A framework for root cause detection of sub-batch processing system for semiconductor manufacturing big data analytics

Abstract

Root cause detecting and rapid yield ramping for advanced technology nodes are crucial to maintain competitive advantages for semiconductor manufacturing. Since the data structure is increasingly complicated in a fully automated wafer fabrication facility, it is difficult to diagnose the whole production system for fault detection. A number of approaches have been proposed for fault diagnosis and root cause detection. However, many constraints in real settings restrict the usage of conventional approaches, due to the big data with complicated data structure. In particular, a batch may not be considered as a run in the present sub-batch processing system for wafer fabrication, in which the processing paths of the wafers in a batch could be different. Motivated by realistic needs, this paper aims to develop a root cause detection framework for the sub-batch processing system. Briefly, the proposed framework consists of three phases: data preparation, data dimension reduction, and the sub-batch processing model construction and evaluation. The proposed approach has been validated by a sequence of simulations and an empirical study conducted in a leading semiconductor manufacturing company in Taiwan. The results have shown practical viability of the proposed approach. Indeed, the developed approach is incorporated in the engineering data analysis system in this case company.