Probing the structure of vanadium tetracyanoethylene using electron energy-loss spectroscopy

As the field of quantum information science progresses, so too does the desire for materials that are both compatible with existing technology frameworks and capable of effectively contributing to quantum-coherent operations.^1,2 Magnon-based hybrid quantum systems are specifically enticing for such application as their low-lying collective magnetic excitations, known as magnons (the quanta of spin waves), can coherently couple to a variety of quantized excitations, including photons, phonons, and localized spin transitions,^3,4 and also their microwave resonance frequencies are commensurate with those of quantum information processing. The traditional material choice for these low-loss magnonics applications is yttrium iron garnet (Y₃Fe₅O₁₂ aka YIG), which has been shown to be an attractive candidate material for magnonic/spintronic applications^5–7 and magnon-based quantum systems.^2,8 However, the processing techniques required to produce high-quality low-loss YIG microstructures on-chip for these magnon-based hybrid quantum systems remain a substantial obstacle.⁹ Solutions to these hurdles may still come to fruition for YIG or other low-loss ferrites; however, reliance on a single material class substantially limits the capabilities and applicability of these integrated systems.

To bypass these challenges, an alternative ferrimagnetic material has been proposed,⁴ the organic-based coordination compound vanadium tetracyanoethylene (V[TCNE]_x). V[TCNE]_x is a low-loss (α ∼ 4 × 10⁻⁵) ferrimagnetic semiconductor with a Curie temperature (T_c) > 600 K⁵ that exhibits magnetic resonance quality factors Q > 7000 (at 6 GHz) and has rapidly gained attention as a material in magnon-based quantum systems.^4,10–12 V[TCNE]_x can be easily deposited as high-quality thin films on a variety of bare or patterned substrates,¹³ and modified electron-lithography techniques have demonstrated that the material can be patterned into microstructures with a negligible increase in its magnetic damping.^11,12 The patterning capability of this low-loss insulating ferrimagnet provides ample opportunity to integrate magnetic microstructures on-chip that extends its use in magnon-based devices.¹⁴ Furthermore, low temperature ferromagnetic resonance (FMR) measurements reveal that V[TCNE]_x retains its low-damping properties even at cryogenic temperatures, demonstrating its potential for use in low temperature quantum information science and engineering (QISE) applications.¹² Established measurements of optically detected FMR (ODFMR) in V[TCNE]_x via diamond NV-centers, combined with its predicted ability to couple NV centers in diamond over micrometer lengthscales,⁴ further bolster V[TCNE]_x as a magnetic material for quantum information in solid-state systems.

Based on these exceptional properties, there has been growing interest in the use of molecular-based materials, including V[TCNE]_x and its class of metal–ligand compounds, as alternatives in quantum systems as they provide significant advantages in their tunability and on-chip integration over more traditional inorganic materials and ferrites, such as YIG.^12,15–22 V[TCNE]_x belongs to a broad class of M[A]_x (M = metal and A = acceptor) organic-based magnetic materials^12,23–28 that remains relatively unexplored. A detailed and thorough understanding of V[TCNE]_x will develop the foundation for a better understanding of other remarkable metal–ligand magnets and, thus, establish a clear path toward their applications in magnon-based QISE.

Despite the canonical role V[TCNE]_x has held in the development and application of this class of material as a result of its excellent magnetic resonance properties, the understanding of its electronic and atomic structure is still largely phenomenological. V[TCNE]_x films typically have high structural disorder or are amorphous, and the material is extremely air-sensitive when not encapsulated, making characterization challenging.^{19–21,29–31} X-ray absorption fine structure (XAFS) studies have shed some light on the environment of vanadium, leading to the prediction of a 3D structure of V²⁺ in octahedral coordination with six surrounding TCNE molecules.^31–34 Previous ab initio calculations supported semiconducting behavior, an octahedral coordination of V²⁺, and suggested that within the band structure, there were effective spin-splittings of ∼2 eV for TCNE molecular orbitals and larger for the vanadium d states.¹⁹ However, because x-ray diffraction (XRD) studies suggest that the material is amorphous, it is less clear which electronic properties from a crystalline calculation survive in the actual material.

