NEW METHOD FOR EMPIRICALLY DETERMINING SURFACE ELECTRONIC SPECIES FROM MULTIPLE-BIAS STM IMAGES - A MULTIVARIATE CLASSIFICATION APPROACH

We introduce the use of multivariate image classification techniques to distinguish and identify surface electronic species directly from multiple-bias scanning tunneling microscope (STM) images. Multiple measurements at each site are used to distinguish and categorize inequivalent electronic or atomic species on the surface via a computerized classification algorithm. Then, comparison with theory or other suitably chosen experimental data enables the identification of each class. We demonstrate the technique by analyzing dual-polarity constant-current topographs of the Ge(111) surface. Without assuming a particular atomic structure of the surface, we find that just two measurements, negative- and positive-bias topography height, permit pixels to be separated into seven different classes. To assess the validity of our classifications and determine the identity of each class, we compare our results with the accepted atomic structure of the c(2X8) reconstruction of Ge(111). Labeling four of the classes as adatoms, first-layer atoms, and two inequivalent rest-atom sites, we find very good agreement with the c(2X8) structure. The remaining classes are associated with structural defects and contaminants. This work represents a first step toward developing a general electronic/chemical classification and identification tool for multivariate scanning probe microscopy imagery.