenziswa se-Samsung]. uhlelo lwabo lokukhiqiza ama-core e-Exynos.
Endabeni ye-Intel, indlela yokusebenza ihlukile kancane. UPat Gelsinger, owayeyi-CEO yayo, wathembisa ukubuyisa ubuholi ezinqubweni zokukhiqiza nge-roadmap enolaka ehlanganisa ama-node afana ne-Intel 4, i-Intel 3, i-20A, kanye ne-18A ngokulandelana okusheshayo. UDave Zinsser, isikhulu sezezimali senkampani, wembule ukuthi i-Intel unqume ukweqa ukumaketha kwe-20A ukonga cishe amaRandi ayizigidi ezingama-500, ngokubeka ezinye zalezo zinsiza ku-18A.
UBen Sell ukuqinisekisile lokho I-18A isifinyelele ukuvuthwa okudingekayo ukuze ingene ekukhiqizweni okukhulu ngo-2025Ngokwemigomo yokumaketha, le node ikububanzi be-2nm (cishe i-1,8nm), futhi ihlanganisa ama-transistors e-nanoleaf GAA nezinhlelo zamandla ze-back-of-chip, okuhambisana kakhulu nalokho i-TSMC ekusikiselayo nge-A16 SPR.
Ku-Intel, inselele enkulu akuyona nje ukuqalisa ama-CPU ayo asekelwe ku-18A, njenge-Panther Lake noma imindeni yesikhathi esizayo, kodwa ukuthola ukwethenjwa kwezinkampani zangaphandle esigabeni sayo sokukhiqiza izisekelo futhi bakhiqize imiklamo yabo enenani eliphezulu lapho. Isihluthulelo kuzoba ukukhombisa isivuno se-wafer esincintisanayo, ukuzinza kokukhiqiza, kanye nezikhathi zokulethwa ezithembekile uma kuqhathaniswa ne-TSMC.
I-Fujitsu, i-Broadcom kanye ne-2nm milestone eku-supercomputing
Ngenkathi izinkampani ezinkulu zabathengi zilungisa amanethiwekhi azo, I-Fujitsu ne-Broadcom benze isitatimende emkhakheni we-HPC kanye ne-AI yezinga eliphezuluNgokubambisana ne-TSMC, sebeqalile ukukhiqiza i-2nm SoC eyaziwa ngokuthi i-FUJITSU-MONAKA, eyenzelwe i-supercomputer enkulu elandelayo yaseJapan, i-FugakuNEXT, eholwa yi-RIKEN.
Le chip yaklanywa ngokukhethekile kucatshangelwa imithwalo yemisebenzi emikhulu ye-supercomputing kanye ne-artificial intelligence. Inama-cores angu-144 kanye nephakheji ye-3,5D (XDSiP).Lokhu kusho izinga lokuhlanganiswa elithuthuke kakhulu kune-2.5D evamile. Umgomo ukuhlanganisa ukusebenza okukhulu kwekhompyutha nokusetshenziswa kwamandla aphansi, okubaluleke kakhulu emishinini yalolu hlobo.
Ukuze kondliwe isilo esinjalo, i-FUJITSU-MONAKA iyahlanganisa uhlelo lwememori oluneziteshi eziyishumi nambili ze-DDR5Ngaphezu kokusekelwa kwe-PCIe 6.0 kanye ne-CXL 3.0, le nhlanganisela iyenza ibe yisinqumo esifanele semithwalo yemisebenzi emikhulu ye-AI kanye nokulingisa kwesayensi okudinga kakhulu, lapho ukubambezeleka kanye nomkhawulokudonsa wememori kubaluleke kakhulu njengamandla ekhompyutha aluhlaza.
Nakuba kungesiyona i-chip yokuqala ye-2nm eyake yamenyezelwa, Kuyagqama ngokuba ngomunye wabokuqala kule node ukungena ekukhiqizeni futhi neklayenti elithile.Esikhundleni sokuhlala esigabeni se-PowerPoint kanye nokukhishwa kwabezindaba, ukuqala kwayo ukusebenza ezikhungweni zedatha kanye nama-supercomputer kulindeleke ngo-2027, okuqinisa isikhundla se-Fujitsu esigabeni se-HPC ngokumelene nabancintisani abanjengo-AMD (Instinct), NVIDIA (Grace, Grace Hopper), kanye ne-Intel (Xeon, Gaudi).
Lesi sinyathelo sibonisa ukuthi i-2nm akuyona nje indaba yamafoni noma ama-PC asezingeni eliphezulu: I-Supercomputing kanye ne-AI ephakeme kungenye yezinkambu zokuqala lapho ama-chip avela kule node azobonakala esebenza ngokugcwele.ngoba yilapho ukhokha khona kakhulu nge-watt ngayinye egciniwe kanye nephuzu ngalinye lephesenti elengeziwe lokusebenza.
