Reliability evaluation of multi-mechanism failure for semiconductor devices using physics-of-failure technique and maximum entropy principle

The physics-of-failure (PoF) technique is a practical approach to evaluate the reliability of semiconductor devices. However, the PoF approaches are usually insufficient in dealing with multi-mechanism failure and fitting the Monte Carlo (MC) sampling data. In our study, we propose an improved reliability evaluation method based on PoF technique and maximum entropy (MaxEnt) principle. The PoF models are used to generate time-to-failure samples of the failure mechanisms. Cumulative damage rules and competing failure rules are adopted to deal with multi-point and multi-mechanism failure and generate lifetime samples of the device. And the lifetime samples are fitted by MaxEnt distributions through the proposed fitting algorithm. The numerical examples given in the paper indicate that the MaxEnt distributions can describe the samples well and have a competitive advantage in dealing with multi-peak samples. A case study about a semiconductor device with multi-mechanism failure is presented to explain the workflow of the proposed reliability evaluation approach. The results show that the proposed MaxEnt distributions can yield reliable reliability evaluation results compared with Weibull and Lognormal distributions in the multi-mechanism failure process.