Generative-adversarial-network-based dimensional measurement of optical waveguides.

Area of Science:

• Optical Engineering
• Materials Science
• Artificial Intelligence

Background:

• Accurate measurement of waveguide dimensions is critical for photonic device performance.
• Traditional methods can be time-consuming or lack the required precision.
• Integrating artificial intelligence with optical microscopy offers potential for enhanced metrology.

Purpose of the Study:

• To propose a high-throughput and precise method for measuring waveguide dimensions.
• To leverage a generative adversarial network (GAN) for image analysis.
• To validate the accuracy of the proposed measurement technique.

Main Methods:

• A novel method combining a generative adversarial network (GAN) with a curve-fitting calculator using sidewall functions.
• Acquisition of low-magnification (LM) and high-magnification (HM) optical microscope images of waveguides at 500× and 2000×, respectively.
• Training the GAN to learn discrepancies between LM and HM images for dimensional analysis.

Main Results:

• Achieved a standard deviation of waveguide widths of approximately 0.8 pixels (∼ 42 nm).
• Demonstrated precise width measurement capabilities.
• Confirmed that the method provides precise measurements at the same imaging throughput as LM microscopy.

Conclusions:

• The proposed GAN-based method offers a high-throughput and precise solution for waveguide dimensional measurement.
• This technique integrates advanced AI with optical microscopy for improved metrology.
• The results show potential for enhanced quality control and device optimization in integrated photonics.