Using Neural Network Based Active Learning for Modelling Integrated Circuits Behavior

This paper presents an efficient sampling active learning algorithm that can be used for modelling integrated circuits behavior by using neural networks regressions. This sampling scheme uses the variance of multiple regression models in order to iteratively select a minimal training sample set that leads to better accuracy regression models when compared to classical fixed sampling methods. We validate the efficiency of this approach by both using a synthetic function and also a simulated Low drop-out voltage regulator.