Simulação do enrugamento cinético no crescimento de interfaces em materiais multicamadas utilizando aprendizado de máquina

Abstract

Due to the increase in the amount of data available globally and the constant evolution of training models and algorithms, machine learning techniques are becoming increasingly popular, whether in industry or academia. In the area of materials science, models involving computational simulation of surface growth in situations outside of thermodynamic equilibrium help to understand and improve the production of new materials. Based on these assumptions, this work has as main objective to use machine learning models to predict the temporal evolution of global roughness, considering the ballistic deposition model. The objective is to perform a prediction of systems with lateral sizes that make it impossible to obtain the steady state for the global roughness via computational simulation. We used the root mean squared error as a validation parameter, in order to assess how successful the learning was performed by the machine. We can highlight as results the access to the temporal evolution of global roughness, as well as its steady state, for lateral sizes that have not yet been observed in the literature, and from these, we explore situations for interfacial growth considering one-dimensional and two-dimensional substrates, with results consistent for the predicted roughness exponents in the thermodynamic limit.