Chii chaitika
Vatsvagiri vakaunza MacroAgent, hwaro hwekuiswa kwekupedzisira kwezvikamu zvakakura zvedunhu, kana macros, mumagadzirirwo akakura kwazvo ekubatanidza. Iyo sisitimu inoshandisa LLMs kuwana akawanda akajairika-anoziva heuristic contour algorithms mukati mezvikamu zvina maitiro anosanganisira kusanganisa, kugadzirwa kwe contour, template yekufananidza uye pakati-masumbu kunatsiridza.
Peji yearXiv inoratidza bepa rine musoro unoti "MacroAgent: Regularity-Aware Macro Legalization with LLM-Agent-Designed Contour Algorithms," yakatumirwa musi waAug. 24, 2026. Pepa rinotarisa macro legalization, iyo vanyori vanotsanangura sedanho rekupedzisira-danho rekugadzirisa zvigadziro muzvikamu zvemazuva ano zvakakura zvikuru dhizaini. Tsime rinoti zvinzvimbo zvemacro zvine mhedzisiro yakakosha pamhando yekupedzisira yemhedzisiro, kana QoR, uye inopokana kuti nzira dziripo dzinogona kushaya kusimba, zvinoda computation yakakura kana kufuratira kugara pakati pemacros. Mukugadzira uku, kunyoreswa zviri pamutemo kwakabatana zvakanyanya nemhedzisiro yekurongeka, pane kuve chiitiko chega chekuisa.
MacroAgent inotsanangurwa senzira ina-nhanho. Iyo inotanga kubatanidza macros, yobva yagadzira contours, inoshandisa template yekufananidza uye inoita inter-cluster kunatsa. Iyo yakasarudzika AI chikamu iko kushandiswa kwemhando dzemitauro mikuru kuwana akawanda heuristic contour algorithms iyo inomiririra nguva dzose. Kunobva kunoratidza maalgorithms akabuda seakasimba uye anoshanda, asi haatarise mhando yemutauro, mupi wayo, kukurumidza kunoshandiswa, maitiro ekutsvaga, kana kuti mhando-yakagadzirwa algorithms inoburitswa kuti vamwe vaongorore kana kuberekazve. Zvikamu saka zvinopa chimiro chese cheiyo nzira, nepo tsananguro iripo inosiya iyo -yekuwana nzira isina kutaurwa.
Vanyori vanorondedzera mhedzisiro inopesana neino-ye-iyo-inoyevedza macro-mutemo basa paTILOS uye Chipyard mabhenji. Zvavo zvisinganzwisisike zvinoti kuvandudzwa kwakapetwa kaviri kusvika kasere mukurongeka kwemaitiro, kuderedzwa kwe3% kusvika 5% murutwayayarengi mushure mekufamba kwepasirese, kusangana kwakafanana uye kusimba kuri nani zvine nguva inogamuchirika. Mukuongorora kwekupedzisira-kusvika-kumagumo uchishandisa Cadence Innovus nzvimbo-nenzira, bepa rinoshuma 2.9% yakaderera wirelength uye 68.3% kuvandudzwa mune yakashumwa TNS metric zvichienzaniswa neiyo DREAMPlace macro-legalization baseline. Iyo zvakare inoshuma 1.8% yakaderera nzira waya kureba apo MacroAgent inosanganiswa muInnovus macro-kuisa kuyerera. Aya ndiwo mhedzisiro yehuwandu yakaunzwa mugwaro, uye inotsanangura ese mabhenji-chikamu mhedzisiro uye yakaparadzana yekupedzisira-kusvika-kumagumo ongororo.
Nei zvichikosha
Iro bepa rinopa AI senzira yekugadzira yakasarudzika engineering algorithms pane kungoburitsa zvinyorwa kana kufanotaura. Kana mibairo yakashumwa ikabata kupfuura mabhenji akaedzwa, nzira yacho inogona kubatsira vanogadzira chip kuvandudza magadzirirwo emuviri uku vachidzora mitengo yenzira uye kuwandisa.
Izvo zvakakosha ndezvekuti iyo LLM iri kushandiswa kutsvaga nzira dzealgorithmic mune imwe nyanzvi yeinjiniya dambudziko. Iyo bepa hairatidze iyo modhi seyekupedzisira yekuisa injini ichiita sarudzo dzisina kumanikidzwa pachayo. Panzvimbo iyoyo, sosi inotsanangura nzira yekuwana yakabatsirwa neLLM iyo inoburitsa seti ye contour heuristics inoshandiswa mukati meiyo yakarongeka yemutemo dhizaini. Musiyano iwoyo une basa rekunzwisisa kuti AI inopinda kupi mukufambiswa kwebasa uye uko kwakajairwa optimization matanho anosara. Iyo zvakare inochengeta iyo yakashumwa mupiro yakanangana nekuwanikwa kweheuristic mukati megadziriro.
