Detection of Spin Polarized Band in VO₂/TiO₂(001) Strained Films via Orbital Selective Constant Initial State Spectroscopy

Abstract

The VO₂ is a 3d¹ electron system that undergoes a reversible metal–insulator transition (MIT) triggered by temperature and characterized by an interplay between orbital, charge and lattice degrees of freedom. The characterization of the MIT features are therefore extremely challenging and powerful investigation tools are required. In this work, we demonstrate how a combination of resonant photoemission and constant initial state (CIS) spectroscopy can be used as an orbital selective probe of the MIT studying three different VO₂/TiO₂(001) strained films. The CIS spectra of the V 3d and V 3p photo-electrons shows sensitivity to different orbital contribution and the presence of a spin polarized band close to the Fermi level.

1. Introduction

The VO₂ is a fascinating system, well know from 60s for its metal–insulator transition (MIT). The transition is triggered by temperature (67 °C in bulk crystals) and since its discovery [1], it continuously attracted the attention of researchers because of possible applications in different areas as energy savings [2,3,4], electron correlation driven electronic devices and ultra-fast switching [5,6,7]. However, the nature of the MIT itself is determined by the complex interplay among lattice, electronic and orbital degrees of freedom [8,9,10,11,12,13,14], which make the transition feature highly dependent on sample quality and synthesis methods [15]. Consequently, different strategies have been applied in order to control and manipulate the MIT features like doping, application of epitaxial strain [11,13,16,17] and exploitation of quantum size effects [18,19,20]. Among them, strain emerges as a powerful method to tune the MIT, since it modifies atomic distances, orbital order and, in particular, electron-electron correlation. X-ray absorption near edge structure (XANES) is particularly suitable to study the complex VO₂ scenario, since it is able to probe at the same time the electronic structure and the local order [21,22,23,24,25,26]. Since, in materials at low dimensions, the electronic properties can be modulated by different parameters, such as surface states and spatial confinement [27,28,29,30,31], the investigation of electronic configuration in strained VO₂ can lead to its exploitation in several applications.

In a previous work, we confirmed the reliability of XANES to probe the orbital strain dynamics across the MIT [17]. We probed the structural and orbital contributions by monitoring the changes in the V L_2,3 edge XANES on single crystalline strained films of VO₂/TiO₂ (001)

In this manuscript, we demonstrate that the combination with resonant photoemission spectroscopy (ResPES) allows us to collect and identify XANES partial electron yield spectra connected to the different screening channel of the VO₂ valence band [32]. This is possible because ResPES offers the possibility to change the photon energy through a resonance of the system, therefore, probing the resonant behavior of photoemission spectra.

The intensity of the photoemission spectra is modulated because of the interference between the direct photoemission channel and the photoabsorption Auger-like channels. Two acquisition methods can be considered to study the resonance effect: constant final state (CFS) and constant initial state (CIS) spectroscopy [33]. In the constant final state, the spectrum is collected when the incident photon energy hv varies while the detected electron kinetic energy is constant [21,33,34]. XANES spectra acquired with the Auger yield is the typical example of the CFS acquisition. The strength of the CFS spectroscopy relies on the possibility to choose the appropriate Auger channel to probe the system. As demonstrated in previous studies selecting the Auger V L₃M₂₃M₄₅ or the O KL₂₃L₂₃, it is possible to optimize the collection at the vanadium L edges, or at the oxygen K edge in any vanadium oxide [17,22,35].

In a CIS acquisition, both the energy of the exciting photon and the energy of the detected photoelectron (Ke) are simultaneously varied, in order to maintain constant the binding energy (BE) = hv − K_e [21,33,34]. ResPES spectra are collected using the CIS mode to probe the same binding energy region for each photon energy. As a consequence, CIS spectroscopy allows studying the resonant enhancement of the cross-section of a selected photoemission feature when the photon energy spans through a core resonance. In synthesis, CIS and CFS are two detection methods, which allow for the collection of an absorption spectrum exploiting different XANES partial electron yield channels.

