Artificial intelligence agitation sweeps across the whole world, and gain ground to expedite AI application, reduce load of work of operation of high in the clouds, realize more innovation application, brim operation demand grows day and day, AI begins from ” high in the clouds ” move toward ” terminal ” , also push litre of ASIC demand; consequently
Move orgnaization Ovum according to city beforehand appraise, 2018~2025 year, the city of ASIC occupies rate general from 11% increase to 48% considerably. Ovum findings report points out, 2016, high in the clouds (include center of enterprise, data to wait) the main battalion that learns chip for deepness controls a territory, occupied 80% . Nevertheless, arrived 2025, this one scale will be changed, change becomes the edge (Edge) was occupied 80% , and the scale of high in the clouds falls for 20% . The brim that indicates here means terminal unit, and it is a center in order to consume sexual product (and rather than is small-sized servo implement or it is a road by implement) , include mobile unit (mobile phone, flat) , head Daishi monitor (HMD) , wait like camera lens of sound box of AR/VR /MR, intelligence, robot, unmanned aircraft, car, safe photography.
Tractica/Ovum studies chief inspector Aditya Kaul expresses, the AI processor of nowadays great majority, be like GPU, multi-purpose Yu Yunduan servo implement, data center, undertake AI training and inference in order to be on high in the clouds. Nevertheless, increase as demand of privacy, security, add break the element such as frequency wide limitation to reduce cost, defer to reach, dispersive type AI subsequently arisen, increasing AI brim applies case to appear. For example A12 copy of the apple gives birth to chip, its have new generation ” nerve network engine ” , learn a skill with instant machine, change the use experience of the smartphone.
Kaul points out, in short, AI turns to the brim from high in the clouds is to have type now, current of course AI is much on brim equipment still give priority to with inference, and rather than trains. As AI nevertheless innovation application increases, have righter and righter the operation efficiency that chip business tries to promote terminal device processor, those who be is need not convey again the data undertakes data operation, inference and training to high in the clouds. Also accordingly, various processor comes out in succession, resembling is CPU, FPGA, GPU, ASIC, NPU or SoC Accelerator.
Among them, the city of ASIC is occupied rate be expected to increase and be climbed apparently as the demand of brim operation litre, from 2018 11% increase to 52% 2025. Kaul explains further, ASIC is favorred, the reason is studying body of processor framework much to in an attempt to at burgeoning deepness (Graph) or Tensorflow are fundamental framework; and afore-mentioned mentioning AI brim operation suffers power comsumption of be confined to and operation efficiency, because this is given priority to with inference more, and rather than trains. However, if arrive suppose when 2021, terminal unit will guide many AI chip, what what need is to be able to be in same a chip is aspirant travel inference and training, can disperse because of answering type operation and the IC that provides low power comsumption again, will go up continuously because of this ASIC demand raise, realize more AI brim to use case.