Modeling of photoemission characteristics for AlGaN nanoarrays photocathode accelerated by machine learning: from aspects of optical properties

Nanoarray photocathodes involve several characteristic parameters, making it difficult to achieve global optimization for optical properties. In this paper, the reflectance characteristics of AlGaN nanoarrays were parametrically scanned by COMSOL (nanorod diameter of 50-350 nm, height of 100-600 nm, substrate thickness of 50-500 nm, and wavelength of 200-400 nm), and a high-precision reflectance database covering multi-dimensional features was established. Based on the random forest regression (RFR) algorithm, a nonlinear mapping model of structural parameters and wavelength to surface reflectance was constructed. Through hyper-parameter optimization and 5-fold cross-validation, the model shows excellent performance on the test set (MSE = 0.0003, MAE = 0.0093). The residual distribution of the model satisfies mean zero with a generalization accuracy of 83.3%. The wavelength affects the reflectivity with a higher weight than the structural parameters. Diameter and substrate thickness have a slight negative effect on reflectivity, and conversely, height and incident wavelength have a slight positive effect. By dynamically coupling the reflectance predicted by RFR to the transmission 3D photoemission model, the problem of large error in the calculation of QE/EQE by the traditional average reflectance approximation is solved. The machine learning-predicted reflectance results are able to reduce the relative error of QE/EQE by 16.86%. This study realizes for the first time the whole chain optimization of structural parameters-optical response-optoelectronic performance for nanoarray photocathodes, which provides a theoretical framework and tool support for the efficient design of deep-ultraviolet optoelectronic devices.