This work focuses on CIS absorption spectra of 3d and 3p electrons in VO₂. Tuning the photon energy through the 2p core resonance, different possible decays of the excited state have been measured [36]. In our case the absorption process results in the excitation

2p⁶3d¹ + hν → 2p⁵3d²

(1)

which can decay via the Coulomb interaction in two configurations with the emission of one electron from different vanadium and oxygen orbitals:

2p⁵3d² → 2p⁶3d⁰ + e_d

(2)

2p⁵3d² → 2p⁶3p⁵3d¹ + e_p

(3)

The processes described by Equations (2) and (3) emit a 3d and a 3p electron, respectively, and in the ResPES spectra, they will constructively interfere with 3d and 3p direct photoemission channels. The samples investigated were three strained VO₂/TiO₂ films (8 nm, 16 nm and 32 nm thick). We demonstrated that, by exploiting the extreme selectivity of the CIS spectroscopy, it has been possible to probe the strain induced orbital dynamic across the MIT with orbital selectivity and, as a consequence, obtained new insights on the spin polarization of the electronic structure of VO₂.

2. Results

2.1. V 3d CIS Spectroscopy

The quality of the three strained films of 8 nm, 16 nm and 32 nm thickness has been verified through temperature dependent resistivity (reported in Figure A1 in Appendix A for sake of brevity).

The resistivity curves show hysteresis loop with a thickness dependent transition temperature ~315 K in the 8 nm film, ~330 K in the 16 nm and ~345 K in the 32 nm film, in agreement with the thickness dependent transition temperature reported in previous studies [11].

The major spectral changes across the MIT in VO₂ occur around the Fermi level (FL) involving mainly V 3d electrons. The oxygen atoms are locally arranged around the central V atom, according to the slightly distorted octahedral geometry. The t_2g and e_g manifold is split in three main bands: σ* and π*, which are generated by the chemical bonds among V 3d and O 2p electrons with σ and π symmetry, respectively, and the $d_{\left|\right|}^{*}$ band, populated by unpaired d electrons [37] (See Figure 1).

Across the phase transition π* and $d_{\left|\right|}^{*}$ collapse to the FL becoming partially populated. Using the TiO₂(001) as the substrate, the effect of the tensile strain is that of increasing the apical oxygen-vanadium distances. This reduces the stability of the π* band which, in the metallic phase, is a downshift in energy, while the $d_{\left|\right|}^{*}$ band is upshifted [11,17]. The overall effect is the inversion of orbital population at the FL (See Figure 1).

The Valence Band (VB) photoemission spectral features of the VO₂ can be described in terms of the final state configurations. In the insulating phase, the V 3d peak is centered at ~1.5 eV, as can be seen from Figure 2 (an example of single VB spectra acquired at different photon energy can be found in Figure A2 in Appendix B).

Its nature can be explained assuming that the V 3d photo-hole is locally screened by the oxygen atoms around the metallic V atom at the center of the octahedral configuration (see Figure 1). Therefore, the final state is depicted as the 3d¹L (ligand hole) [32]. In the metallic state, we assist the population of the FL. Since electrons are free to move within the solid, an additional screening mechanism and, therefore, an additional final state is available, and can be recognized in the VB: the 3d¹C (coherent hole). This is generated by the non-local screening channel of the photo-hole provided by the surrounding vanadium atoms, and it is centered at about 0.4 eV [32].

Since C and L VB features are generated by two different screening mechanisms, they couple differently with the resonant excited state and exhibit a different CIS lineshape. For the metallic phases (bottom panels of Figure 2) it is easily recognized the population at the FL at BE = 0 eV. The maximum of the resonance is located in the energy range 518 to 519 eV. At hv ~516 eV a weaker resonant enhancement is also observed. The photon energy of the V 3d electrons is more pronounced across the L₃ edge with respect to the L₂, because of the presence of Coster–Kronig decay (V L₂L₃M_4,5) [38]. Therefore, we limited the study of the spectral changes to the L₃ edge.

