Ultra-Low Power Computing Toolbox

This lecture introduces the IcySoC project, focusing on inexact sub- and near-threshold systems for ultra-low power devices. It covers the development of a toolbox for tera-scale computing in nano-scale devices, addressing challenges such as energy efficiency, process variations, and reliability. The lecture explores the concept of approximate computing, which leverages application tolerance to inaccuracies for complexity reduction. It also discusses the design techniques for deterministic approximation of arithmetic primitives to optimize area, delay, and power consumption. Various demonstrations and applications of the IcySoC project, including near-sensor processors and chips with shared LNUs, are presented, showcasing significant energy efficiency gains and power savings.