Inexact Arithmetic Operators

Approximate implementations of arithmetic circuits have been developed to find the best trade-offs between the key circuit parameters (such as energy, area, and delay) and the error of arithmetic operations. This chapter discusses various methodological aspects of developing approximate arithmetic circuits, including design abstractions, number representation, error analysis methods, and particular design methods. We survey problem-specific methods proposed for the manual design of approximate adders and multipliers. The circuit approximation problem is also formulated as a multi-objective optimization problem that can be solved by a suitable automated circuit design method. A comprehensive open-source library EvoAppoxLib of approximate circuits that was automatically generated by one of the automated methods is introduced. We stress the importance of correct evaluation and comparison of approximate implementations and a proper benchmarking methodology. Finally, two case studies demonstrate some frequently overlooked issues related to the selected error analysis approaches.