Global Lifetime Estimation in Survival Analysis: Application to Solder Joint Reliability

In this paper, survival analysis is used to provide a global accelerated lifetime estimate for solder joints in electrical components. This method relies solely on global covariates remaining constant within specific test groups-meaning that individuals of the same group share the same geometry, solder material, and accelerated stress conditions. Two classical survival models are used-Cox proportional hazards (CoxPH) and Weibull accelerated failure time (Weibull AFT) -as well as a deep learning-based survival model, LogisticHazard. To assess their performance, a novel concordance metric is introduced, the inter-group concordance index, which evaluates concordance of lifetime ranking between test conditions. The survival curves predicted by each model are compared with the corresponding field-accurate Kaplan-Meier (KM) non-parametric model. Interpretation of the outputs for all three survival models is provided, giving access to the per-covariate acceleration factors (instead of the global acceleration factor), thus informing on key variables for reliability and giving insights for the development of more durable electronics.