High-level synthesis for approximate computing: a survey

The approximate computing paradigm allows designers to design efficient hardware and software by leveraging the inherent tolerance of error-tolerance applications such as signal and multimedia processing, computer vision, and machine learning. A recent research focus is incorporating approximate computing techniques in high-level synthesis (HLS), providing a new perspective on hardware design by facilitating trade-offs between performance, energy efficiency, and accuracy. However, in the high-level synthesis for approximate computing (AHLS) field, there is currently a lack of systematic review papers and in-depth analysis of the latest methodologies. In this work, we present a comprehensive summary of the latest technologies in AHLS, with particular focus on error estimation, approximation techniques, and design space exploration (DSE). Additionally, we analyze the current research gaps in AHLS. This survey aims to provide researchers, engineers, and scholars with a comprehensive theoretical and practical framework for AHLS, fostering academic exchange and technological innovation in this field.