20184_ C˹(Artificial Intelligence, AI)oǽһƼaITԒ}˿ƼI^oͶ֮ڵȷIߌ˹ҲչFdȤIAIgPעҲԒPPIguλr_ʼЌH
첻zAƼ˞иIṩMQԼaھҲ_ʼ˹ҪC_OĠBɜy/\ࡢԭ/ԴʹàraƷƷ̵˹\(Arm)Ĺؔ(IP)cSoCِ`˼(Xilinx)ĬFɾT(FPGA)gtAƄaAIăɴ
AIoT.SENSELAgLcSL֖|̽ӑAAIIoTȄ˳µĽQߣ錣LժҪ
A gL
A SL ֖|
AIMٶ@ˡIÃrֵ^
AƼgLʾ˹WgоI}Ŀ֮lcиIVĿҪԭMٶ̫^@ˣIrֵֻWgо}Ŀ
ᘌIAIgҲߵķԡDeepmindµGo()õQAlphaԓҲձҽB@o˹ܷlչһҪ̱
ڿݻcNIrֵr˹ܳ鮔ǰĿĿƼh}⡣ӑՓwӑՓڸиI˹FaIܻ߀кܶ}Ҫ˷
˹
IfKaƷΣIÓxˡC@ԪָTCtָNߙC_ָNԭϼԴtƳ̷ԹIIaƷ׃Óx@Ԫع@ԪtIÿ춼Ҫn}
˹ҪKPIҪչF˹Ϸăcf׃ɿM˵UɢAIһҪĿ
^ҪFĴҪ߀IAI̶ԼѼ|ÉġI߱ҪAI_JR֪AIm̎}кΣAIՓģ͵ӖɹģͱOӋӖ|ҲҪ|}ӖģAIՓYFrF
MĻҲ{AI֮ǰaQ߶^AIKQmȻ߀ֻ{^ĸX^yӋW˸C֮gهPϵҪһrgȻAIҲҪmMAyĿɿȸʴ_
ͬ̎LArmܘmՓ\
Mһ˹ܿԷֳģӖ(Training)cՓ(Inference)ɂaFfDzӖõģиNՓֱ߅MģӖģӖҪ\Чܸ^mĻƶMС
ҲՓ\Чܵ^SFɵ̎Q܄x86 CPUGPUcArmܘSoC̎P\eֻڳɱcɢǷMFOҎơ
ͼgǶȶGPUĿǰmMģӖ̎ܘģՓծȻҲbb࣬GPUijɱĸS֮ɢᆖ}s@̎߅cFO䑪x86 CPUҲк\ЧܣܘOӋĿMN\/AI㷨rЧʲGPU
@}AIı|PAIֻͨNָһָ̎WeWj(Convolutional Neural Network, CNN)W^cfǾ\LD\ʮGPUȻ@샞x86 CPUtLָ(Multiple Instruction, Multiple Data, MIMD)\龳̫^rrͱlԶļ̼ܘ푪
þָ(RISC)Arm̎tGPUx86 CPU֮gϽArm̎һָ(Single Instruction, Multiple Data, SIMD)ЧܲAI\rӵ֡mȻĿǰҪArm̎ģӖЧ߀DzܸGPUȣՓՕrsǹġɱЧƽķ
¶A\к\aƷlչ{DҲһաδ\ᘌAI\ƳػЧʸõ̎ĸ߅IP@߅\AIõռǺܴAҲ\^mкĻPϵ
߅\Mչ١˹MvF
oץ˹ϷĴҪأA_ʼArmܘSoCِ`˼FPGAģKӲAԼҵau˹
AƼSL|ܱʾĿǰAaAIMAIfxԭʼYϵAڹIWrֻeaęC_aS^ĻAOҲɿ^Ҫ˹x@Щ䱳x]rЧЧ
A̎hǵ͵aBһMaƷҎһaкܴIJͬaĹҲs@ҲA˹rϣܽQʹc֮һ
|ܱʾڼgϵĿǰ߀oFȫϵyxԭʼKOĿ@AδŬķ
wfδAϣFĿһaOFϣеęC_Oֹ֧I4.0FɼcܛĽҪӵϵy(MES)aƷڹ(PLM)ϵyCҕXcWMһUƷhС
ᘌһc|ܲMFЙC_ͨҪrͣeҪԭSṩֻ֧Լָĵr^ijЩr£FЙC_^AԼ_lɼģK@
ڙCҕXWUĿǰAcԺ_lzyNͬaƷęCҕXϵyϣAʹùWәzy(AOI)кLһrgFеAOIϵyHm塢·MzymzyK˳Ʒ㲿
һ棬AaƷҲʹĿǰϵęCҕXҪAayĿǰϵęCҕX˴aƷzyOӋAmNaƷBęCҕXzyAQcԺ_lƻW㷨ԱCҕXϵyܸmͬBaƷ
FPGAģKFCҕX㷨
CҕXFPGAģKչֵ_ҲAFPGAlչFwɹĿ֮һ^FPGAģKAQЩӰRhҪӲҕXzyϵy\Ч
ָCPUGPUʹõӲоƬAIϵyЧՓҲһlе·^ĿǰAI㷨߀ڿMASICܿϼglչ_FPGAtЧcԵ\ԪYԿƻMض㷨ٵԣ㷨ҪĻµr_һwоƬֻҪOӋa
FAFPGA팍FAI㷨ٵ뷽֮һAҲFPGA_lFδ^mͶY@g