AI-Based Prediction of Warpage in Organic Substrates
In substrate fabrication, thermal mismatch between materials and structural configurations can readily induce warpage deformation, which severely impacts subsequent chip mounting and solder joint alignment processes, ultimately compromising the reliability of packaging integration. Therefore, invest...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11108242/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In substrate fabrication, thermal mismatch between materials and structural configurations can readily induce warpage deformation, which severely impacts subsequent chip mounting and solder joint alignment processes, ultimately compromising the reliability of packaging integration. Therefore, investigating the effects of material types and structural layouts on warpage is essential for design optimization. However, traditional experimental approaches incur high costs, while case-by-case finite element method (FEM) modeling presents computational bottlenecks. This study proposes an artificial intelligence (AI)-based method to predict the warpage behavior of organic substrates with diverse material and structural configurations. Through material preparation, substrate manufacturing, and comprehensive characterization testing, process-matched material parameters and warpage data were obtained. Combined with thermomechanical simulation, a high-precision dataset comprising 972 cross-case studies was constructed, achieving a prediction error of less than 10%. Utilizing this dataset, the network architectures and hyperparameters of Multi-Layer Perceptron (MLP), Extreme Gradient Boosting (XGB), and Gradient Boosting Machine (GBM) algorithms were optimized, and their performance was evaluated in terms of loss convergence, learning rate adaptability, training efficiency, and robustness. The results indicate that MLP demonstrates rapid loss decay, a training time of 135 seconds, a fitting slope of 0.98, minimal dependency on dataset size, and outperforms the other algorithms. This approach reduces warpage evaluation from traditional day scale to AI-driven second scale, facilitating data-driven rapid design iterations for multi-material variable-structure substrates. |
|---|---|
| ISSN: | 2169-3536 |