isiZulu
Indlela Amakhamera eNeural Azosusa Ngayo Ama-Module e-CMOS Ajwayelekile: Ukushintsha Kwepharamitha Ekuthwebuleni
Kwadalwa ngo 2025.12.29
Isingeniso: Ukuphela Kokubusa Kwe-CMOS Akuvumi—Kukhona Lapha
Lapho imoto ezishayelayo iphuthelwa umgibeli ekukhanyeni okuphansi noma i-microscope ingakwazi ukulandela ama-spike e-neural ngesikhathi sangempela, umphumela akuwona nje umkhawulo wezinsiza—kuyisimo sokuthwebula esineminyaka engama-30. Traditional Amamojula e-CMOS, i-backbone ye-ejensi yedijithali yanamuhla, yakhelwe umhlaba lapho “kuhle okwanele” kusho ukuthwebula izithombe ngezikhathi ezithile. Kodwa njengoba imboni ifuna izinhlelo zokubona ezisheshayo, ezihlakaniphile, nezisebenza kahle, izithiyo zokwakhiwa ze-CMOS zibe nzima ukuhamba phambili. Ngena ezithombeni ze-neural: ama-sensors akhuthazwe yizinto eziphilayo angakrecordi kuphela ukukhanya—aphinde akuchaze. Lokhu akusikho ukuphuculwa okuncane; kuyisithombe esisha ngokuphelele sokuthi sithwebula kanjani idatha yokubona. Ngonyaka ka-2030, ochwepheshe balindele ukuthi izithombe ze-neural zizothatha u-45% wemakethe yokuthwebula ephezulu, kusukela ezimotweni ezizimele kuya ezivivinyweni zezokwelapha. Nansi indlela—nokuthi—zishintsha kanjani ama-modules e-CMOS ngokuphelele.
Iphutha Elifihlekile ku-CMOS: Lakhiwe Ngempumelelo Ephukile
Ngeminyaka engamashumi amabili, abakhiqizi be-CMOS baphonse izinjongo ezimbili eziphikisanayo: isixazululo esiphezulu kanye nezinga elisheshayo lokuhamba. I-CMOS ehlanganisiwe (ukuphuma kwamuva, esetshenziswa kumafoni aphakeme njenge-iPhone 15 Pro) izame ukuxazulula lokhu nge-TSV (Through Silicon Via) ubuchwepheshe, ihlukanisa izingqimba ze-pixel ezivela kumasistimu okucabanga ukuze ikhulise ibhendi. Kodwa le ndlela yokulungisa ibhande yadala izinkinga ezintsha: ama-TSV asebenza njengemithombo yokushisa, ekhuphula izinga lokushisa le-pixel nokwandisa umsindo. Kubuhlungu, i-CMOS ehlanganisiwe isaqhubeka nokulandela imodeli "esekelwe kumafreyimu"—i-pixel ngayinye ithatha ukukhanya isikhathi esifanayo, iphoqa ukwehlukaniswa phakathi kwesivinini nesilinganiso sokuphuma kumasignali (SNR).
Cabanga ngocwaningo lwezengqondo olufundisa ukusebenza kobuchopho: ukuze balandele ama-voltage spikes akhulu kumasekhondi, badinga amaframu angaphezu kuka-1,000 ngaleso sikhathi. Kodwa ama-sensor e-CMOS ngaleso sikhathi athola ukukhanya okuncane kangangokuthi izimpawu zikhishwa yizwi. Ngakolunye uhlangothi, ukuvezwa okude ukuze kutholakale i-SNR engcono kubhidliza izinjongo ezihamba ngokushesha. Lokhu akuyona iphutha ku-CMOS—kuyisici sokwakhiwa kwaso. Njengoba umcwaningi we-MIT uMatthew Wilson ebeka kanje: “Ukuvezwa kwe-CMOS okungafaneleki kuzo zonke kuyisithiyo esiyisisekelo uma uzama ukuveza izigcawu ezishintshashintshayo, eziyinkimbinkimbi.”
Amanye amaphutha ajulile:
• Ukuphindaphinda Kwedatha: I-CMOS iqopha iphiksela ngalinye kuwo wonke amafreyimu, ngisho nezizinda ezimile, iphula u-80% wezinsiza.
