Conclusions and Outlook

Abstract

In this book, algorithms and methodologies for the approximate computing paradigm have been proposed. Approximate computing hinges on cleverly using controlled inaccuracies (errors) in the operation for performance improvement. The key idea is to trade off correct computation against energy or performance. Approximate computing can address the growing demands of computational power for the current and future systems. Applications such as multi-media processing and compressing, voice recognition, web search, or deep learning are just a few examples where this novel computational paradigm provides huge benefits. However, this technology is still in its infancy and not widely adopted to mainstream. This is because of the lack of efficient design automation tools needed for approximate computing. This book provides several novel algorithms for the design automation of the approximation circuits. Our methodologies are efficient, scalable and significantly advance the current state-of-the-art of the approximate hardware design. We have addressed the important facets of approximate computing—from formal verification and error guarantees to synthesis and test of approximation systems.