Outlier Labeling Method for Univariate Data for Module Test and Die Sort

Qorvo is a global leader in scalable and dynamic RF solutions for mobile, infrastructure, and defense. To ensure quality and compliance to customer specifications, a large number of product devices and die on wafers are tested every day. Typical data sets will contain unusually large or small values when compared with others in that data set, which may or may not pass specification limits. Such outliers will have negative effects on data analyses such as cpk-analysis, ANOVA, and regressions. More important, outliers, particularly those passing specification limits, may indicate defective parts or pose a reliability concern. For engineering teams in development and production, it is standard practice to identify or label outliers in their data sets, so that they can be removed to improve data analyses and to protect customers against receiving defective parts. The aim for this paper is to provide product engineering teams at Qorvo with reliable outlier labeling methods particularly for skewed distributions. While the ultimate goal is to label and to remove outliers, we also need to balance this with unnecessary yield loss for the financial bottom line.