Data science framework for variable selection, metrology prediction, and process control in TFT-LCD manufacturing

TFT-LCD panel manufacturers rely on experimental design and engineering experience for process monitoring and quality control throughout the production line. To shorten production and reduce the cost of labor resources, this study proposes a three-phase data science framework embedded with several data mining and machine learning techniques, which can identify the variables affecting yield, predict the metrology result of photo spacer process, and suggest the process control in the color filter manufacturing process. An empirical study of Taiwan's leading TFT-LCD manufacturer is conducted to validate the proposed framework. The results indicate that the proposed framework effectively and quickly selects the important variables, predicts the metrology result with higher performance, and identifies the main effect and interaction effect of the selected variables for yield improvement.