Umthelela we-2nm kuma-PC, amaselula kanye nehadiwe yasekhaya
Uma sibheka umsebenzisi ojwayelekile, namuhla ama-CPU amaningi edeskithophu nama-laptop asetshenziswa ekhaya asasebenza kuma-node afana nalawa I-5 nm kanye ne-7 nm uma kwenzeka i-AMD Ryzenkanye nezinqubo ezilinganayo noma ezindala kancane kweminye imigqa yomkhiqizo we-Intel. Kumadivayisi eselula, ububanzi obuphezulu kakade bucishe bube ngu-3 no-4 nm, ngezibonelo ezifana nama-Apple chips noma ama-SoC athile avela ku-Qualcomm kanye ne-MediaTek.
Ngalo mongo, Ngeke sibone ama-CPU "aphelele" edeskithophu ekhiqizwa ngokuphelele ngenqubo ye-2nm ngobuningi esikhathini esifushane.Izindleko nge-wafer ngayinye zikhuphuka cishe ngo-50% noma ngaphezulu uma zisuka ku-5nm ziye ku-3nm bese ziya ku-2nm, futhi inzuzo ifaneleka kuphela ezigabeni lapho i-margin nge-chip ngayinye iphezulu kakhulu, njenge-AI GPUs, ama-server processors noma ama-ultra-premium SoCs.
Emvelweni yasekhaya, umkhuba osheshayo uwukuthi khetha imiklamo ehlukahlukene enezindawo eziningana zokukhiqiza kumkhiqizo ofanayoI-Intel isivele isebenzisa ama-lithographies ahlukene kuphakheji eyodwa ukuhlanganisa ama-cores asebenza kahle, ama-cores asebenzayo, ihluzo ezihlanganisiwe, kanye nabalawuli be-I/O ngendlela elinganiselayo. I-AMD ilandela indlela efanayo: i-lithography ethuthukisiwe yama-chiplets ayinhloko (ama-CCD) kanye ne-evuthiwe kakhulu ye-I/O die.
Lokhu kusho ukuthi, esikhathini esimaphakathi, cishe sizobona imikhiqizo exubile lapho ezinye izingxenye ze-chip (njenge-main compute block, i-NPU, noma ezinye izingcezu ze-cache) empeleni zenziwa ngama-2 nmNgenkathi ezinye zizohlala kuma-node ashibhile njenge-3nm, 4nm, noma ngisho ne-6nm. Umgomo ukulinganisela isibalo sezindleko, ukusebenza, kanye nokusetshenziswa kwamandla ngaphandle kokukhulisa kakhulu intengo yokugcina yomthengi.
Ukwanda kwe-AI kuzo zonke izingxenye kuye kwanezela ingcindezi eyengeziwe: manje Cishe yonke i-SoC noma i-CPU yesimanje idinga ukuhlanganisa i-NPU noma i-accelerator ezinikezele. ukusebenzisa amamodeli e-AI ngempumelelo. Lokhu kwandisa ubunzima bokuklama futhi kudinga i-logic ekhethekile kakhulu ku-chip, into i-2nm ekhanga kakhulu kuyo, kodwa futhi ebiza kakhulu.
Izindleko, isivuno nge-wafer ngayinye, nokuthi kungani kungebona bonke abantu abazoshintsha okwamanje
I-lithography ethuthukisiwe ibilokhu ibiza kakhulu, kodwa nge-3 nm kanye ne-2 nm Lo mthethosivivinywa ukhuphuke kakhulu waze wafika emazingeni akwenza ucabange kabili.Kuthiwa i-wafer eyodwa engu-2nm ingabiza phakathi kuka-$25.000 no-$30.000, kuye ngokuthi ingakanani imali etholwe yi-TSMC, futhi ngaphezu kwakho konke, inzuzo etholwe.
Ekuqaleni kokuphila kwe-node, isivuno nge-wafer ngayinye ngokuvamile sibonakala singconoKunama-chips amaningi anephutha, ama-wafer amaningi angasetshenziswa, kanye nokulungiswa kwenqubo okudingekayo. Abakhiqizi abakhulu bahlose ukufeza okungenani isivuno esingu-70% kuma-node abo e-2nm ukuze bawenze abe nenzuzo yangempela futhi akhange kumakhasimende. Kuze kube yilapho lowo mkhawulo ufinyelelwe, amanani kanye nemikhawulo yamandla iphakeme kakhulu.
Lokhu kuchaza ukuthi kungani imiklamo eminingi iqhubeka nokuthembela kuma-node afana ne-3 nm, 4 nm, noma i-5 nm, lapho Ibhalansi phakathi kokusebenza, ukusetshenziswa kanye nezindleko isivele ihle kakhulu.Ukuqinisekisa ukweqa ku-2nm komkhiqizo onevolumu ephakathi noma ephansi, njengeprosesa ephakathi noma i-SoC yamadivayisi angewona awe-premium, kuyinkimbinkimbi lapho izindleko ze-wafer zicishe ziphindwe kabili.
