Approximate Computing: Evolutionary Methods for Functional Approximation of Digital Circuits

Approximate computing deviates from long-held paradigm and attracted towards intrinsic application resilience to enhance the efficiency by relaxing the parameter of full accuracy. The specifications of digital circuits are violated while explicitly designing. The benefits are in terms of area, performance, power consumption, and speed. For approximating circuit, synthesis tools depend on available technique to check whether the synthesized circuit meets the set parameters and provides the quality. Mostly used approach is circuit simulator for analyzing responses received from all input variables but it is an exhaustive approach and can be used only for a small number of inputs. So this paper focuses on techniques, methods used to approximate circuits and to use formal methods for solving challenges faced by traditional methods. The use of evolutionary methods for circuits has led to the promising results. Major challenges in these methods are the ability to automatically synthesize approximation circuits without relying on the skill of designers. Strategies for generating approximate circuits, methodologies for evaluating the error with benchmarks used are tabulated in paper. The exact circuit, error and their approximate threshold are required for simplification which leads to selecting synthesis techniques for circuit transformations. The criteria of selection of approximate logic synthesis method are based on error analysis, number input and desired output, use of types of application and digital circuits like adder, multiplier, FIR, FFT etc. Introduction, principle, methods of approximate computing with major sub-areas of research work conducted in field of functional approximation which are apropos for designing and testing of circuits are main concern of paper. (C) 2022 Elsevier Ltd. All rights reserved.