Lifetime Prediction in Power Semiconductor Devices: A Comparative study between Analytical Modeling and Artificial Neural Network

In this work the reliability of TO-247 IGBT devices is investigated in the case of power cycling stress. A large range of test conditions is considered, varying the junction temperature cycling from 40°C to 150°C and the heating time from 0.3s to 32s. Under this wide range of test conditions, different failure modes determine the end of life of components. This work investigates the suitability of artificial neural networks and conventional analytical models to predict the lifetime of components where different failure mechanisms occur. Both models are compared against experimental data and root mean square relative errors are evaluated to quantify their accuracy.