Plasmonic metamaterials allow the engineering of the photonic density of states and thus enable control over spontaneous and stimulated emission. Plasmonic nanomaterials are particularly interesting for applications in sensing, imaging, and catalysis where they can produce dramatic performance improvements due to the local electromagnetic field enhancement effect at metal-dielectric interfaces. While plasmonic nanospheres are ubiquitous in nanoscience and nanotechnology, other shapes such as nanoprisms, nanocubes, nanoshells, and nanorods are becoming increasingly common. There has been considerable progress in tuning the shape of plasmonic nanomaterials.

This family includes Ag, Au, Cu, Al, HfN, ZrN, TiN, TaN, Cu 2S, and a few other compounds. There exists a relatively small family of materials that exhibit localized surface plasmon resonances (LSPR) at visible and near-infrared wavelengths in their nanostructured forms. Our results suggest that the DL network learns the underlying physics governing the optical response of plasmonic core-shell nanocylinders, which in turn builds trust in the use of DL methods in materials science and optoelectronics. We probed deeper into these networks’ architecture to explain how the optimized network could predict the final results. In addition, we addressed an issue related to DL methods, namely explainability. A combination of a CNN and MLP was designed to accept the cross-sectional images of cylindrical plasmonic core-shell nanomaterials as input and rapidly generate their optical response. Nanotubes of Si, Ge, and TiO 2 coated on either their inner wall or both their inner and outer walls with a plasmonic noble metal (Au or Ag) were thus modeled. Here, we used the combination of a computer vision technique, namely convolutional neural network (CNN), with multilayer perceptron (MLP) to obtain the far-field optical response at normal incidence (along cylinder axis) of concentric cylindrical plasmonic metastructures such as nanorods and nanotubes. The application domain of deep learning (DL) has been extended into the realm of nanomaterials, photochemistry, and optoelectronics research.

Lumerical fdtd examples