Atomic mechanism of polarization-controlled surface reconstruction in ferroelectric thin films

At the ferroelectric surface, the broken translational symmetry induced bound charge should significantly alter the local atomic configurations. Experimentally revealing the atomic structure of ferroelectric surface, however, is very challenging due to the strong spatial variety between nano-sized domains, and strong interactions between the polarization and other structural parameters. Here, we study surface structures of Pb(Zr0.2Ti0.8)O3 thin film by using the annular bright-field imaging. We find that six atomic layers with suppressed polarization and a charged 180° domain wall are at negatively poled surfaces, no reconstruction exists at positively poled surfaces, and seven atomic layers with suppressed polarization and a charged 90° domain wall exist at nominally neutral surfaces in ferroelastic domains. Our results provide critical insights into engineering ferroelectric thin films, fine grain ceramics and surface chemistry devices. The state-of-the-art methodology demonstrated here can greatly advance our understanding of surface science for oxides.


Amorphous layer on the surface.
The specimens are inevitably coated by a thin amorphous layer after ion milling.
To minimize the thickness of amorphous layer, our cross-sectional STEM specimens were cleaned by ion milling at 0.1 kV. The typical amorphous layer thickness in our samples is less than 3 nm. Owing to a very strong channeling effect on crystalline structure, we can visualize the atomic columns as dark dot contrast in ABF STEM images. While, the amorphous layer consists of random atom distribution and therefore the layers can contribute to simply increase the background noise level and not significantly change the atomic column positions. To determine the atomic position, we have implemented 2D Gaussian fitting, where we use more than 500 pixels per column. Therefore, the statistics is very high and the background noise attributed to the amorphous layer should be negligibly small.

Specimen tilt.
During experiments, we deliberately reduce the misalignment by means of: i. The STEM mode was carefully aligned. All the data was recorded from the prototype JEM ARM300CF that has a very friendly and efficient alignment system. The aberration therefore can be corrected and minimized before recording data needless of standard specimen.
ii. We use CCD to align both specimen orientation and illumination aperture location on a big monitor by watching the Kikuchi patterns. We marked different angle contours such as 24 mrad, 6 mrad and 3 mrad on the monitor by home-made Digital Micrograph script.
iii. We used smaller illumination aperture (8 mrad) to doubly check the alignment.
With the small aperture, a small misalignment can be more readily observed and corrected. iv. We used larger illumination aperture (24 mrad) to record ABF images. Our multislice simulations and previous study 1 suggest that the effects of misalignment are less significant with large illumination aperture.
By these methods, we can easily make the misalignment as small as 3 mrad (equivalent to ~0.17°). In fact, in our experimental images in Figures 1a, 3a, and 4a, the atom columns appear "round" shape confirms that the alignment was very close to perfect zone axis. We note that our previous simulation 2 indicated ABF image is more sensitive to the misalignment compared to HAADF and small misalignment could significantly alter the shape of atomic columns.
Furthermore, we have checked the misalignment effects by multi-slice image simulation in Figure R1 that is shown below, and we are sure that the change of surface atom positions is an intrinsic behavior in this material rather than the artifacts from misalignment, which is discussed below.
i. From the simulation in Figure R1 the lattice constants that are calculated from interdistance of Pb-Pb columns remain unchanged regardless of the misalignment, which is not consistent with our results in Figures 2 and 4, confirming the presence of surface reconstruction instead of misalignment effects in our study. The constant interdistance of Pb-Pb columns can be interpreted by the fact that all the Pb columns show the same deformation behavior with specimen tilt and therefore the relative interdistance between them does not change at all regardless of misalignments.
ii. From the simulation in Figure R1, the misalignment indeed can influence on the length of short Pb-O bond, i.e., short Pb-O increases when tilt angle α>0 and it decrease when α<0. Note that only the former case α>0 is possible in our experiments because short Pb-O bond length in the surface reconstructions never decreases in Figures 2, 3 and 4. Furthermore, within a misalignment of ±6 mrad, the increasement in short Pb-O bond length is less than 4.6% (at +6 mrad) from the simulation in Figure R1. Such subtle elongation (4.6%) is much smaller than that we observed in the Figures 2f   and 4g, where the elongation of short Pb-O bond length is as large as 31%.
In other words, the specimen misalignment caused increment in short Pb-O bond length (4.6%) can be even smaller than the error bar (as large as ±8.5%) in Figures 2f and 4g. In this sense, the effects of small specimen misalignment would not alter our conclusion.