Measuring grain boundary segregation by microscopy: numerical comparison of different methods for planar interfaces

Session

PST.4 - Spectroscopies in Electron, X-ray and Ion Microscopy

Authors

Dr Thomas Walther (1)

Affiliations

1. University of Sheffield

Keywords

Gibbsian excess, compositional profiles, grain boundaries, segregation

Abstract text

Grain boundary segregation is an important phenomenon in both metallurgy and in semiconductor technology. For a comparison between different experimental methods as well as for setting up appropriate computer models for atomic-scale simulations it is necessary to measure the interfacial area density of excess atoms (Gibbsian excess) per unit area segregated to grain boundaries or incorporated at interfaces between thin films as accurately as possible.

While there are many methods of chemical analysis of buried interfaces based on surface chemical analysis (e.g. photoelectron spectroscopy, Auger electron spectroscopy, high resolution electron energy-loss spectroscopy) or spectroscopic depth profiling (e.g. secondary ion mass spectroscopy, laser ablation coupled mass spectrometry) non-imaging techniques without spatial resolution in at least two dimensions always lack a direct identification of the specific microstructure and/or chemistry of a given interface and implicitly assume that all interfaces can be treated in the same way. Two prominent methods to measure locally grain boundary segregation to an individual buried interface are analytical scanning transmission electron microscopy (STEM) and atom probe tomography (APT). The first gathers spectroscopic data as function of two spatial coordinates, averaged over a third coordinate (the electron beam direction), the latter provides real 3D information.
Simulations have been run to evaluate numerically compositional profiles of modelled interfaces in three different ways, namely by
A: standard line profile integration,
B: the ladder approach developed for APT [1],
C: the conceptEM approach developed originally for analytical TEM [2] and later transferred to STEM geometry [3].

Applying all methods to the exact same simulated data allows a direct comparison of the evaluation methods, independent of experimental issues, and explores the ultimate accuracy obtainable with each of them. Methods B and C are shown to be generally superior to method A, in particular when finite solute solubility and counting noise make it difficult to define and eliminate the background doping level. For simulations ran for 0.25 of an effective monolayer coverage, method B was best when the apparent width of the interface was varied, while C was better when solubility or total atom number were varied. The root-mean-squared scatter was of the order of 0.01-0.03 monolayers in theses cases.

References

[1]     BW Krakauer and DN Seidman Phys Rev B 48:9 (1993) 6724-6727

[2]     T Walther T (2004) J Microsc 215:2 (2004) 191-202

[3]     T Walther T (2006) J Microsc 223:2 (2006) 165-170