The following information was submitted:
Transactions: INTERNATIONAL JOURNAL of MECHANICS
Transactions ID Number: 17-220
Full Name: Nittaya Kerdprasop
Position: Associate Professor
Age: ON
Sex: Female
Address: Suranaree University of Technology, Computer Engineering
Country: THAILAND
Tel:
Tel prefix:
Fax:
E-mail address: nittaya@sut.ac.th
Other E-mails: nittaya.k@gmail.com
Title of the Paper: A Data Mining Approach to Automate Fault Detection Model Development in the Semiconductor Manufacturing Process
Authors as they appear in the Paper: Kittisak Kerdprasop and Nittaya Kerdprasop
Email addresses of all the authors: KittisakThailand@gmail.com, nittaya@sut.ac.th
Number of paper pages: 9
Abstract: In the semiconductor manufacturing process, fault detection is a major step of process control aiming at constructing a decision tool to help detecting as quickly as possible any equipment or process faults in order to maintain high process yields in manufacturing. Traditional statistical based techniques such as univariate and multivariate analyses have long been employed as a tool for creating model to detect faults. Unfortunately, modern semiconductor industries have the ability to produce measurement data collected directly from sensors during the production process and such highly voluminous data are beyond the capability of traditional process control method to detect fault in a timely manner. We thus propose the techniques based on the data mining technology to automatically generate an accurate model to predict faults during the wafer fabrication process of the semiconductor industries. In such process control context, the measurement data contain over 500 !
signals or features. The feature selection technique is therefore a necessary tool to extract the most potential features. Besides the feature selection method, we also propose an over-sampling technique to handle the highly imbalance situation of fail versus pass test cases. Such imbalance cases refer to rare class prediction in the data mining context. The experimental results support our assumption that choosing the right features and over-sampling rare cases can considerably improve detection accuracy of fault products and processes.
Keywords: Fault detection model, Semiconductor manufacturing process, Data mining, Rare class mining, Feature selection, Over-sampling technique.
EXTENSION of the file: .pdf
Special (Invited) Session: Data Preparation Techniques for Improving Rare Class Prediction
Organizer of the Session: 658-342
How Did you learn about congress:
IP ADDRESS: 203.158.4.227
