Abstract
The lengthy manufacturing processes of thin film transistor-liquid crystal displays (TFT-LCDs) are complex, in which many factors can cause different types of defects on the panel and result in low yield. Examples are line defects, point defects, and Mura defects. Engineers rely on personal experience for trouble shooting during TFT-LCD manufacture, which does not quickly locate possible fault root causes using their own domain knowledge or rules of thumb. In a fully automated manufacturing environment in TFT-LCD factories, large amounts of raw data are increasingly accumulated from various sources, automatically or semi-automatically, for fault diagnosis and process monitoring. This study aims to propose a data mining framework for diagnosing the root causes of defects in factories. The extracted information and knowledge is helpful to engineers as a basis for trouble shooting and defect diagnosis. To examine the validity of this approach, an empirical study was conducted in a TFT-LCD company in Taiwan, and the results demonstrated the practical viability of this approach.