Electron Energy Loss Spectroscopy of 2D Materials in a Scanning Electron Microscope

Abstract

This work demonstrates electron energy loss spectroscopy of 2D materials in a 1-30 keV electron microscope, observing 100-times stronger electron-matter coupling relative to 125 keV microscopes. We observe that the universal curve relating beam energy to scattering holds for the transition from bulk graphite to graphene, albeit with a scale factor. We calculate that optimal coupling for most 2D materials and optical nanostructures falls in this range, concluding that spectroscopy of such systems will greatly benefit from use of this previously unexplored energy regime.

Figures (7)

• Figure 1: Rendering of our experiment. (a) Schematic of our SEM column, secondary electron detector (SED), electron spectrometer, and sample. After interaction, the spectrometer plates deflect each electron proportional to its energy, which is then filtered by a slit and detected. (b) Rendering of our sample plane, in which our specimen (shown in Fig. 1c) generates an excitation in the material with likelihood $|\beta|^2$. This results in an attenuated probability of measuring an initial energy electron (our ZLP) with probability $|\alpha|^2$, and a $|\beta|^2$ chance of the electron losing $\hbar\omega_\pi$ energy. (c) SEM micrograph of our sample, which consists of graphene on lacey carbon. (d) Graphene loss spectrum at various energies, showing an increase in the loss peak as the energy is decreased.
• Figure 2: (a) Calculation of the thickness to IMFP ($t/\lambda_i$) versus energy for a variety of 2D materials. The graphene bulk curve was adapted from Jablonski et al. jablonski_calculations_2023, and is based on optical data. The gold, amorphous carbon, and various other 2D material curves were estimated using the Bethe approximation, with details in the supplement. We plot measurements of the IMFP for LEEM geelen_nonuniversal_2019 and TEM persichetti_folding_2011 energies from the literature with triangles. Our measurements are the blue circles with error bars. While the slope is consistent, the coupling measured in all of these experiments is approximately five times stronger than predicted.
• Figure 3: Simulation of an electron passing distance $\rho$ from a gold nanoparticle of diameter $d$. (a) Plot of the coupling factor versus electron velocity for an electron passing at $\rho$ = 5 nm from a nanoparticle excited with 800 nm light. For 20-100 nm particles, optimal coupling ($\beta_{opt}$) occurs at SEM energies between the vertical dotted lines. The value of $\beta_{opt}$ when $\rho<<\lambda$ follows a simple linear relationship, shown by the dashed line. (b) Plot of the optimal $\beta_{opt}$ for an optical nanostructure for a given thicknesses and excitation wavelength. Again, the white dotted lines divide zones corresponding to boundaries between LEEM, SEM, and TEM energies.
• Figure 4: Effect of aperturing on the electron beam energy distribution. (a) Standard focusing alignment of the electron beam in the SEM leads to multiple peaks in the spectrometer, which we hypothesize is due to multiple facets of our electron source leading to a well-defined, incoherent probe, but multiple energy peaks. (b) If we aperture our beam further, we can isolate single electron peaks from the source and thus achieve high resolution imaging. In this case, the top spectrometer plate is held at 704 V, meaning the energy FWHM corresponds to 1.75 eV. After this step further alignment of the slit is required
• Figure 5: Fast scanning in our spectrometer. By rapidly sweeping the beam faster than instabilities in our system, we can compensate them out be measuring the ZLP position. (a) The spectrometer configuration, in which the bottom plate has a bias tee configuation to couple in a fast modulated trangle wave. This results and a repeated spectrum shown below for four periods. (b) The raw spectrum, showing drift in the peak position. If we naively integrate the signal along un-adjusted time bins, we can get the appearance of spurious sidebands, even in short integration windows of a few seconds. (c) Compensated signal, in which we find the ZLP peak, and shift our data on each row to this value. Note that this is not perfect and leads to a random spread of our beam. Because of the summing of these random errors our spectrum appears closer to a normal distribution rather than the expected Maxwell-Boltzman distribution given this energy.