A photoinduced metal-like phase of monoclinic VO2 revealed by ultrafast electron diffraction

See allHide authors and affiliations

Science  24 Oct 2014:
Vol. 346, Issue 6208, pp. 445-448
DOI: 10.1126/science.1253779

How to make vanadium dioxide metallic

At about 70°C, the material vanadium dioxide (VO2) switches from being a semiconductor to a metal. The switch happens so fast that it may be useful in electronic devices, but it is not clear whether the switch is primarily caused by enhanced interactions between electrons or by a change in the crystal structure. Morrison et al. shone laser light on a sample of VO2, initially in a semiconducting state. They used electron diffraction to monitor the changes in the material's crystal structure and simultaneously measured its optical properties to monitor the electronic state. For certain laser powers, VO2 switched to a long-lived metallic state even though it preserved its initial crystal structure.

Science, this issue p. 445


The complex interplay among several active degrees of freedom (charge, lattice, orbital, and spin) is thought to determine the electronic properties of many oxides. We report on combined ultrafast electron diffraction and infrared transmissivity experiments in which we directly monitored and separated the lattice and charge density reorganizations that are associated with the optically induced semiconductor-metal transition in vanadium dioxide (VO2). By photoexciting the monoclinic semiconducting phase, we were able to induce a transition to a metastable state that retained the periodic lattice distortion characteristic of the semiconductor but also acquired metal-like mid-infrared optical properties. Our results demonstrate that ultrafast electron diffraction is capable of following details of both lattice and electronic structural dynamics on the ultrafast time scale.

The central problem of condensed matter physics involves determining the pathways through which microscopic interactions lead to the emergent properties of materials. Collective phases affected by multiple collaborating or competing interactions (13) have provided a great challenge in this respect. Understanding the properties of such phases depends on the development of methods to address and separate the roles of several active interacting degrees of freedom. Here, we demonstrate that recent improvements in ultrafast electron diffraction (UED) instrumentation (47) provide such a capability by exploring the nature of the semiconductor-to-metal transition in VO2 (8).

At ~343 K, VO2 undergoes a first-order transition between two crystalline phases (Fig. 1A). This structural phase transition (SPT) is accompanied by a marked change in conductivity (fig. S3), as much as five orders of magnitude in high-quality single crystals (9). The high-temperature phase (Fig. 1A, left) is metallic with a rutile crystalline structure (R, P42/mnm). The low-temperature phase (Fig. 1A, right) is characterized by semiconducting electronic behavior (with a band gap energy Eg ≈ 0.6 eV) and monoclinic structure (M1, P21/c). The SPT may be understood roughly as the formation of a charge density wave (CDW) along the rutile c axis with wave vector 2cR, which leads to a doubling of the unit cell along this direction. This periodic lattice distortion (PLD) dimerizes vanadium atoms along the cR direction, spaced by 2.85 Å in the high-temperature phase, into alternating V-V separations of 2.62 Å and 3.16 Å. The dimers are also rotated slightly with respect to cR.

Fig. 1 Structure of VO2.

(A) The structure of rutile VO2 (left) and monoclinic VO2 (right). (B) Transmission electron microscopy image of the pulsed laser deposition–grown VO2 sample used in these studies. (C) Example of electron powder diffraction pattern of the monoclinic phase.

A challenge to understanding this semiconductor-to-metal transition has been to determine the relative contributions of electron-lattice interactions (lattice and charge order) and electron-electron interactions (dynamical correlations and orbital selection) to the change in properties and the nature of the semiconducting phase (1015). Here, we interrogated both structure and electronic properties by making use of the beam brightness enhancement provided by radio-frequency compressed ultrafast electron diffraction (1618) in combination with time-resolved infrared (IR) transmittance measurements. The combined approach makes it possible to map the reorganization of the VO2 unit cell during the optically induced transition (19, 20) while simultaneously determining the electronic properties of the material. The results demonstrate a photoinduced transformation to a long-lived state with metal-like mid-IR optical properties and the PLD (or CDW order) of the semiconducting M1 phase intact. This metastable state differs from the equilibrium rutile metal crystallographically, and only involves a one-dimensional (1D) reorganization of charge density rather than a transition to the isotropic 3D electronic state of the high-temperature phase (21, 22).

