A Comprehensive Analysis of Approximate Computing Techniques: From Component- to Application-Level

A new design paradigm, Approximate Computing (AxC), has been established to investigate how computing systems can be more energy efficient, faster, and less complex. Intuitively, instead of performing exact computation and, consequently, requiring a high amount of resources, AxC aims to selectively violate the specifications, trading accuracy off for efficiency. It has been demonstrated in the literature the effectiveness of imprecise computation for both software and hardware components implementing inexact algorithms, showing an inherent resiliency to errors. This tutorial introduces basic and advanced topics on AxC. We intend to follow a bottom-up approach: from component- up to application-level. More in detail, we will first present the main concept and techniques (e.g., functional approximation, voltage over-scaling). We then move to present some compile-time results in terms of energy-efficiency, area, performance versus accuracy of computations when using customized arithmetic (fixed-point, floating-point) and also try to derive some conclusions by comparing the different paradigms. The algorithmic-level approximation methods are then presented. Energy consumption can be reduced by approximating or skipping part of the computation. The concept of incremental refinement, early termination and fast decision will be detailed.