Investigating software and hardware codesign for approximate computing

Dnr:

SNIC 2017/7-88

Type:

SNAC Small

Principal Investigator:

Christos Sakalis

Affiliation:

Uppsala universitet

Start Date:

2017-05-23

End Date:

2018-06-01

Primary Classification:

10206: Datorteknik

Webpage:

Allocation

Abstract

Approximate computing is a eld of Computer Science focusing on computations where accuracy can be traded o for higher e ciency, be it performance, energy, or area e ciency. Many computations are by nature approximate, such as physics simulations, image processing, and neural networks, and this can be taken advantage of in the design of the algorithms algorithms used, as well as the software and the hardware architecture. Our focus will be on the latter, aiming to develop the key insights and technologies needed to build systems in which computational accuracy can be adjusted to meet the requirements of an application, as set by the programmer, and in doing so not spending more time or energy than is required to achieve adequate results.