Development of ANN-Based Warpage Prediction Model for FCCSP via Subdomain Sampling and Taguchi Hyperparameter...

Area of Science:

• Materials Science and Engineering
• Computational Mechanics
• Artificial Intelligence in Engineering

Background:

• Process-induced warpage in flip-chip chip-scale packages (FCCSP) is a critical reliability concern in electronic packaging.
• Accurate prediction of warpage is essential for mitigating manufacturing defects and ensuring device performance.
• Existing prediction models often lack the efficiency and accuracy required for complex thermo-mechanical behaviors.

Purpose of the Study:

• To develop a highly accurate and efficient artificial neural network (ANN) based prediction model for process-induced warpage in FCCSPs.
• To enhance ANN model performance through a novel subdomain-based sampling strategy and Taguchi hyperparameter optimization.
• To validate the developed model against experimental measurements and compare its effectiveness with existing approaches.

Main Methods:

• A process modeling approach incorporating the viscoelastic behavior of epoxy molding compound, with properties determined via dynamic mechanical measurement.
• Assessment of temperature-dependent thermal-mechanical properties using thermal-mechanical analysis and dynamic mechanical analysis.
• Development and application of a novel subdomain-based sampling strategy and Taguchi hyperparameter optimization within the ANN algorithm.

Main Results:

• The developed ANN model accurately predicts FCCSP warpage, with results validated against experimental measurements.
• Parametric analysis identified key factors influencing warpage behavior, informing the model's construction.
• The proposed sampling and hyperparameter tuning methods demonstrated superior performance compared to existing models.

Conclusions:

• The study successfully established a robust ANN-based deep learning model for predicting FCCSP warpage.
• The novel sampling strategy and hyperparameter optimization significantly enhance prediction accuracy and efficiency.
• The validated model provides a reliable tool for optimizing FCCSP design and manufacturing processes.