Sensitivity based parameter reduction for statistical analysis of circuit performance

If we view a circuit performance such as delay as a function of underlying parameters, e.g. gate length and gate width, the difficulty of performance analysis lies in two aspects: The first one is the high dimension of the parameter space as well as the statistical relationship among these parameters. The second one is the complex mapping between the performance space and the parameter space where commonly no direct analytical form exists and numerical methods are used instead. In this paper, we propose a new method for reducing the dimension of the parameter space while taking the first order sensitivity information into account. Experimental results have verified that the proposed method can achieve lower standard deviation error of the performance in comparison with the traditional method PCA.