Iindaba zeShishini: Ukudibana kunye nokuthengwa kweshishini le-semiconductor kwihlabathi liphela kuyanda kwakhona

I-MCU+AI: umkhwa ongenakuphepheka

I-STMicroelectronics ifumene i-Deeplite, igxile kwi-AI edge

Ngo-Epreli kulo nyaka, i-STMicroelectronics (ST) ithenge inkampani entsha ye-AI yaseKhanada iDeeplite, eyatsala ingqalelo kushishino. Njengoko sonke sisazi, umngeni omkhulu ojongene neemodeli zokufunda nzulu ekusetyenzisweni kwezorhwebo sisikali sazo sokusebenza, iimfuno zeprosesa, kunye nobungakanani bokusetyenziswa kwamandla. I-Deeplite isombulula le ngxaki ngokubonelela ngenjini yesoftware ezenzekelayo yokuphucula iimodeli ze-DNN (deep neural network), ivumela i-AI ukuba yenze i-edge computing kuyo nayiphi na isixhobo.

Yasekwa ngo-2017, iDeeplite yaziwa ngesisombululo sayo se-AI esisecaleni iDeepSeek, igxile ekwenzeni ngcono, ukulinganisa kunye nokucinezela iimodeli ze-AI. I-optimizer yayo entsha eqhutywa yi-AI iNeutrino inokucinezela iimodeli ezinkulu zokufunda okunzulu ukuya kwishumi lobukhulu bazo bokuqala ngelixa igcina ukuchaneka okungaphezulu kwama-98%. Ngeteknoloji ezintathu eziphambili - ukuthena ubunzima (ukususa iiparameter ezingafunekiyo), ukulinganisa (ukunciphisa iimfuno zokuchaneka kokubala) kunye nokwandisa (ukwandisa umlinganiselo wobunzima obungenaxabiso), iimodeli ezinkulu ze-AI zinokusebenza ngokukhawuleza, ezincinci kwaye zisebenzise amandla amaninzi kwizixhobo ezisemacaleni. Izicelo ezazifuna amandla e-cloud computing ngaphambili ngoku zinokusebenza kakuhle kwizixhobo ezisemacaleni ezifana neekhamera zefowuni kunye neesensa zemizi-mveliso.

I-Deeplite itsale ingqalelo enkulu kwiintsuku zayo zokuqala kwaye ithiwe igama lokuba ngumvelisi we-AI ophambili nguGartner, Forbes, Inside AI, kunye ne-ARM AI. Olu kuthengwa lunxulumene kakhulu notshintsho olucwangcisiweyo lwe-STMicroelectronics ukuya kwi-AI edge, oludibanisa i-hardware kunye nesoftware ngendlela "ephindwe kabini". Itekhnoloji yokwenza ngcono imodeli ye-Deeplite idityaniswe ngokunzulu ne-STM32 series MCUs ze-STM32 kunye nee-NPU ezizinikeleyo ukuxhasa ukwakhiwa kwezisombululo ze-AI ezisuka ekupheleni ukuya ekupheleni. Umzekelo, kwiimeko zefektri ezikrelekrele, iikhamera ezixhotyiswe ngeetships ze-STMicroelectronics zinokubona ngokuthe ngqo iziphene ngaphandle kokulayisha idatha kwilifu, kwaye isantya sempendulo sinyuswa ngama-40 amaxesha.

Kwelinye icala, iDeeplite ineqela leenjineli ze-algorithm ze-AI ezikumgangatho wehlabathi, apho i-ST iza kudibanisa izixhobo zophuhliso ze-AI ezingaphezulu kwama-200 ukwenza inkqubo yophuhliso edibeneyo "yeqonga lelayibrari-optimizer-hardware". Ngamafutshane, ukufunyanwa kweDeeplite akupheleli nje ekugqibezeleni iqhekeza lokugqibela lephazili ye-ST kwinqanaba lesoftware ye-AI, kodwa kukwaphawula utshintsho oluqhelekileyo kwishishini le-semiconductor ukusuka "ekwenzeni iitships" ukuya "ekwenzeni iingqondo".