In order to probe the structure of this highly disordered material, higher spatial resolution studies are needed. Electron energy-loss spectroscopy (EELS) performed in a scanning transmission electron microscope (STEM) is a powerful characterization technique that can be used to study various aspects of a material, including the composition, oxidation state, and bonding/coordination environment of atoms in the specimen, as well as the optoelectronic properties of the material. Here, we report the study of V[TCNE]_x films grown by chemical vapor deposition (CVD) using STEM EELS. Low-loss EELS data backed by density functional theory (DFT) calculations provide insights into the electronic, magnetic, and spintronic properties of V[TCNE]_x. EELS spectra calculated directly from the ab initio structure based on the crystalline electronic structure match remarkably well with those measured, down to small details. Clear support is obtained for a revised schematic picture of the electronic structure, which distinguishes apical and equatorial TCNE molecular orbitals and vanadium d states. Furthermore, EELS data collected on multiple plan-view specimens indicate the high-quality CVD growth of V[TCNE]_x and provide previously unknown information on the oxidation mechanism and an additional thermodynamic structural change as noted by the relaxation of the structure over time.

V[TCNE]_x films are deposited via ambient-condition chemical vapor deposition (CVD) in a custom CVD reactor inside an argon glovebox (O₂ < 1 ppm, H₂O < 1 ppm) in accordance with the literature.²¹ Argon gas flows over TCNE and V(CO)₆ precursors that react to form a V[TCNE]_x thin film on the substrates. The pressure inside the CVD reactor for all growths was 35 mmHg, and TCNE, V(CO)₆, and the substrates are held at 65, 10, and 50 °C, respectively. All substrates were cleaned with the solvent chain [acetone, methanol, isopropanol, and deionized (DI) water (x2)] and dried with N₂, followed by a 10 min UV/ozone clean in a UVOCS T10× 10/OES to remove any residual organic contaminants.

For focused ion beam (FIB) foil samples, nominally, 400 nm thick V[TCNE]_x films were deposited on commercially available 400 μm-thick p-doped Si substrates. These samples were, then, capped with 30 nm of Al deposited via thermal evaporation with a BOC Edwards Auto 500 deposition system housed inside an argon glovebox so that the V[TCNE]_x films were not exposed to oxygen between the V[TCNE]_x and Al deposition. The 30 nm cap served as an oxygen barrier for loading the samples into the FIB and was deposited in two 15 nm depositions to prevent V[TCNE]_x films from overheating during the deposition. After capping with Al, samples are transferred under argon and exposed momentarily to air while loading into the FIB microscope. Cross-sectional FIB specimens were prepared using a FEI Helios NanoLab 600 DualBeam FIB/SEM. FIB foils were, then, loaded as quickly as possible into the microscope to avoid oxidation confirmed by the EELS spectra in Fig. S3.

For the plan-view samples, nominally, 100 nm thick V[TCNE]_x films were deposited on 3 nm-thick Pt and Pd Substratek^TM TEM grids purchased from Ted Pella. For each growth, 3 Pd and 3 Pt Substratek TEM grids were used, resulting in six potential TEM specimens, the best of which were loaded into a Gatan vacuum transfer holder inside the glovebox. This TEM sample holder allowed for transportation from the glovebox into the microscope without exposure to air.

In each growth, commercially available 430 μm-thick C-plane sapphire substrates were included as calibration samples for ferromagnetic resonance (FMR) characterization.