• Imikhawulo ye-Dynamic Range: I-CMOS ejwayelekile iphakama ku-80–100 dB, iphumelela ezindaweni eziphakeme zokuphikisana (isb. ukushona kwelanga phezu kwehlathi).
• Ukulibaziseka: Ukuguqula izimpawu zokukhanya ezihambisanayo zibe idatha ye-digital nokuthumela ku-processor kudala ukuhamba kancane—okubulalayo ezinhlelweni ezifana nokushayela okuzimele.
Lezi akuzona izinkinga ezingalungiswa ngokukhiqiza okungcono. I-CMOS iyisigameko sokwakhiwa kwayo. Amakhamera e-neural, ngokuphambene, akhiwe ukuze akhiphe lezi zinkinga.
Amakhamera e-Neural: Izinguquko Ezintathu Eziphumelelayo
Amakhamera e-neural athola ugqozi kwi-retina yomuntu, ethumela izimpawu kuphela uma ukukhanya kushintsha—akukho idatha eyengeziwe, akukho zikhathi ezinqunyiwe zokukhanya. Nansi indlela ababhala ngayo imithetho kabusha:
1. Ama-Pixels Angalungiseka: I-Pixel Ngayinye Isebenza Ngomsebenzi Wayo
Ukuqhamuka okukhulu kuvela ku-intelligence ye-pixel-level. I-MIT’s Programmable Exposure CMOS (PE-CMOS) sensor, eyethulwe ngo-2024, ivumela ukuthi i-pixel ngayinye ibeke isikhathi sayo sokukhanya ngokuzimela. Ngokusebenzisa ama-transistor ayisithupha kuphela nge-pixel (ukuhlanganiswa kwemiklamo yangaphambilini), ama-pixel aseduze angasiza omunye nomunye: ama-pixel anokukhanya okusheshayo alandelela ukuhamba okusheshayo (isb., ama-neural spikes), kanti ama-pixel anokukhanya okuphuthumayo abamba imininingwane ezindaweni ezimnyama—konke kulolu hlobo olufanayo.
Ezihlolweni, i-PE-CMOS ifinyelele isixazululo se-single-spike ku-neural imaging, umphumela i-CMOS engakufinyeleli ngaphandle kokuphonsa isivinini. “Asikabambi nje ukukhanya—siyathuthukisa indlela i-pixel ngayinye ethintana nayo,” kuchaza umcwaningi ophambili u-Jie Zhang. Le ndlela yokusebenza ikhipha ukushintshashintsha kwesivinini-SNR okuphazamisayo i-CMOS.
2. Ukuthwebula Okushayelwa Yizenzakalo: Idatha kuphela lapho ibalulekile
Amakhamera emicimbi (uhlobo lwamakhamera e-neural) athatha lokhu phambili: akhiqiza idatha kuphela uma iphiksela ibona ushintsho ekukhanyeni. Esikhundleni samafreyimu, akhipha "imikhosi"—amaphakethe amancane edatha anemikhawulo, izikhathi, kanye ne-polarity (ukukhanya kukhula noma kwehla).
Imiphumela iyashintsha:
• 120+ dB Ububanzi Obuphakeme: Amakhamera ezenzakalo aphatha ukukhanya kwelanga ngqo kanye nezithunzi ezimnyama ngesikhathi esifanayo.
• Microsecond Latency: Akukho i-frame buffer kusho ukuthi kukhona ukuphuma kwedatha okusheshayo—kubalulekile kumakhiya azimele agwema ukuhlinzwa.
• 90% Ukuncipha Kwedatha: Ngokungabheki izigcawu ezimile, amakhamera ezenzakalo anciphisa izidingo zebhendi, ehla ukusetshenziswa kwamandla ngama-70% uma kuqhathaniswa ne-CMOS.
Abacwaningi be-Indian Institute of Science basebenzise ikhamera yezenzakalo ye-iniVation ukuze bathwebule ama-nanoparticles amancane kune-50 nanometers—ngaphezu komkhawulo wokuphazamiseka kwamakhamera ajwayelekile. Umjikelezo wedatha ophumayo wekhamera uvumele ama-algorithms e-AI ukuthi agxile kumasignali anencazelo, aguqule umsindo ube ulwazi olusetshenziswayo.