I-NXP ithenge inkampani ye-NPU iKinara ukuze itshintshe i-smart edge

NgoFebruwari kulo nyaka, i-NXP ibhengeze ukuthengwa kwenkampani entsha ye-US edge AI chip iKinara ngemali eyi-$307 yezigidi. IKinara yasekwa ngo-2013 kwaye ekuqaleni yabizwa ngokuba yiCore Viz, kamva yabizwa ngokuba yiDeep Vision, yaza yabizwa ngokuba yiKinara ngo-2022. I-NPU yeKinara ehlukeneyo (kubandakanya i-Ara-1 kunye ne-Ara-2) ikhokela ishishini ekusebenzeni kakuhle nasekusebenziseni amandla, okwenza ukuba ibe sisisombululo esikhethwayo kwizicelo ze-AI ezisakhulayo eziqhutywa ngumbono, ilizwi, isenzo kunye nezinye iindlela ezahlukeneyo zokuphumeza i-AI, kwaye ukucwangciswa kwayo kuqinisekisa ukuba inokuziqhelanisa ne-algorithms ye-AI eguqukayo.

I-NXP ithe oku kuthengwa kuya kudibanisa i-NPU ezimeleyo yeKinara kunye nepotifoliyo yayo yesoftware yeprosesa, uqhagamshelo kunye nokhuseleko, eya kunceda ekuboneleleni ngeqonga le-AI elipheleleyo nelinokwandiswa ukusuka kwi-TinyML ukuya kwi-AI evelisayo ukuhlangabezana neemfuno ze-AI ezikhula ngokukhawuleza kwiimarike zoshishino kunye neemoto. Oku kuya kunceda ekudaleni iinkqubo ezintsha eziqhutywa yi-AI kwiinkalo zoshishino kunye ne-IoT, kuncede abathengi ukuba benze lula ubunzima, bakhawulezise ixesha lokuthengisa, kwaye baphucule izakhono zobugcisa kwiindawo ezifana neemoto ezikrelekrele, besiya kwiindawo ezinexabiso eliphezulu.

I-Edge AI: Indawo yokulwa yabavelisi be-MCU

Kudala kukho ingcamango ephosakeleyo kwicandelo lobukrelekrele bokwenziwa yokuba "isikali ngamandla". Nangona iimodeli ezinkulu zisebenza kakuhle, zijongene nemingeni ekusetyenzisweni kwazo - ukusetyenziswa kwazo kwamandla aphezulu kuchasene neemfuno ezilula kwicala lomphetho. Iingcali zoshishino ziye zaphawula ngokuphindaphindiweyo imida yendalo yeemeko zokusetyenziswa kweemodeli ezinkulu: kwelinye icala, ukuqeqesha nokusebenzisa iimodeli ezinkulu kufuna izixhobo ezinkulu zekhompyutha; kwelinye icala, iindawo eziphambili zokukhuthaza uphuhliso lwezentengiselwano zobukrelekrele bokwenziwa zizixhobo zekhompyutha ezisemgangathweni kunye nezixhobo zesiphelo ezibuthathaka ngakumbi ekusetyenzisweni kwamandla kunye nokulibaziseka.

Akunzima ukuqonda ukuba oku kufunyenweyo apha ngasentla kubonisa ukuba inkundla ephambili ye-MCU itshintshela kwi-edge AI computing. Kulindeleke ukuba ngo-2025, i-75% yedatha iya kucutshungulwa emphethweni, nto leyo egxininisa amandla amakhulu emarike ye-edge AI MCU. Oku kubonisa ukuba imfuno ye-edge AI computing ikhula ngokukhawuleza, kwaye i-MCU, njengecandelo eliphambili lezixhobo ze-edge, iya kudlala indima ephambili kule ndlela.