Ferromagnetic resonance (FMR) measurements were performed in an x-band (9.8 GHz) Bruker Elexsys E500 CW electron paramagnetic resonance (EPR) spectrometer. The frequency of the microwave source is tuned to match the resonant frequency of the cavity before each scan to ensure optimal cavity tuning. All scans had a 0.03 G modulation field at 100 kHz modulation frequency and were performed at the lowest possible microwave power (0.2 μW) to prevent sample heating and non-linear effects distorting the FMR line shape. V[TCNE]_x films deposited on sapphire were mounted on a custom Macor ceramic sample holder and loaded into sealed quartz EPR tubes for FMR measurements. Samples were rotated in-plane (IP) to out-of-plane (OOP) for FMR measurements in 10° increments.

Electron energy-loss spectroscopy data were collected on a FEI image corrected Titan3 G2 60–300 kV S/TEM equipped with a Gatan Quantum spectrometer and a K2 direct electron detector. Dual EELS spectrum images were collected at an operating voltage of 300 kV using a monochromator. The data were collected at a convergence semi-angle of 5.54 mrad and a collection semi-angle of 10.71 mrad. The exposure times were 0.0001 s for the low-loss region and 0.3 or 5 s for the high-loss region.

Low-loss EELS data were collected at an operating voltage of 60 kV to avoid Cherenkov radiation. The 60 kV data were collected at a convergence semi-angle of 9.3 mrad and a collection semi-angle of 18.04 mrad. In order to observe low eV transitions, the zero-loss peak (ZLP) was shifted slightly off the detector by setting the drift tube to 3 eV, and the exposure time for these datasets was 0.3 s. Low-loss EELS processing included the removal of the zero-loss peak (ZLP) using the reflected tail method and deconvolution using the Fourier-log method. High-loss data processing included a background subtraction before each edge of interest and deconvolution using the Fourier ratio method.

The pseudopotentials used are default options from the official projector augmented wave (PAW) potential set of the Vienna ab initio Simulation Package (VASP), with five valence electrons per vanadium, four per carbon, and five per nitrogen.^35,36 For the rest of the calculation, we used 400 eV for the energy cutoff and a Γ centered 5 × 5 × 3 k-mesh sampling. The plots in Fig. S3 show the detailed partial density of states for majority and minority spins, associated with the vanadium d electrons and the equatorial (planar) and apical carbon or nitrogen p orbitals. These partial densities of states suggest that there is an effective U of ∼5 eV between majority and minority vanadium d levels and of ∼2 eV between majority and minority p levels of carbon and nitrogen. They also suggest a ligand field splitting of 1 eV between the equatorial and apical TCNE molecule p states.

Following the CVD growth of V[TCNE]_x films as described in Sec. II, the magnetic properties were measured to determine the quality of the material. The ultra-narrow, single-peak FMR spectra throughout the angular measurements confirm that the V[TCNE]_x films are of high-quality. The in-plane (IP) and out-of-plane (OOP) peak-to-peak (full width at half max, FWHM) linewidths are 0.54 Oe (0.94 Oe) and 0.69 Oe (1.14 Oe), respectively (Fig. 1, insets). These linewidths correspond to FMR quality factors $Q= H R Δ H$ > 3000 for all field orientations, where H_R is the FMR resonant field and ΔH is the FWHM linewidth. The angular FMR resonance positions for the thin film shown in Fig. 1 are fit to the following equation:^21,37,38

where ω is the resonance frequency, γ is the gyromagnetic ratio, H_R is the external field, M_eff is the effective magnetization of the V[TCNE]_x film, and θ is the field angle with respect to the film normal. This equation assumes that the magnetization and external magnetic field are parallel, which is appropriate since M_eff ≪ H_R; therefore, we do not need to consider separate angles for the external field and magnetization.¹¹ Fitting to this equation yields 4πM_eff = 99.5 G, which is consistent with the previous literature values.^{11–14,21,39} We do not see any IP anisotropy in these films, consistent with the previous work.¹¹ 

