摘要
AI芯片是被专门设计用于加速人工智能计算任务的集成电路。在过去几十年里,AI芯片经历了持续的演进和突破,促进着人工智能领域的发展。文章探讨了AI芯片的发展史、主流技术和应用场景,以及面临的挑战和问题。进而提出采用Chiplet技术,将不同的功能模块独立集成为独立的Chiplet,并融合在一个AI芯片上,从而实现更高的计算能力。该设计不仅允许独立开发和升级各个模块,还可在封装过程中将它们巧妙组合起来,使得AI芯片能够随着人工智能技术的不断优化而持续发展。
AI chips are integrated circuits specifically designed to accelerate AI computing tasks.Over the past few decades,AI chips have undergone continuous evolution,contributing to the development of AI.In view of this,the article explores the development history,mainstream technologies,and application scenarios of AI chips,as well as the challenges they face.Furthermore,it is proposed to adopt chiplet technology,which integrates different functional modules into independent chiplets and fuses them on a single AI chip to achieve higher computing power.This design not only allows independent development and upgrading of each modules,but also cleverly combines them during the packaging process,enabling AI chips to be continuously optimized with the continuous optimization of AI technology.
作者
张志伟
田果
王世权
Zhang Zhiwei;Tian Guo;Wang Shiquan(Sun Moon New(Suzhou)Semiconductor Co.,Ltd.,Suzhou 215000;Asia Pacific Silicon Valley Technology Research Institute(Hong Kong)Limited,Hong Kong 999077,China)
出处
《中阿科技论坛(中英文)》
2023年第11期90-94,共5页
China-Arab States Science and Technology Forum