摘要
本文主要讨论了现有基于CNN的AI芯片及相关软硬件关键技术的发展情况,在此基础上基于现有加速单元模块和FPGA资源开发形成了可用于CNN硬件加速的运算电路平台;结合经典人脸识别算法进行了验证,并对该加速电路平台的算力、功耗和电路性能优化方面提出了优化和改进策略。
The paper mainly discusses the development situation of the CNN AI chip and the key software/hardware technologies,and then designs the computing circuit platform which can be used on the CNN hardware acceleration,the platform is formed by CNN acceleration Unit and FPGA resources.The paper verifies the circuit platform combined with the classical face recognition algorithm,at the same time the paper takes optimization and improvement strategies about computing power,power dissipation and performance of the platform.
作者
常迎辉
李斌
武唯康
刘朋
CHANG Ying-hui;LI bin;WU Wei-kang;LIU peng(The 54 th Research Institute of CETC)
出处
《中国集成电路》
2020年第11期50-54,共5页
China lntegrated Circuit
关键词
CNN
AI
可重构
算力
人脸识别
Convolutional Neural Networks
Artificial Intelligence
reconfigurable
computing power
face recognition
