Nanometer Resolution Elemental Mapping in Graphene-Based TEM Liquid Cells

We demonstrate a new design of graphene liquid cell consisting of a thin lithographically patterned hexagonal boron nitride crystal encapsulated on both sides with graphene windows. The ultrathin window liquid cells produced have precisely controlled volumes and thicknesses and are robust to repeated vacuum cycling. This technology enables exciting new opportunities for liquid cell studies, providing a reliable platform for high resolution transmission electron microscope imaging and spectral mapping. The presence of water was confirmed using electron energy loss spectroscopy (EELS) via the detection of the oxygen K-edge and measuring the thickness of full and empty cells. We demonstrate the imaging capabilities of these liquid cells by tracking the dynamic motion and interactions of small metal nanoparticles with diameters of 0.5–5 nm. We further present an order of magnitude improvement in the analytical capabilities compared to previous liquid cell data with 1 nm spatial resolution elemental mapping achievable for liquid encapsulated bimetallic nanoparticles using energy dispersive X-ray spectroscopy (EDXS).

. Schematic showing successive steps in EGLC fabrication process. The right column shows optical images of a specific EGLC sample at the corresponding fabrication stage, (a) hBN flakes are exfoliated onto a silicon oxide substrate. (b) hBN flakes are patterned using electron beam lithography and RIE etching. (c) 2 separate few layer graphene flakes are exfoliated, one onto a polymer layer supported by a silicon substrate and one directly onto oxidised silicon. (d) The few layer graphene flake is lifted from the silicon using the polymer film and is deposited on top of the hBN flake. The inset in the optical image (for this and subsequent steps) shows the locations of the two flakes, with hBN shown in blue/purple and graphene shown in grey. (e) The graphene + hBN spacer flake stack is transferred onto the few layer graphene on silicon oxide. During the transfer a liquid sample is introduced between the sheets. The optical image was captured prior to the removal of the polymer support film. (f) The whole EGLC sample is transferred onto a TEM compatible support. The optical image shows that some of the sample has folded during the transfer, reducing the useful area for imaging.
The top flake and patterned hBN were then transferred onto a mechanically exfoliated few layer graphene flake on silicon oxide which forms the bottom window of the cell. During transfer, the solution of interest was introduced between them using a pipette. Under the effect of van der Waals forces the bottom graphene flake and hBN spacer gradually contact, trapping liquid inside the well areas etched in the hBN spacer ( Figure S1e). After the pockets are sealed, the sample is slowly dried at low temperature (~40 °C), to avoid bursting of the pockets.
The final step was to transfer the whole sample onto the support grid, in a second wet transfer process ( Figure S1f). After this last transfer, the supporting PMMA can be dissolved using organic solvents, and dried supercritically to avoid damage to the graphene membranes due to surface tension. 3 Critical point drying was performed in a CO 2 atmosphere using a Bal-Tec CPD030 critical point dryer.

Pressure testing of the EGLCs
Hencky's solution 4,5 describes the approximate central deflection of a doubly clamped circular thin membrane under uniform pressure loading. The solution for the pressure drop across the membrane ∆P as a function of the perpendicular deflection δ of the central point is 6,7 given by: where T is the intrinsic (zero deflection) membrane tension (typically not more than 0.1 N m -1 for exfoliated/transferred monolayer graphene 8 ), a is the membrane radius, E 2D is the 2D modulus (units N m -1 ), and v is the Poisson's Ratio of the membrane material. The constant C 1 and function f(υ) are unitless and can be determined using numerical techniques. [9][10][11][12] The membrane parameters are illustrated schematically in Figure S2.
