Pattern-Aware Intelligence Enables Nondestructive, Rapid Quantification of High-Aspect-Ratio Silicon Etching

Area of Science:

• Materials Science and Engineering
• Artificial Intelligence in Manufacturing
• Semiconductor Device Fabrication

Background:

• High-aspect-ratio (HAR) structures are crucial in microelectromechanical systems (MEMS), electronics, and advanced packaging.
• Current characterization methods are destructive and fail to capture complete 3D profiles, hindering process control and device reliability.
• Existing metrology lacks the ability to reconstruct the full three-dimensional profile of HAR structures.

Purpose of the Study:

• To develop a pattern-aware intelligent model for rapid, minimally destructive 3D reconstruction of HAR structures.
• To integrate pattern-dependent etching behavior as a physical prior for enhanced prediction accuracy.
• To enable near-real-time process feedback for intelligent manufacturing.

Main Methods:

• A YOLO-Pose-based model was developed to extract subpixel geometric features from scanning electron microscopy (SEM) images.
• Pattern-dependent etching effects were incorporated as a physical prior to create a topography network.
• Minimal destructive observations were used for training and validation.

Main Results:

• The YOLO-Pose model achieved 95.11% quantification accuracy for key dimensions (R² >0.98).
• The framework demonstrated 93.96% prediction accuracy for critical parameters (etch depth, sidewall angle, scallop texture) using non-destructive retrieval.
• The model showed robust cross-scale generalization across varying layout densities and reduced data acquisition time to 1 minute.

Conclusions:

• The proposed pattern-aware intelligent model offers a highly interpretable, cost-effective, and intelligent solution for HAR structure characterization.
• This approach enables rapid, minimally destructive 3D reconstruction and near-real-time process feedback, advancing intelligent manufacturing.
• The study highlights the potential of AI and physical priors in revolutionizing semiconductor metrology and process control.