1D ResNet for Fault Detection and Classification on Sensor Data in Semiconductor Manufacturing

With much attention being placed on reducing manufacturing costs and improving productivity, maintaining process tools in good operating conditions is one of the most important objectives. A huge amount of data is collected during manufacturing processes and the challenge nowadays is to efficiently make use of this massive data. In this paper, a multivariate time-series fault detection method, based on the 1D ResNet algorithm is proposed. The objective is to analyze the raw data, collected via various sensors during the semiconductor manufacturing process in order to detect abnormal wafers. For this, a set of features derived from specific tools in the manufacturing chain are selected and evaluated to characterize the wafer status. Two distinct data sets are used to validate the proposed approach. The results obtained highlight the strengths of the proposed method, which could serve as a valuable decision-making support for abnormal wafer detection in the semiconductor manufacturing process.