Accelerating the discovery of bimetallic materials for sensing applications via DFT and ML

The use of metal alloys is a promising route to improve the performance of gas-, water-, and biosensors due to their ability to engineering physical properties such as the dielectric function. To determine the best composition for a particular application, traditional methods use repeated rounds of material synthesis and characterization with high costs. Alternatively, simulation and modeling methods have enabled the expansion of databases that cover the calculated properties of known and hypothetical systems. However, evaluating the alloys' dielectric response by the independent particle approximation using DFT with adjustable smearing parameters for the inter- and intraband transitions is a rather time-consuming task. To overcome this challenge, we designed an artificial neural network trained to predict an Al-Au system's dielectric response. To confirm our prediction, we fabricated bimetallic films with different compositions and measured their optical properties at different temperatures. We find a 97% accuracy for the ML, with a time reduction of the calculation form weeks to less than one hour. Moreover, we show that all alloys outperform their pure counterparts in sensitivity, with Au_0.85Al_0.15 being the best candidate for replacing pure gold in SPR sensors.