Based on Monte Carlo simulations of X-ray generation by fast electrons we calculate curves of effective sensitivity factors for analytical transmission electron microscopy based energy-dispersive X-ray spectroscopy including absorption and fluorescence effects, as a function of Ga K/L ratio for different indium and gallium containing compound semiconductors. For the case of InGaN alloy thin films we show that experimental spectra can thus be quantified without the need to measure specimen thickness or density, yielding self-consistent values for quantification with Ga K and Ga L lines. The effect of uncertainties in the detector efficiency are also shown to be reduced.

SummaryWe have applied our previous method of self-consistent k*-factors for absorption correction in energy-dispersive X-ray spectroscopy to quantify the indium content in X-ray maps of thick compound InGaN layers. The method allows us to quantify the indium concentration without measuring the sample thickness, density or beam current, and works even if there is a drastic local thickness change due to sample roughness or preferential thinning. The method is shown to select, pointby-point in a two-dimensional spectrum image or map, the k*-factor from the local Ga K/L intensity ratio that is most appropriate for the corresponding sample geometry, demonstrating it is not the sample thickness measured along the electron beam direction but the optical path length the X-rays have to travel through the sample that is relevant for the absorption correction.

Energy-dispersive X-ray mapping in a scanning transmission electron microscope is a method to visualize the spatial distribution of chemical elements in a sample. Quantification of the signal intensities depends on proper background elimination and correction of the self-absorption of X-ray lines in the sample. Here we show that our previously developed method of self-consistent effective absorption factors works well even with extremely noise elemental maps of a few net counts only where the human eye can hardly discern any pattern and the background signal is typically less than a single count in each spectrum channel. Correcting the background intensity to sub-pixel accuracy is then necessary for reliable quantification.