Track single adsorption atoms in liquids with transmission electron microscope
Article source: Nick Clark, Daniel J. Kelly, Mingwei Zhou, Yi-Chao Zou, Chang Woo Myung, David G. Hopkinson, Christoph Schran, Angelos Michaelides, Roman Gorbachev, Sarah J. Haigh. Tracking single adatoms in liquid in a transmission electron microscope. Nature 2022, 609, 942-947.
Abstract: Single atoms or ions on the surface of affect the process from nucleation to electrochemical reactions and heterogeneous catalysis. Transmission electron microscopy is the main method for observing single atoms on various substrates. It usually requires high vacuum conditions, but in-situ imaging in liquid and gas environments has been developed with combined resolution of space and time, which is unmatched by any other method despite concerns about the impact of electron beams on the sample. When imaged in a liquid using commercial techniques, electrons are scattered in the windows surrounding the sample and in the liquid, often limiting the achievable resolution to a few nanometers. On the other hand, graphene liquid pools have achieved atomic resolution imaging of metal nanoparticles in liquids. Here, the authors show a bigraphene liquid pool consisting of a central molybdenum disulfide monolayer separated by two closed graphene windows, which makes it possible to monitor the dynamics of Pt atoms adsorbed on the monolayer in the water salt solution at atomic resolution. By imaging over 70,000 adsorption sites, the authors compared the position preference and dynamic motion of adsorbed atoms in a fully hydrated state and in a vacuum state. The authors found that the adsorption position distribution of adsorbed atoms in the liquid phase changed compared with the adsorption position distribution in vacuum, and the diffusion coefficient was higher. This approach paves the way for in-situ liquid phase imaging of chemical processes with single atomic accuracy.
Graphene is an ideal window material for in situ transmission electron microscopy due to its extremely thin, high mechanical strength, low atomic number, chemical inertness, strong impermeability and ability to remove aggressive radicals. The original graphene liquid pool (GLC) design relies on liquid bags formed randomly between two graphene sheets, so under prolonged electron exposure, low yield and poor stability. The more advanced design includes patterned spacers of SiNx or hexagonal boron nitride (hBN) to define the liquid bag, improving control of GLC geometry and experimental conditions.
The authors developed a dual graphene liquid cell (DGLC) to study the movement of single solvated metal atoms on atomic films. This was inspired by the non-in-situ STEM study, which showed that the choice of liquid environment can change the distribution of metal atoms, from nanoclusters to single atoms, but experimentally detecting this behavior in situ is not feasible, and even in earlier studies, imaging of single atoms in liquids has proven difficult to achieve. The authors focus on Pt on MoS2, the abundant existing data makes it an ideal model system for exploring the limitations and potential of atomic resolution liquid pool microscopy.
DGLC, as shown in Figure 1a, consists of two hBN spacers, each layer several tens of nanometers thick, with a single-molecular layer of molybdenum disulfide (MoS2) sandwiched in the middle. Both hBN spacers contain holes pre-imprinted with electron beam lithography and subsequent reactive ion etching. Use a small amount of layers of graphene (FLG) on the top and bottom of the stack to trap the liquid sample in the void. The atomic plane hBN crystal forms a seal with graphene and MoS2; this prevents leakage, transfer of liquid between individual cells and complete loss of liquid if the cell is locally ruptured. This design has a highly controlled total pool thickness (less than 70 nm), retaining the atomic resolution imaging and analysis capabilities of TEM. At the same time, the presence of liquids on and off the substrate makes the DGLC powerful enough to continuously image for more than 10 minutes, with an electron flux of about 2.8 × 106 e·s-1·nm-2; i.e., atomic resolution imaging at 200 kV. The design can also directly mix separated liquid samples under the beam by directed ablation of the MoS2 layer.
