Minimizing the Number of Process Corner Simulations during Design Verification
Michael Shoniker, Bruce F. Cockburn, Jie Han and Witold Pedrycz
University of Alberta, Edmonton, AB T6G 2V4, Canada
Integrated circuit designs need to be verified in simulation over a large number of process corners that represent the expected range of transistor properties, supply voltages, and die temperatures. Each process corner can require substantial simulation time. Unfortunately, the required number of corners has been growing rapidly in the latest semiconductor technologies. We consider the problem of minimizing the required number of process corner simulations by iteratively learning a model of the output functions in order to confidently estimate key maximum and/or minimum properties of those functions. Depending on the output function, the required number of corner simulations can be reduced by factors of up to 95%.
Keywords: Adaptive algorithms, Circuit simulation, Design automation, Function approximation, Gaussian processes, Robustness, Unsupervised learning.
Full Text (PDF)