A framework for self-reconfigurable DCTs based on multiobjective optimization of the Power-Performance-Accuracy space

Abstract

We present a framework for the implementation of self-reconfigurable 2D Discrete Cosine Transforms (DCTs). Dynamic Partial Reconfiguration (DPR) and Dynamic Frequency Control lead to a multi-objective optimization scheme that generates Pareto-optimal realizations from the Power-Performance-Accuracy (PPA) space. The PPA space is created by evaluating the 2D DCTs realizations in terms of their power consumption, performance, and accuracy. Dynamic PPA management can then carried out by selecting Pareto-optimal realizations that meet time-varying PPA constraints.