Discovery of New Plasmonic Metals via High-Throughput Machine Learning

The field of plasmonics aims to manipulate and control light through nanoscale structuring and choice of materials. Finding materials with low-loss response to an applied optical field while exhibiting collective oscillations due to intraband transitions is an outstanding challenge. This is viewed as a materials selection problem that bridges the gap between the large number of candidate materials and the high computational cost to accurately compute their individual optical properties. To address this, online databases that compile computational data for numerous properties of tens to hundreds of thousands of materials are combined with first-principles simulations and the Drude model. By means of density functional theory (DFT), a training set of geometry-dependent plasmonic quality factors for ≈1000 materials is computed and subsequently random-forest regressors are trained on these data. Descriptors are limited to symmetry, quantities obtained using the chemical formula, and the Mendeleev database, which allows to rapidly screen 7445 candidates on Materials Project. Using DFT to compute quality factors for the 233 most promising materials, AlCu₃, ZnCu, and ZnGa₃ are identified as excellent potential new plasmonic metals. This finding is substantiated by analyzing their electronic structure and interband optical properties in detail.