AI Acceleration with RISC-V for Edge Computing

Abstract

Convolutional neural networks (CNNs) have been successfully applied to many Al applications and even demonstrate beyond-human capability in some cases. By implementing CNNs on edge devices, less energy dissipation, higher security, and lower latency can be achieved. In this talk, 1 will present a design framework that optimizes the dataflow of CNN by leveraging the data reuse. Memory access times can be minimized through proper memory partitioning and allocation. The proposed methodology is demonstrated by a system with an Al accelerator and a RISC-V core.