Izindaba Zemboni: Ukuhlanganiswa kanye nokuthengwa kwezimboni ze-semiconductor zomhlaba wonke kuyanda futhi

I-MCU+AI: umkhuba ongenakugwenywa

I-STMicroelectronics ithola i-Deeplite, igxile ku-AI enqenqemeni

Ngo-Ephreli kulo nyaka, i-STMicroelectronics (ST) ithenge inkampani entsha ye-AI yaseCanada i-Deeplite, eyadonsela ukunaka embonini. Njengoba sonke sazi, inselele enkulu ebhekene namamodeli okufunda okujulile ekusetshenzisweni kwezentengiselwano yisilinganiso sawo sokusebenza, izidingo zeprosesa, kanye nokusetshenziswa kwamandla ngamandla. I-Deeplite ixazulula le nkinga ngokunikeza injini yesofthiwe ezenzakalelayo ukuze ithuthukise amamodeli e-DNN (inethiwekhi ye-neural ejulile), okuvumela i-AI ukuthi yenze i-edge computing kunoma iyiphi idivayisi.

Yasungulwa ngo-2017, i-Deeplite yaziwa ngesisombululo sayo se-AI esinqenqemeni i-DeepSeek, igxile ekwenzeni ngcono, ukulinganisa kanye nokucindezelwa kwamamodeli e-AI. I-optimizer yayo entsha eqhutshwa yi-AI i-Neutrino ingacindezela amamodeli amakhulu okufunda okujulile abe yingxenye yeshumi yobukhulu bawo bokuqala ngenkathi igcina ukunemba okungaphezu kuka-98%. Ngobuchwepheshe obuthathu obubalulekile - ukunqunywa kwesisindo (ukususa amapharamitha angafuneki), ukulinganisa (ukunciphisa izidingo zokunemba kokubala) kanye nokuhlukanisa (ukwandisa isilinganiso sezisindo ezingenalo inani), amamodeli amakhulu e-AI angasebenza ngokushesha, amancane futhi ngendlela ephumelelayo kumadivayisi anqenqemeni. Izinhlelo zokusebenza ezazidinga amakhono e-cloud computing ngaphambilini manje zingasebenza kahle kumadivayisi anqenqemeni njengamakhamera e-smartphone kanye nezinzwa zezimboni.

I-Deeplite idonsele ukunaka okukhulu ezinsukwini zayo zokuqala futhi ibizwe ngokuthi umsunguli we-AI ohamba phambili nguGartner, Forbes, Inside AI, kanye ne-ARM AI. Lokhu kutholwa kuhlobene kakhulu nokuguqulwa kwesu kwe-STMicroelectronics kube yi-AI edge, okuhlanganisa ihadiwe nesofthiwe ngendlela "ephindwe kabili". Ubuchwepheshe bokwenza ngcono imodeli ye-Deeplite buhlanganiswe ngokujulile nama-STM32 series MCU e-STM32 e-STMicroelectronics kanye nama-NPU azinikele ukusekela ukwakhiwa kwezixazululo ze-AI ezisuka komunye kuya komunye. Isibonelo, ezimweni ezihlakaniphile zasefektri, amakhamera ahlonyiswe ngama-chip e-STMicroelectronics angathola ngqo amaphutha ngaphandle kokulayisha idatha efwini, futhi isivinini sokuphendula sanda izikhathi ezingu-40.

Ngakolunye uhlangothi, i-Deeplite inethimba lomhlaba wonke lonjiniyela be-algorithm ye-AI, lapho i-ST izohlanganisa khona amathuluzi okuthuthukisa i-AI angaphezu kwama-200 ukwakha uhlelo oluhlanganisiwe lokuthuthukiswa "lwepulatifomu ye-library-optimizer-hardware". Ngamafuphi, ukutholwa kwe-Deeplite akugcini nje ngokuqeda ingxenye yokugcina yephazili ye-ST ezingeni lesofthiwe ye-AI, kodwa futhi kuphawula ushintsho olujwayelekile lwemboni ye-semiconductor kusukela "ekwenzeni ama-chip" kuya "ekwenzeni ubuchopho".

