Computer Vision Applied to In-Situ Specimen Orientation Adjustment for Quantitative SEM Analysis

Abstract Quantitative analysis methods based on the usage of a scanning electron microscope (SEM), such as energy-dispersive X-ray spectroscopy, often require specimens to have a flat surface oriented normal to the electron beam. In-situ procedures for putting microscopic flat surfaces into this orientation generally rely on stereoscopic methods that measure the change in surface vector projections when the surface is tilted by some known angle. Although these methods have been used in the past, there is no detailed statistical analysis of the uncertainties involved in such methods, which leaves an uncertainty in how precisely a specimen can be oriented. Here, we present a first principles derivation of a specimen orientation method and apply our method to a flat sample to demonstrate it. Unlike previous works, we develop a computer vision program using the scale-invariant feature transform to automate and expedite the process of making measurements on our SEM images, thus enabling a detailed statistical analysis of the method with a large sample size. We find that our specimen orientation method is able to orient flat surfaces with high precision and can further provide insight into errors involved in the standard SEM rotation and tilt operations.