Individual heteroatom identification with X-ray spectroscopy

The aberration-corrected scanning transmission electron microscope (STEM) enables unparalleled capability for direct imaging and spectroscopy at the single-atom scale. For thin samples with simple elemental compositions, individual heteroatoms can often be identified on the basis of atomic number contrast in annular dark field (ADF) images alone.¹ For samples with more complex compositions, individual heteroatoms can be spectroscopically identified with electron energy-loss spectroscopy (EELS).² However, such single-atom sensitivity ADF and EELS measurements are limited to specific host material and heteroatom combinations. In ADF images, the contrast is a function of variation in thickness and atomic number; thus, for example, individual O and C impurities provide enough contrast to be visible in single layer boron nitride (BN), but not in multilayer BN. Similarly, EELS often lacks the sensitivity to detect single heteroatoms when the heteroatom absorption edge is at higher energy than the edges of the major elements of the host material, because the heteroatom signal is superimposed on a large inverse power-law background.

Energy dispersive X-ray spectroscopy (EDXS) has the potential to greatly increase the range of materials and species of heteroatoms for which single-atom-sensitivity spectroscopy is possible. Toward that goal, single-atom sensitivity EDXS has been demonstrated for constrained sample geometries with known species of heteroatoms, e.g., a single Er atom in a carbon cage inside a nanotube³ and a single Si or Pt atom on monolayer graphene.⁴ While these experiments did prove that single-atom EDX is possible, they did not provide a significant advance in material analysis capability over what could be done previously with EELS and ADF. In fact, these experiments required spectral acquisition times of ∼1–6 min,^3,4 significantly longer than typical single-atom EELS acquisition times of under a second.² This difference in collection time is the primary reason that EDXS has lagged EELS as an analytical method for atom-by-atom analysis, although it has been implemented at the atomic-column scale for thicker samples.⁵ For thin samples, the quantum efficiency of the EELS is higher, i.e., the intensity of the signal collected per incident electron is greater.⁶ This is due to both more signal emitted per incident electron, and the greater collection efficiency of the EELS detectors, which approaches 100% in some instruments, compared to <1% for typical EDXS detectors.

Despite the clear advantage of EELS instrumentation in collection efficiency, there is a significant fundamental disadvantage of EELS for atom-by atom analysis: some a priori knowledge of the identity of the heteroatoms is required in order to choose the relevant experimental conditions. Core-loss EEL spectra follow a power law decay in the signal intensity as a function of energy that makes it impossible to optimize an EELS measurement for simultaneous detection of most elements. As a consequence, the typical EELS experiment is optimized for the elements most expected to be present, i.e., host material and/or dopants rather than the unexpected, i.e., impurities. In EDXS experiments, however, the signal intensity and background levels remain relatively constant over the typical collection range of 0–20 keV, which means little adjustment to collection conditions is required, and no a priori information about composition is necessary. Thus, if the collection efficiency of EDXS can be improved, it has a natural advantage for samples with complex, and variable chemistry and/or topography that inhibit direct quantitative interpretation of the ADF image intensity or pre-optimization of the EELS collection parameters. Relevant investigations include nearly all nanoscale materials, including doped semiconductors, nanostructured catalysts, and nanoscale minerals.

The microscope employed in these studies was a Nion UltraSTEM200 equipped with a Bruker XFlash 6|100 UHV compatible, windowless silicon drift detector. The geometric solid angle of the detector is calculated to be 0.7 sr. The microscope was operated at 60 kV, with a nominal probe size of 0.15 nm and probe current of 50–120 pA. The samples were deposited as aliquots onto lacey-carbon-film-coated copper TEM grids. The grids were held in sample cartridges with Cu-Be sample cups that provide low X-ray background signal and the ability to tilt the sample towards the EDXS detector, to minimize shadowing of the X-ray signal by the grid and holder. Spectra were collected with a sample tilt of ∼10° with the Espirit 1.93 software and exported as EMSA files for analysis. Additional details of the experimental details are provided in the supplementary material.

The samples used for this study consisted of a two-phase mixture of amorphous carbon and ∼2 nm nanodiamonds, obtained by acid dissolution of the matrix materials from the Murchison meteorite. Prior study of this residue with STEM-ADF and EELS revealed the presence of abundant impurity atoms distributed over both phases.⁷ Although most of the impurities are likely residual components from the acid dissolution of the meteorite, a few rare impurities incorporated into the nanodiamonds and/or amorphous carbon are signatures of the origin of some of the material as products of interstellar and/or circumstellar processes. More detailed knowledge of the distribution of impurities in the two carbonaceous phases could help constrain their origins. For the purpose of this study, the important characteristics are that the samples have a wide range of impurity species and are thin, but vary in thickness from 1 to 10 planes of carbon. In this case, EELS is well suited to spatially resolve sp² and sp³ regions,⁷ but an inefficient way to identify individual heteroatoms.

For the initial impurity atom EDXS measurements, we followed the method of Lovejoy et al.,⁴ in which a reduced raster scan box was centered on the atom of interest, and moved manually by the researcher to follow the motion of the atom, during collection of an X-ray spectrum.⁸ To verify that the atoms were in fact single atoms, rather than pairs or clusters, we used multiple lines of evidence. First, we observed the motion of the impurity over the sample: atom pairs and clusters separated and reformed while diffusing across the surface, whereas single atoms remained fixed, or moved as single units. Second, we measured the ADF intensity due to the impurity, which depends on atomic number, ∼Z^1.65, and then compared this to possible sources of the collected X-ray peaks for self-consistency. Using this method, individual Ca, Si, and S atoms were identified. Example parameters for these experiments are summarized in Table I. The manual tracking method is robust to moderate motion of the atom, but required long collection times. The need for a large sampling area compared to the size of the atom limits the fraction of the collection time that the beam dwells on the atom and excites X-ray emission.

