Deep learning-enabled optical scatterometry technique for high-precision and non-destructive measurement of grating...

Area of Science:

• Optical Metrology
• Nanoscale Science
• Advanced Manufacturing

Background:

• Conventional techniques struggle with non-destructive, high-precision nanoscale grating microstructure parameter measurement.
• Advanced optical manufacturing requires robust metrology solutions for grating characterization.

Purpose of the Study:

• To develop a novel optical scatterometry system for non-destructive, high-precision measurement of grating microstructure parameters.
• To integrate a custom deep learning architecture for rapid and accurate analysis of diffraction spectra.

Main Methods:

• A custom-designed dual-beam scatterometer with a reference-beam design for stability.
• An Adaptive Self-calibrating Physics-constrained Convolutional Neural Network (ASPCNN) for parameter retrieval.
• ASPCNN incorporates Adaptive Receptive Field Fusion (ARFF) and Self-Calibrating Residual Attention (SCRA) modules, utilizing a physics-constrained loss function.

Main Results:

• Demonstrated sub-nanometer accuracy in grating parameter measurement.
• Achieved a coefficient of determination (R²) > 0.99, outperforming established deep learning models.
• Realized a single-sample inference time of 9.07 ms, enabling real-time metrology.

Conclusions:

• The proposed system establishes a new paradigm for intelligent, high-speed, non-destructive optical metrology.
• Offers a robust solution for industrial metrology of grating microstructures.
• Advances the field of nanoscale metrology with high accuracy and efficiency.