AI-driven feature recognition of SEM profiles in deep reactive ion etching based on physics-constrained variational...

Area of Science:

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

Background:

• Deep reactive ion etching (DRIE) is crucial for microelectromechanical systems (MEMS) but faces optimization challenges due to its complex, parameter-dependent nature.
• Current scanning electron microscopy (SEM) analysis for DRIE outcomes is labor-intensive, prone to errors (15-20%), and hinders high-throughput manufacturing.
• Existing automated methods struggle with accuracy (70-80%) in noisy SEM images and fail to capture the dynamic evolution of etched structures.

Purpose of the Study:

• To develop an automated SEM sectional-profile analysis method for DRIE processes.
• To address the limitations of manual and existing automated SEM analysis in terms of accuracy, efficiency, and data acquisition for AI-driven optimization.
• To enable precise extraction of critical dimensions from DRIE profiles for improved process control and manufacturing.

Main Methods:

• Proposed a physics-constrained variational level set autoencoder (VLSet-AE) for automated SEM sectional-profile analysis.
• Integrated physical etching constraints and a three-dimensional framework (time, linewidth, etching depth) into the VLSet-AE model.
• Evaluated VLSet-AE's performance against manual annotation and state-of-the-art models using key dimensional metrics and computational efficiency.

Main Results:

• VLSet-AE achieved high accuracy in extracting nine critical dimensions, with an average error of 3.65% and an overall model accuracy of 94.3%.
• The model significantly outperformed manual annotation and current alternatives in accuracy and efficiency.
• VLSet-AE demonstrated the shortest training time (20s), fastest inference time (1.2s), and highest recognition accuracy (96%) compared to seven other models.

Conclusions:

• VLSet-AE provides an efficient and accurate solution for automated SEM analysis in DRIE processes, enabling large-scale data acquisition for AI-optimized manufacturing.
• This physics-constrained AI approach empowers scalable, intelligent manufacturing and unlocks the potential for advanced microfabrication technologies.
• The study presents a forward-looking framework for AI-driven MEMS process design and manufacturing, paving the way for future AI-assisted microfabrication advancements.