Physics-of-failure based performance modeling of critical electronic components

Sammanfattning: Reliability prediction of the electronic components used in industrial safety systems requires high accuracy and compatibility with the working environment. The traditional reliability prediction methods that draw on standard handbooks such as MIL-HDBK 217F, Telcordia, PRISM etc., are not appropriate to determine the reliability indices of these components. For one thing, technology is constantly advancing; for another, the empirical data do not always match the actual working environment. The newest reliability prediction methodology, the physics-of-failure (PoF), emphasizes the root cause of failure, failure analysis, and failure mechanisms based on the analysis of parameter characteristics. It involves a focused examination of failure point locations, considering the fabrication technology, process, materials and circuit layout obtained from the manufacturer. This methodology is capable of providing recommendations for the increased reliability of components using intuitive analysis. However, there is a limitation: it is sometimes difficult to obtain manufacturer’s details for failure analysis and quality information. Several statistical and probability modeling methods can be performed on the experimental data of these components to measure the time to failure. These experiments can be conducted using the accelerated-testing of dominant stress parameters such as Voltage, Current, Temperature, Radiation etc. In this thesis, the combination of qualitative data from PoF approach and quantitative data from the statistical analysis is used to create a modified physics-of-failure approach. This methodology overcomes the limitations of the standard PoF approach as it involves detailed analysis of stress factors, data modeling and prediction. A decision support system is created to select the best option from failure data models, failure mechanisms, failure criteria and other factors to ensure a growth in reliability. In this study, the critical electronic components used in certain safety systems from different technologies are chosen for reliability prediction: Optocoupler, Constant Fraction Discriminator, BJT Transistor, Voltage Comparator, Voltage Follower and Instrumentation amplifier. The study finds that the modified physics-of-failure methodology provides more accurate reliability indices than the traditional approaches using field data. Stress based degradation models are developed for each of the components. The modified PoF models developed using Response Surface Regression and Support Vector Machine (SVM) show better performance.