Optimization of photonic crystal nanocavities based on deep learning

Area of Science:

• Photonics and optical engineering
• Artificial intelligence and machine learning
• Materials science

Background:

• Two-dimensional photonic crystals (2D-PCs) are crucial for advanced optical devices.
• Optimizing nanocavity Q factors is essential for improving device performance.
• Traditional optimization methods struggle with high-dimensional parameter spaces.

Purpose of the Study:

• To develop a deep learning-based approach for optimizing the Q factors of 2D-PC nanocavities.
• To demonstrate the effectiveness of this AI-driven method in achieving ultra-high Q factors.
• To enable efficient exploration of complex design parameters in photonic structures.

Main Methods:

• Generation of a dataset of 1000 nanocavities with random air hole displacements.
• Calculation of Q factors using a first-principles method.
• Training a four-layer convolutional neural network to predict Q factors and their gradients.
• Utilizing back-propagation for rapid gradient estimation in high-dimensional spaces.

Main Results:

• The trained neural network estimates Q factors with a 13% standard deviation error.
• An optimized nanocavity achieved an ultra-high Q factor of 1.58 × 10⁹.
• The optimized Q factor is over ten times higher than the base cavity and significantly surpasses existing records.

Conclusions:

• Deep learning offers a powerful and efficient method for optimizing 2D-PC nanocavities.
• This AI approach overcomes limitations of traditional methods for large parameter spaces.
• The demonstrated technique can be applied to enhance other optical characteristics of photonic devices.