### Publication Type:

Thesis### Source:

Laboratoire de Physique des Solides, , Université Paris-Sud, Volume Ph.D., Orsay, p.181 (2010)### Keywords:

curve-Fitting, data analysis, EELS, ICA, MVA, PCA### Abstract:

Modern analytical transmission electron microscopes are able to gather a large amount of information from the sample in the form of spectrum image (SI). Until recently the results of electron energy loss spectroscopy (EELS) analysis were considered as doubtful in terms of accurate quantification. Furthermore, although the analytical procedures developed for single spectra

can be extended to the analysis of SIs, for an optimal use of this highly redundant information, more advanced techniques must be deployed. In this context, we investigate alternatives to the standard quantification methods and seek to optimise the experimental acquisition for accurate analysis. This addresses the current challenges facing the EELS community, for whom beam damage and contamination are often the limiting factors.

EELS elemental quantification by the standard integration method is limited to well-behaved cases. As an alternative we use curve fitting which, as we show, can overcome most of the limitations of the standard method. A sample of known composition (magnesium oxide cubes) is used to demonstrate the procedure and it is shown that the correct composition is obtained from the analysis when all the relevant factors are taken into account. The same procedure is applied to characterising ZnO nanorods, of which we accurately determine the non-stoichiometric composition. In the characterisation of boron nitride we extend the method to obtain, in addition to elemental maps, the first bonding maps at the nanoscale. The same approach can be useful

for other spectroscopies, and we have applied it to photoemission spectroscopy (PES), to achieve the first spatially resolved analysis of the distribution of the silicon sub-oxide structure at the SiO2/Si interface as a function of underlying doping. We will show that recent advances in instrumentation, in particular, the alternating core-loss-low-loss acquisition camera (ACLAC),

enable accurate quantitative analysis when combined with recent advances in signal analysis. A multivariate analysis (MVA) method, principal component analysis (PCA), is rising in popularity in the EELS community as a noise reduction method. However, in general the EELS SI do not comply with PCA’s linearity requirement. We investigate, the range of validity of PCA for EELS data and the effects that energy instabilities produce in the output. A major difficulty when analysing SIs of samples of unknown composition is that the quantitative methods require as an input the composition of the sample. This knowledge is normally acquired by a visual inspection of the SI but this way of proceeding is prone to errors because important features can be hidden by noise or be simply difficult to observe. We show that by combining PCA with independent component analysis (ICA) it is possible to analyse fully an SI without defining any model. In optimal conditions this method is capable of extracting signals from the SI corresponding to the different chemical compounds in the sample and their distribution.

We have successfully applied this approach to different materials science problems. For example, we have obtained one of the first examples of complex atomic structures resolved by EELS from CexZr1−xO2 catalyst nano-particles, pushed forward the detection limit by obtaining elemental maps of trace amounts of Sb in GaxIn1−xAsySb1−y quantum dots and quantified the

distributions of TiO2 and SnO2 in a ferroelectric multi-layer.