Figure 1b shows HAADF STEM images of a typical experimental DGLC sample, a dual liquid pool structure (yellow shaded portion) was found at the overlap of the voids of the two hBN layers, allowing the MoS2 layer to be suspended between the two liquid pockets.Through electron energy loss spectrum imaging, the oxygen K-side characteristics of water are mapped to the double-cell area, verifying the existence of liquid in the cavity. The samples are supported by a custom SiNx TEM support grid with large circular holes as shown in the green section in Figure 1b. The upper liquid pot of
DGLC is filled with Pt saline solution (10 mM H2PtCl6), while the lower part is filled with deionized water . The Pt solution produces atomically dispersed Pt species on the MoS2 separation membrane, which is visible as a highlight in the HAADF STEM (Z-contrast) image because the atomic number of Pt is relatively high compared to other materials in the pool. Whether Pt is adsorbed on top graphene, encapsulated MoS2 monolayer or bottom graphene layer can be inferred by the limited focus depth (about 10 nm) of the STEM probe (Figure 1c-1e). High density single Pt atoms and nanoscale Pt nanocrystals can be seen on both the top graphene window and the submerged MoS2 film (Fig. 1c and 1d, respectively), while some Pt atoms are also found on the bottom graphene window (the area in the upper right corner of Fig. 1e, which may overflow to the outer pool surface during sample preparation). Relative defocus values indicate that the upper liquid layer is 42 nm thick and the bottom layer is 28 nm; very consistent with the hBN spacer measured by atomic force microscope as part of the cell manufacturing (32 nm and 30 nm, respectively). The special spatial resolution of the
DGLC system allows the clear identification of highlights corresponding to the electron scattering of a single Pt nucleus and their relative positions relative to the underlying lattice . Whether Pt is adsorbed or replaced, MoS2 can only be inferred through analysis of dynamic motion. Nevertheless, the complexity of the solid-liquid interface environment means that it is very challenging to verify whether Pt exists in the form of atomic, ionic or molecular complexes. The authors recognize this uncertainty, but refer to the Pt species studied here as “adsorption atoms” to distinguish them from nanoclusters attached to the surface and ions that move freely in the solution. This unparalleled high spatial and temporal resolution imaging offers a large number of new measurement possibilities for a single metal species surrounded by liquids.
Fig. 1
First, the adsorption site distribution of Pt adsorption atoms relative to MoS2 is concentratedly analyzed. Figure 2 shows a video sequence image focused on a MoS2 film, where the hexagonal lattice of the MoS2 monolayer is decorated with brighter points corresponding to the single Pt species. The average area density of the sample is approximately constant, about 0.6 nm-2. The lattice position supporting Pt on the MoS2 crystal is identified by image processing method: the Pt position is identified from the Fourier filtered image, and then the template matching algorithm is used to enhance local self-similarity and remove spatially uncorrelated noise, thereby identifying the Mo lattice position. Then compare these two coordinate sets to find the position of Pt adsorption atoms relative to the MoS2 lattice.
analysis showed that Pt ions tend to occupy one of the three highly symmetric MoS2 lattice positions shown in Figure 2b. Here it refers to: "Mo site", that is, Pt is located directly above the Mo lattice site; "S site", that is, Pt is located directly above the S lattice site; and a hexagonal center or "HC site", where Pt is located above the hexagonal lattice site, equidistant from the three S and three Mo positions. Figures 2c-2d show the experimental original and average images, showing a good qualitative match to the image simulations shown in Figure 2e.
