Autofocus control using adaptive region selection and reinforcement learning applied in wafer micro-imaging.

Area of Science:

• Semiconductor Manufacturing
• Computer Vision
• Artificial Intelligence

Background:

• Wafer micro-imaging requires precise autofocus, but inconsistent focal distances across regions pose a challenge.
• Existing autofocus algorithms often lack generalization for diverse wafer inspection areas.

Purpose of the Study:

• To propose a generalized reinforcement learning (RL) approach for personalized autofocus control in wafer micro-imaging.
• To enhance focusing quality and eliminate manual adjustments in wafer inspection.

Main Methods:

• Developed a deep network within an RL framework to estimate focal distances from image frames.
• Utilized engineer feedback to fine-tune personalized models predicting optimal focal distances for regions of interest.
• Employed the Gaussian policy gradient algorithm for policy network updates.

Main Results:

• Created a dataset of wafer images for training and validation.
• The proposed network demonstrated improved generalization across different wafer regions.
• Achieved an average improvement of approximately 4.0% in focusing quality compared to existing methods.

Conclusions:

• The generalized RL approach effectively addresses inconsistent focal distances in wafer micro-imaging.
• Personalized autofocus control enhances imaging quality and reduces the need for manual intervention.
• Offers novel insights for advancing wafer micro-imaging techniques.