Model Discovery for Analog/Mixed-Signal Circuits


Advantages of formal verification are offset by the difficulties in the generating of good system models. This paper presents an improvement to the existing method, which generates models from a set of simulation traces. The proposed methodology aims to improve model precision by introducing new derivative based discretization method. Furthermore, a fine control over model fitness is provided via a notion of data rule mining, a novel approach to finding recurring patterns in the input data.

FAC 2017; Frontiers in Analog CAD