Reliability analysis and condition monitoring of SAC+ solder joints under high thermomechanical stress conditions using neuronal networks

Abstract

The thermo-mechanical fatigue of different SAC+ solders is investigated using transient thermal analysis (TTA) and predicted using artificial neural networks (ANN). TTA measures the thermal impedance and allows detection of solder cracks and delamination of material interfaces. LEDs soldered to printed circuit boards using seven different solders were aged within passive air-to-air temperature shock tests with TTA measurements every 50 cycles with the increase of the thermal resistance as failure criterium. A SnAgCuSb solder showed the best performance improvement over the SAC305 reference under the test conditions. In addition to standard evaluation by the cumulative failure-curve and Weibull plot, new approaches for reliability assessment are investigated to assess the reliability of the solder joint of the individual LEDs. A hybrid approach to predict failures in the solder joints of the individual LEDs during accelerated stress testing is set-up which processes the TTA data using artificial neural networks with memory, specifically LSTM, where the memory allows full use of the measurement history. Two ANN approaches, regression and classification, are used. Both approaches are shown to be quite accurate. The greater information gained from the regression approach requires more processing using external knowledge of the problem requirements, whereas the categorical approach can be more directly implemented. The results demonstrate the advantages of integrated approaches for assessment of the remaining useful life of solder joints.