A physics-based machine learning approach for modeling the complex reflection coefficients of metal nanowires

Metal nanowires are attractive building blocks for next-generation plasmonic devices with high performance and compact footprint. The complex reflection coefficients of the plasmonic waveguides are crucial for estimation of the resonating, lasing, or sensing performance. By incorporating physics-guided objective functions and constraints, we propose a simple approach to convert the specific reflection problem of nanowires to a universal regression problem. Our approach is able to efficiently and reliably determine both the reflectivity and reflection phase of the metal nanowires with arbitrary geometry parameters, working environments, and terminal shapes, merging the merits of the physics-based modeling and the data-driven modeling. The results may provide valuable reference for building comprehensive datasets of plasmonic architectures, facilitating theoretical investigations and large-scale designs of nanophotonic components and devices.

Automated summary

In this study, we propose a method to convert the specific reflection problem of metal nanowires into a universal regression problem, efficiently and reliably determining both the reflectivity and reflection phase of the nanowires. Our approach combines the advantages of physics-based modeling and data-driven modeling.