Résumé
Approximate computing has been adopted as a promising approach to energy-efficient design of digital systems. Actually the straightforward idea is to lower the supplied voltage level until errors appear. These errors can be "implicitly" tolerated since the application accepts some loss of quality or optimality in the computed result(s). This paper aims at targeting applications that cannot accept loss of quality. We propose an architecture able to auto-adapts the supplied voltage level in order to fit the actual running application. The key concept is a preliminary characterization phase that exploits the benefit of approximate computing.