Straining VO2 provides new insight into Mott Physics
Metal oxides display abrupt metal-to-insulator transitions (MITs), and for this reason might lead to the design of new electronic switches in the crucial search for ‘More-than-Moore’ technologies. In this context, vanadium dioxide (VO2) is a promising testing ground for improving the understanding of the mechanisms responsible for MITs that are thought to be driven cooperatively by Peierls and Mott physics – where the former relates to the crystal structure of the material, whereas the latter is concerned with electron correlations.
The development of a theoretical framework described how the application of strain on VO2 thin films enhances electron correlations and leads to orbital anisotropy. These results suggested that it is possible to engineer an orbital-selective Mott transition (OSMT) within the metallic phase of VO2. This is relevant to multi-orbital systems and might play a role in iron-based superconductors. Soft X-ray Absorption Spectroscopy (XAS) and Hard X-ray Photoelectron Spectroscopy (HAXPES) measurements using the Surface and Interface Structural Analysis beamline (I09) at Diamond Light Source provided experimental evidence confirming the predicted effect of strain on VO2 thin films near room temperature.
I09 was particularly suited for this study as it enables soft and hard X-ray measurements, which were both necessary to characterise relative orbital occupancy and electron correlations for the three considered strain scenarios. This theoretical and experimental investigation of strain-induced OSMT identifies a way to tune the Mott contribution to MITs in metal oxides, with potential applications beyond silicon-based electronics.
Engineering quantum materials with tunable metal-insulator transitions (MIT) is an important subject in condensed matter physics. Coherent oxide/oxide film growth provides the opportunity to tailor the MIT by facilitating large percent level strains well beyond their bulk counterparts. A classic example is strain engineering the thermally-induced transition temperature of VO21. In its bulk form, VO2 exhibits an abrupt MIT near room temperature (65ºC) coinciding with a structural transition from rutile (R) to monoclinic (M1) involving the formation of V-V dimers (Fig. 1). The origin of the MIT has been hotly debated since the 1970s as either arising primarily from the lattice (i.e. a Peierls distortion) or from electron correlations (i.e. Mott transition).2 When epitaxial VO2 is grown on isomorphic rutile TiO2 the transition is modified by ± 50 K depending on how the epilayer distorts itself to match the larger substrate lattice1. This has generally been considered to demonstrate that the transition to the insulating phase is driven mostly by the change of crystal structure (i.e. Peierls) instead of the Mott physics. However, this view has now been revised by new studies revealing increased electron correlation effects when biaxial strain is employed to raise the transition temperature i.e. strained-induced Mott physics3.
Mott physics is usually illustrated by the ratio of U/W, where U is the onsite Coulomb interaction and W is the bandwidth of the conduction band. When a band is half-filled, the expected metallic phase can transition to an insulating phase due to the Mott gap dynamically generated from the electron correlations if the U/W is sufficiently large. Although it is well-accepted that varying W can lead to a MIT, such a mechanism has never been experimentally demonstrated in a single system. In the single band Hubbard model, the Mott insulating state can be tuned by arbitrarily changing U/W, but this is very hard to do experimentally because U and W are fixed materials properties and thus usually not tunable. Furthermore, real systems are usually multi-orbital and thus do not exhibit a single exactly half-filled band. The orbital selective Mott transition (OSMT) describes the situation for multi-orbital systems. The basic idea of the OSMT is that the electron correlation can in principle have different effects on different orbitals in a multi-orbital system. Thus, the OSMT is a new model describing an intriguing state in which some of the orbitals become Mott-insulating while others do not. The OSMT could then explain many of the puzzling metallic states exhibiting insulator-like behaviour. However, because the ingredients resulting in an OSMT strongly depend on the materials, OSMT phases are rarely seen and consequently hard to be engineered by design.
To examine a strain-induced OSMT, 10 nm epitaxial VO2 films were grown on three different orientations of TiO2 to modify the electron correlation strength within the same crystal phase. The transition temperatures varied between the three strain cases per established trends1, confirming they were coherently strained. The samples were all held ~20ºC above their respective transitions (i.e. the rutile phase). In this manner, the strain-engineered Mott physics could be isolated from Peierls contributions in one system. The measurements were performed at Diamond’s I09 beamline employing both soft X-ray absorption spectroscopy (XAS) and hard X-ray photoelectron spectroscopy (HAXPES). The benefit of I09 is its almost unique ability to use hard and soft spectroscopy measurements in tandem. For these studies, HAXPES of the valence band was required to measure the U/W and XAS of the V L-edge was necessary for studying preferential orbital filling. Both techniques were required to confirm evidence of a strained-induced OSMT.
