Corinne BERGES, Six Sigma Black Belt, NXP Semiconductors

A customer in the automotive industry reported an issue with some parts, and failure analysis highlighted several failing tests in valve management functions. New tests on the products revealed an area in the die edge in which the parts were more likely to fail. Analyses were designed in order to understand the difference between this weak area and the rest of the die. These analyses used unit probe and class probe data: the former enables investigation at the die level; the latter supports study of the reticles on the wafer. The reticles are specific structures embedded on the wafers between the dies and picture each individual manufacturing step; their test enables the identification of the failing step.

Data analysis in JMP consisted of univariate analysis with test distribution plotting, multivariate analysis, and also prediction algorithms using the comparison between the two areas on the die on seven wafer lots. The analyses were conducted at the die and reticle level to be able to highlight the failing manufacturing steps.

This case study was used in training materials to promote machine learning. In the context of the automotive industry, where zero defects is the goal, everyone has to be trained in such analyses in order to achieve the highest possible level of quality.