Ngaphezu kwakho konke lokhu, kusukela ku-14 nm kuya ku-5 nm, imboni ihlangabezane nakho isigaba sentuthuko esheshayo kakhulu futhi "ehlanzekile"Lokhu kuyiqiniso ikakhulukazi nge-TSMC, eyashintsha ngempumelelo yaba yizinqubo ze-7nm kanye ne-5nm. Ijubane loshintsho, kanye nokukhula kwezikhungo zedatha ze-AI kanye nokukhula okukhulu kwe-electronics yabathengi, kusunduze amandla okukhiqiza kanye nokutshalwa kwezimali okudingekayo emafektri amasha kwaze kwaba sezingeni lawo.
Ngokuphambene nalokho, ukuguquka kwe-Intel kusuka kuma-node ayo angu-10nm kuya kuma-node alingana no-7nm kwakunezinkinga eziningi, ngokubambezeleka nezinkinga zokusebenza okwabiza isabelo sayo semakethe. Lokhu kungafani kukwenze kwacaca ukuthi I-node ngayinye entsha ethuthukisiwe inzima futhi ibiza kakhulu kuneyangaphambilini.Futhi ukuthi akwanele ukunciphisa ama-transistors: izakhiwo, ukupakisha, kanye nezinhlelo zokuphakelwa kukagesi kudingeka ziphinde zakhiwe kabusha kusukela ekuqaleni.
Yini umsebenzisi wokugcina angalindela kuma-chip e-2nm
Uma sibheka lokhu ngokombono wokusetshenziswa kwansuku zonke, Ama-chip e-2nm ngeke amele ukugxuma "okumangalisayo" njengalokho esakubona eminyakeni engaphezu kweshumi edluleLapho isizukulwane esisha sinciphisa ukusetshenziswa kwaso ngesigamu, isikhathi sezimangaliso ezinkulu ze-miniaturization siphelile; manje sikhuluma ngentuthuko eqhubekayo kodwa eklanywe ngokucophelela.
Eminyakeni ezayo, umthelela we-2 nm uzozwakala kuqala ku izingxenye eziphezulu nezingochwephesheAmaseva, izindawo zokusebenza, izisheshisi ze-AI, ama-GPU ajwayelekile, kanye nama-smartphone nama-laptop e-premium kamuva. Yilapho yonke intuthuko encane ekusebenzeni nge-watt ngayinye ihumushela khona ezinkontilekeni zezigidi zamaRandi kanye nenzuzo yangempela yokuncintisana.
Kumsebenzisi, lokho kusho ukuthi amaqembu akwazi ukusebenzisa amamodeli amakhulu e-AI endaweniImidlalo eyinkimbinkimbi kakhulu enamazinga ozimele angcono kanye nokusetshenziswa kwamandla okuphansi, izikhathi zokukhiqiza nokuhlanganisa eziphansi, kanye namadivayisi eselula agcina ukusebenza okuphezulu isikhathi eside ngaphandle kokushisa ngokweqile.
Esikhathini esimaphakathi, into ethakazelisa kakhulu izoba inhlanganisela yalezi zingqimba nokupakisha kwe-3D kanye nemiklamo esekelwe ku-chipletAma-died amaningi ahlanganisiwe, inkumbulo isondele kakhulu kumayunithi ekhompyutha, ukuxhumana kwangaphakathi kwe-bandwidth ephezulu kakhulu kanye nokwandisa ukukhethekile kwe-chiplet ngayinye ngokuya ngomsebenzi wayo (i-CPU, i-GPU, i-NPU, i-cache, i-I/O…).
Zonke lezi zindlela zenza umyalezo ofinyelela kumthengi ("i-chip entsha, isheshe futhi isebenze kahle") Lokho kungase kube yiqiniso, kodwa kufihla ngemuva kobunjiniyela obuyinkimbinkimbi ngokwengeziwe.I-2 nm iyingcezu eyodwa nje yephazili enkulu kakhulu, lapho yonke imininingwane ibalulekile ukuze kuqhubeke kucindezelwa wonke ama-watt wamandla atholakalayo.
Njengoba ukukhiqizwa ngobuningi sekuseduze kakhulu futhi amaphrojekthi okuqala angempela aseqalile, Ama-chip e-2nm ahloselwe ukuba yinhliziyo yesizukulwane esilandelayo sama-CPU kanye nama-acceleratorNgeke sibone ushintsho oluzumayo kusukela onyakeni owodwa kuya kolandelayo, kodwa kunalokho sizobona inqubekela phambili eqhubekayo lapho i-TSMC, i-Intel, ne-Samsung zincintisana khona ngobuholi kwezobuchwepheshe kuyilapho amakhasimende afana ne-Apple, i-NVIDIA, i-AMD, i-Fujitsu, ne-Broadcom ecindezela yonke ithonsi lokugcina lokusebenza kusukela ekuphindaphindweni ngakunye okusha; ekugcineni, kumsebenzisi, konke lokhu kuzohunyushwa kumadivayisi asebenza kangcono, asebenzise amandla amancane, futhi avule umnyango wokuhlangenwe nakho kokubala okubonakala sengathi kuyindaba yesayensi muva nje.