Figure 2
To quantitatively compare the distribution of Pt adsorbed atoms in liquid pools and vacuum environments, the authors calculated the spatial resolution histogram by analyzing the positions of more than 70,000 Pt adsorbed atoms relative to the underlying MoS2 substrate (Figure 3a-3j). This vacuum data was obtained by two methods: 1) intentionally pierce the graphene windows of different pools, and then the sample was placed in TEM vacuum overnight to allow the liquid to escape; 2) preparation of in situ Pt on MoS2 samples for comparison. In vacuum, experimental data show that Pt atoms are clearly inclined to be above the S position in the MoS2 lattice, although theoretical calculations show that the preferred position of Pt atoms in the original MoS2 is above the Mo position, while metastable is above the S position and HC position.The strong preference for S sites in the experimental data is consistent with previous non-in-situ analysis and is attributed to the presence of S vacancies, which provide more favorable energy positions. The author's density functional theory (DFT) calculation predicts that the binding energy of Pt on the S vacancies is 6.1 eV, while the binding energy of Pt on the Mo and S sites is 3.5 and 3.1 eV. In fact, by superimposing all occupied lattice positions in the image series, the authors can see that the positions of Pt adsorbed atoms are distributed more evenly in the liquid cell, while the vacuum data is concentrated at specific locations. This cluster may also be related to Pt staples at less moving S vacancies, which have relatively low mobility compared to Pt adsorbed atoms on the perfect MoS2 lattice (the diffusion barrier of the S vacancies is 0.8-2 eV, while the Pt adsorbed atoms is about 0.5 eV).
compares the results of the vacuum with the measured Pt site distribution in the liquid unit (Figure 3a-3e), and the authors can see significantly different behaviors, and when the liquid is present, the S and Mo points show similar occupancy (Figure 3a). This may be because the oxygen in the water replaces the S vacancies, making them less attractive to Pt. In fact, the author's DFT calculations show that the S vacancy binding energy of O atoms (9.1 eV) is larger than that of Pt atoms (6.1 eV), which suggests that in places with aerobicity, it preferentially replaces the S vacancy. Once oxygen replaces an S vacancy, this position becomes a position that is relatively unfavorable for Pt to adsorb atoms. In addition, Cl in the salt can also replace the S vacancy. Measurable native S vacancies are often present in
MoS2, which is considered key to the production of high-performance hydrogen evolution reactions. As the author's material was mechanically stripped, S vacancy dominated in non-in-situ characterization, which is likely the result of induced knock damage by electron beam irradiation , expected to occur at measurable atomic resolution imaging levels even at lower acceleration voltages. In addition, the electron beam effect may play a role in the filling of the S vacancies, because the products dissolved by water radiation (such as O2, H2O2) can react with the S vacancies, causing O to enter MoS2. Given these views, it is surprising that the behavior observed by the authors has a high degree of structural stability for local environmental changes induced by electron beams. In liquid and vacuum environments, when clear defect areas were removed from the analysis, the authors did not observe changes in the relative Pt distribution with imaging time (up to 10 seconds per minute), or with electron flux (more than 0.7-2.8 × 106 e·nm-2·s-1, the beam current details are shown in Figures 3b-3d and 3f-3h). The authors hypothesize that for non-in-situ imaging, the number of S vacancies required is generated in the first few frames of the image, which are sufficient to govern the behavior of Pt adsorption atoms. During prolonged imaging, these vacant areas diffuse in MoS2 to form visible holes, and these areas are excluded from the analysis. In a liquid environment, the beam-induced S vacancies may be generated at a similar rate, but will soon be filled with native or radioactive O or Cl elements available in the liquid, just as quickly as they are produced. This led to the author's experimental observation that in the presence of liquid, the adsorption behavior of Pt adsorption atoms is not dominated by strong S vacancy position interactions, and is closer to the predicted distribution of original MoS2 by DFT.
