A Bidirectional Design Method for Through-Glass Vias with Selective Laser Wet Etching Based on the Cross-Modal...

Area of Science:

• Materials Science
• Optical Engineering
• Semiconductor Manufacturing

Background:

• Through-Glass Vias (TGVs) are essential interposers in advanced packaging, particularly for Co-packaged optics (CPO).
• Predicting TGV parameters and etching morphology accurately remains challenging due to complex laser wet-etching process parameters.
• Existing methods struggle with bidirectional prediction, hindering optimal TGV design.

Purpose of the Study:

• To propose a novel bidirectional design method for TGVs.
• To enable precise prediction of TGV parameters and etching morphology.
• To enhance TGV quality for advanced packaging applications.

Main Methods:

• Integration of a Cellular Automaton Etch-Diffusion (CAED) physical model with a Stable Diffusion (SD) deep learning architecture.
• Application of Contrastive Language-Image Pre-training (CLIP) for efficient inverse design.
• Utilizing cross-modal learning for TGV parameter and morphology prediction.

Main Results:

• Successful accurate forward prediction from laser parameters to TGV morphology.
• Achieved efficient inverse design of TGVs using the CLIP model.
• Demonstrated significant robustness and stability of the generative model, confirming generalization ability.

Conclusions:

• The proposed cross-modal learning method provides an efficient approach for TGV design.
• This deep learning framework enhances TGV quality and manufacturing predictability.
• The bidirectional design method offers a robust solution for challenges in TGV fabrication.