In these experiments, polycrystalline VO2 films grown by pulsed laser deposition (Fig. 1, B and C) (23), initially in the low-temperature M1 phase (at ~310 K), are subject to optical (800 nm) excitation with 35-fs laser pulses. The time dependence of the changes in structure and electronic properties after optical excitation of the material are determined using pump-probe UED and time-resolved IR transmittance measurements (24). Raw (background-subtracted) ultrafast powder electron diffraction data taken in a transmission geometry (fig. S1) exhibit several weak reflections (Embedded Image, Embedded Image, and Embedded Image), indicated by vertical red lines in Fig. 2A. These peaks are allowed in the M1 phase, thanks to the PLD and the doubling of the unit cell along cR, but not in the R phase. At this intermediate pump fluence of 20 mJ/cm2, the intensity of these peaks clearly decreases with time after photoexcitation, indicative of the optically induced SPT that occurs in some of the (polycrystalline) sample. Similar observations were made in previous ultrafast structural measurements on VO2, which identified a time scale of <500 fs for aspects of the SPT (25, 26).

Fig. 2 Structural dynamics during the semiconductor-to-metal transition in VO2.

(A) Raw, background-subtracted UED data from 0 to 20 ps. Red vertical lines indicate several weak reflections allowed in the M1 phase but not allowed in the R phase. Blue vertical lines indicate several peaks present in both equilibrium phases. Gray vertical lines indicate peaks for which hM = 0, that are only affected by structural changes orthogonal to cR. (B) Overall diffraction difference spectrum from –0.5 to 20 ps. (C) Time-resolved diffraction peak intensity showing fast (~300 fs) and slow (~1.6 ps) dynamics, respectively, for peaks indicated by red and blue vertical lines in the diffraction spectra in (A), (B), (E), and (F). (D) Fluence dependence of the fast and slow signal amplitudes as measured for the (Embedded Image) and (220) peaks shown in (C). The range of fluences for which no SPT is observed is indicated by the hatched region. Inset: Time-resolved IR (5 μm, 0.25 eV) transmissivity in the hatched fluence region displays a persistent decrease to a very long-lived plateau (>100 ps). The amplitude of this decrease reaches >99% at 3.7 mJ/cm2, indicating a closing of the semiconducting gap and a transition to a metallic-like state. (E) Diffraction difference spectrum for the fast dynamics from –0.5 to 1.5 ps. The change in diffracted intensity is plotted with respect to –1 ps. (F) Diffraction difference spectrum for the slow dynamics. The change in diffracted intensity from 2 to 10 ps (referenced to 2 ps) is shown.

Beyond these specific peaks, however, photoexcitation induces changes in diffracted intensity over the entire scattering vector range shown. This is clearly evident in Fig. 2B, which shows the time-dependent difference in diffracted intensity with respect to negative pump-probe delays (i.e., before photoexcitation). The presence of multiple time scales can be seen clearly in Fig. 2C, which shows the time-dependent intensity of several diffraction features labeled in Fig. 2A. After photoexcitation at 20 mJ/cm2, there is a fast (time constant 310 ± 160 fs) decrease in the intensity of diffraction peaks associated with the PLD (e.g., Embedded Image), followed by a slow increase (time constant 1.6 ± 0.2 ps) in the intensity of most peaks in the range s < 0.45 Å–1 that are present for both phases (e.g., the 220 and 200 features). These are the only two ultrafast time constants observed in the data up to 10 ps.

The amplitudes of these two qualitatively distinct diffraction signatures each scale linearly with fluence, but they have different slopes and threshold fluences: 9 mJ/cm2 for fast pump-induced changes to peaks associated with the PLD in the M1 phase, and 2 mJ/cm2 for the slow changes. At pump fluences below ~9 mJ/cm2 (hatched region in Fig. 2D), there is no change in the intensity of diffraction peaks associated with the PLD in the M1 phase. The inset in Fig. 2D shows time-resolved IR transmittance curves at 5 μm (0.25 eV) for the VO2 film at several pump fluences below 9 mJ/cm2. These curves show a persistent decrease in IR transmissivity that increases with pump fluence and reaches an amplitude of >99% by 3.7 mJ/cm2. This observation is in quantitative agreement with previous experiments using multi-THz spectroscopy on VO2 films grown by pulsed laser deposition (27, 28) and is indicative of a (partial) transition to a state with metal-like ac conductivity at a threshold pump fluence of ~2 mJ/cm2.

