Temperature acceleration models in reliability predictions: Justification & improvements

Reliability predictions have been for a long time a difficult task, due to the conflict between high reliability requirements and the lack of component manufacturer data. As the data available during the development phase is the product bill of materials, reliability prediction methods have developed component reliability models based on in-service field return data and/or physics of failure. The repartition of these two approaches has changed over time, where predictions were mainly based on empirical data at the beginning (MIL-HDBK217), but more recently attempts have been made to incorporate some form of physics of failure (MIL217+, FIDES). Note, that even in these attempts the acceleration factor coefficients are still based on empirical data. For our aeronautic applications, with the recent reliability prediction methods, steady-state temperature is the main reliability driver and any wrong assumption in the model can be catastrophic. Our methodology is based on several recent works such as Pecht's model, new mechanisms of failure (hot carrier, delamination, etc) due to greater component integration, and physics of failure simulations (Bernstein). We propose some improvements on this model in order to have some more realistic reliability predictions. For example, we introduce a non Arrhenius model for steady-state temperature for active electronic components and explain in detail why these formulations don't match correctly to real data.