Memory Bandwidth Optimization Strategy of Coarse-Grained Reconfigurable Architecture

Abstract

With the increasingly efficient streaming of multimedia standard and the computational resources in the coarse grained reconfigurable architecture (CGRA), how to provide high data throughput for the CGRA for multimedia application in real time becomes a bottleneck challenge. In this paper, a data reuse method is proposed based on the pre-parsing mechanism of MB information in advance. The pre-parsed information guides the replacement process of the data buffer. By employing the data reuse method, the required bandwidth can be reduced by nearly 50% on average and the CGRA can support 1080p@30fps video decoding at the frequency of 200MHz.