The CIS spectra of the C valence band feature are compared in Figure 3. All spectra have been normalized to the maximum of the L₃ resonance intensity. In the metallic phase, two main features are clearly detected, the σ* resonance around 518.4 eV and the small pre-edge peak at ~516 eV.

The CIS spectra resemble the more commonly studied XANES spectra that also exhibit a strong σ* resonance and a pre-edge peak feature [17]. In XANES spectra collected across the phase transition boundaries, the most evident spectral changes are observed in the pre-edge peak region, which is mostly generated by a superposition of π* and $d_{\left|\right|}^{*}$ bands [22,39,40], as demonstrated for this set of film in our previous study [17]. However, since XANES is less selective than CIS, the line-shape is usually broader, and the pre-edge peak is not well isolated, as in Figure 3. Since CIS is a XANES partial electron yield, across the MIT, we expect to observe the same behavior as XANES, i.e., spectral changes around the pre-edge energy region. However, an increased sensitivity to different spectral feature as a function of the CIS channel is expected.

To understand the nature of the CIS pre-edge peak, it is necessary to follow its evolution vs. thickness, i.e., the strain contribution. Increasing the strain, the intensity of the pre-edge peak increases relatively to the σ* resonance, and concurrently, the valley amongst them decreases its deepness. In addition, a broadening of the pre-edge peak width occurs. In the metallic phase of the VO₂, π* and $d_{\left|\right|}^{*}$ are partially degenerate at the FL and, hence, are both populated, although with a different occupation. In bulk VO₂, because of the strong V-O hybridization, the FL is mostly occupied by the $d_{\left|\right|}^{*}$ band with a small component of π* (see Figure 1). In strained samples the $d_{\left|\right|}^{*}$ band empties respect to the π* band, opposite to bulk samples. As a consequence, it is possible to assign the pre-edge peak of the C CIS spectra in the metallic phase to the $d_{\left|\right|}^{*}$ band (516 eV). In fact, its intensity and width increase with the strain, indicating that the increase of the empty states in agreement with the strain-induced orbital dynamics described above. In addition, within our experimental resolution, any spectral feature linked to the π* band are not detectable. It is important to notice that, with respect to standard XANES spectra, the increased selectivity of CIS spectroscopy make it possible to collect spectra with sharper features.

The CIS spectra of the L valence band feature (BE = 1.5 eV) are compared in Figure 4. The overall line shape is broader in respect to the C CIS spectra, and the individual orbital contributions are not distinguishable. In the metallic phase, the feature at ~516 eV is evident.

Increasing the strain, its intensity decreases. This trend is compatible with the increase of the population (and the depletion of the empty states) of the π* band as a function of the strain, which suggests an increased π* sensitivity in respect to the CIS spectra of the C feature. In the insulating phase, the feature around 515.5 eV is assigned to the π* band. The $d_{\left|\right|}^{*}$ band is not detectable, because of the broadening of the σ* features, which might cover this weak signal.

The selective sensitivity to different empty bands for the C and L partial electron yield is due to their nature. The C final state is generated by the cooperative screening of the 3d photo hole by the nearest vanadium atoms. In metallic vanadium dioxide, the $d_{\left|\right|}^{*}$ band is directed along the rutile c axis, i.e., along the V-V bonds. These intermetallic bonds are the same as those involved in the non-local screening mechanism of the C final state, therefore, it is very likely that the core-hole excited state 2p⁵3d² wave function (where the extra 3d electron lies in the $d_{\left|\right|}^{*}$ band) has a greater overlap with the 3d¹C final state wave function. The same consideration can be applied to explain the higher sensitivity of the L CIS spectra to the π* empty band. In the L final state, the 3d photo hole is locally screened by oxygen atoms of the octahedral coordination. Since the π* band is formed by V 3d electrons, which form bonds with π character with oxygen atoms, an excited electron lying in the π* band is more likely to be spatially close to oxygen atoms. Then, the Auger transition of the excited state with an extra electron in the π* band is favored in the L final state.

