Efficient Stochastic Collocation Based Variability Analysis Using Model-Order Reduction Techniques

Predicting the effect of the variability of design parameters on the performance of high-speed integrated circuits is crucial to a successful design. The conventional Monte Carlo technique is computationally expensive due to the large number of simulations and a slow convergence rate. To address the above difficulties, a novel method is presented in this thesis for time-domain stochastic analysis of large active/passive circuits with multiple stochastic parameters. The new approach reduces the computational cost of variability analysis by using the Stochastic Collocation technique. The Sparse Grid algorithm is applied to limit the growth of the computational cost with an increase in the number of stochastic parameters. In addition, the proposed method is based on the Model Order Reduction algorithms coupled with the Numerical Inverse Laplace Transform approach.