Innovative methodology in automotive semiconductors for statistical correlation study between reliability accelerated stress tests and real field applications

In semiconductor manufacturing for automotive as for many other industries, reliability tests are designed and implemented in order to predict failure rate in real life and applications. Physics-of-failure is used on the rejects observed in field and during the reliability tests to check them to stress the components as in life applications. Besides this qualitative study between field and reliability tests, a quantitative comparison is possible on the failure rates observed in field and predicted by reliability, but this type of comparison is often limited to the maximum rates as not enough read points are implemented in reliability to allow a fine estimation or modeling. In the event conditions on reliability are implemented to estimate more precisely failure rate and even to be able to model reliability test results, a more precise quantitative comparison between field and reliability may be thought, which is revealed sometimes very important in case of critical quality issue for specific rejects observed in field. In this paper, an innovative correlation method between field and reliability is designed for a specific field quality issue. In a first step, modeling technique is described, used on field rejects as for reliability tests that a specific design makes possible. Then, difference between field and reliability sample sizes, and between mileages covered in field and test hours in reliability laboratory are bypassed: the method takes into account acceleration factor and car usage. Finally, a statistical test establishes correlation.