P‐46: Exploration of AI Applications of Neural Networks in TFT‐LCD Film Thickness Prediction

This paper presents a virtual model for predicting the thickness of TFT LCD films based on neural networks. The Python programming environment, coupled with neural networks, was utilized to conduct machine learning training on over 100,000 sets of FGI coating parameters and their corresponding film thicknesses. The output of this training is an FGI film thickness prediction model, and a VM (Virtual Model) management system was developed to implement the model's application. The results indicate that the residuals between the model's predicted values and the actual measured values remain within 30 (with an FGI tolerance range of ±629), and the standardized residuals are 0.48, meeting the ±2 range, demonstrating high prediction accuracy of the model. This virtual model enables the prediction of film thickness within 10 seconds of panel output from the CVD coating unit, significantly improving the monitoring efficiency of film thickness compared to traditional measurement methods that require 10 minutes to output measured values. Additionally, the VM management system visually displays the trends of coating parameters, allowing technical departments to quickly adjust parameters based on trend fluctuations, effectively reducing film thickness fluctuations. Through monitoring with the virtual model, we discovered periodic variations in FGI film thickness, which are related to chamber cleanliness. This discovery breaks through the limitations of conventional monitoring. This case successfully demonstrates the application value of neural networks in predicting FGI film thickness in TFT LCD production.