Modified variational autoencoder for inversely predicting plasmonic nanofeatures for generating structural color.

Area of Science:

• Materials Science
• Computational Physics
• Optics

Background:

• Plasmonic composites offer tunable structural colors based on their nanoscale geometry.
• Inverse modeling is crucial for designing these materials to achieve specific colors.
• Conventional methods often lack the accuracy and flexibility required for precise color control.

Purpose of the Study:

• To develop and evaluate a modified variational autoencoder (VAE) regressor for inverse design of plasmonic composites.
• To compare the performance of the VAE-based inverse model against conventional tandem networks.
• To enhance the VAE model's accuracy through strategic data filtering.

Main Methods:

• A modified variational autoencoder (VAE) regressor was employed for inverse retrieval of topological parameters.
• A multilayer perceptron regressor linked electromagnetic response (structural color) to latent space geometrical dimensions.
• A simulated dataset was filtered prior to training to improve model performance.

Main Results:

• The VAE-based inverse model demonstrated superior accuracy compared to conventional tandem inverse models.
• The model successfully linked electromagnetic response to geometrical dimensions.
• Data filtering significantly improved the performance of the VAE model.

Conclusions:

• The modified VAE regressor offers a more accurate and efficient approach for the inverse design of plasmonic composites for structural color generation.
• This AI-driven method provides a powerful tool for tailoring material properties to desired optical outputs.
• The findings suggest a promising direction for advanced material design in optics and photonics.