3D structure and chemical composition reconstructed simultaneously from HAADF ‐ STEM images and EDS ‐ STEM maps

Abstract

Electron tomography (ET) is nowadays commonly used in materials science to obtain a three dimensional (3D) structural characterization of nanomaterials. Typically it is based on tomographic reconstruction from high angle annular dark field (HAADF) scanning transmission electron microscopy (STEM) images yielding Z‐contrast in final reconstructions. However, when investigating heteronanostructures with small differences in Z, spectroscopic techniques such as Electron Energy Loss Spectroscopy (EELS) and Energy Dispersive X‐ray Spectroscopy (EDS) should be used. Here, we focus on EDS‐STEM spectroscopic imaging. Recently, EDS in combination with electron tomography was demonstrated [1,2], but the quality of these reconstructions is limited because of the small signal‐to‐noise ratio (SNR) in the acquired elemental maps compared to HAADF‐STEM projection images. In this study, we propose to combine HAADF‐STEM tomography with EDS‐STEM tomography instead of processing both signals independently. This combination has not yet been completely explored except for using HAADF‐STEM images for aligning EDS‐STEM maps [3] and for estimating density to correct X‐ray absorption [4]. Here, we introduce the concept of multi‐modal tomography to ET by proposing a novel HAADF‐EDS bimodal tomographic reconstruction technique. The technique is based on the physical model that both types of projection images are linearly related to the projections of chemical compositions. Based on this assumption, HAADF‐STEM images can be approximated as a linear combination of EDS‐STEM maps. By estimating the linear relation, we can scale EDS‐STEM maps to the same physical unit as HAADF‐STEM images. The two types of images are related and used together as the input data for one single reconstruction process. As a result, we are able to reconstruct 3D elemental distributions with reduced noise levels compared to conventional EDS‐STEM tomography. To evaluate the technique, it has been applied to two Au‐Ag nanoparticles. The samples were imaged by an electron microscope (Tecnai Osiris, FEI) equipped with four silicon drift detectors (SuperX system, FEI). The first sample was tilted from −75° to 75° (from −70° to 70° for the second sample) with a step of 5°. At each tilt, a Z‐contrast image was recorded by HAADF‐STEM and two elemental maps for Au and Ag were generated from X‐rays spectrum images acquired by EDS‐STEM. Both the HAADF‐STEM and the elemental maps were aligned using cross correlation algorithms. The first sample that is investigated consists of a Ag nanoparticle with an embedded Au octahedral core. As indicated in Figure 1, Au and Ag are well separated in this sample. Consequently, a segmention based on the Z‐contrast of a HAADF‐STEM reconstruction can be considered as ground truth for the distributions of the chemical elements (Figure 1 (a)). As illustrated in Figure 1 (c), we are able to determine the 3D elemental distributions using our novel HAADF‐EDS bimodal reconstruction technique. This figure indicates a significant improvement in comparison to conventional EDX tomography (Figure 1 (b)) where the raw elemental maps are used as input for a tomographic reconstruction. The second sample that is investigated is an alloy of Au and Ag. Since no clear boundaries exist between the two compositions, we are not able to segment their 3D distributions based on the HAADF‐STEM tomographic reconstructions. Although 3D compositional distributions can be reconstructed from EDS‐STEM maps, the 3D image is very noisy and difficult to interpret (Figure 2 (b)). In comparison, the 3D compositional distributions reconstructed by HAADF‐EDS bimodal tomography (Figure 2 (c)) provides more information on the concentration of the different elements while the outer shape of the nanoparticle agrees well with the 3D shape reconstructed from HAADF‐STEM images (Figure 2 (a)).