Abstract
Semiconductor fabrication involves highly complex and lengthy processes in which a large number of variables are interrelated. Process control and monitoring are necessary in a semiconductor fab to ensure the yield and thus the profitability of huge investments. This study aims to develop a framework for Fault Detection and Classification (FDC) and statistical models to be embedded in the FDC system to monitor the semiconductor fabrication process in which multi-faults may exist. The objective of the proposed framework is to structure the process operation from a large number of correlated variables, to detect faults, diagnose them by clustering and classifying the abnormal events, eliminate the cause of the faults, and then improve the performance of the process. The information and knowledge are discovered and extracted by the multivariate statistical methods and data mining approaches form the historical process data, and can be an aid of fault diagnosis and recovery. Simple rules can be generated to classify and predict the wafers. Also, an empirical study is conducted in an advanced 300mm DRAM fab for validation. The results showed the practical viability of this approach. The developed model can be embedded in a Fault Detection System for process or equipment monitoring. As a result of routine monitoring process and the extracted information from process data, critical variables of the process are identified, and the faults can be removed, process excursion is decreased. Moreover, the monitoring process are simplified by reducing the number of variables in the system, fewer key variables are monitored by the FDC instead of all the variables in the system.