Physical chip marongero anosanganisira zvinokwikwidza zvinangwa. Iyo sosi inobatanidza macro zvinzvimbo kusvika kuQoR yekupedzisira uye inoshuma kuvandudzwa kwenguva dzose uye yakafambiswa waya kureba ichitsanangura kusangana sekufananidza. Inoshumawo zvakawanikwa mumagumo-kusvika-kumagumo nzvimbo-uye-nzira yekuongorora, kusanganisira yakaderera yakarongwa waya kureba uye yakagadziridzwa TNS mhedzisiro. Zvikumbiro izvi zvinopa chinongedzo chinogoneka pakati pekutsvaga kweAI-yakabatsirwa uye yakadzika marongero zvabuda, kunyangwe sosi yacho isingapi humbowo hwakakwana, nguva dzekuvimba kana kudzokorora kwakazvimiririra. Saka kukosha kwacho kunoenderana nekuti izvo zvakashumwa mhedzisiro zvinofananidzwa sei pane ese magadzirirwo uye ekuongorora mamiriro.
Kana yakasimbiswa zvakanyanya, nzira yacho inogona kubatsira kune chip-dhizaini zvikwata zvinoshandisa simba rakawanda kugadzira domain-chaiyo heuristics yekuisa uye nzira. Inotaridzawo gwara rekutsvagisa rakafara: kushandisa mamodheru emitauro kupa mazano anogoneka ekugadzirisa tekinoroji pane kuabata semichinjiro chete. Mamiriro eruzhinji haana kunanga asi angangove akakosha nekuti kuvandudzwa kwekugadzirwa kwemuviri kunogona kukanganisa kushanda uye kunaka kwemachipi emangwana. Iyo sosi hairatidze kuti MacroAgent yakagamuchirwa mune zvekutengesa, yakashandiswa mune tapeout kana kuvandudza chigadzirwa chakaiswa. Hukoshi hunobvira hunoramba hwakasungirirwa kune kusimbiswa kupfuura zvakashumwa preprint zvawanikwa.
Interactive Mechanism: Iyo Inonyatsoshanda
Ongorora ari pasi tekinoroji kuseri kwekusimudzira uku uchipindirana.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
Zvekutarisa zvinotevera
Mibvunzo mikuru yakavhurika ndeyekuti iyo LLM-yakagadzirwa algorithms inogadzirwa uye kusarudzwa, yakawanda sei computation iyo maitiro inoda, uye kuti mhedzisiro yacho inowanza kune mamwe chip dhizaini uye kugadzira workflows. Kwainobva hakuzivise iyo LLM yakashandiswa, inopa mhedziso yekumisikidza kukosha kana kutsanangura yakagadzirwa chip mhedzisiro.
Mubvunzo wakakoshesesa wekuongorora ndeyekuberekazve. Iyo sosi inopa misoro yekuvandudza muzana asi haitaure hukuru hwebhenji, Hardware yakashandiswa, yakadzama yekumhanyisa kuyerwa, kusiyanisa kwenhamba kana iyo chaiyo-ye-iyo-maitiro masisitimu anosanganisirwa mukuenzanisa kwese. Izvo zvakare hazvitaure kana izvo zvakashumwa zviwanikwa zvinoenderana neyakagadzika dhizaini iyo inogona kusaoneka mune mamwe chip marongero. Izvo zvakasiirwa zvinoita kuti zviome kuongorora kuti zvinogara sei zvabuda zvichienda kune mamwe marongero ekuongorora.
Basa reLLM rinoda kunyatsoongororwa. Iyo abstract inotaura kuti modhi inowana akawanda anoshanda heuristic kugara-aware contour algorithms, asi haitsanangure kuti maalgorithms emumiriri anogadzirwa sei, akaedzwa, anosefa kana kusanganiswa. Izvo hazvizivikanwe zvakare kana iyo LLM inodiwa chete panguva yekuwanikwa kwegorgorithm kana kuti inoramba iri chikamu chekushanda kwekugadzira kuyerera. Idzo ruzivo rwaizosarudza iyo nzira yekuenzanisa mutengo, kuberekazve uye kukodzera kwekushandiswa kwekugadzira. Vaizojekesawo kuti ndezvipi zvikamu zvemaitiro zvinoda modhi uye zvikamu zvipi zvinogona kuita seyakajairwa dhizaini.
Humwe humbowo hungasanganisira ongororo yemamwe anowanikwa pachena uye magadzirirwo emaindasitiri, kuenzanisa neakasimba kana kuti ichangoburwa nheyo, uye kuzvimiririra kudzokororwa kwemigumisiro yeInnovus. Sosi yacho inoshuma kuvandudzwa pamusoro peDREAMPlace baseline uye mhedzisiro yakaparadzana kana yasanganiswa neInnovus, asi haitsanangure chip yakagadzirwa, dhizaini yekugadzira kana ongororo yakazvimirira. Iro bepa rinowanikwa searXiv preprint, saka zvakashumwa zvakawanikwa zvinofanirwa kubatwa sezvikumbiro kubva kuvanyori zvakamirira kusimbiswa kwakawanda. Kusimbiswa ikoko kwaizobatsira kuona kana kuvandudzwa kwakataurwa kuchidzokororwa uye kunoshanda zvakanyanya.