Integrating Cluster Analysis and Artificial Neural Networks to Predict Warpage in Fan-out Panel-Level Packaging

Fan-Out Panel-Level Packaging (FO-PLP) offers high efficiency in area utilization, which helps reduce production costs. This study establishes a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$320 ~\text{mm} \times$</tex> 320 mm three dimension FO-PLP model using the Finite Element Method (FEM), combined with process modeling technology to simulate the actual manufacturing process closely, ultimately obtaining the warpage values after the debonding process. To reduce the high computational cost required for simulation, this study employs FEM to create FO-PLP models with various feature sizes, which are used as training and testing data for machine learning. Machine learning is then utilized to predict the overall warpage behavior of the FO-PLP after processing. In order to reduce the time cost of training the model and optimize its prediction accuracy, this study adopts Cluster Analysis to simplify the data points in the training dataset, selecting representative points as new training data. Subsequently, an Artificial Neural Network (ANN) is used to predict the warpage results of the package, ensuring the quality of the representative points selected by the clustering method. Additionally, these representative points effectively reduce the complexity of the model structure while maintaining good accuracy, and they help decrease the training time cost. Compared to regularization, representative points reflect similar effects in terms of model performance.