In its bulk form, the rutile VO2 phase is a metal with near equal orbital filling in its topmost bands, in contrast strain-engineered rutile VO2 films should be strongly correlated metals with preferential orbital filling i.e. the OSMT phase (Fig 1). VO2 grown on TiO2(001), where the c-axis is compressed, acted as a reference for the bulk since the lowering of its transition temperature can be described within Landau-Ginzburg theory i.e. no change in Mott physics3. The HAXPES measured valence band of VO2/TiO2(001) displayed the expected line shape of rutile metallic VO2 from band theory3. Furthermore, the lack of significant angular dependence of the V L-edge XAS was also consistent with bulk VO2, where all topmost orbitals are nearly equally filled4. VO2 grown on TiO2(100) and (110) cases reflect elongating the c axis to match the underlying larger TiO2 lattice constant, albeit through different means. In both cases, the valence band spectra revealed a reduction of W (i.e. increasing U/W) along with a loss of Fermi-liquid behavior consistent with a strongly correlated metallic phase. The corresponding XAS revealed evidence of preferential orbital filling remarkably like that observed for the insulating monoclinic phase of bulk VO24. Taken together the data from Diamond confirmed that an OSMT phase could be modulated by biaxial strain without requiring chemical doping.
A generalised 2-band OSMT model was employed to interpret the spectra for VO24. The conventional method of describing the MIT of bulk VO2 involves the formation of tilting V-V pairs along the rutile c-axis upon going from the metallic rutile phase to the insulating monoclinic phase2. In the rutile phase, two partially filled bands cross the Fermi level, referred to as π* and d|| (Fig. 1). The V-V dimerising is largely responsible for splitting of the d|| and giving rise to its insulating 1D orbital character i.e. preferentially orbital filling of the d|| band. A two-band OSMT model is adequate for capturing the topmost bands of VO2. The generalised model revealed that when a multi-orbital system is close to the required half-band filling for a Mott transition (as for VO2), strain can be employed to preferentially fill one orbital at the expense of the other. In doing so, one of the orbitals can satisfy the Mott criterion of half-filling but not both. In contrast to a true Mott transition, in the OSMT phase a Mott gap only forms in one orbital and so a true insulating phase is never achieved i.e. a strongly correlated metal results. In rutile VO2, elongating the c-axis preferentially fills the d|| state at the expense of the π* (Fig. 1). The resultant strained-induced OSMT mimics the Peierls transition, such that the net effect is to remove the spectral weight of the d|| state from the Fermi energy. Importantly, a strain-induced OSMT phase occurs in the absence of crystal structure change (i.e. all strained films were in the rutile phase). Furthermore, an OSMT can only be engineered by strain if the bulk is already correlated. Thus, the strain-induced OSMT reveals the underlying cooperative Mott-Peierls nature of the MIT of bulk VO2. More generally, the observed strain-induced OSMT in one system (i.e. no chemical effects) should revise our understanding in Mott physics and may explain non-Fermi liquid behaviors observed in other correlated systems where the bands are not strictly half-filled.
Figure 1: A schematic highlighting the differences between the insulating monoclinic phase of bulk VO2 from the OSMT phase of suitably strained rutile VO2. The V-V dimers of monoclinic VO2 largely facilitate the true insulating phase by preferentially filling and splitting the d|| orbital. The strain-induced OSMT phase mimics the monoclinic phase but a Mott gap can only form in one orbital and so a true insulating phase cannot occur. The fact that a strain-induced OSMT phase can be induced confirms that Mott Physics must contribute to the MIT of bulk VO2.
- Muraoka, Y. et al. Metal–insulator transition of VO2 thin films grown on TiO2 (001) and (110) substrates. Applied Physics Letters 80, 583, doi:10.1063/1.1446215 (2002).
- Eyert, V. The metal-insulator transitions of VO2: A band theoretical approach. Annalen der Physik 514 (9), 650-704, doi:10.1002/1521-3889(200210)11:9<650::AID-ANDP650>3.0.CO;2-K (2002).
- Quackenbusch, N.F. et al. Stability of the M2 phase of vanadium dioxide induced by coherent epitaxial strain. Physical Review B 94, 085105,doi:10.1103/PhysRevB.94.085105 (2016).
- Haverkort, M.V. et al. Orbital-Assisted Metal-Insulator Transition in VO2. Physical Review Letters 95, 196404, doi:10.1103/PhysRevLett.95.196404 (2005).
National Science Foundation under DMR-1409912.
Dr Nicholas F Quackenbush, Professor Wei-Chen Lee, Professor Louis Piper, State University of New York at Binghamton, lpiper (at) binghamton.edu
Mukherjee S, Quackenbush NF, Paik H, Schlueter C, Lee TL, Schlom DG, Piper, LFJ, Lee, WC. Tuning a strain-induced orbital selective Mott transition in epitaxial VO2. Physical Review B 93, 241110, doi:10.1103/PhysRevB.93.241110 (2016).
Metal-insulator transitions; Vanadium dioxide; X-ray absorption spectroscopy; Hard X-ray photoelectron spectroscopy