Provocation tests, design of experiments and advanced statistical modeling to estimate product sensitivity to a defect: delamination failure case study for automotive semiconductors

During a short period, a malfunctioning happened on a manufacturing equipment in which about twenty semiconductor parts for automotive have been submitted on a process step. A same defect of inter-layer dielectric delamination was generated but each product showed different sensitivity to this defect. Failure that was generated was a latent one and occurred in field, which is fitting with the most serious situation for a final customer. Knowing how to predict quantity of failures expected to happen is very critical. Two semiconductor products for automotive have been specifically studied: the first one was a valve driver, the second one was an accelerometric sensor. Purpose was to highlight the main factors on sensitivity to defect. Was it possible to build a model from these factors to estimate sensitivity of any impacted product among the twenty ones ? For that, many technics have been implemented, from most typical reliability and provocation tests, until advanced design of experiments and modeling. Some strong results were obtained, as a model validated, but above all, in order to fully analyze a failure, to understand product sensitivity, and to assess risk, complementarity of typical reliability tests with advanced statistical technics was definitively proved.