Fault detection for semiconductor quality control based on Spark using data mining technology

The aim of this paper is to discuss how to apply data mining technology to semiconductor manufacturing process quality control. The significance of this paper is that it solves a practical engineering problem and is not limited to theoretical analysis. This paper proposes and completes a complete semiconductor quality control program, including the problem analysis, the field semiconductor data collection, the data preprocessing, the feature selection, the classification model selection, the model building, the model testing, the model contrast, and the model improvement. The data preprocessing includes data cleaning, data standardization, data formatting, and so on. And the paper uses the Fisher criterion algorithm to select useful features. In addition, this paper selects two data mining algorithms (SVM and Random Forest) based on the distributed computing platform (Spark) and establishes the corresponding models for analysis. After analyzing and comparing the models, it is found that Random Forest has stronger anti-overfitting ability and is more suitable for the semiconductor quality control.