This solution is generally a good approximation for circular membranes thinner than a few nm 13 . It depends only on the membrane deformation, δ, and pressure, although the membrane radius, a, can change due to partial delamination, as in Figure S2. The value of the pretension term, T, is typically 0.1 Nm - 1 8,14 or lower for freshly exfoliated graphene flakes, and is only observed due to partial adhesion of the flake around the sidewalls of the drum causing additional tension 16,17 . As our samples show a 'flat' membrane geometry with no sidewall adhesion, and the pretension is typically lower for the same flakes after transfer due to strain relaxation, we suspect the values for pretension will be smaller for our samples. We use reported values 18 for the 2D modulus and Poisson's ratio of graphene of E 2D = 340 ±20 N m -1 per layer, and ν = 0.19 ± 0.02 respectively. We assign C 1 = 4 and f(v)= 8/3 following previous studies where these values were shown to accurately predict the shape of graphene bubbles. 19,20 The curvature of the cell windows can therefore be used to estimate the pressure inside the cells. Figure  To test the robustness of our EGLCs, the sample was coated coated by a 300 nm thick Au layer using e-beam evaporation under vacuum conditions ( Figure S3). After Au evaporation the cells appear to remain intact but the upper graphene membrane had deflected by 90nm for the largest well (as shown by the optical image in Figure   S3c, the SEM micrographs in Figure S3d-f and the AFM data in Figure S3g and S3h. The measured values of deflection (90 nm) and diameter (1250 nm) correspond to an internal pressure of 116 ± 46 bar. The high pressure is attributed to beam induced heating of the cell during Au deposition. Nevertheless it is encouraging that our cells remain intact for large internal pressures. To investigate the response of the EGLC to prolonged external vacuum, we subjected a EGLC sample consisting of a ~30 nm thick hBN flake capped using ~3 nm thick graphene flakes on a silicon oxide support to progressive vacuum cycling.
AFM images of the flakes (recorded at ambient external pressure) were captured at various stages during the vacuum cycling process. A set of these images along with an optical reference image of the EGLC sample are shown in Figure S5.
For the larger cells the low aspect ratio (height 30nm, diameters over 1µm) meant that these were usually either 'ring filled' (where the top graphene window was able to deflect and adhere to the bottom window in the center of the cell, Figure S4) or empty. For smaller cells (diameters less than 600 nm) this collapsing of the cells was rarely seen, although the top flake would often partially adhere to the sidewalls of the well, creating a 'half-filled' structure (see Figure S4c).
AFM images were captured immediately after transfer, but imaging conditions were inhibited by the polymer residue visible on the surface (see left panel in Figure S5d).
The sample was therefore exposed to a total of 50 seconds of low power oxygen cleaning to remove this residue. The sample was then repeatedly exposed to high vacuum conditions (≤ 10 −7 mbar) and imaged. No membrane rupture or leakage was observed after a cumulative 24 hours in vacuum, although changes in some of the well geometries were observed, suggesting that leaking of liquid between cells may have occurred. For example, one of the small holes in the upper part of panel (d) goes from a 'half-filled' to a 'filled' geometry between the central two panels. Smaller trapped pockets of liquid are visible between the flat hBN surface and the upper graphene window away from the patterned areas, as shown in Figure S5e. They were not observed to move or change shape during the course of the experiment.
The total yields of cells of various geometries after 24 hours in vacuum conditions sorted by the cell diameter are presented in Table 1. For the smallest cells (with diameters ≤ 100 nm) no empty or ring-filled cells were observed and the cells were believed to be all filled within measurement accuracy. High-magnification TEM imaging was performed using cells with diameters between 400 nm and 600 nm. Table 1. Fabrication yield of EGLCs sorted by cell diameters. The location of the cells recorded here is shown in Figure S5b. Smaller cells were found to be all filled within measurement accuracy.    Figure S7 shows further examples of HAADF STEM imaging of graphene liquid cells.

Scanning Transmission Electron Microscopy Imaging
Where nanoparticles were sputtered on one of the graphene sheets as in Figure S7c the cells were found to be less stable for prolonged imaging. We attribute this to a reduction in the quality of the hermetic seal able to form between the graphene and the hBN spacer.
