Fisher Discriminant Analysis for Semiconductor Batch Tool Matching

We propose in this paper how Fisher discriminant analysis can be applied for differentiating classes of semiconductor data from different tools or chambers. The tool and chamber matching analysis can be useful not only from a process characterization standpoint, but also for identifying the proper fault detection and classification strategy. If the FDA analysis shows that the chambers or tools are not matched, it is likely that either preventive maintenance will be needed, or a separate fault detection and diagnosis model will be required for each tool or chamber. Alternatively, if the match fraction is large enough, a global model should be considered for monitoring a plurality of tools or chambers to reduce the model maintenance effort. Contribution analysis is also applied to identify variables responsible for tool mismatching. Copyright © 2007 International Federation of Automatic Control All Rights Reserved.