Data-and State-Dependent Power Characterisation and Simulation of Black-Box RTL IP Components at System Level

Abstract

Due to the increasing algorithmic complexity of todays embedded systems, the consideration of extra-functional properties becomes even more important. Extra-functional properties such as timing, power consumption, and temperature need to be validated against given requirements on all abstraction levels. For timing and power consumption at RT- and gate-level, several techniques are available, but there is still a lack of methods and tools for power estimation and analyses at electronic system level (ESL) and above. Existing ESL methods in most cases use state-based methods for power simulation. This may lead to inaccurate results, especially for data-dependent designs. In this paper, we extend the Power State Machine (PSM) model for black-box RTL IP components with a mechanism that employs data-dependent switching activity using the Hamming distance (HD). In pipelined designs, we do not only consider the input HD but also the HDs of the internal pipeline stage registers. Since these registers of black-box IP are not observable from the outside, our model derives the internal HDs from previous input data. The results show that our extension achieves up to 38% better results than the previous PSM approach and up to 35% better results compared to a model considering only the input HD.