Using Evolutionary Algorithms for Certifying Behavioral Models

If behavioral models are to be used for speeding up analog simulations by co-simulating with analog circuit elements, the models must be fairly accurate in representing the circuit that they are going to swap. As circuits get more mature, the corresponding behavioral model has to be adjusted to capture in-band behavior, loading effects and any non linearities of the circuit. This needs a constant resource allocation in terms of an engineer who will be responsible for maintaining model-to-circuit synchronization. In this paper, I will present a successful approach of automatically fine-tuning the model behavior using Evolutionary Algorithms. This method is being extended towards developing a regression based automatic model generation and maintenance.