Avaliação da rugosidade de filmes de Molibdênio fabricados por RF Sputtering através do uso de Deep Learning

Abstract

One of the main factors responsible for the advancement of many technologies produced since the beginning of the 20th century were thin films. Directly or indirectly, these have become essential for human life as they are present in the most diverse applications in the optics, electronics, medicine, energy and many other industries. For these applications to be developed it was necessary to manufacture them beforehand. That said, this work proposes to evaluate an ANN for the development of a Regression model to optimize the fabrication of Thin Mo Films by RF Magnetron Sputtering from surface roughness. Three manufacturing parameters were then selected to optimize: surface treatment (cleaning with Hexane, Electropolishing), deposition time (5, 10, 20 and 30 minutes) and power (40W and 60W). By varying these parameters, 12 samples were developed, and from a profilometer the roughness (Ra) and the morphological profile of each film were calculated. These results were then applied to ANN where a Regression model was developed. It was observed that the type of surface treatment was the most influential parameter in RA, where electropolishing reduced by approximately 83.3% when compared to cleaning with Hexane. Then, the power reduction caused a reduction in the value of Ra. Regarding the deposition time, for Electropolishing, the increase in time caused a reduction in roughness, on the other hand, the increase in time for substrates cleaned with Hexane caused an increase in roughness. Finally, the regression model was evaluated by the value of R2, which presented a high value of 98.76%, and with this optimization curves were created. Therefore, RNA proved to be an excellent tool to optimize the process of manufacturing Mo films by RF Magnetron Sputtering.