Approximate computing architectures

Approximate computing is an emerging computing paradigm for improving the efficiency of error-tolerant applications. It allows designers to trade a negligible amount of accuracy for significant efficiency gains. This chapter provides an overview of approximate computing and how it can be exploited to offer improved efficiency while satisfying the user-defined accuracy/quality constraints. First, an overview of techniques for approximating arithmetic hardware modules is presented. Then, methodologies for efficient design space exploration of approximate modules and for building approximate accelerators are covered. Apart from hardware-level approximations, the chapter also discusses different software-level approximations and how they can be integrated with hardware approximations in a cross-layer design flow for building efficient systems.