Towards probabilistic analog behavioral modeling

Analog behavioral models are widely used to reducethe complexity in hierarchical analog circuit design and verification.In the presence of process variations and atomic-levelfluctuations, however, these models have to be extended to takevariability into account. In this paper, we present a probabilisticsolution that treats the behavioral model coefficients as multidimensionalrandom variables and supports non-Gaussian aswell as correlated parameters. A voltage divider and a bandgapvoltage reference demonstrate the capabilities of our modelingapproach in terms of accuracy and efficiency.