Machine learning-based leaky momentum prediction of plasmonic random nanosubstrate

Area of Science:

• Optoelectronics and Nanophotonics
• Machine Learning Applications
• Plasmonics

Background:

• Surface plasmon polariton (SPP) propagation is crucial for nanophotonic devices.
• Leakage radiation offers a method for visualizing and analyzing SPP modes.
• Current methods for leakage radiation characterization can be complex and equipment-intensive.

Purpose of the Study:

• To develop a machine learning framework for constructing leakage radiation characteristics from bright-field images of nanoislands.
• To simplify the process of leakage radiation image generation.
• To enable efficient analysis of SPP propagation.

Main Methods:

• Utilized a plasmonic random nanosubstrate.
• Employed a fast-learning two-layer neural network.
• Trained the network to predict leakage radiation characteristics from bright-field images using limited data.

Main Results:

• Successfully demonstrated the prediction of leakage radiation characteristics from bright-field images.
• Showcased the effectiveness of a two-layer neural network for this image-to-image prediction task.
• Achieved this with a limited number of training samples.

Conclusions:

• The proposed machine learning approach significantly simplifies leakage radiation image construction.
• This method reduces the reliance on sophisticated experimental equipment.
• The framework has potential for broad applications in nanophotonics and image prediction.