Challenges for Scaling Up Electron-Beam Manipulation of Graphene Impurities

Since the possibility to manipulate lattice impurity atoms using a focused electron beam in scanning transmission electron microscope was discovered in 2014 [1] and first experimentally demonstrated in graphene in 2017 [2, 3], progress has been made to extend manipulation to different elements and even to other materials including single-walled carbon nanotubes and bulk silicon, as discussed in a recent review [4]. However, despite advances in automation and machine-learning, the scaling up of the technique to build atomically precise patterns out of multiple impurity atoms has proven challenging. Electron-beam manipulation is challenging for two reasons: the same focused electron beam must be used to both manipulate and visualize the structures, and energy transfer from the beam electrons to the target atoms is stochastic. Together these mean that it is not possible to prevent unwanted interactions whereby an impurity moves in an undesired direction [2, 3], or worse yet, a carbon atom is knocked out instead of desired movement [1]. However, although the interaction does not appear to be fully elastic [5], tuning the beam energy is effective in increasing the number of successful manipulations [6]. The greatest initial challenge was the fabrication of samples with a high local concentration of impurities in an otherwise undamaged lattice, but techniques such as vacancy-mediated substitution of silicon adatoms [3,7] is able to produce highly suitable samples with more than a dozen potentially manipulable impurities in close proximity (Fig. 1a). However, attempts to manipulate these led to a surprising and disappointing result: most impurity atoms were quickly replaced by carbon (Fig. 1b), preventing the creation of multi-atom structures. This so-called carbon replacement process was first observed with germanium [8], and was particularly problematic for phosphorus where its probability was measured to be on par with other interactions [9]. The most common silicon impurities were at first not thought to be as susceptible to the same problem, but it turned out that the cleaner the samples, the more likely it became – precisely the issue mentioned above and shown in Fig. 1. We note that such data of failed experiments are usually not published. Although the atomistic details of this replacement mechanism are still murky, the process cannot be purely thermally activated. In our own experiments, phosphorus- and silicon-doped graphene samples have been heated to temperatures of up to 1000 °C for tens of cumulative hours, but lattice impurities can still be easily found, placing stringent constraints on possible related reaction barriers and suggesting that electron irradiation must play a pivotal role. Recent studies of the radiation stability of graphene at elevated temperatures have shed light on the source of the carbon atoms: these appear to be thermally diffusing adatoms whose migration barrier was estimated to be 0.33 eV [10]. It does not seem possible to prevent this process by heating the samples, and although the source of the adatoms is unclear, they seem to be ubiquitous. However, considering the value of the migration barrier, experiments with ultra-stable cryogenic stages could be expected to provide a solution, as recently proposed [4]. A second challenge to be tackled before the technique can be scaled up is sample drift. Although active drift compensation can be enabled by small subscan windows [3], we believe this is not ideal as it leads to a significant part of the irradiation dose missing the target atom and even triggering unwanted processes. Further, machine learning structure recognition we use for automation cannot easily keep track of drift. Active drift compensation with a parked electron beam can instead be enabled by real-time monitoring of the converged-beam electron diffraction pattern, whose center-of-mass displacement results from the real-space displacement of the beam. This can be compensated via opposing beam shifts, keeping the irradiation focused on the target atom, detecting atom jumps, and correcting the drift for the following scan frames. Image simulations and work on practical implementation will be discussed [11]. Medium-angle annular dark-field STEM images of laser-cleaned graphene recorded at 60 keV, cropped and aligned to match the field of view. a) Large atomically clean area with 16 silicon lattice substitutions. b) Same area about 30 min later after attempts to manipulate many of the impurities, with six or seven of them having been replaced by carbon (marked by white dashed circles). Scale bar is 2 nm.