关键词: 关键词: 统计模量, 局部近邻标准化, 局部离群因子, 多工况, 半导体过程

Abstract: A fault detection method based on statistics pattern and local nearest neighbor normalization (SP-LNS-LOF) is proposed to deal with the problem of unequal batch length, center drift, and different batch structure of multi process data. First, calculate statistics pattern of the sample; secondarly standardize statistical sample using the set of local neighborhood samples; finary calculate the local outlier factor of standard sample, and as a detection index for online fault detection. Statistics pattern extracts the information of the process of data, eliminate the impact of unequal batch length; local neighborhood normalization overcomes the difficulties of the center shift and different batch structure of process data; local outlier factor measurements similarity of samples, separates the fault samples and the normal samples. SP-LNS-LOF improves the rate of fault detection in multiple mode process. Through the simulation experiment of fault detection in semiconductor etching process, compared with PCA, KPCA, FD-kNN, LOF and other methods, the validity of the method is verified.

Key words: Keywords: statistics pattern analysis, local neigborhood standardization, local outlier factor, multimode , semiconductor process