Implementation of CNN Heterogeneous Scheme Based on Domestic FPGA with RISC-V Soft Core CPU

Abstract

Field Programmable Gate Array (FPGA) has the characteristics of low power consumption, high performance and flexibility. Research on FPGA neural network acceleration is emerging, but most of the researches are based on foreign FPGA devices. In order to improve the current situation of domestic FPGA, a novel Convolutional neural networks (CNNs) accelerator for domestic FPGA equipped with lightweight RISC-V soft core is proposed. The peak performance of the proposed accelerator reaches 153.6 GOP/s, occupying only 14K LUTs (Look-Up-Table), 32 DRMs (Dedicated RAM Modules) and 208 APMs (Arithmetic Process Modules). The proposed accelerator has enough computing power for most of the Edge-AI applications and embedded systems, providing a possible AI inference acceleration solution for domestic FPGA.