Figure 3
Now turn to discuss the dynamic motion of Pt adsorption atoms, comparing the behavior in the liquid pool with the behavior observed in the vacuum. Quantitative data are obtained by connecting atomic positions in adjacent video frames using the minimum displacement method. As shown in Figures 4a-4b, the authors observed that most single Pt-bonded atoms are movable, and Figures 4a-4b show the trajectory of several representative Pt-bonded atoms over a period of time. Statistical analysis of these atomic dynamics and a certain range of electron fluxes under liquid tanks and vacuum conditions can obtain a histogram of atomic displacements of radial averages and frames, as shown in Figure 4c. The two distributions peak at the Pt transition distance of about 0.05 nm, which may be related to the motion around the selected lattice position. This motion is not caused by tracking uncertainty or background motion of the underlying MoS2, after correcting for sample drift, tilt and rotation, the displacement of the Mo site was tracked and its peak was found to be much smaller (below 0.02 nm under both liquid and vacuum conditions, see Figure 4c).For this system, the high mobility of Pt atoms is consistent with the fact that they are the main surface adsorption atoms: those located on the MoS2 lattice, rather than being replaced. The only exception is the small number of completely fixed Pt atoms in the liquid pool. They are located at the Mo site, indicating that they are Pt substitutions into the Mo vacancy. In liquid pools, although the surface density of adsorbed atoms is relatively stable, adsorbed atoms occasionally appear and disappear spontaneously during imaging, which may be due to the dynamic equilibrium of Pt deposited onto the MoS2 surface in the salt solution and the Pt adsorbed atoms dissolved back into the solution. The authors also observed that in liquid, the Pt bond located above the S site is unlikely to be stuck in multiple frames, supporting the author's hypothesis that the binding of the Pt bond to the S vacancies in the liquid is inhibited compared to the vacuum condition.
compared the liquid and vacuum data sets, and the atomic displacement distribution in the liquid pool data was significantly wider, indicating that larger inter-frame displacements (0.1-0.5 nm) occur more frequently between different positions. Taking into account the mean square displacement (MSDs) shown in Figure 4d, it is obvious that a larger displacement occurs in the liquid pool relative to the samples imaged in vacuum, with the surface diffusion coefficient of Pt on the liquid and vacuum imaging D, greater than 0.25 and less than 0.2 nm2·s-1, respectively. Under vacuum conditions, the barrier for the diffusion of Pt adsorption atoms between adjacent Mo sites on the MoS2 lattice was 0.5-0.82 eV. Calculating the diffusion barrier in liquids with DFT is challenging, but the diffusion barrier of charged Pt surface ions (including reduction and oxidation) is reduced by about 0.2 eV, which may lead to an increase in D of the liquid pool. These calculated diffusion barriers are below the typical energy that can be transferred to Pt-adsorbed atoms under 200 kV (2.69 eV) light, suggesting that energy transfer from electron beams may play a role, similar to TEM studies on all atomic surface dynamics. Previous reports on beam-induced motion show the dependence of beam scanning and probe position, which will be observed in the author's statistical and directional analysis of site jumps. However, the near perfect symmetry seen in all crystallographic equivalent lattice directions shows that electron beam scanning has no strong effect on the jump direction.
In addition, studying dynamic motion as a function of electron flux (Fig. 4c and 4d) found that the difference between liquid and vacuum conditions was significantly greater than the dependence on electron flux. Larger motion of adsorption atoms can be seen at lower fluxes, meaning the motion observed by the authors is not simply beam-driven. The higher beam-induced S vacancy production, changes in Pt hydration state, or being nailed by other radiation-dissolved substances may limit the movement of Pt at higher electron fluxes, even in liquids. A full computational understanding of this experimental behavior would be an important task, requiring detailed modeling of possible Pt species and differences in local binding energy due to the presence of hydrated shells. Parameter space increases further, as it is often impossible to verify the local charge state of a single atom with S/TEM, even if it is not in situ studies. Nevertheless, the experimental data presented here provide key inputs for theoretical calculations, involving not only adsorption positions and kinetics; for example, the analysis of distribution function shown here by Pt shows that Pt adsorption atoms prefer to be adjacent to each other, suggesting that hydration is incomplete and there is no evidence that long-distance Coulomb repulsion.
Figure 4
In summary, the results presented here demonstrate the ability to measure adsorbed atomic motion at the solid-liquid interface, although the importance of understanding the electron beam effect and performing supplementary theoretical research on atomic behavior in complex hydration systems. The experimental technology is widely applicable to different material systems and provides a way to obtain previously unavailable atomic analytical, dynamic and structural information in different environments, suitable for many different physical science systems.
