Deep learning approach for an interface structure analysis with a large statistical noise in neutron reflectometry

Abstract

Abstract Neutron reflectometry (NR) allows us to probe into the structure of the surfaces and interfaces of various materials such as soft matters and magnetic thin films with a contrast mechanism dependent on isotopic and magnetic states. The neutron beam flux is relatively low compared to that of other sources such as synchrotron radiation; therefore, there has been a strong limitation in the time-resolved measurement and further advanced experiments such as surface imaging. This study aims at the development of a methodology to enable the structural analysis by the NR data with a large statistical error acquired in a short measurement time. The neural network-based method predicts the true NR profile from the data with a 20-fold lower signal compared to that obtained under the conventional measurement condition. This indicates that the acquisition time in the NR measurement can be reduced by more than one order of magnitude. The current method will help achieve remarkable improvement in temporally and spatially resolved NR methods to gain further insight into the surface and interfaces of materials.