While V[TCNE]_x has excellent low-loss magnetic properties, its electronic properties still leave much to explore and understand. Due to its sensitivity to air and high temperatures,^18,40 its electronic properties have predominately been measured via electric transport methods. Transport studies suggest that V[TCNE]_x is a semiconductor with 0.5 eV bandgap and spin-polarized subbands; however, there are systematic limitations in these measurements, which reduce the precision and accuracy in extracting the bandgap.^41–43 Mott variable-range hopping (VRH) has been proposed in some studies, but more recent measurements suggest that this mechanism does not accurately describe electronic transport in V[TCNE]_x.^41–43 Thus, the mechanism for charge transport remains elusive. A more complete understanding of the electronic transitions in V[TCNE]_x would provide a framework upon which to design future studies to investigate electronic transport, the optoelectronic properties, and the long-range magnetic ordering of V[TCNE]_x films.

For example, the low-loss EELS spectrum, which contains energy-losses up to 50 eV, can be used to study the optoelectronic properties of a material. The single scattering distribution [$J 1 E$] of the EELS spectrum is related to the complex dielectric function [ɛ(E)] by

where I₀ is the intensity of the zero loss peak (ZLP), t is the specimen thickness, a₀ is the Bohr radius, m₀ is the electron rest mass, v is the velocity of the primary electrons, β is the collection semi-angle, and θ_E is the characteristic scattering angle for specific energy-loss.⁴⁴ The raw EELS spectrum is a convolution of single and multiple scattering events; therefore, to extract $J 1 E$⁠, a deconvolution is performed in DigitalMicrograph (DM),⁴⁵ as described in Sec. II. The real [$ε 1 E$] and imaginary [$ε 2 E$] parts of the complex dielectric function can, then, be extracted from $J 1 E$ by performing Kramers–Kronig (KK) analysis.^44–46 

Figure 2(a) shows low-loss data of a plan-view V[TCNE]_x specimen that was collected at 60 kV to avoid Cherenkov radiation.^44,47 The ZLP is the most intense peak in the EELS spectrum and is comprised of electrons that have passed through the specimen losing zero to a very little energy. The ZLP is needed to (i) calibrate the EELS spectrum and (ii) perform deconvolution and KK analysis. However, due to the intensity of the ZLP, it is difficult to observe low energy transitions without saturation. We, therefore, collected a low-loss spectrum with the ZLP partially shifted off the detector in order to increase exposure times (Fig. S1a). The right tail of the ZLP was, then, fit to an exponential function to remove the ZLP contribution from the low-loss spectrum (Fig. S1b). We compared this method to additional datasets in which KK analysis was performed in DM after the ZLP was removed using the reflected tail method, and the results do not differ (Fig. S1c).

After the removal of the ZLP, multiple single electron transitions can be observed in the low-loss EELS spectrum in Fig. 2(a). The onset of intensity provides a measure of the bandgap of the material, and here, we measure a bandgap of ∼0.6 ± 0.18 eV. These measurements are consistent with previous bandgap EELS measurements collected at 60 kV from a FIB foil (Fig. S1c).

Ab initio density functional theory calculations provide further insights into the nature of the electronic transitions in Fig. 2(a). The electronic structure and dielectric function of V[TCNE]_x was calculated using the Vienna ab initio Simulation Package (VASP) (version 5.4.4) with a plane wave basis, projector-augmented-wave pseudopotentials,^48–51 and a generalized gradient approximation of Perdew, Burke, and Ernzerhof (PBE).⁵² The hybrid Heyd–Scuseria–Ernzerhof (HSE06) functional⁵³ with range separation parameter ω = 0.2 incorporates an admixture of Hartree–Fock energies with PBE to improve the accuracy of the bandgap calculations. The structure is very similar to that obtained in Ref. 19 although the details of the electronic structure, including the bandgap and band ordering, appear to differ from those reported there. Detailed plots of the partial densities of states are shown in the supplementary material (Fig. S2). From this, we calculate the loss function averaged over all momentum transfers, Tr(1/ε), shown in Fig. 2(b). The general features show remarkable agreement with the experimentally measured low-loss data. It should be noted that while the energetic positions of the peaks align well, the peak intensities do not due to an assumption of an infinite periodic lattice in the DFT model. Thus, the matrix/symmetry elements likely differ between the calculation and experimental data as the DFT calculation is performed for a single crystal that is oriented; however, as-grown films are not single crystal and are randomly oriented.

