Signal Activity Analysis for High-Level Power Estimation in Time-Shared Linear Systems

摘要

Power estimation techniques able to handle high levels of abstraction are strongly required to face up the increasing complexity and the stringent power budgets of current designs. This work addresses the modeling of signal activity and inter-wire coupling effects in data-intensive linear DSP architectures that share one or more resources. In this context, a rigorous mathematical model for the evaluation of those metrics is described. Furthermore, as a practical evaluation of the proposed technique, the work addresses the problem of power-aware binding during high-level synthesis. Experimental results show that the proposed estimation technique provides an excellent accuracy with errors smaller than 10% and a reduction in estimation time of two orders of magnitude respect two simulative approaches.