关键词: 混合特征选择, 超参优化, 随机森林, 支持向量机, 序列模型优化, 晶圆, 蚀刻缺陷预测

Abstract: To improve the defect prediction accuracy in semiconductor wafer manufacturing process,a defect prediction method based on hybrid feature selection and Sequential Model-Based Optimization(SMBO)was introduced.Based on random forest model and support vector machine model of which were both have good applicability to high-dimensional,multi-noise,multi-modal and linearly inseparable data.The stability selection based on random forest model was used to evaluate features,then the features in descending order were searched based on sequential forward selection method,and the support vector machine classification model was constructed accordingly.SMBO method was used to optimize the model,and the best performance model with minimum number of features was chosen for defect prediction.To verify the effectiveness and advantage of the proposed method,the actual engineering data of residue defect and pits defect of Etch process were used to compare and analyze,and the result showed that the proposed method had excellent recognition ability for defects in wafer manufacturing process.

Key words: mixed feature selection, hyper-parameter optimization, random forest, support vector machine, sequential model-based optimization, wafer, etch defect prediction