Three-Dimensional Imaging by Large-Angle Illumination STEM

Following the developments of geometric aberration correction, atomic-resolution imaging and spectroscopy become versatile tools in the fields of physics, chemistry, and materials science. The spatial resolution in scanning transmission electron microscopy (STEM) is now in deep sub-angstrom realm of 40.5 pm [1]. However, the achieved high spatial resolution is only valid in the projected lateral two dimensions, and the depth resolution along the axial direction is still far from atomic resolution. Since the depth resolution of STEM is λ/α², one of solutions is to use a larger illumination angle. Recently, we have recently installed a Delta-type 5^th order geometric aberration corrector in JEM ARM300CF, and the flat region of Ronchigram is significantly increased up to 70 mrad in semi-angle at 300 kV. However, to utilize such large illumination angles, it is prerequisite to accurately correct lower orders of geometric aberrations such as 2-fold, 3-fold astigmatisms and axial coma. We therefore developed a new method to efficiently minimize residual aberrations by using atomic-resolution STEM images, rather than the conventional Ronchigram algorithms [2]. By implementing such automated geometric aberration correction, it becomes possible to perform STEM observations at sub-angstrom lateral resolution, even with the large illumination angle of 63 mrad.<br/>To evaluate the depth resolution with the larger illumination angles, we observed isolated single cerium (Ce) dopants embedded within a crystal of cubic boron nitride (<i>c</i>-BN) via ADF-STEM depth sectioning, where sequentially acquires atomic-resolution ADF-STEM images as a function defocus. On a basis of a statistical evaluation of single dopant Z-contrast distribution along the depth, it turns out that the depth resolution is improved as 2.14 nm with the illumination angle of 63 mrad [3]. Combining such high depth resolution with statistical evaluation, we also determine the Ce-Ce partial pair distribution function with a considerably longer distance up to 8.5 nm, which is much larger than the coherent conventional diffraction experiments (2 or 3 nm at most).<br/>Even with 63 mrad, the achievable depth resolution is still far from atomic depth resolution. To overcome the poor depth resolution, we statistically analyze 3D atomic-resolution image data set with a Bayesian framework fitting algorithm. Utilizing such statistical approach into materials’ surface analysis, we achieved ±0.9 Å depth resolution at the entrance surface, which makes possible to observe surface atomistic defects such as vacancies, adatoms, kinks and ledges [3].<br/>[1] S. Morishita et al, Microscopy 67 46 (2018).<br/>[2] R. Ishikawa et al, Ultramicroscopy 222 133215 (2021).<br/>[3] R. Ishikawa et al, Phys. Rev. Appl. 13 034064 (2020).<br/>[4] R. Ishikawa et al, ACS Nano 15 9186 (2021).