Fault detection based on difference locality preserving projections for the semiconductor process

Some existing fault detection methods for the semiconductor process, such as principal component analysis and locality preserving projections (LPP), are linear algorithms, so they degrade the performance of fault detection in a nonlinear process. In addition, they are not effective for fault detection in a multimodal process. To solve the problems caused by nonlinear and multimodal characteristics in the semiconductor process, a new difference (DIF) pre-processing strategy is proposed to normalize the nonlinear and multimodal data. After detailed analysis of DIF, a new method called DIF-LPP is developed for fault detection in the semiconductor process. The nonlinear and multimodal data can be transformed into data sets that follow a Gaussian and single mode distribution, respectively. The proposed method contains a model without prior knowledge of the nonlinear and multimodal process. To demonstrate the proposed method's effectiveness, it is applied to 2 numerical examples and the semiconductor process. Simulation results verify that the proposed method is effective for fault detection in the nonlinear and multimodal process. To solve the problems caused by nonlinear and multimodal characteristics in the semiconductor process, a new difference (DIF) preprocessing strategy is proposed. After detailed analysis of DIF, a new method called DIF-LPP is developed for fault detection in the semiconductor process. The nonlinear and multimodal data can be transformed into data sets that follow a Gaussian and single mode distribution, respectively. To demonstrate the proposed method's effectiveness, it is applied to 2 numerical examples and the semiconductor process.