3. Ku-Sensor AI: Ukucubungula, hhayi nje Ukuthwebula
Ngokwe-CMOS, ethanda ukuncika kumaprosesa angaphandle ukuze ahlaziye izithombe, amakhamera e-neural ahlanganisa i-AI ngqo ku-sensor. Ama-sensor amasha e-Samsung asevele afaka ama-modules e-AI ayisisekelo wokunciphisa umsindo, kodwa amakhamera e-neural athatha lokhu kube sezingeni elisha: ab processing idatha njengoba ithathwa.
Isibonelo, i-Metavision sensor ye-Prophesee isebenzisa amanethiwekhi e-neural ku-chip ukuze ithole izinto ngesikhathi sangempela, ithumela kuphela idatha efanele kumaprosesa amakhulu. Ekubhekeni kwezimboni, lokhu kusho ukuthola amaphutha emgqeni wokukhiqiza ngaphandle kokugcina ama-terabytes wezithombe ezingasizi. "Amakhamera e-neural awasiyo nje ama-sensor ezithombe—ngokuyinhloko, yizinjini zokubona," kusho u-Chetan Singh Thakur, umbhali-mbumbulu wesifundo se-nanotechnology.
Izikh replacements zeMvelo: Lapho Amakhamera e-Neural Esevele Ewinile
Ukushintsha kusuka ku-CMOS kuya kumakhamera e-neural akukhona okucabanga—kwenzeka namuhla, kuqala ngezicelo ezinomthelela omkhulu lapho amaphutha e-CMOS ebiza kakhulu:
I-Neuroscience & Ukuthwebula Izithombe Zokwelapha
I-PE-CMOS ye-MIT isivele isetshenziswa ukulandela imisebenzi ye-neural ezilwaneni ezihambayo, okungukuthi i-CMOS ayikwazi ukukwenza ngaphandle kokuphazamiseka noma umsindo. Ku-endoscopy, izithombe ze-event cameras ezinezinga eliphansi lokubambelela nesikhala esikhulu se-dynamic zivumela odokotela ukuba babone ngaphakathi emzimbeni ngaphandle kokukhanya okukhulu, kunciphisa ukungakhululeki kwabaguli.
Izimoto Ezizimele
I-Tesla ne-Waymo zihlola izithombe ze-event cameras kanye ne-CMOS ukuze ziqede izindawo ezingenakubona futhi zinciphise isikhathi sokuphendula. Ikhamera ye-neural ingakwazi ukubona ingane egijimayo emgwaqweni ngokushesha okuphindwe kabili kune-CMOS, okungavimbela izingozi.
I-Nanotechnology & Isayensi Yezinto
I-microscope ye-neuromorphic ye-IISc manje isithengiswa, ivumela abacwaningi ukuba bahlolisise ukuhamba kwemolekyuli ngokuqonda okungakaze kubonwe. Lokhu akusikho kuphela ukuthuthukiswa—kuyithuluzi elisha elwandisa lokho okungenzeka ocwaningweni lwezesayensi.
Izinto Zokusetshenziswa Kwabathengi (Isiteshi Esilandelayo)
Ngenkathi amakhamera e-neural manje ephakeme kakhulu kune-CMOS, izindleko ziya phansi. Ukuklama kwe-pixel okulula kwe-MIT kunciphisa ubunzima bokukhiqiza, futhi ukukhiqizwa okukhulu kuzosheshisa izindleko ziye ezingeni le-CMOS ngonyaka ka-2027. Amafoni aphakeme kungenzeka athathe izinhlelo ezihlanganisiwe kuqala—amakhamera e-neural ukuze kuqoshwe ividiyo nangesibani esiphansi, i-CMOS ukuze kuqoshwe izithombe—ngaphambi kokuthi ashintshe ngokuphelele i-CMOS ngonyaka ka-2030.
Indlela Yokushintsha: Ukuthuthukiswa, hhayi Uguquko
Amakhamera e-neural awasoze asuse i-CMOS ngasikhathi sinye. Ukushintsha kuzolandela izigaba ezintathu:
1. Ukusetshenziswa Okuphumelelayo (2024–2026): Amakhamera e-neural akhulisa i-CMOS ezinhlelweni eziphezulu zokusebenza (isb. izimoto ezizimele, imifanekiso yesayensi).
