A study of stratified sampling in variance reduction techniques forparametric yield estimation

摘要

The Monte Carlo (MC) method exhibits generality and insensitivitynto the number of stochastic variables, but is expensive for accuratenyield estimation of electronic circuits. In the literature, severalnvariance reduction techniques have been described, e.g., stratifiednsampling. In this contribution the theoretical aspects of thenpartitioning scheme of the tolerance region in stratified sampling isnpresented. Furthermore, a theorem about the efficiency of this estimatornover the primitive MC (PMC) estimator versus sample size is given. Tonthe best of our knowledge, this problem was not previously studied innparametric yield estimation. In this method we suppose that thencomponents of parameter disturbance space are independent or can bentransformed to an independent basis. The application of this approach tona numerical example (Rosenbrock's curved-valley function) and a circuitnexample (Sallen-Key low-pass filter) are given