No Trouble Found (NTF) Customer Return Analysis

This paper analyzes NTF customer returns using production test data to verify if they can be screened. A total of 70 NTF customer returned units from multiple lots spanning three different product families across different customer applications and market segments were examined using three statistical methods - Principal Component Analysis (PCA), Mahalanobis Distance, and binary Support Vector Machine (SVM) in search of potential outliers. The data shows the NTF customer returns are not marginal units and cannot be screened without significant yield penalty.