P‐45: Development of an AI Model for Defect Detection Considering Manufacturing Variability

To evaluate the risk points and levels of display panels caused by manufacturing variations, a numerous number of simulations are required. Traditional reliability assessment method, such as Finite Element Analysis (FEA) are time‐consuming for this purpose. Therefore, this study has developed an AI‐based model that can predict panel defects, enabling efficient simulations for numerous scattered cases within a short timeframe, serving as a practical alternative to FEA. Initially, structural analysis are conducted following the reliability evaluation technique through conventional FEA simulations. The likelihood and location of defects are determined by examining strain contours derived from the FEA results. In the panel layout, manufacturing tolerances that may occur include critical dimension (CD), overlay, thickness, and taper angle. To secure data, all manufacturing tolerances are assumed to randomly occur in each layer, and 100 cases of panel structures are modeled and analyzed. The solving time for a FEA simulation takes about 3 hours. The finite element (FE) model used in FEA simulation is preprocessed to be input data into the neural network model. And strain contours derived from FEA simulation results are preprocessed to be output data into the neural network model. We developed and optimized an AI model capable of taking input from the preprocessed data. Input and output data are highdimensional, and a fully convolutional network (FCN) model is employed to handle these data. After completing the training of the AI model, we input a panel with a new structure into the AI model and predicted the strain contour. We compared the predicted contours with the ground truth strain contours by performing FEA on a panel of the same structure. The results revealed an inference accuracy of 92.02% for the AI model, with a prediction time of around 17 seconds. This indicates a significant time‐saving effect of 99.84% compared to FEA. It is expected that a faster structural reliability assessment will be conducted through this method.