A schematic band ordering from our calculations is shown in Fig. 2(d), which suggests an approximate effective U for the vanadium d states of ∼5 eV and for the π orbitals of ∼2 eV, along with a ligand field splitting between the equatorial and apical TCNE molecules of ∼1 eV. The V–N bond lengths of the apical and equatorial TCNE molecules are very similar, preserving the octahedral point symmetry of the V ion; the apical/equatorial molecule distinction emerges because only two of the apical TCNE nitrogen molecules bond to vanadium, whereas all four in an equatorial TCNE molecule bond to vanadium. The result is a small (∼1 eV) ligand field splitting for the C–N π bond. We note that while the local V-cyano coordination is similar for all the six TCNE bonded to central V, the apical TCNE molecules are not fully coordinated to adjacent V, e.g., only two cyano groups for each apical TCNE are coordinated to a neighboring V as opposed to fourfold coordination for equatorial TCNE. Additionally, the DFT calculations predict an optical bandgap of 0.8 eV, providing theoretical support for our experimentally measured bandgap of ∼0.6 ± 0.18 eV. These results are, to the best of our knowledge, the first quantitative comparison of DFT calculations and experimental measurements of the electronic structure of V[TCNE]_x.

In addition to providing a measurement of the optoelectronic properties of a material, STEM EELS can also be used to probe the structural information. As stated in the Introduction, the general structure of V[TCNE]_x is believed to be a 3D network of V²⁺ in octahedral coordination with six surrounding TCNE molecules, and the above calculations were performed using a V[TCNE]_x crystal with the same parameters [Fig. 2(c)]. The high-loss EELS data collected from 260 to 570 eV contain the C K-edge, N K-edge, V L_2,3 edge, and the O K-edge if present [Fig. 3(a)]. The absence of the O K-edge indicates that the material was not oxidized during the sample transfer to the microscope. The V oxidation state of the V can be determined by the L₃ maximum position along the x axis of the EELS spectrum. According to the x-ray absorption near edge fine structure (XANES) data by Kortright et al.,³² the L₃ peak for V²⁺ occurs at ∼517 eV. Fitting a Gaussian peak to the L₃ edge in the deconvolved spectrum in Fig. 3(b) indicates that the center of this peak lies at 516.8 eV, confirming that the oxidation state of the CVD films is V²⁺. While the V oxidation state has been measured previously, the methods used were spatially averaging techniques and it was unknown whether the V oxidation state varied within the V[TCNE]_x film. Exploiting the high spatial resolution of STEM EELS spectrum imaging, spatial variations in the V oxidation state can be probed. Using a cross section of the thin film prepared using FIB (see Sec. II), it was demonstrated that there is no shift in the V L₃-peak maximum through the foil thickness, indicating that the oxidation state is homogeneous throughout V[TCNE]_x films (Fig. S3).

EELS can also be used to provide insights on the local bonding environment by analysis of the energy-loss near edge fine structure (ELNES). For V²⁺ in a perfect octahedral environment, the L₂ peak (and to a lesser extent L₃) exhibits a complex multiplet structure arising from the interactions of crystal field splitting of the d orbitals, spin-orbit effects, and core hole effects that have been observed in previous XANES measurements of V[TCNE]_x.³² Simulations of the ELNES using crystal field multiplet calculations confirm that V²⁺ in an octahedral environment exhibits a distinct “doublet” structure on the L₂ peak (Fig. S4). The experimental data [Fig. 3(b)] show the presence of this doublet structure on the L₂ peak and asymmetry of the L₃-peak. This is consistent with simulation and indicates that V in our CVD grown films is in a perfect octahedral environment. The difference in relative intensities of the doublet features in the L₂-peak has been attributed to back-bonding effects that are not included in the multiplet calculation.³² 