We isolated the full spectrum of diffracted intensity changes that correspond to the fast and slow components in the time domain by choosing reference time points for computing the intensity differences that separate these dynamics: t = –1 ps in Fig. 2E (fast dynamics) and t = 2 ps in Fig. 2F (slow dynamics). Unlike the fast dynamics in Fig. 2E, the slow dynamics are dominated by increases in peak intensity over a limited range of scattering vectors (s < 0.45 Å–1) for which electron scattering is known to be particularly sensitive to the valence charge distribution (29, 30). In addition, these slow dynamics are absent for reflections whose reciprocal lattice vector is perpendicular to cR. That is, peaks with a zero first index in the monoclinic system ([hMkMlM] = [0kMlM]) as indicated by gray vertical lines in Fig. 2. This observation establishes that electron structure factors orthogonal to cR are largely unaffected by the slow process. The slow process corresponds to a 1D modification of the electrostatic crystal potential in the octahedrally coordinated vanadium chains oriented along cR.

The two diffraction signatures described above represent qualitatively distinct structural reorganizations within VO2 after photoexcitation. This can be understood by computing the pump-induced changes to the radial pair distribution function (PDF) (31) directly from the observed changes in diffracted intensity shown in Fig. 2, E and F. These curves (Fig. 3, A and B) represent the time-dependent difference in the radial autocorrelation function of the crystal potential with respect to the reference time point. The computed difference PDF for the fast dynamics (Fig. 3A) provides a straightforward structural interpretation for this signal. The positive growing feature (II) corresponds to increased correlation at the R-phase V-V bond length, 2.85 Å, while the adjacent negative features (I, III) represent a reduction at the dimer (2.62 Å) and unpaired (3.16 Å) distances of the M1 phase (Fig. 1). Thus, the fast dynamics correspond to nonthermal melting of the PLD in a fluence-dependent fraction of crystallites. In these crystallites, the vanadium atomic positions relax to their equilibrium R-phase separation on a time scale of 300 fs (26). Extrapolating the linear scaling of these dynamics with fluence indicates that ~43 mJ/cm2 is required to melt the PLD (CDW order) in the entire film. Pump fluences less than ~9 mJ/cm2 are insufficient to initiate this nonthermal SPT in any crystallites, and leave the PLD and M1 crystal structure completely intact. Below this ~9 mJ/cm2 threshold, only the slow dynamics are observed (Fig. 2F and fig. S4). The diffraction signature of these dynamics is identical below and above the SPT threshold, demonstrating that the slow and fast components represent distinct transitions occurring in different crystallites as a result of the heterogeneity of these pulsed laser deposition–grown samples.

Fig. 3 Difference pair distribution functions for fast and slow dynamics.

(A) Difference PDF from –0.5 ps to 1.5 ps, referenced to –0.5 ps. (B) Difference PDF from 2 ps to 10 ps, referenced to 2 ps. The roman numerals correspond to the distances labeled in Fig. 1.

In contrast to the above, the slower dynamics do not correspond to a structural rearrangement of the lattice (which result in a conservation of diffracted intensity like that seen for the fast dynamics in Fig. 2E). The difference PDF (Fig. 3B) for the slow changes is dominated by negative features at 1.3 Å (IV) and 4.4 Å (VI), equal to half the V-V dimer bond length and the undimerized V-V separation plus half the V-V dimer bond length, respectively (Fig. 1A). Positive changes are also observed at around 1.9 Å (V), the average V-O separation in the octahedron, and at <0.8 Å. These observations are consistent with a collective reorganization of valence charge density in the M1 phase that increases the electron density in the vanadium dimer bonds while decreasing the electrostatic potential on primarily the oxygen atoms—an effective modification of the atomic scattering factors.

Previous theoretical work on VO2 has focused on the behavior and occupancy of the three bands formed from hybridized V-3d/O-2p states of t2g symmetry (Fig. 4A) as the determining factor in its electronic properties (10, 1315). Figure 4B shows the orientation of the localized d orbitals from which these bands are formed. The dxy (also referred to as d||) and dxz orbitals mediate σ- and π-type interactions between vanadium atoms along cR, respectively. The dyz orbital is oriented orthogonal to cR. There is broad agreement (1315) that in the high-temperature phase these three bands almost completely overlap at the Fermi level (Fig. 4A, i). This results in roughly equal occupancy in these bands and a nearly isotropic electronic state (21, 22). It has been suggested that the PLD in the M1 phase splits the d|| states into bonding and antibonding combinations sufficiently to open an insulating gap (10), but density functional theory calculations using the local density approximation maintain density of states at the Fermi level in the M1 phase (13, 14), as shown in Fig. 4A, ii (32, 33).

Fig. 4 Effective band diagrams for states of t2g symmetry.

