Quantization of HLS Designs Using Value Range Analysis
Algorithm developers usually use double precision data types to be able to focus on the mathematical functionality of the algorithm. When this algorithm is implemented as a hardware module, the data accuracy must be reduced to a minimum number of bits that still fulfills the system performance requirements. The process of converting the floating-point algorithm to bit-level optimized model is complicated and requires special knowledge. This paper introduces a simple and robust quantization methodology for HLS designs based on value range analysis.