The C and N K-edges are not influenced by multiplet effects, and the sharp, intense peaks on both edges [Fig. 3(a)] arise from 1s-π* transitions between well-defined atomic/molecular orbitals in the TCNE molecule.³² Figure 3(c) shows non-linear least squares (NLLS) fitting of the C K-edge with four Gaussian peaks, making up the total 1s-π* peak. The first two Gaussians at 284 and 285 eV are associated with C1s $→ π C = C *$ transitions, while the higher energy Gaussian peaks at 286.5 and 287.8 eV are associated with C1s $→ π C ≡ N *$⁠. The ∼1 eV split between the π* molecular orbitals (MOs) associated with both the C≡N and C=C bonds suggests that not all TCNE molecules are identical. This is consistent with the previously described DFT calculations that indicate the MOs associated with the apical ligands are shifted to higher energy relative to those on the equatorial ligands. Figure 3(d) shows the NLLS fitting of the N K-edge with three Gaussians at 396, 399, and 401 eV, which are all associated with N1s $→ π C ≡ N *$ transitions. Multiple peaks arise for N due to the p–d mixing with V d orbitals. Additionally, one peak is likely associated with the small fraction of N atoms on the apical TCNE molecules that are not bonded to V²⁺.

We have also established that the V L₂ doublet structure evolves as a function of both sample aging and oxidation. Thin films are deposited on multiple TEM grids during a CVD growth, which means samples grown under identical conditions can be compared. In Fig. 4(a), spectra from three plan-view samples grown under identical conditions are compared: (1) fresh (examined within a few hours of growth), (2) aged (stored for 1 week at room temperature in an argon glovebox (O₂, H₂O < 1.0 ppm), and (3) oxidized (exposed to air for >3 weeks). The V L_2,3 peaks of the fresh sample show asymmetry on the L₃ peak and a clear doublet splitting on the L₂ peak. In the aged sample, the fine structure is still present although less pronounced than the fresh specimen. Crucially, there is no evidence of an oxygen K-edge in either dataset. This indicates that the reduction of the splitting is due to aging rather than oxidation. As discussed earlier, V is in a perfect octahedral environment in the fresh sample, but it appears that over time, structural relaxation in the V[TCNE]_x film occurs, which lowers the point symmetry at the V site, changing the local electronic structure, and, thus, suppresses the multiplet fine structure. This relaxation of the structural order in V[TCNE]_x films is also reflected in Raman studies of V[TCNE]_x films comparing fresh vs aged samples.⁵⁴ Raman spectroscopy of V[TCNE]_x films directly validates the distortion about V as the V–N stretching mode features vanish upon aging, which is attributed to a change in the bonding between V and TCNE molecules.⁵⁴ We also note that the V L₂ peak from the cross-sectional FIB foils does not show a well-defined split in most cases, but rather matches that of the aged plan-view films, showing a hint of asymmetry. The C and N K-edges of the cross-sectional foils also lack the fine structure we see in the plan-view data, suggesting that the FIB process damages the films to some extent. Furthermore, the cross-sectional samples all have an Al capping layer to reduce the speed of oxidation. This Al capping layer was deposited using thermal evaporation, which could also contribute to thermal degradation of the cross-sectional samples, as indicated in Fig. S3b, in which we see more asymmetry on the L₂ peak in regions further from the Al capping layer.