(A) i: Band diagram for the rutile, metallic phase. ii: Band diagram modified as a result of the PLD. iii: The effect of electron-electron correlations as described in (15). UHB and LHB are upper and lower Hubbard bands, respectively. iv: Schematic band diagram indicating partial Mott melting of the dxz band. (B) Illustrations of the dxy, dxz, and dyz molecular orbitals.

Recent work using cluster dynamical mean-field theory points to dynamical electron-electron correlations acting in collaboration with the PLD as being responsible for the insulating properties of the M1 phase (Fig. 4) (14, 15). Our results support this view. We have demonstrated that optical excitation can induce a long-lived state with IR transmissivity like that of the metallic phase (i.e., collapse of the optical band gap to below 0.25 eV) even without melting the CDW order. In this state the PLD of the insulating phase remains intact, but the valence charge distribution is altered. The nature of the changes in charge density can be understood from the symmetry of the changes in diffraction (Fig. 2F), the orbitals (Fig. 4B), and the difference PDF (Fig. 3B). The negative features at 1.3 Å and 4.4 Å in Fig. 3B suggest an increase in the filling of the dxy subshell that contributes most strongly to the CDW along cR. The positive features at 0.8 Å and 1.9 Å suggest reduced filling of the dxz subshell, which reduces charge density on the V and O atoms in the octahedral chains. The dyz states oriented orthogonal to cR, which are understood to be unoccupied in semiconducting VO2 (15, 22), remain unchanged. Thus, optical excitation with fluences below the threshold required to melt the PLD drives a 1D redistribution of occupancy in the dxy and dxz subshells, not a transformation to the isotropic state of the equilibrium metal. Suppression of correlation-induced splitting into upper (UHB) and lower (LHB) Hubbard bands, either preferentially in the dxz band (Fig. 4A, iv) or in both dxz and dxy shells (Fig. 4A, ii), could lead to such a reorganization. The first case represents an optically induced orbital-selective transition with a mixture of localized (dxy) and itinerant (dxz) behavior (34).

The picosecond time scale of this transformation, in addition to its long-lived nature, suggests that the increased vibrational excitation of the lattice due to carrier relaxation (electron-phonon coupling) is a key factor in both inducing and maintaining the reorganization. The nonequilibrium population of excited carriers relaxes within 1 ps in thin VO2 films grown by pulsed laser deposition (27). However, our measurements cannot rule out the possibility of other mechanisms affecting this stability, including kinetic trapping of the valence charge reorganization. Earlier work identified metal-like phases of VO2 with properties distinct from that of the rutile, high-temperature metal. This includes nanoscale correlated metallic domains with unidentified lattice structure that were observed near the transition temperature (35, 36), as well as the observation that the thermally initiated semiconductor-to-metal transition and the SPT can occur noncongruently, suggesting the presence of a metal-like M1 phase of VO2 (37). The correlated metallic state observed in the thermally activated phase transition and the M1 metastable state accessed optically here may be related.

The profound decoupling of the semiconductor-to-metal transition in mid-IR optical properties and the SPT induced through optical excitation indicates that the PLD of the M1 phase is insufficient to fully explain the semiconducting gap. From the perspective of the striking change in electronic properties, the principal role of the PLD is to alter the accessibility of the bands formed by states of dxy symmetry. With the PLD in place, these states are depopulated, and the highest-energy occupied bands have a 1D character and are susceptible to further electronic ordering. The isotropic electronic character of the equilibrium rutile metal (22) cannot be realized with the PLD intact. Finally, the large threshold excitation fluence for the SPT relative to that for the observed electronic reorganization demonstrates that the latent heat of the first-order phase transition at ~340 K is dominated by the SPT rather than by the electronic transition.

Our study shows that UED is able to provide deep insights into the nature of other strongly correlated materials through the disparate concurrent responses of active degrees of freedom in the time domain. Further, our results have relevance to the study of the interplay between valence charge and lattice structure in molecular and materials chemistry.

Supplementary Materials

Materials and Methods

Figs. S1 to S4

References (38, 39)

References and Notes

  1. See supplementary materials on Science Online.
  2. Acknowledgments: Supported by Natural Sciences and Engineering Research Council of Canada (NSERC), the Canada Foundation for Innovation, the Canada Research Chairs program, NSERC PGS-D and CGS-D fellowships (R.P.C. and V.M.), and Fonds de Récherche du Québec–Nature et Technologies. We thank C. Weber for insightful discussion regarding the contemporary theoretical treatment of the electronic structure of VO2.
View Abstract

Stay Connected to Science

Navigate This Article