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

NgoFebhuwari kulo nyaka, i-NXP yamemezela ukuthengwa kwenkampani entsha ye-US edge AI chip iKinara ngemali engu-$307 million. IKinara yasungulwa ngo-2013 futhi ekuqaleni yaqanjwa ngokuthi iCore Viz, kamuva yaqanjwa kabusha ngokuthi iDeep Vision, futhi yaqanjwa kabusha ngokuthi iKinara ngo-2022. I-NPU yeKinara ehlukanisiwe (kufaka phakathi i-Ara-1 ne-Ara-2) ihola imboni ekusebenzeni kahle nasekusebenziseni kahle amandla, okwenza kube yisisombululo esithandwayo sezinhlelo zokusebenza ze-AI ezisafufusa eziqhutshwa umbono, izwi, isenzo kanye nezinye izinhlelo zokusebenza ze-AI ezahlukahlukene, futhi ukuhleleka kwayo kuqinisekisa ukuthi ingazivumelanisa nama-algorithms e-AI ashintshashintshayo.

I-NXP ithe lokhu kuthengwa kuzohlanganisa i-NPU ezimele yeKinara nephothifoliyo yayo yesofthiwe yokucubungula, yokuxhumana kanye neyokuphepha, okuzosiza ekuhlinzekeni ngeplatifomu ye-AI ephelele nengandiswa kusukela ku-TinyML kuya ku-AI ekhiqizayo ukuze kuhlangatshezwane nezidingo ze-AI ezikhula ngokushesha zezimakethe zezimboni nezezimoto. Lokhu kuzosiza ekudaleni izinhlelo ezintsha eziqhutshwa yi-AI emikhakheni yezimboni neye-IoT, kusize amakhasimende ukuthi enze kube lula ukuba yinkimbinkimbi, kusheshiswe isikhathi sokumaketha, futhi kuthuthukiswe amakhono obuchwepheshe ezindaweni ezifana nezimoto ezihlakaniphile, kuqhubekele emasimini anenani eliphezulu.

I-Edge AI: Indawo Yokulwela Abakhiqizi be-MCU

Sekuyisikhathi eside kunombono ongalungile emkhakheni wobuhlakani bokwenziwa wokuthi "isikali singamandla". Nakuba amamodeli amakhulu esebenza kahle kakhulu, abhekene nezinselele ekusetshenzisweni kwangempela - ukusetshenziswa kwawo kwamandla aphezulu kuphambene nezidingo ezilula ohlangothini lomphetho. Ochwepheshe bemboni baveze ngokuphindaphindiwe imikhawulo engokwemvelo yezimo zokusetshenziswa kwamamodeli amakhulu: ngakolunye uhlangothi, ukuqeqesha nokusebenzisa amamodeli amakhulu kudinga izinsizakusebenza ezinkulu zekhompyutha; ngakolunye uhlangothi, izindawo ezibalulekile zokukhuthaza ukuthuthukiswa kwezimboni zobuhlakani bokwenziwa yi-edge computing kanye namadivayisi e-terminal azwela kakhulu ekusetshenzisweni kwamandla kanye nokubambezeleka.

Akunzima ukuqonda ukuthi ukuthengwa okungenhla kukhombisa ukuthi inkundla yempi eyinhloko ye-MCU ishintshela ekubeni yi-edge AI computing. Kulindeleke ukuthi ngo-2025, u-75% wedatha uzocutshungulwa onqenqemeni, okugqamisa amandla amakhulu emakethe ye-edge AI MCU. Lokhu kukhombisa ukuthi isidingo sokuba yi-edge AI computing sikhula ngokushesha, kanti i-MCU, njengengxenye eyinhloko yamadivayisi e-edge, izodlala indima ebalulekile kulo mkhuba.