Nanocrystal motion in the HAADF images was analysed using the spot detection and tracking capabilities available in Icy software, 21 where the positions of nanocrystals over a HAADF time-series were recorded and compiled to form particle trajectories.
The algorithms used for particle detection involve application of a Laplacian of a Gaussian (LoG) filter for edge detection, followed by segmentation and thresholding to determine particle properties such as size and position. Tracking scripts then link particles with similar features across consecutive frames using user-defined distance for a few frames). These two particles coalesce later in the video to form a larger particle (t =500s). A further analysis of particle coalescence seen in these videos is shown in Figure   S8. Huang thresholding by minimization of fuzziness 25   By 100 s the small yellow particle has dissolved and the blue and green particles have coalesced into the largest (red) particle. Inset right is the final frame at t = 498 s where, only the red particle remains of the four particles originally measured. Schematic diagrams illustrating the coalescence process are shown above. Interestingly, after the initial coalescence the area of the largest (red) particle reduces from 150 s until a steady state is reached at 400 s. We attribute the size reduction to a period of densification after the particles initially touch. The data for the area of the red particle has been averaged over the nearest 5 frames to reduce noise. Scale bar is 2 nm.
Energy Dispersive X-ray Spectroscopy Figure S9 compares summed spectra for areas of the spectrum image inside and outside the liquid well region of the EGLC. As expected outside the well there is a strong nitrogen signal due to the presence of the hBN spacer layer. Inside the well we see an enhanced oxygen signal, consistent with the presence of water. The ratio of oxygen counts in and outside the cell was measured to be O wet /O dry = 3.05. Figure S10 shows STEM EDX spectrum imaging of nanoparticles inside the EGLCs.
A smoothing filter was applied to the elemental maps in Figure 4 of the main text. All other spectral maps are unprocessed. Line profiles in Figure S10 have the width shown on the elemental map above and were horizontally averaged over 3 pixels to improve signal to noise ratio. The particles were monitored during spectrum imaging and mapping ceased if noticeable changes in shape or morphology occurred. In some cells the graphene windows were sputtered with gold prior to closing the cell, in order to produce metal nanoparticles which provide nucleation sites for the formation of core-shell particles (main text Figure 4, Figure S11). For these cells a greater degree of leakage was observed, both during prolonged imaging and when stored for months in ambient conditions. After 5 months stored at room temperature all wells appeared dry. We explain this by the presence of sputtered particles which deleteriously affect the graphene-boron nitride seal, creating microchannels which allow liquid to escape. This effect could be reduced by lowering the seed crystal Drying could also be produced by piercing the graphene window with an intense electron beam (Supplementary video 2). Interestingly, cells where the liquid had been removed appeared less stable to prolonged imaging with the electron beam ( Figure S11). The as made EGLC was imaged for extended periods (30 minutes at 80 kV, beam current 240 pA) without any signification morphological changes to the particles, while particle sintering was observed after just 5 minutes, (200 kV, beam current 100 pA) in dry cells. Particle bridging was seen across iron-coated faces creating single elongated nanocrystal. Here imaging of the empty cells was performed at 200 kV to achieve higher spatial resolution with our microscope).
Nevertheless, similar sintering behavior was observed in dried cells imaged at 80kV. The electron beam quickly induced rapid changes in the nanoparticles' morphology causing coalescence to form nanorods. This change in behavior can be attributed to passivation of nanocrystals by a layer of water molecules, possibly a solvation shell or electrostatic layer that provides a barrier to sintering.
Initial iron beam induced deposition in the cells appeared to occur preferentially on vertices rather than facets of the Au particles, particularly at vertices in close proximity to other Au particles (main text Figure 4, Figure S11). During sintering, redistribution of the iron occurred such that it still appeared to be associated with vertices and the nanoparticle surface. Understanding the preferential colocation of different metallic phases in solution is very relevant to colloidal nanoparticle synthesis but has previously been impossible to investigate without dismantling the liquid cell. This is a rich area of further work made possible by our new EGLCs.