The determination of unknown crystal defect structures by iterative digital image matching (“quantitative HREM”) has previously been successfully applied to grain boundaries (e.g. in niobium [1]) or phase boundaries with small misfit (e.g. Nb-sapphire [2]), which offer several advantages: (i) the supercell is small leading to fast cycle times in iterative structure refinement and thus to a secure global optimum, (ii) noise-filtering is possible by translation symmetry, (iii) the interface structure is mostly planar (i.e. perfectly columnar). For dislocations and other aperiodic defects all these three issues are missing and curved atomic columns from surface strains especially cause problems [3]. Any progress in this area must therefore address (i) - (iii) separately and estimate the change of the confidence level of the retrieved structure when switching from interface-refinement to dislocation-refinement. First results of this project address questions of convergency and speed.
Since each HREM-Lab is only concerned with a limited number of materials (and elements) as well as with a few microscope-voltages only, it is advisable to calculate transmission functions for single atoms for each element and voltage a single time, and store these data files in a library.