In the oxidized sample [Fig. 4(a)], spectral intensity in the 530–550 eV range is attributed to the oxygen K-edge excitation and is consistent with oxidized specimens previously analyzed using XAFS.³² In addition to the presence of the O K-edge, the L₂ doublet structure has completely disappeared, indicating that the local point symmetry at the V site is further reduced. However, there is no change in energy-loss of the V edge, indicating that the V oxidation state remains as V²⁺. This intriguing observation is consistent with changes observed at the C K-edge upon oxidation. Figure 4(b) shows EELS data of the fresh and oxidized specimens. It is apparent that the peak associated with 1s-π* transitions at the C K-edge is significantly reduced on oxidation. In contrast, the 1s-π* peak at the N K-edge remains sharp and intense. The broad peak before the N K-edge is the Pd M_4,5 edge from the Pd Substratek TEM grids. There is no Pd peak in the fresh spectrum as that specimen was grown on a Pt Substratek TEM grid; however, both samples are from the same growth as described in the methodology section. These observations give clear mechanistic insights into the oxidation process, indicating that oxygen attacks the C=C double bond in the TCNE molecules. This will result in transforming C from sp² to sp³ coordination, and thus, the TCNE molecule is no longer planar. The sp³ environment allows for (relatively) free rotation around the new C–C single bond, which will distort the local point symmetry at the V site, therefore reducing or eliminating the observed fine structure on the V L₂ peak. These TEM results may also provide additional insights into the progression of the C=C stretching mode peaks seen in the Raman spectra of aged films, which see a significant broadening and a change in the relative intensity between the two dominant peaks.⁵⁴ How these features change with age may reflect not only the structural relaxation in the material but perhaps also the oxidation, targeting C=C naturally occurring in encapsulated films as they age in atmospheric conditions. With these new insights into the oxidation mechanism of V[TCNE]_x films and the success of this DFT model, the origin of these spectral changes may be revealed and identified both experimentally and theoretically.

We used electron microscopy and density functional theory methods to study and reveal previously unknown properties of the quantum information relevant material, V[TCNE]_x. Low-loss EELS measurements align quantitatively with DFT calculations, concretely establishing the optoelectronic properties of V[TCNE]_x. We measure a bandgap of 0.6 eV, which is close to the 0.8 eV optical gap of the DFT band structure calculations and ∼0.5 eV bandgap from past transport methods. Our DFT calculations identify multiple single electron transitions observed in the low-loss EELS data, including features from Hubbard U on the p and d states and the apical/equatorial ligand field splittings. Through analysis of high-loss EELS data, we have shown that our CVD grown films of V[TCNE]_x contain V in an oxidation state 2+, which does not change spatially. Splitting on the V L₂ peak is observed, indicating that V is in a perfect octahedral environment. EELS measurements on oxidized and aged films show that this V splitting decreases upon aging and oxidation independently of each other, therefore establishing two distinct degradation mechanisms of V[TCNE]_x. The reduction of the C K-edge 1s-π* peak suggests that the carbon–carbon double bond of the TCNE molecules is targeted during oxidation. This change from sp² to sp³ forces the carbon to a tetrahedral geometry, therefore distorting the octahedral environment of the V and eliminating the doublet structure on the V L₂ peak. These findings help elucidate the structure of V[TCNE]_x and establish the theoretical framework necessary to predict and study the family of related metal–ligand magnetic materials.^55–60 

See the supplementary material for EELS data collected from a FIB foil, the partial density of states for majority and minority spins, and the simulated L_2,3 V edge.

The authors would like to thank Dr. Jessica Alexander for her contributions to the FIB and EELS procedures used in this work. V[TCNE]_x synthesis (S.W.K., M.C., and E.J.-H.) was supported by the NSF under Grant No. DMR 1808704. The electron microscopy experiments (A.H.T. and D.W.M.) were supported by the Center of Emergent Materials, an NSF MRSEC, under Award Nos. DMR-2011876 and DMR-1420451. Theoretical calculations (M.E.F. and Y.S.) were supported by the NSF under Grant No. DMR-1808742.

The authors have no conflicts to disclose.

The data that support the findings of this study are available from the corresponding author upon reasonable request.