Parameter estimation for pattern formation induced by ion bombardment of solid surfaces using deep learning

The nanostructures produced by oblique-incidence broad beam ion bombardment of a solid surface are usually modelled by the anisotropic Kuramoto-Sivashinsky equation. This equation has five parameters, each of which depend on the target material and the ion species, energy, and angle of incidence. We have developed a deep learning model that uses a single image of the surface to estimate all five parameters in the equation of motion with root-mean-square errors that are under 3% of the parameter ranges used for training. This provides a tool that will allow experimentalists to quickly ascertain the parameters for a given sputtering experiment. It could also provide an independent check on other methods of estimating parameters such as atomistic simulations combined with the crater function formalism.

Automated summary

A deep learning model has been developed to estimate all five parameters in the anisotropic Kuramoto-Sivashinsky equation using a single image of the surface, with root-mean-square errors under 3% of the parameter ranges used for training. This provides a tool for experimentalists to quickly determine the parameters for a given sputtering experiment and can serve as an independent check on other methods of estimating parameters.