In order to increase the collection efficiency by limiting the oversampling, we used a “point and shoot” method. We first recorded a single frame high-angle ADF image, and then placed a fixed raster box that tightly framed the atom, and collected the spectrum from that raster area (Table I “fixed” raster area). Collection continued until the impurity atom X-ray count rate plateaued, which indicated that the signal-to-noise had reached a maximum. Any subsequent decrease in the X-ray count rate of the heteroatom was a sign that the atom had moved out of the sampling box, at which point the collection was terminated. Figure 1 illustrates the identification of an Si heteroatom with this method. In comparison, the spectrum from an adjacent area, in which no visible impurities were present, shows only peaks from the underlying C, and Cu from the grid, holder, and pole piece shields. Thus, X-ray counts obtained in the Si spectra are not due to spurious system peaks. Similarly, we identified S and P impurity atoms (Figs. 2 and 3). Compared to the manual tracking, this method was up to an order of magnitude faster, with sufficient counts for unambiguous atom identification under 10 s in most cases.

The alternative methods we tested include single multi-frame spectrum imaging of areas containing one or more visible atoms.⁸ Here, a full spectrum is collected for each pixel in the sampling area, summed over one or more frames. This method has promise for robustly bonded, isolated heteroatoms. However, it was found to be less useful for cases in which the heteroatoms were mobile, because the beam induced motion of the atoms resulted in the detection of X-rays from the atom in multiple locations in the spectrum image. As a consequence, it was not possible to determine if additional heteroatoms diffused into the sampling area, or whether all the signals were due to motion of the originally targeted atom(s).

The spectra shown here demonstrate that single atom EDXS is both possible and practical, even for relatively low Z volatile elements, such as S and P, despite the two orders of magnitude lower collection efficiency of EDXS compared to EELS. This raises the question of how much further can a single atom EDXS be improved before reaching a fundamental physical limit. Two critical limiting factors for the analysis are the motion of the heteroatom and the ratio of dwell time on and off the atom in the sampling window. In the case of surface adsorbed species, some diffusion under the beam is to be expected and can even be an important part of the sample properties, e.g., supported catalysts. Thus, eliminating the heteroatom diffusion is not always an option. Manual active tracking of the atom can be employed,⁸ but compared to the “point and shoot” method, it leads to lower fractional dwell times on the atom, and longer collection times, due to the need to increase oversampling area of the tracking window. For example, if we compare the relative efficiency of the two Si atom detection experiments as the X-ray count rate normalized to the probe current: 0.007 Si counts s⁻¹pA⁻¹ for manual tracking vs. 0.096 Si counts s⁻¹pA⁻¹ for point-and-shoot. The order of magnitude difference can be accounted in part by the 4.4-fold relative decrease in the sampling area compared to the atom size for the point-and-shoot measurement. The remaining difference is likely due to the size of the probe tails relative to the sampling area, which affects the effective dwell time of the beam on the atom. Future computer automation of the atom tracking could offer more efficient spectral collection, through the reduction of the sampling area and more accurate centering of the probe on the atom in question. Ultimately, the physical limit is the probe current the atom can withstand before full ionization and irretrievable loss into the microscope vacuum. Lower probe currents could extend the possible measurement time, at the cost of count rate. We estimate that for the detector geometry used here, it should be possible to decrease the sampling area and increase the count rate such that an Si atom can be identified in 1 s. Some additional improvement might also come from changes to the sample holder or detector geometry, though these are unlikely to yield 10-fold increase.

The implications for the practical atom-by-atom EDXS analysis in terms of possible new measurements include the ability to detect impurities or dopants, including “invisible” atoms,⁹ in single quantum dots or other nanoparticles. Although the proof of concept examples all pertained to surface heteroatoms with visible contrast in ADF images (Figs. 1–3), image contrast is not required if the heteroatom is robustly incorporated into the low dimensional host material. For example, an EDX spectrum collected from an individual nanodiamond shows the presence of N and O atoms (Fig. 4), which are too close in atomic number to C to be seen in ADF images of multilayer C materials. The N atoms are detected more easily with EDXS than EELS in this case, because the N is 1% of the sample (10 atoms), compared to >95% C (>950 atoms). The corresponding N K-edge (402 eV) in the EEL spectrum is a very small peak on top of a large background from the C K edge (285 eV). In the future, the whole nanoparticle approach to the measurement of embedded heteroatoms should allow analysis of a wide range of heteroatoms and host particle combinations. The limiting factors here will again include the size of the sampling window relative to the atom or atoms in question, which will be determined by the host particle size, as well as the background count rate compared to the heteroatom count rate, which depends on the cross-section of the heteroatom and the inherent noise in the EDX detector electronics. The limiting factor for accurate quantification of whole particle results in terms of integer numbers of atoms will be the knowledge of the probe–atom interactions that accounts for the probe tails, raster parameters, and diffusion of host and impurity atoms under the beam.

This work was supported in part by the NRL base program and the NASA Cosmochemistry program. Support for the construction of the microscope and EDX detector was received from NASA LARS, and the NRL capital equipment fund. The Murchison nanodiamond residue was received from C.M. O'D. Alexander.