Lifetime Estimation of Discrete IGBT Devices Based on Gaussian Process

Discrete package insulated gate bipolar transistor (IGBT) devices are a popular choice for low-power converters. Although IGBT power modules used in high-power applications have recently been studied in the literature, there are major knowledge gaps regarding reliability and lifetime estimation of discrete devices. In this paper, on-state collector–emitter voltage drop $(V_{{\rm{ce}},{\rm{on}}})$ variations are characterized for discrete IGBT devices exposed to cyclic thermal stresses. Variations in $V_{{\rm{ce}},{\rm{on}}}$ are carefully identified and classified depending on different aging mechanisms, stress levels, and device structures. A probabilistic framework for remaining useful lifetime (RUL) estimation based on the knowledge obtained by accelerated aging experiments for real-time RUL estimation has been proposed. Specifically, the proposed model uses Gaussian process regression (GPR) for applying a Bayesian inference (BI) on RUL estimation of the device under test. Using BI reduces the uncertainty associated with RUL estimation and leads to more accurate results. This concept is also tested by comparing the classical maximum-likelihood estimation and GPR estimation results with the ones obtained by accelerated aging tests.