Distortion removal in wafer surface interference fringe patterns based on large language models.

Area of Science:

• Optical Metrology
• Artificial Intelligence in Manufacturing
• Semiconductor Inspection

Background:

• Wafer surface inspection using interference fringes is crucial for semiconductor manufacturing.
• Accuracy is compromised by optical aberrations, speckle noise, and unmodeled distortions.
• Current distortion removal methods lack generalizability.

Purpose of the Study:

• To develop a robust method for restoring distorted fringe patterns using large language models (LLMs).
• To enhance the accuracy and reliability of wafer surface inspection.
• To provide a generalizable solution for diverse optical distortions.

Main Methods:

• Utilized a large language model (LLM) for fringe pattern restoration.
• Employed a cross-modal guidance strategy to align fringe features with LLM's semantic latent space.
• Incorporated a fringe-fidelity constrained mechanism for high-quality restoration.
• Generated a large-scale synthetic dataset using physics-based modeling.

Main Results:

• Demonstrated superior performance in restoring distorted fringe patterns under various conditions.
• Achieved robust restoration against optical aberrations, speckle noise, and combined distortions.
• Showcased strong generalization capabilities to unseen distortion types.
• Validated performance in controlled and real-world scenarios.

Conclusions:

• The proposed LLM-based method offers a reliable preprocessing solution for high-precision wafer surface profilometry.
• The approach significantly improves the accuracy of wafer inspection by effectively handling distortions.
• This method enhances the practical deployment of optical metrology in semiconductor manufacturing.