Meshless optical mode solving using scalable deep deconvolutional neural network.

Area of Science:

• Photonics and Optical Engineering
• Computational Electromagnetics
• Artificial Intelligence in Engineering

Background:

• Accurate optical mode solving is crucial for advancing photonic device design and discovery.
• Existing methods can be computationally intensive and lack resolution scalability.
• The need for efficient, versatile, and portable optical simulation tools is growing.

Purpose of the Study:

• To propose a novel deep deconvolutional neural network (DNN) architecture for meshless and resolution-scalable optical mode calculations.
• To develop a method that is arbitrary in wavelength and applicable to diverse photonic materials and dimensions.
• To demonstrate the model's capability for rapid edge computing solutions.

Main Methods:

• A two-stage deconvolutional neural network architecture is employed.
• The first stage projects photonic geometrical parameters into a higher-dimensional space.
• The second stage deconvolves this vector into a mode image, utilizing scaling blocks for adjustable resolution and transfer learning for efficient training.

Main Results:

• The proposed method achieves meshless and resolution-scalable optical mode calculations.
• The model demonstrates applicability across various wavelengths, photonic materials, and dimensions.
• The deep learning approach results in a light, portable solution suitable for edge computing.

Conclusions:

• The developed deep deconvolutional neural network offers a significant advancement in optical mode solving.
• This meshless, scalable method accelerates photonic design and discovery for components like waveguides, photonic crystals, and metasurfaces.
• The model's portability and speed enable rapid deployment of edge computing-ready photonic solutions.