11 Electron Tomography
273
this technique with a highly accurate ET reconstruction method (i.e. CS), able to
retrieve fine structural details in a 3D distribution from fewer projection images and,
thus, reducing sample damage, enables the recovery of the complete EELS-SV with
access to oxidation state information.
To test this possibility, an experiment was proposed, aiming to recover the EELSSV reconstruction of an iron oxide nanocube, with a FeO x /FeO y core-shell structure
(Torruella, Pau et al. 2016. 3D Visualization of the Iron Oxidation State in FeO/Fe3O4
Core–Shell Nanocubes from Electron Energy Loss Tomography. Nano letters, 16(8),
5068–5073) [34]. This particular problem poses a challenge due to the fact that the
contrast difference between shell and core was not enough, neither in HREM nor in
STEM-HAADF, to accurately segment the image.
In order to obtain a quantitative 3D oxidation state map of the particle, a EELS-SI
tilt series was acquired in low voltage conditions (80 keV) from −69° to +67°, every
4° each containing 64 × 64 pixels with spectra in the energy range 478−888 eV.
The high spectral resolution (given a dispersion of 0.2 eV/pixel and 0.015 s/pixel),
much needed to analyze ELNES features, meant severe sample damage by beam
focus during the second half of the tilt series acquisition, due to accumulated dose.
Furthermore, some SI were also discarded because the evidences of the significant
coherent diffraction occurring at specific angles. Fortunately, the high symmetry of
the nanocube allowed the reconstruction assuming a mirror effect in the projections
for positive tilting angles.
The procedure followed in MVA for noise reduction (namely PCA) and BSS was
applied only to the area of interest in the EEL spectra, i.e. around the iron L 2,3 edges.
This effectively reduced the computational time. The BSS algorithm chosen was
Fast-ICA, implemented in Hyperspy [47]. Fast-ICA was performed over the first
derivative of the six remaining spectral components after PCA (denoising). Among
the six components, two were related to the Fe ionization edge, labelled as C 1 and
C 2 . They show clear ELNES features that can be identified as the Fe L 2,3 white lines
(Fig. 11.6). The position of the maximum of the Fe L 3 of C 2 is shifted +1.9 eV with
respect to Fe L 3 maximum in C 1 . For comparison, Fig. 11.6 also shows reference EEL
spectra corresponding to the Fe L 3,2 edges for wüstite (Fe 1−x O), Fe
2+ and haematite
(Fe 2 O 3 ) Fe
3+ . This allowed the extraction of the Fe oxidation state maps, depicted as
the weight of the respective components in each pixel as shown in Fig. 11.6 (right) for
the 0° projection. Similar maps are obtained for the whole SI angular range, giving
rise to two set of images suitable for 3D reconstruction.
For the tomographic reconstruction of the SV, the obtained components fulfilled
the projection requirement, given that the nanocube size ranged between 35 and
40 nm and no cupping artefact was observed. CS was chosen as the reconstruction
algorithm, given the low number of projections available and the precision expected
in the SV. The sparsity promoting transform was the one described in the previous
section (both image and gradient spaces where sparsely transformed).
The results of the reconstruction are shown in Fig. 11.7. The orthoslice for the
Fe
2+ Fig. 11.7d reveals the presence of this oxidation state in the core and the shell
of the nanoparticle, whereas the presence of Fe
3+ Fig. 11.7e is confined to the shell.
This is consistent with the shell Fig. 11.7b being Fe 3 O 4 and the core Fig. 11.7a FeO.
273
this technique with a highly accurate ET reconstruction method (i.e. CS), able to
retrieve fine structural details in a 3D distribution from fewer projection images and,
thus, reducing sample damage, enables the recovery of the complete EELS-SV with
access to oxidation state information.
To test this possibility, an experiment was proposed, aiming to recover the EELSSV reconstruction of an iron oxide nanocube, with a FeO x /FeO y core-shell structure
(Torruella, Pau et al. 2016. 3D Visualization of the Iron Oxidation State in FeO/Fe3O4
Core–Shell Nanocubes from Electron Energy Loss Tomography. Nano letters, 16(8),
5068–5073) [34]. This particular problem poses a challenge due to the fact that the
contrast difference between shell and core was not enough, neither in HREM nor in
STEM-HAADF, to accurately segment the image.
In order to obtain a quantitative 3D oxidation state map of the particle, a EELS-SI
tilt series was acquired in low voltage conditions (80 keV) from −69° to +67°, every
4° each containing 64 × 64 pixels with spectra in the energy range 478−888 eV.
The high spectral resolution (given a dispersion of 0.2 eV/pixel and 0.015 s/pixel),
much needed to analyze ELNES features, meant severe sample damage by beam
focus during the second half of the tilt series acquisition, due to accumulated dose.
Furthermore, some SI were also discarded because the evidences of the significant
coherent diffraction occurring at specific angles. Fortunately, the high symmetry of
the nanocube allowed the reconstruction assuming a mirror effect in the projections
for positive tilting angles.
The procedure followed in MVA for noise reduction (namely PCA) and BSS was
applied only to the area of interest in the EEL spectra, i.e. around the iron L 2,3 edges.
This effectively reduced the computational time. The BSS algorithm chosen was
Fast-ICA, implemented in Hyperspy [47]. Fast-ICA was performed over the first
derivative of the six remaining spectral components after PCA (denoising). Among
the six components, two were related to the Fe ionization edge, labelled as C 1 and
C 2 . They show clear ELNES features that can be identified as the Fe L 2,3 white lines
(Fig. 11.6). The position of the maximum of the Fe L 3 of C 2 is shifted +1.9 eV with
respect to Fe L 3 maximum in C 1 . For comparison, Fig. 11.6 also shows reference EEL
spectra corresponding to the Fe L 3,2 edges for wüstite (Fe 1−x O), Fe
2+ and haematite
(Fe 2 O 3 ) Fe
3+ . This allowed the extraction of the Fe oxidation state maps, depicted as
the weight of the respective components in each pixel as shown in Fig. 11.6 (right) for
the 0° projection. Similar maps are obtained for the whole SI angular range, giving
rise to two set of images suitable for 3D reconstruction.
For the tomographic reconstruction of the SV, the obtained components fulfilled
the projection requirement, given that the nanocube size ranged between 35 and
40 nm and no cupping artefact was observed. CS was chosen as the reconstruction
algorithm, given the low number of projections available and the precision expected
in the SV. The sparsity promoting transform was the one described in the previous
section (both image and gradient spaces where sparsely transformed).
The results of the reconstruction are shown in Fig. 11.7. The orthoslice for the
Fe
2+ Fig. 11.7d reveals the presence of this oxidation state in the core and the shell
of the nanoparticle, whereas the presence of Fe
3+ Fig. 11.7e is confined to the shell.
This is consistent with the shell Fig. 11.7b being Fe 3 O 4 and the core Fig. 11.7a FeO.