Design Space Exploration Tools

The Approximate Computing design paradigm has repeatedly shown to be well suited to the needs of modern applications, especially those that interact with the physical world and process large amounts of data. By leveraging the presence of error-tolerant data and algorithms and the perceptual limitations of end-users, it allows to selectively relax the correctness requirements, achieving great performance enhancement and admitting a negligible output quality loss. Unfortunately, applying Approximate Computing to its full potential requires addressing several challenges: there is neither a generic methodology for identifying approximable code or circuit parts nor an approach for selecting the most suitable approximate techniques to apply. However, several tools have been proposed that seek to automate or at least guide part of the approximation process. In this chapter, we first discuss the state of the art for automatic tools for Approximate Computing, targeting digital circuits and software applications. We then introduce I DEA , an extendible tool suite that allows to describe Approximate Computing techniques, apply them to C/C++ code, and explore the design space of the obtained approximate variants to find an estimate of the Pareto front.