76‐3: TEG Electrical Virtual Measurement and Monitoring Based on Interpretable Machine Learning Method

In the process of semiconductor display manufacturing, many testing devices are used to perform sampling inspection of panel characteristics, monitor the quality of the production process, and ensure the product yield. With the development of artificial intelligence (AI), data‐driven virtual Measurement (VM) AI models can predict the characteristics of display backplane devices in real time based on process parameters, and predict the device characteristics of display backplane devices in advance (usually > 2 days). In this paper, an AI prediction method based on the fusion of random forest and deep learning is designed, which can be used to predict the on‐state current Ion & threshold voltage Vth values that affect the performance of the display panel in real time. The prediction accuracy of our proposed model is more than 95%, the MAPE is less than 10%, and the R 2 is between 0.5 and 0.6. Also, the proposed model has good interpretability, and can assist in locating the candidate parameter factors that cause TEG values to exceed control. The model is deployed in the production environment and linked to the SPC system, which can achieve 100% virtual measurement and anomaly monitoring and interception. The model can be automatically updated regularly on the platform to alleviate the decline of prediction effect caused by equipment data drift and model degradation.