A Method to Group Reliability Data by Hierarchical Clustering

We present an approach of reliability data grouping by the hierarchical clustering method, which is useful for the analysis of reliability result for semiconductor devices. The purpose of this method is to effectively group a series reliability test result based on its characteristics in multiple dimensions. The Ward's method is well recommended to be used for distance definition. According to the grouping result by hierarchical clustering method, the analysis can help identify what data is the baseline performance, and the rest outlier can be further analyzed for special handling, such as a process tool's excursion or a test abnormality. A series of reliability test data of the EM (Electro Migration) for one technology in an 8-inch semiconductor wafer manufactory as an example shows the effectiveness on grouping based on 4-dimension clustering analysis. Moreover, a reliability monitoring flow is developed to feasibly fulfil the clustering method.