Esikhathini esizayo, ama-MCU ngeke esakwazi ukukhawulelwa emisebenzini yokulawula yendabuko, kodwa azohlanganisa kancane kancane amakhono okucabanga e-AI futhi asetshenziswe ezimweni ezifana nokuqashelwa kwezithombe, ukucubungula izwi, kanye nokugcinwa kwemishini kusengaphambili. Ama-MCU anamakhono e-edge computing azoba ngumthwali obalulekile wamandla e-edge computing ngokusetshenziswa kwawo kwamandla aphansi, ukusebenza kahle okuphezulu, kanye nokuphendula okusheshayo, enikeza ukwesekwa okuqinile kwamadivayisi nezinhlelo ezihlakaniphile.

Abanye abakhiqizi abakhulu be-MCU nabo bathenga futhi bancintisana ngenkuthalo kulo mkhakha, njengokuthengwa kwe-Reality AI yi-Renesas Electronics, ukuthengwa kwe-Imagimob yi-Infineon yaseSweden, kanye nokwethulwa kwe-NXP kwesoftware yokufunda komshini i-eIQ kanye ne-AI tool chain NANO.

Kungaphethwa ngokuthi i-edge AI izoba yindawo yokulwa ebalulekile yama-MCU eminyakeni embalwa ezayo.

I-elekthronikhi yezimoto: ukugxila komncintiswano wezimali

Muva nje, ukuhlanganiswa kwe-semiconductor kanye nokuthengwa okuhlobene nezinhlelo zokusebenza zezimoto kuye kwavela kaningi. Ngaphezu kwamandla ekhompyutha, ukuvela kwamandla ezimoto, ukuxhumana kwenethiwekhi emotweni, umsindo emotweni kanye nobunye ubuchwepheshe nakho kuye kwaqhuba ukuphindaphinda kanye nokuvuselelwa kobuchwepheshe be-semiconductor, okwenza izinkampani ezihlobene ukuthi zengeze ukwakheka kwazo kobuchwepheshe ngokuhlanganiswa kanye nokuthengwa.

Imboni ye-semiconductor iyimboni ejwayelekile edinga ubuchwepheshe kanye nemali eningi. Uma sibheka emuva emashumini ambalwa eminyaka edlule, ukuhlanganiswa kanye nokuhlanganiswa kube yinto engenakugwenywa ekuthuthukisweni kwemboni.

Ama-giant e-AI avame ukuthenga ngomzamo wokuthuthukisa ukwakheka kwawo kobuchwepheshe nokwakha inzuzo egcwele "ye-chip + system + ecosystem". Abakhiqizi be-MCU abajwayelekile bashintsha kancane kancane baba yi-AI edge, bezama ukuthatha imakethe ye-smart terminal ngokusetshenziswa kwamandla aphansi kanye nokuguquguquka okuphezulu. Emkhakheni wezimoto, ukubala ngaphakathi kwezimoto, ukushayela okuzenzakalelayo kanye nokuxhumana kwedatha kube yizindawo ezibalulekile zokuncintisana ngemali. Ngesikhathi esifanayo, imboni ye-EDA ishintsha kusukela ekuhlinzekeni ngamathuluzi kuya ekwakheni i-ecosystem. Ama-giant ahlanganisa izinqubo ze-IP kanye nokuklama, futhi akha ukubusa kwemakethe ngokwakhiwa "kwethuluzi-ukwakheka-okujwayelekile".

Kuleli gagasi lokuhlanganiswa nokuthengwa kwezinkampani, ukubambisana kwezobuchwepheshe, ukwanda kwemakethe kanye nokubusa kwezinhlelo zemvelo sekuyinto eyinhloko. Izinkampani zidinga ukulinganisela ukuhlanganiswa kwesikhathi esifushane kanye nocwaningo lwesikhathi eside kanye nentuthuko phakathi kokungena kwemali. Njengoba kunezithiyo zobuchwepheshe kanye nesimo sezimali esidinga kakhulu embonini ye-semiconductor, lolu shintsho aluyona "indlela emfushane" kodwa "i-marathon" edinga ukutshalwa kwezimali kwesikhathi eside.