10www.microscopy-today.com • 2010 Marchdoi: 10.1017/S1551929510000040Risks of “Cleaning” Electron Backscatter Diffraction DataL.N. Brewer* and J.R. MichaelSandia National Laboratories, P.O. Box 5800, Albuquerque, NM 87185* [email protected] Collecting good data is an important task, but handling the data correctly is important also. How to handle data largely depends on what the analyst is going to do with it. Electron backscatter diffraction (EBSD) is no exception. Electron backscatter diffraction is a widely used technique for collecting crystallographic information at micrometer and even nanometer scales [1, 2]. An EBSD orientation map, or inverse pole figure (IPF) map, is acquired in a scanning electron microscope (SEM) by scanning the electron beam over an area of interest. An orientation map is produced by placing the beam sequentially at a series of points on the surface of the samples and performing several operations: collection of an electron backscatter diffraction pattern, detection of the lines in the pattern using a Hough transform, calculation of the orientation of the crystal based on candidate crystal structures for the specimen, and storage of this orientation information (Figure 1). The electron beam is then moved and the process repeated until a map of orientation data is complete. Currently, EBSD can be used to map crystallographic parameters such as phase type and fraction, grain size, orientation distribution, and deformation content. These measurements can be made rapidly, routinely in excess of 100 patterns per second. A good example of standard orientation mapping of an iron-based, magnetic alloy is shown in Figure 2. This image represents 585,000 pixels and 95.7 percent of those pixels were successfully indexed by the software. The unindexed pixels mainly reside on grain boundaries and at second phase particles (small black areas) that were not included in the list of phases to be considered. The orientation map in Figure 2 is an inverse pole figure, and its colors are explained by the color key on the stereographic triangle. An inverse pole figure can be plotted with respect to any physical direction. Normally, either x, y, or z is chosen—the direction that has the most physical significance, for example, rolling direction, growth direction, etc. The choice of indices is based on a given stereographic triangle. Because the crystal structure in Figure 2 is cubic, the choice of a particular stereographic triangle is arbitrary. Once a given triangle is chosen, however, all of the indices must be consistent. For example, in Figure 2 the IPF was plotted with respect to the x-axis and shows the stereographic triangle with the 001, 111, and 101 poles. In Figure 2, any pixels that have a <001> parallel to the x-direction in the microscope (horizontal in the map) direction are colored red, those with the <111> direction parallel to the horizontal direction in the image are colored blue, and pixels that have a <101> direction parallel to the horizontal direction in the image are colored green. This image represents what we would consider to be good data, as-acquired, and we would expect that measures of grain size and orientation parameters would be reliably interpretable. Unlike most other analytical techniques in the SEM, there are pixels in the acquired data set that have no useful or reliable interpretation, leading to data maps that have missing pixels or speckled noise in them. In order to make the resulting orientation maps more visually appealing, various data- cleaning routines have been made available by the manufacturers. These algorithms maybe based on sound crystallographic and microstructural reasoning, but it is easy to misapply them, resulting in data with introduced artifacts. S.I. Wright describes a particularly clear example of artifacts generated in orien-tation maps of copper interconnect lines in devices with the overuse of these algorithms [3]. Unfortunately, many authors perform this data modification routinely without ever mentioning the process in published papers. This article examines two common scenarios in which EBSD data quality may create problems for the analysis and where one must be careful in doing noise reduction: unindexed pixels and systematically misindexed pixels. This article is not a review of de-noising or filtering routines, nor is it a comprehensive examination of the Figure 1: Basic geometry and data of the EBSD experiment. A) Chamberscope image of the internal geometry of the SEM during an EBSD experiment. B) Raw EBSD pattern from silver metal. C) Overlay of indexed solution to the EBSD pattern in B.112010 March • www.microscopy-today.comRisks of “Cleaning” Electron Backscatter Diffraction Dataone may be given the opportunity to reduce the noise in the data set until, for example, ninety-five percent of the pixels are said to be “correctly indexed.” However, if we do this noise reduction, what are the consequences on how we use the data? Can we still do quantitative grain size measurements? Are the pole figures still usable?Problem 1: Unindexed and Randomly Misindexed Pixels In the literature when EBSD orientation maps are presented, it is not uncommon to observe maps that are absolutely beautiful and are completely devoid of pixels that are misindexed or unindexed. In our experience, these maps seldom exist as collected from the SEM, yet in published articles they appear frequently, with no mention of any post-processing. Even with simple, well-polished samples, it is uncommon to collect data with 100 percent of the pixels correctly indexed. Unindexed pixels are often most apparent at grain boundaries where overlapping patterns cause the indexing algorithms to fail. In addition, most orientation maps have a small percentage of unindexed pixels randomly distributed throughout the map. Higher-speed data collection can increase the percentage of these pixels. There are sometimes isolated, single pixels that have an abrupt orientation change, which may be suspect for random misindexing. The standard procedures used to clean or filter unindexed or misindexed pixels require two steps. Step one will