FDTD Simulation Dataset (5 nm and 2 nm mesh) for Physics-Informed Inverse Design of Plasmonic Bowtie Nanoarrays

This repository contains the structured Finite-Difference Time-Domain (FDTD) simulation datasets used for training, validating, and testing machine learning models in the study: Physics-Informed Parametric Machine Learning Approach for Inverse Design of Gold Bowtie Nanoarrays at 785 nm Excitation The dataset comprises 7,101 geometrically distinct bowtie unit-cell configurations simulated on a 5 nm mesh, spanning gold thickness h = 100, 120 and 150 nm, triangle side length S = 180–340 nm, corner radius r = 15, 35 and 50 nm, and nanogap width δ = 0–200 nm (2 nm steps) on a fixed 780 nm square lattice, together with 789 refined top-candidate configurations re-simulated on a fine 2 nm mesh. Reflectance is provided over the 300–1100 nm range at normal incidence, with the optimization target at the Raman-relevant wavelength of 785 nm. Simulations were performed with Lumerical FDTD; file structure and naming conventions are documented in the accompanying Read_Me.txt. This research was funded by the Research Council of Lithuania (LMTLT), grant agreement No. S-PD-24-115.