3D modeling of nanomaterials from 2D STEM images acquired along a single viewing direction

Abstract number
Virtual Early Career European Microscopy Congress 2020
Presentation Form
Submitted Oral
Corresponding Email
[email protected]
DHA.2 - Advances in 3-dimensional image reconstruction
Ece Arslan Irmak (1, 2), Ivan Lobato (1, 2), Annick De Backer (1, 2), Pei Lui (1, 2), Sara Bals (1, 2), Sandra Van Aert (1, 2)
1. EMAT, University of Antwerp
2. NANOlab Center of Excellence, University of Antwerp

nanomaterials, quantitative ADF STEM, 3D atomic structure, atomistic simulations

Abstract text

In this talk, a robust methodology for the characterization of the 3D atomic structure of nanomaterials from 2D scanning transmission electron microscopy (STEM) images acquired along a single viewing direction will be presented. Use is made of atom counting combined with relaxation studies by molecular dynamics. Advantages of this method and limitations will be shown.

High angle annular dark field STEM (HAADF STEM) is a valuable method for the structural characterization of nanomaterials since it can provide high resolution 2D images that are sensitive to thickness and atomic number of the sample. Nonetheless, these 2D images are usually inadequate to analyze the structure-property relation of nanomaterials because they only provide a projected image of a 3D structure. Electron tomography is one of the most known and powerful methods to retrieve the 3D atomic structure but it is not straightforward to apply the technique to beam sensitive materials or to study dynamic processes that occur during in situ heating or gas experiments. The reason is that conventional electron tomography requires multiple exposures along different viewing directions. 

As an alternative method, atom counting using statistical parameter estimation theory has been combined with relaxation studies leading to a 3D characterization of crystalline nanomaterials based on 2D images acquired along a single zone axis orientation [1–3]. In this contribution we will compare this approach to the outcome of a conventional electron tomography experiment and discuss the applicability of the method for different particle geometries. Our results indicate that the approach is reliable for symmetrical systems along the direction of the electron beam which is shown in Figure 1. Next, we focus on asymmetric nanoparticles. We hereby note that as compared to ab initio approaches, molecular dynamics is computationally efficient for the relaxation studies. However, the relaxation procedure may easily end up in a local energy minimum, instead of the global energy minimum, which makes it sensitive to the interatomic potentials and initial atomic positions. Due to this sensitivity, the 3D reconstruction of asymmetrical nanomaterials is therefore still a challenging task since different initial atomic configurations, size of the particle and different parameterization of the interatomic potential lead to different relaxed structures which is presented in Figure 2. 

To overcome these limitations and to present a robust methodology for the characterization of the 3D atomic structure of asymmetrical nanomaterials from a single 2D STEM image, we need to optimize the initialization of the molecular dynamics simulations. Therefore, we first focused on the post-processing of the HAADF STEM images. After the restoration of the images with a neural network, the experimental distortions are corrected. Then, by applying rigid and non-rigid registration [4] and averaging the 2D images acquired along a single viewing direction, the signal to noise ratio is improved which allows determining atomic column positions and image intensities precisely, especially when studying the surface diffusion mechanism of nanostructured materials at low electron dose. 

After correct post-processing, the intensities in a HAADF STEM image can be modeled as a superposition of Gaussian functions. Then, the unknown parameters and the volume under the Gaussian peak, i.e. the scattering cross-section, are estimated by fitting this model to the experimental images. In a subsequent analysis, the distribution of the scattering cross-sections of all atomic columns is decomposed into overlapping normal distributions. This allows us to count the number of atoms in a particular atomic column [5]. However, in the absence of a priori knowledge, the results need to be compared with reference scattering cross-section values resulting from accurate image simulations for nanocrystals of arbitrary shape and size. Hereby it is important to take the detector sensitivity and temperature effect into account for the frozen phonon calculations for which we made use of MULTEM [6]. In this manner, the number of atoms in each atomic column could be determined with single-atom sensitivity (Figure 3a-c).

From the atom counting results, a starting 3D atomic configuration of the relaxation studies is obtained by positioning the atoms based on their crystalline structure. Next, we applied the method to the evolution of Au nanoparticles on a support during heating and the non-equilibrium structural evolution of Au nanorods during femtosecond laser irradiation. Modeling the nanoparticles unsupported under static conditions make the simulations simpler. However, during catalysis, the presence of a support provides stability to the catalyst particles and certainly affects the particle morphology and the structural evolution of the nanomaterial. Therefore, the interaction between the support and the particle has also been taken into account when simulating nanoparticle under realistic conditions (high temperature and pressure) and we could successfully observe the surface dynamics and the morphology change of the supported Au nanoparticles at high temperatures (Figure 3d). 

Figure 1. Comparison of atom counts from the 3D reconstruction by conventional high-resolution tomography and atom counting/relaxation. (a) Reconstruction of a Pt nanoparticle using high-resolution tomography [2]. (b) Reconstruction based on the atom-counting and relaxation approach. (c,d) Comparison of atom-counts from slice-1 and slice-2

Figure 2. Initialization effect on the relaxation by molecular dynamics simulations. (a) The relaxation study of the gold nanoparticle with 887 atoms and different initial shapes in the z-direction. (b) The relaxation study of the gold nanoparticle with 3927 atoms and different initial shapes in the z-direction.

Figure 3.  (a) HAADF STEM detector scan and the radial sensitivity curve of the detector which is used for image simulations. (b) Comparison of the experimental scattering cross-section values and the simulated values. (c) The number of Au atoms per column at different temperatures. (d) The reconstructed 3D atomic model of Au nanoparticle at 21 °C, 300 °C, 700 °C, respectively. The atoms are presented in different colors, according to the type of surface facet: light blue = {100}, dark blue = {110}, green = {111}.


[1]  A. De Backer, L. Jones, I. Lobato, T. Altantzis, B. Goris, P. D. Nellist, S. Bals, S. Van Aert, Nanoscale 9 (2017), 8791-8798 

[2] T. Altantzis, I. Lobato, A. De Backer, A. Béché, Y. Zhang, S. Basak, M. Porcu, Q. Xu, A. Sánchez-Iglesias, L. M. Liz-Marzán, G. Van Tendeloo, S. Van Aert, S. Bals, Nano Letters 19 (2019), 477-481. 

[3] L. Jones, K. E. MacArthur, V. T. Fauske, A. T. J. van Helvoort, P.D. Nellist, Nano Letters 14, 11, (2014), 6336-6341. 

[4] L. Jones, H. Yang, T. J. Pennycook. M. S. J. Marshall, S. Van Aert, N. D. Browning, M. R. Castell, P. D. Nellist, Adv Struct Chem Imag. 18 (2015). 

[5] A. De Backer, K. H. W. van den Bos, W. Van den Broek, J. Sijbers, and S. Van Aert, Ultramicroscopy 171 (2016), 104-116. 

[6]  I. Lobato, S. Van Aert, and J. Verbeeck, Ultramicroscopy 168 (2016), 17-27. 

[7] This work was supported by the European Research Council (Grant 770887 PICOMETRICS to SVA and Grant 815128 REALNANO to SB, Grant 823717 ESTEEM3). The authors acknowledge financial support from the Research Foundation Flanders (FWO, Belgium) through project fundings and a  postdoctoral grant to ADB.