Benchmarking the Accuracy of PCA Generated Statistical Compact Model Parameters Against Physical Device Simulation and Directly Extracted Statistical Parameters

Statistical variability is a major challenge for CMOS scaling and integration. In order to achieve variability aware design, it's critical important to reliably transfer device characteristics statistical variability information into compact models. A PCA based statistical compact modeling strategy is benchmarked against 'atomistic' device simulation and direct statistical parameter extraction strategy. The results indicate that PCA based approach may introduce considerable error in tail of distribution, which in turn may generate pessimistic or optimistic conclusions in statistical circuit simulation.