Skip to Main content Skip to Navigation

Approximate computing for high energy-efficiency in internet-of-things applications

Abstract : Approximate computing explores methods to trade-off the quality of result and the computation costs, e.g. energy consumption. One of the proposed methods is to reduce the width of the computation units. To date, such units have been mostly evaluated separately, i.e. not evaluated in a complete application. In this thesis, we evaluate the global energy reduction vs quality of output trade-offs of applications. These applications are executed on a RISC-V processor extended with reduced width computation and memory units. In these units, only a number of most significant bits, configurable at runtime, is active. The results indicate in average that the energy can be reduced by up to 14% for an error ≤ 0.1%. Moreover we propose a generic energy model that indicates that both software parameters (e.g. fraction of approximable code) and hardware architecture ones (e.g. degree of approximation) impact the applications energy reduction.
Complete list of metadatas

Cited literature [114 references]  Display  Hide  Download
Contributor : Arnaud Tisserand <>
Submitted on : Friday, September 20, 2019 - 2:06:17 PM
Last modification on : Monday, April 6, 2020 - 9:22:34 AM
Document(s) archivé(s) le : Sunday, February 9, 2020 - 2:21:04 AM


Files produced by the author(s)


  • HAL Id : tel-02292988, version 1


Geneviève Ndour. Approximate computing for high energy-efficiency in internet-of-things applications. Hardware Architecture [cs.AR]. Université Rennes 1, 2019. English. ⟨tel-02292988⟩



Record views


Files downloads