Identifying surface elements in STM images using neural networks

Scanning Tunneling Microscopy (STM) can easily image hydrogen-passivated silicon(001) with atomic resolution, but images often contain artifacts such as double tips, blunt tips or unstable tips. Nevertheless, a trained human eye can recognize surface details such as dimer rows, step edges, depassivated silicon, etc. A Neural Network could classify the surface better than the human eye. One advantage of a deep learning algorithm is that it can analyze, in parallel, information from multiple channels such as topography, tunneling current, forward and reverse scans. Our scanning software also collects Tunnel Barrier Height information that contributes extra information about the electronic properties of the surface at each point. This identification will be integrated into ZyVector – our STM controller product -- to provide more accurate depiction of the surface than is available in the STM image, and to automate Hydrogen Depassivation Lithography (HDL) patterning.