Application of artificial neural network for switching loss modeling in power IGBTs

The modeling of switching loss in semiconductor power devices is important in practice for the prediction and evaluation of thermal safety and system reliability. Both simulation-based behavioral models and data processing-based empirical models are difficult and have limited applications. Although the artificial neural network (ANN) algorithm has often been used for modeling, it has never been used for modeling insulated gate bipolar transistor (IGBT) transient loss. In this paper, we attempt to use the ANN method for this purpose, using a customized switching loss test bench. We compare its performance with two conventional curve-fitting models and verify the results by experiment. Our model is generally superior in calculation speed, accuracy, and data requirement, and is also able to be extended to loss modeling for all kinds of semiconductor power devices.