Transfer Learning for Modeling Plasmonic Nanowire Waveguides

Area of Science:

• Plasmonics
• Nanophotonics
• Computational Physics

Background:

• Numerical simulations for plasmonic metal nanowires (MNWs) are computationally intensive, especially for complex geometries.
• Deep learning models struggle with generalization and require extensive training data for MNW analysis.

Purpose of the Study:

• To develop an efficient and accurate method for predicting the waveguiding properties of MNWs.
• To overcome the limitations of traditional simulations and direct deep learning approaches.

Main Methods:

• A physics-guided transfer learning approach was employed, leveraging knowledge from simpler MNW models.
• Basic plasmon mode knowledge was learned from free-standing circular MNWs and transferred to complex configurations.

Main Results:

• The transfer learning model achieved significant reductions in errors (~23-61%), trainable parameters (~42%), and training data (~50-80%).
• Computational time was reduced by five orders of magnitude compared to numerical simulations.
• The approach demonstrated higher accuracy and more comprehensive characterization than non-deep learning methods.

Conclusions:

• The proposed transfer learning framework offers an effective and efficient method for investigating MNWs.
• This approach can facilitate the design of polaritonic components and devices by enabling rapid property prediction.