Performance Variability Modeling of Analog Circuits Using Improved Orthogonal Matching Pursuit

Abstract

Process variations have a significant impact on performance and yield in the nanoscale design of integrated circuits (ICs). Monte Carlo simulation is useful in predicting these variations due to its simple implementation. However, the simulation cost of this method increases exponentially as the number of random parameters increases. In this paper, we solve these problems by using an Improved OMP algorithm based on sparse regression. The proposed algorithm solved heuristically L-0 regularization problem and effectively model performance based on a limited number of simulation samples. Numerical results for circuit designed in a 28nm process demonstrate that IOMP achieves up to 1.625x speedup compared to the traditional OMP method.