A Neural Network based Approach to Simulate Electro-thermal Device Interaction in SPICE Environment

An innovative modelling methodology for the simulation of electro-thermal interaction in power devices, based on Neural Networks (NNs), is shown. The suitability of NNs in modelling the complicated non-linear, temperature dependent characteristic that power electronics devices feature, is shown. The proposed methodology is particularly suited to be implemented in electrical simulators. The approach can be divided in two parallel steps: firstly, NNs are used to describe the complex, highly non-linear electro-thermal characteristic of the considered device; secondly, a non-linear RC-based thermal model is generated, with a method published in a previous work. These two sub-systems are coupled together in order to achieve a self-consistent electro-thermal model. The modelling results are validated against experiments with very satisfactory results. The technique is explained in detail; advantages and drawbacks of the method are then discussed.