Accurate estimation of the Remaining Useful Lifetime (RUL) of Insulated-Gate Bipolar Transistors (IGBTs) is critical to early estimate the reliability of power transistors. This work presents an Artificial Intelligence (AI) RUL based on Radial Basis Function Neural Networks (RBF-NN). It was trained and validated on the NASA IGBT accelerated aging dataset. Unlike conventional AI models, RBF-NN can be executed by an Intelligent Sensor Processing Unit (ISPU) and packed with the IGBT. Leveraging its lightweight memory architecture, the RBF-NN achieves comparable accuracy to costly state-of-the-art recurrent neural networks (RNNs). Furthermore, it significantly reduces training and inference times, making it highly suitable for super integration with the IGBT.
