Multiple-parameter optimization of AlGaN nanoarrays based on optical absorption accelerated by machine learning

Abstract

AlGaN nanorod arrays (NRAs) have considerable significance in optoelectronic devices. However, due to the specificities of their structure, it is imperative to address the global optimization of material parameters, geometrical parameters, and system parameters. In this study, a combination of COMSOL simulation and machine learning is utilized to investigate the absorption under global parameter adjustment. Firstly, the optical absorptivity of Al_0.35Ga_0.65N modified by a single parameter is simulated and analyzed. The absorption of NRAs exceeds that of thin films. Moreover, the absorption is notably enhanced under forward incidence (FI) of photons compared to reverse incidence (RI). The optimal parameters for NRAs under FI are a diameter of 250 nm, a height of 600 nm, a substrate thickness of 50 nm, and an arrays period of 250 nm. Using the simulated results as dataset for machine learning, three regression forecasting models (RFR, DTR and NNR) are adopted. These models are meticulously trained and subsequently utilized to predict the absorption. The results indicate that RFR owns the highest accuracy with R² of 0.97 and MAE of 3.66. The validation set is additionally employed to verify the accuracy of RFR with 96 % area within the differential heatmap closed to zero. This research is expected to offer valuable insights for the structural design and optimization methods of AlGaN NRAs in the application of photocathodes, solar cells and photodetectors.