Detection and elimination of initial transient for accurate power analysis

In this paper, we propose a statistical technique for detecting and eliminating initial transient in power analysis. While most power analysis approaches are based on the assumption of stationarity, artificial nonstationarity, or initial transient, may be introduced in early simulation period due to improper initialization. Left unattended, initial transient can seriously undermine the power estimation accuracy. As a remedy, we devise a statistical test for nonstationarity detection based on the Brownian bridge, the asymptotic behavior of a power dissipation sequence. Using the test, a stopping criterion is designed to determine if the circuit is warmed up enough for normal power analysis to proceed. The experimental results show the proposed technique efficiently detects and eliminates initial transients, thus maintaining power estimation accuracy