Towards approximating Multiphysics Simulations with Artificial Neural Networks
A complete analysis of an electronic design requires not only circuit simulations but also the simulation of the thermal and electromagnetic interaction of the different components, packaging, etc. There are several standard methods to perform those simulations but all come at large computational costs, especially for complex systems. In this talk, I will explain how machine learning techniques can be used to approximate these simulation results and, thus, reduce simulation times. As a proof of concept, results for the heat transfer in electronic systems are presented.