Bayesian Network Model with Application to Smart Power Semiconductor Lifetime Data.

Area of Science:

• Materials Science
• Statistical Modeling
• Semiconductor Physics

Background:

• Semiconductor lifetime data often exhibits complex behavior, influenced by multiple failure mechanisms and censoring.
• Understanding interactions between device parameters and test conditions is crucial for accurate reliability prediction.
• Existing methods may not fully capture the intricate relationships within semiconductor lifetime data.

Purpose of the Study:

• To apply Bayesian networks for modeling semiconductor lifetime data from cyclic stress tests.
• To investigate the interactions between test settings, geometric designs, material properties, and physical parameters.
• To provide a reliable alternative for semiconductor reliability analysis.

Main Methods:

• Utilized Bayesian networks to model log-normal distributed semiconductor lifetime data, accounting for censoring.
• Extended MATLAB® statistical toolboxes for Bayesian network structure learning and parameter estimation.
• Employed Markov chain Monte Carlo (MCMC) simulations for posterior distribution determination with censored data.
• Applied Automatic Relevance Determination (ARD) and various goodness-of-fit criteria for model selection.

Main Results:

• The Bayesian network model successfully captured the complex lifetime behavior of semiconductor devices.
• Identified significant interactions between covariates influencing semiconductor reliability.
• Demonstrated the effectiveness of Bayesian networks in providing actionable insights into reliability.

Conclusions:

• Bayesian networks offer a powerful framework for analyzing semiconductor reliability data.
• The developed methodology provides valuable information on covariate interactions, enhancing predictive accuracy.
• This approach serves as a reliable and informative alternative to conventional semiconductor reliability assessment techniques.