Kwixesha elizayo, ii-MCU azisayi kuphelela kwimisebenzi yokulawula yendabuko, kodwa ziya kudibanisa kancinci kancinci amandla okuqiqa e-AI kwaye zisetyenziswe kwiimeko ezifana nokuqondwa kwemifanekiso, ukucubungula ilizwi, kunye nokugcinwa kwangaphambili kwezixhobo. Ii-MCU ezinezakhono zokubala ezikwi-edge ziya kuba ngumthwali obalulekileyo wamandla e-edge computing ngokusebenzisa amandla aphantsi, ukusebenza kakuhle, kunye nokuphendula kwangoko, zibonelela ngenkxaso enamandla kwizixhobo kunye neenkqubo ezikrelekrele.

Abanye abavelisi abakhulu be-MCU nabo basebenza nzima ekufumaneni nasekukhuphisaneni kweli candelo, njengokuthengwa kweRenesas Electronics kweReality AI, ukuthengwa kwe-Infineon kwe-Imagimob yaseSweden, kunye nokuqaliswa kwe-NXP kwesoftware yokufunda koomatshini i-eIQ kunye ne-AI tool chain NANO.

Kungagqitywa kwelokuba i-edge AI iya kuba yindawo ebalulekileyo yokulwa nee-MCU kwiminyaka embalwa ezayo.

Izixhobo ze-elektroniki zeemoto: ingqwalasela yokhuphiswano lwemali

Kutshanje, ukudibana kwee-semiconductor kunye nokuthengwa kwee-semiconductor ezinxulumene nezicelo zeemoto kuye kwavela rhoqo. Ukongeza kumandla ekhompyutha, ukuvela kwamandla eemoto, uqhagamshelo lwenethiwekhi yemoto, isandi semoto kunye nezinye iitekhnoloji kukwaqhubele phambili ukuhlaziywa kunye nokuhlaziywa kwetekhnoloji ye-semiconductor, oku kukhokelela iinkampani ezinxulumene noko ukuba zongeze uyilo lwazo lwetekhnoloji ngokudibanisa kunye nokuthengwa kweemoto.

Ishishini le-semiconductor lishishini eliqhelekileyo elisebenzisa ubuchwepheshe kakhulu kwaye lisebenzisa imali eninzi. Xa sijonga emva kwiminyaka embalwa edlulileyo, ukuhlanganiswa kunye nokudibana kube yinto engenakuphepheka ekuphuhlisweni kweshishini.

Iinkampani ezinkulu ze-AI zihlala zithenga izinto ngeenjongo zokuphucula uyilo lwazo lwetekhnoloji kwaye zakhe inzuzo epheleleyo "ye-chip + inkqubo + i-ecosystem". Abavelisi be-MCU abaphambili batshintsha kancinci kancinci babe yi-AI, bezama ukubamba imakethi ye-smart terminal ngokusetyenziswa kwamandla aphantsi kunye nokuguquguquka okuphezulu. Kwicandelo leemoto, ukubalwa kwekhompyutha ngaphakathi kwimoto, ukuqhuba okuzimelayo kunye nokunxibelelana kwedatha kuye kwaba ziinkalo eziphambili zokhuphiswano lwemali. Kwangaxeshanye, ishishini le-EDA liyatshintsha ukusuka ekuboneleleni ngezixhobo ukuya ekwakheni i-ecosystem. Iinkampani ezinkulu zidibanisa iinkqubo ze-IP kunye noyilo, kwaye zakha ulawulo lwemarike ngoyilo "lwezixhobo-uyilo-olusemgangathweni".

Kule nkqubo yokudibana nokuthengwa kweenkampani, intsebenziswano yetekhnoloji, ukwanda kweemarike kunye nolawulo lwenkqubo yendalo ziye zaba yingcamango ephambili. Iinkampani kufuneka zilinganisele ukuhlanganiswa kwexesha elifutshane kunye nophando lwexesha elide kunye nophuhliso phakathi kokungena kwemali. Ngenxa yemiqobo yetekhnoloji kunye nohlobo lweshishini le-semiconductor olusebenzisa imali eninzi, olu tshintsho aluyondlela imfutshane kodwa yi-"marathon" efuna utyalo-mali lwexesha elide.