移动智能终端人工智能芯片算力评测方法  

Computing Power Evaluation Method of Artificial Intelligence Chip in Intelligent Mobile Terminal

作　　者：庞涛[1] 沙通 崔思静 杨婷婷 PANG Tao;SHA Tong;CUI Sijing;YANG Tingting(Research Institute of China Telecom Co.,Ltd.,Guangzhou 510630,China)

机构地区：[1]中国电信股份有限公司研究院移动与终端技术研究所,广东广州510630

出　　处：《移动通信》2022年第5期89-93,118,共6页Mobile Communications

摘　　要：随着AI芯片软硬件架构的迭代升级,移动智能终端的AI计算能力得到了迅速提升。重点研究移动终端AI芯片的算力评估方法,首先介绍了端侧AI加速技术的发展现状和评测工具,深入研究影响AI芯片算力的各种影响因素,随后结合近两年主流手机AI芯片的算力评估结果,研究得出算力评测的具体方法。With the iterative upgrade of both hardware and software architectures of artificial intelligence (AI) chips,the computing power of intelligent mobile terminals has been rapidly improved.This paper focuses on the computing power evaluation method of AI chips in mobile terminals.Firstly,the development status of terminal-side AI acceleration technology and measurement tools are introduced.Then,the factors affecting the computing power of AI chips are comprehensively analyzed.Finally,based on the computing power evaluation results of mainstream mobile phone AI chips in the past two years,the specific method of computing power evaluation is studied.

关 键 词：移动终端 人工智能 深度学习 神经网络模型 算力 

分 类 号：TN92[电子电信—通信与信息系统]