A Markov Model for Low-Power High-Fidelity Design-Space Exploration

Abstract

We use a Markov model to specify the behaviour of a protocol, and show an analysis of this model can generatea high-level design space that an engineer can explore. The behaviour that we study is the leakage power and the area complexity. The design space generated from the Markov model is shown to have high fidelity, which means it faithfully reflects the corresponding ‘implementation space’, and the lowest-power design will synthesise to the lowest-power implementation. In effect, the high-level Markov-based analysis we carry out allows low-level behaviour to be predicted, and this diminishes the need for extensive, time-consuming simulation. We also compute the theoretical lower and upper bounds of power, and in so doing, can determine how close our high-level designs are to being optimal. To test fidelity, we apply two different simulation tools, and measure the correlation between our high-level estimates and the results produced by simulation. In a case study, we predict which design of an AMBA protocol will consume least total power and cover least area.