Approximate Computing Technique Using Memoization and Simplified Multiplication

In embedded systems, approximate computing can strongly promote reduction of execution time and energy consumption in exchange for some output errors. We focused on “Fuzzy memoization”, which is one of the approximate computing techniques. We improved it by using simplifying multiplication. By using this approach, we have developed a novel technique to reduce execution time and energy consumption while keeping output precision. Then, we applied it to grayscale filters on the Zynq system that contains ARM-based processor and field-programmable gate array (FPGA). Evaluation results from the implemented system showed that our proposed technique can reduce the execution time by up to 28% and reduce the energy consumption by 11% in spite of very high-quality output images.