Energy minimization for dynamic supply voltage scaling using data dependent voltage level selection

Abstract

In this paper we propose a workload distribution based quantized voltage level selection method called data-dependent level selection (DDLS) for minimizing energy in data-dependent computations using dynamic supply voltage scaling. In previous works, voltage levels have been selected by dividing maximum normalized workload by the number of discrete voltage levels, and translating workloads to voltage levels. The existing techniques place no emphasis on workload characteristics of data in voltage level selection. Our DDLS technique is workload distribution based and therefore selects minimum energy yielding voltage levels for a given data sequence. Our analysis of Akiyo video sequence shows energy savings of up to 55% when compared to existing methods of voltage level selection.