A RISC-V Based Coprocessor Accelerator Technology Research for Convolution Neural Networks
Shuhua Zhang, Jie Tong, Jun J. Zhang, Yuqing Lei, Minghao Zhang, Dang Li, Lanruo Wang · Journal of Physics Conference Series · 2020
Abstract The advancements of neural networks have significantly improved the accuracy of a variety of intelligent applications, such as image processing, voice recognition and so on. However, one of the challenges is to accelerate the speed of inference with the networks designed to be deeper. In this paper, we first realize the algorithm involved in software. Then, combined with the expanded characteristic of RISC-V architecture, the acceleration of convolution operation taking up the largest proportion of computation in CNN is realized by coprocessor expansion mode on open-source Hummingbird E203 processor. Through tests, convolution coprocessor improves the performance of CNN while ensuring the accuracy in function. Finally, the Cifar-10 image classification, a common benchmark in machine learning, is used to verify the feasibility in functions and comparatively analysis the pure software-based and coprocessor-based implementations.