2.2. V 3p CIS Spectroscopy

In addition to the 3d electrons, we investigated the resonating behavior of the V 3p_1/2 and 3p_3/2 electrons (Equation (3)) across the V L₃ edge. V 3p electrons can be found at ~41 eV of BE (see Figure 5 and Figure A3 in Appendix B). The lineshape is quite broad, since the multiplet effect splits V 3p levels in two components: V 3p_1/2 and 3p_3/2, similarly to V 2p electrons. In this case, however, the spin-orbit splitting is quite small, and a fitting procedure is necessary to separate the two contributions. The V 3p_1/2 and 3p_3/2 components have been modelled using a pseudo-Voigt curve [41] fixing their intensity ratio to 0.5, while for the background, a Shirley curve has been used. The fit returns the V 3p_1/2 component at BE = 43.6 eV and the 3p_3/2 at BE = 41.2 eV. The measured spin–orbit splitting (2.4 eV) is higher in respect to the value of 1 eV estimated by Zimmerman and co-workers [42].

Exploiting these results, the CIS spectra of the 3p_1/2 (BE = 43.6 eV) and 3p_3/2 (BE = 41.2 eV) features have been obtained. For the sake of brevity, the CIS spectra of the 3p_1/2 and 3p_3/2 components are called CIS_1/2 and CIS_3/2, respectively. The CIS_3/2 spectra are compared in Figure 6, and the contribution to the intensity of the V 3p CIS spectra of the incoherent Auger channel V L₃M₂₃M₄₅ has been minimized using the procedure in [43].

The V 3p_3/2 spectra exhibit a strong enhancement of the pre-edge feature respect to the V 3d CIS spectra. This is due to the large overlap among 3p and 3d orbitals, which may favor the decay 2p⁵3d² -> 2p⁶3p⁵3d¹ + e_p of the photo-excited state. According to the analysis of the V 3d CIS spectra, the two features of the CIS_3/2 spectra of the metallic phase can be assigned to the $d_{\left|\right|}^{*}$ (515.8–516 eV) and σ* (518.4 eV) empty bands. In the metallic phase, their intensity ratio ($d_{\left|\right|}^{*}$/σ*) increases with the strain, in agreement with the V 3d (C) CIS spectra and the strain-induced orbital occupancy at the Fermi level, by shifting up the $d_{\left|\right|}^{*}$ while lowering the π* band. In addition, the $d_{\left|\right|}^{*}$ and σ* features are separated by about 2.5 eV, again, in good agreement with the $d_{\left|\right|}^{*}$-σ*split observed in the 3d C CIS spectra.

In the insulating phase, $d_{\left|\right|}^{*}$ (515.9 eV) and σ* (518.3 eV) bands are separated by 2.4 eV, in agreement with XANES results available in the literature [44].

The CIS_1/2 spectra are compared in Figure 7. An unresolved feature is detected at ~516 eV and in the insulating phase, all spectra are dominated by two main features at 518.1 eV and at 516.7 eV. The energy position does not match that of the $d_{\left|\right|}^{*}$ and σ* features observed in the CIS_3/2 spectra.

The pre-edge peak of the CIS_1/2 spectrum is shifted toward higher energies respect to the CIS_3/2 spectra, indicating that the excited states associated to the Auger decay of the 3p³_3/2 and the 3p¹_1/2 final states are different. At the resonance of the L₃ edge, the only electrons available for the XANES excitation are the 2p_3/2. As a consequence, because of the dipole selection rules, the total angular momentum of the electron excited in the resonant state can only be J = 5/2 or J = 1/2. Depending on the J value, electrons can be excited in the specific empty band with the proper symmetry. This may result in a preferential decay of the excited state in the V 3p_3/2 or 3p_1/2, depending on the excited and final state symmetry.

