Abstract
Statistical lifetime models for high-power IGBTs are developed based on results from power-cycling experiments, and relate lifetime expectancy to the well-defined conditions of a laboratory experiment. In most cases, predefined cyclic-stress conditions are repeatedly applied, until the power device under test reaches its end of life. However, in real applications, power modules are exposed to non-repetitive stress patterns that can be very different from the power-cycling conditions. A well established lifetime-estimation method suggests to decompose the stress pattern into individual components, whose damage can be calculated using a lifetime model. These damage contributions are then summed up in order to estimate the consumed or the remaining lifetime of a device. When comparing the estimation results to field measurements though, they often fail to match the real behaviour of a power device. This paper points out the uncertainties that appear in this process of applying a lifetime model on stress patterns deriving from field applications. The main purpose is to present typical sources of errors, and discuss how severely these may impact the lifetime estimation.