Multi-Feature Shuffle Algorithm for Root Cause Detection in Semiconductor Manufacturing

The semiconductor industry has achieved complete automation in the manufacturing process where sensors collect a large number of process parameters online for fault detection and classification. Usually those data with a large number of parameters are complex and the process is multivariate, making it difficult for traditional methods to effectively perform root cause detection. In recent years, machine learning technology has been widely used to extract useful information from complex multi-features manufacturing process data and to detect the key process parameters that lead to relatively abnormal product quality. This paper proposes a key feature detection algorithm in the semiconductor manufacturing process based on machine learning. The classifier model in machine learning is used to map the causal relationship between features and wafer labels, and multiple actual key features are detected based on changes in the classification accuracy of the classifier before and after shuffle. Based on the simulation data set, the influence of threshold parameters, data set size, and key parameter significance on the model detection performance is analyzed. Based on multiple actual case data from a chip factory in China, the recall rate of the proposed algorithm is 100%, which has been applied to actual production.