A Novel Approach to Design Space Realization and Exploration of Mathematical Kernels for Automated FPGA Implementation

In this paper we describe a novel approach that automates the process of producing and optimizing FPGA designs. Our approach produces a comprehensive design search space comprising performance, power, and resource utilization estimates for these designs that may be searched for any optimal combination of theses characteristics. The design search space consists of both parallel and serial designs as well as heterogeneous combinations of parallel and serial designs. We are able to reduce the large search space determined by Catalan numbers by reducing the space modulo equivalent designs. Our approach also generates structural VHDL for any of the designs using FPGA vendor- specific VHDL libraries. In this paper we describe the approach and its application to the problem of accelerating algorithmic performance in stereoscopic vision applications.