2. Ukushintshwa Okukhethwayo (2026–2028): Njengoba izindleko ziya phansi, amakhamera e-neural azothatha izimakethe ezikhethekile zabathengi (isb. amakhamera ezenzakalelayo, ukuthwebula izithombe nge-drone) lapho isivinini nokusebenza kahle kokukhanya okuphansi kubaluleke kakhulu.
3. Ukubusa Okukhulu (2028–2030): Amakhamera e-neural aba yizinto ezijwayelekile kumaselula, kumalaptop, nasezisetshenzisweni ze-IoT, kanti i-CMOS iyagcinwa ezimakethe eziphansi.
Le ndlela ibonisa ushintsho oluvela ku-CCD luya ku-CMOS eminyakeni yama-2000—olwenziwa ukusebenza, hhayi kuphela izindleko. "I-CMOS ithathe indawo ye-CCD ngoba yayinokuguquguquka okwengeziwe," kuchaza umhlaziyi wezomnotho uSarah Chen. "Amakhamera e-neural athatha indawo ye-CMOS ngenxa yesizathu esifanayo: ahambelana nesimo, hhayi ngendlela ehlukile."
Izinselelo Zokudlula
Naphezu kwesithembiso sabo, amakhamera e-neural abhekene nezithiyo:
• Izindinganiso Zezimboni: Akukho phrothokholi evamile yokwenzeka kwedatha okwenza kube nezinkinga zokuhambisana phakathi kwezinsiza nezinhlelo zokusebenza.
• Ukuthinteka Kwamalanga Aphansi: Ngenkathi amakhamera ezikhathi ethokozisa ekuphikisaneni, awakwazi kahle emdarkness edlulele—nokho ucwaningo e-MIT lukhuluma ngalokhu ngokuthuthukiswa kwe-photodiodes.
• Ukuqonda Okungafanele: I-AI ku-sensor ingafaka ukungafanele uma ingaqeqeshwanga kahle, ingozi ezinhlelweni ezibalulekile zokuphepha.
Lezi zinkinga zingaxazululwa. Ama-consortium afana ne-IEEE athuthukisa izindinganiso zamakhamera ezikhathi, futhi ama-startups afaka imali ekuthuthukiseni ukukhanya okuphansi. Ibhari enkulu akuyona ubuchwepheshe—kuyindlela yokucabanga: abakhiqizi nabaqambi badinga ukujolisa emhlabeni lapho amakhamera engathathi kuphela izithombe, kodwa aqonda lokho abukayo.
Isiphetho: Ikusasa Lokuthwebula Izithombe Li-Neural
Amamojula e-CMOS ajwayelekile aguqule ukufotografia ngokwenza amakhamera edijithali atholakale. Kodwa abavumelani nomqondo ophathelene nezithombe obekungakwazi ukuhamba kahle nezidingo ze-AI, ukuzimela, kanye nocwaningo lwezesayensi. Amakhamera e-neural awathuthukisi kuphela i-CMOS—aphinde achaze ukuthi yini engaba yisithonjana.
Ngokuhlanganisa ama-pixels angalungiseka, idatha eqhutshwa yimicimbi, kanye ne-AI esithombeni, amakhamera e-neural akhipha izinkinga ezibambe imifanekiso emuva kweminyaka. Asheshisa, aqonda kahle, futhi asebenza kahle, futhi asevele athatha indawo ye-CMOS ezinhlelweni ezibaluleke kakhulu. Njengoba izindleko zehla futhi ubuchwepheshe bukhula, amakhamera e-neural azoba yinto ejwayelekile njengokuthi i-CMOS namuhla—aguqula hhayi kuphela indlela esithatha ngayo izithombe, kodwa nendlela esisebenzisana ngayo nezwe.
Umbuzo akuwona ukuthi amakhamera e-neural azoshintsha i-CMOS—kukhona indlela esheshayo yokuthi uzamukele. Kubhizinisi, impendulo ingasho ukuhamba phambili kokuncintisana. Kubathengi, kusho izithombe ezingcono, izimoto eziphephile, nezobuchwepheshe esingakaze sicinge ngazo. Ikusasa lokuthwebula izithombe liwukuthi i-neural—futhi lifika ngokushesha kunalokho ocabanga.
Uxhumane
Sicela uxhumane nathi uhambele
WhatsApp
WeChat