In-Service Diagnostics for Wire-Bond Lift-off and Solder Fatigue of Power Semiconductor Packages

• Published in: IEEE transactions on power electronics Vol. 32; no. 9; pp. 7187 - 7198
• Main Authors: Eleffendi, Mohd Amir, Johnson, C. Mark
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.09.2017; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Converters; Degradation; Electrical resistance; Estimates; Fatigue; Heat transfer; Insulated Gate Bipolar Transistor (IGBT); Insulated gate bipolar transistors; kalman filtering; Kalman filters; least squares methods; life estimation reliability; Mathematical models; Monitoring; Packages; Parameters; Solders; Temperature measurement; Thermal resistance; Tolerances; Uncertainty; Wire
• ISSN: 0885-8993; 1941-0107; 1941-0107
• DOI: 10.1109/TPEL.2016.2628705

Wire-bond lift-off and Solder fatigue are degradation mechanisms that dominate the lifetime of power semiconductor packages. Although their lifetime is commonly estimated at the design stage, based on mission profiles and physics-of-failure models, there are many uncertainties associated with such lifetime estimates, emerging, e.g., from model calibration errors, manufacturing tolerances, etc. These uncertainties, combined with the diverse working environments of power semiconductor packages result in inaccurate lifetime estimates. This paper presents an approach for estimating the extent of degradation in power semiconductor packages based on online monitoring of key parameters of the semiconductor, namely, the thermal resistance R thja and the electrical resistance R CE . Using these two parameters, solder fatigue and wire-bond lift-off can be detected during normal converter operation. In order to estimate these two parameters, two techniques are introduced: a residual obtained from a Kalman filter, which estimates the change in the thermal resistance R thja , and a recursive least squares algorithm, which is used to estimate the electrical resistance. Both methods are implemented online and validated experimentally.