Behavioral compact models of IGBTs and Si-diodes for data sheet simulations using a machine learning based calibration strategy

Abstract:
Power device customers make use of compact models to evaluate their designs and to investigate the device behavior in circuit simulation. Since there is often the requirement to perform simulations for large time scales including a high number of switching events, models have to be fast, but also accurate and stable with regard to convergence. This paper introduces new behavioral compact models developed for power Si-diodes and IGBTs enabling shorter simulation times as well as enhanced convergence stability in comparison to existing physics-based models. The model development aims at a highly flexible implementation in order to ensure an accurate calibration of the characteristics of several IGBTs and Si-diodes in more than a dozen different technologies. The key element of the model parameter calibration is a machine learning algorithm, which focuses on the representation of the data sheet content. The highly efficient calibration strategy reduces the effort of a human-based calibration procedure significantly, and offers the possibility to parametrize 100 products within one day.