AI-driven atomic manipulation and characterization in the STEM

Control over matter has allowed physics, condensed matter, and materials science communities to realize a wide range of properties useful in many applications. Both industry and academia have approached the few-nanometer level of control, used largely in the realm of semiconductor fabrication. The scanning transmission electron microscope (STEM), like scanning probe microscopy (SPM), routinely enables direct visualization of the atomic nature of many material systems. Electron-matter interaction in the STEM typically causes undesirable effects that microscopists collectively term “beam damage,” however may be precisely what is needed for atomic fabrication. While manipulation of matter at the level of single atoms has been demonstrated in both STEM and SPM, it has predominantly been performed manually without reproducibility or useful precision. Beam-induced effects in the STEM have been notoriously difficult to control, and the - ordinarily stochastic - changes have also been challenging to characterize analytically via electron energy loss spectroscopy (EELS) or 4D-STEM.

It will be discussed how artificial intelligence can be leveraged to guide the microscope in a variety of styles. In terms of atomic fabrication, the atomic coordinates must be extracted from image data as soon as possible. To overcome this and ensure robust predictability, deep ensembles are used, which simultaneously handle the issues - also encountered in self-driving vehicles - of so-called “out of distribution” shifts.

With fast and reliable extraction of atomic coordinates, one can begin to experiment with beam-induced effects precisely at the single-defect level. We show that different strategies can realize structures – in both graphene and MoS₂ - that cannot be fabricated by any other means. It is also discussed how one can autonomously measure and discover different phenomena by utilizing deep kernel learning in both EELS and 4D-STEM modalities.