New Method of Screening Out Outlier; Expanded Part Average Testing During Package Level Test

Abstract

To achieve the automotive quality for semiconductor devices represented by “zero-defect,” part average testing (PAT) is actively applied to outgoing test. PAT has been generally used on wafer level test. This time we expanded it to package level test and established the method of detecting the characteristic variations and screening out the outliers. Moreover, we suggest a way of thinking about the outliers and consideration on the basis of the actual data analysis. We were especially conscious of the statistical meaning of the outliers which were detected. It has been demonstrated that the expanded PAT to package level test can reject the outliers of characteristics variation even within spec. The outliers in distribution of characteristic variations were detected by maintaining chip traceability. The new method can screen out outliers outside the 6-sigma within spec, and there was a clear inflection point between the distribution of the abnormal data and that of the population. It was assumed that hidden failure devices which had not been rejected as early failure were screened out. Detecting the outlier with the new method can reduce the failure rate to the target.