High-precision misalignment sensing for lithography using a joint space-frequency regression network.

Area of Science:

• Semiconductor Manufacturing
• Metrology
• Image Processing

Background:

• Achieving sub-0.2 nm precision in lithography misalignment sensing is critical for semiconductor fabrication.
• Existing regression algorithms struggle with moiré fringe analysis due to limitations in single-domain information processing by convolutional networks.
• This deficit in feature data hinders high-precision regression for accurate misalignment measurements.

Purpose of the Study:

• To develop an advanced convolutional regression network for precise lithography misalignment sensing.
• To overcome the limitations of single-domain analysis in current regression algorithms.
• To synthesize spatial and frequency domain information for enhanced fringe pattern analysis.

Main Methods:

• Engineered a novel convolutional regression network.
• Integrated spatial and frequency domain information from fringe patterns.
• Applied the network to analyze moiré fringes for misalignment detection.

Main Results:

• Achieved a misalignment measurement accuracy of 0.12 nm at a 3σ confidence level.
• Demonstrated significant improvements over existing methods requiring sub-0.2 nm precision.
• The method proved robust against system errors and environmental noise.

Conclusions:

• The developed convolutional regression network effectively synthesizes multi-domain fringe information for high-precision lithography misalignment sensing.
• This approach successfully addresses the limitations of single-domain analysis, enabling sub-0.2 nm accuracy.
• The technique's robustness makes it suitable for critical applications in semiconductor manufacturing.