On estimation accuracy for guiding low-power resynthesis

There are three properties which define the global effect on power of a resynthesis step: change in function, change in spurious dynamic activity, and change in delay. In this paper, it is shown that the prediction of change in power due to change in functionality and change in spurious dynamic activity is accurate at high levels of estimator abstraction. Change in power due to change in delay is not. However, the magnitude of the effect of delay on power during resynthesis is shown to be the least critical of the three properties.