Neural network evaluation of geometric tip-sample effects in AFM measurements

Abstract

Abstract The geometrical design of probe-tips used in atomic force microscopy (AFM) can have a significant impact on the quality of scanned measurement images which are directly related to the roughness of the sample surface being investigated. When a non-suitable probe-tip is selected to scan a given sample surface, this can give rise to imaging artifacts which result from geometrical tip-sample interactions as the true image of the surface being scanned will be morphologically dilated by the tip. In this paper, the impact of these imaging artifacts on AFM measurement accuracy can be predicted by using a multi-layered neural network, by involving surface roughness and the geometry of a given AFM rigid probe-tip. The neural network is trained by using data produced from simulated AFM measurements, and shows a good level of fit for both training and test sets. Evaluation of tip-sample effects using such a machine learning approach could allow for a fast and efficient probe-tip selection process that can reduce the operational costs of an AFM measurement.