In the CIS_3/2 spectra a huge increase in the $d_{\left|\right|}^{*}$ band sensitivity is observed, while the CIS_1/2 does not show any resolved feature correlated to the MIT. This result suggests the occurrence of a spin polarization of the $d_{\left|\right|}^{*}$ band, which favors the decay in the final state with a hole in the V 3p_3/2 level.

3. Discussion

Our work demonstrates how the selectivity of CIS spectroscopy can be exploited, in order to track the strain induced orbital dynamic of VO₂ across the MIT and extract important information regarding the band structure.

The use of ResPes allowed us to isolate the contribution in the VB spectra from two different screening channel L and C.

We showed that the CIS spectra of these two features are orbital selective probes of the empty bands structure of VO₂ as a function of strain and across the MIT. In particular, the C CIS spectra are more selective toward the $d_{\left|\right|}^{*}$ band while the L spectra to the π*.

The C screening channel is generated by the cooperative screening of the photo-hole from itinerant electrons in metallic VO₂. Accordingly, the C CIS spectra exhibits a pronounced sensitivity for the $d_{\left|\right|}^{*}$ band, which are related with the intermetallic bond. This allowed us to observe a progressive depopulation of the $d_{\left|\right|}^{ }$ band at FL (increase in $d_{\left|\right|}^{*}$ band intensity) as a function of strain.

On the other hand, for the L channel the photo-hole is screened by the nearest oxygen atoms. Consequently, the CIS spectra of the L feature show an increased sensitivity for the π* band. As a function of strain, the reduction of the intensity of the π* related feature is observed, which confirms the strain induced orbital evolution of the π* band in the metallic phase of VO₂.

Finally, from the analysis of the CIS spectra of the V 3p electrons, we suggest the occurrence of the spin polarization of the $d_{\left|\right|}^{*}$. The use of CIS spectroscopy as orbital selective probe of the MIT in VO₂ demonstrated in this study can be extended to all transition metal oxides and, in general, to many other complex or correlated systems.

4. Materials and Methods

VO₂ films 8 nm, 16 nm and 32 nm thick were deposited on a clean substrate of TiO₂ (001) using RF-plasma assisted oxide-MBE instrument with a base pressure < 4 × 10⁻⁹ mbar. Using a growth rate of 0.1 Å/s, the thickness was controlled via the deposition time. During the growth, the substrate was heated at the temperature of 550° C. Additional details of the epitaxial film preparation are reported elsewhere [11,45].

CIS and ResPes experiments were performed at the ANCHOR end-station of the ALOISA beamline [46] at the Elettra synchrotron radiation facility. The photoelectrons were collected at normal emission by a PSP Vacuum 120 mm electron analyzer equipped with a 2D delay line detector. The photons were linearly polarized in the scattering plane and the sample was irradiated at the magic angle (~35° respect to the sample surface). Measurements were performed using the constant pass energy of E_p = 20 eV with an overall resolution of 0.25 eV.

The ResPes maps of the 16 nm and 32 nm films were acquired using two different intervals of photon energy 511–514 eV (step 1 eV) and 514.2–528.2 eV (step 0.2 eV). The photoelectrons were acquired in the kinetic energy (KE) range 502–508 eV (step 0.1 eV) to collect the V 3d photoemission signal, while to detect V 3p photoelectrons, the KE 460–470 eV (step 0.2 eV) was used. For the 8 nm sample, the photon energy range was 511–528.2 (step 0.2 eV), V 3d KE range 502–508 eV (step 0.25 eV) and V 3p (step 0.25 eV). In the next sections of this work, the spectra presented as metallic and insulating VO₂ have been acquired at temperatures of 90 °C and 25 °C, respectively, unless specified otherwise.