Memory synthesis for FPGA-Based reconfigurable computers

Abstract

For data intensive applications like Digital Signal Processing, Image Processing, and Pattern Recognition, memory reads and writes constitute a large portion of the total design execution time. With the advent of on-chip memories, a rich hierarchy of physical memories is now available on a Reconfigurable Computer (RC). An intelligent usage of these memories can lead to a significant improvement in the latency of the overall design. This paper presents an automated heuristic-based memory mapping framework for RCs. We use a Tabu search guided heuristic, Rectangle Carving, to map a single data structure onto several instances of a memory type on the RC. We also introduce control logic to resolve potential memory access conflicts and to make the details of memory mapping transparent to the accessing logic.