A CNN Hardware Accelerator Designed for YOLO Algorithm Based on RISC-V SoC

Abstract

YOLO (You Only Look Once) has been widely used in the field of object detection because of its extremely fast real-time calculation speed and good migration ability. In recent years, the design of artificial intelligence systems with high real-time and low energy consumption has become a research hotspot. In this paper, we propose a CNN hardware accelerator specifically designed for YOLOv3-Tiny to increase the calculation parallelism while reducing the frequency of memory access. The design is configured and controlled by T-Head C910, a state-of-art open source multi-core processor based on RISC-V architecture. Experimental results show that the design can provide effective throughput improvement for small embedded systems with limited resources.