Cycle-accurate macro-models for RT-level power analysis

摘要

In this paper, we present a methodology and techniques forngenerating cycle-accurate macro-models for register transfer (RT)-levelnpower analysis. The proposed macro-model predicts not only thencycle-by-cycle power consumption of a module, but also the movingnaverage of power consumption and the power profile of the module overntime. We propose an exact power function and approximation steps tongenerate our power macro-model. First-order temporal correlations andnspatial correlations of up to order three are considered in order tonimprove the estimation accuracy. A variable reduction algorithm isndesigned to eliminate the “insignificant” variables using anstatistical sensitivity test. Population stratification is employed tonincrease the model fidelity. Experimental results show our macro-modelsnwith 15 or fewer variables, exhibit <5% error for average power andn<20% errors for cycle-by-cycle power estimation compared to circuitnsimulation results using Powermill