ACRE – Approximate Computing Reducing Energy

PI: Oscar Gustafsson (LiU); co-PI: Joachim Rodrigues (LU)

Trading energy for accuracy is a promising approach to reduce the computing energy dissipation. Especially, for classes of applications with inherent resiliency, including AI/ML, significant reductions can be obtained. We focus on the novel and visionary area of modifying exact arithmetic operators to also perform more complex operations approximately with a focus on floating-point representations. This brings energy savings from approximate computing into general purpose processors.

Project number: D3