Ultrafast X-ray imaging of the light-induced phase transition in VO₂

The study investigates how light induces and evolves transient electronic and structural phases in quantum materials, focusing on the insulator-to-metal transition in VO₂. Prior ultrafast studies have shown sub-100 fs disordering of vanadium dimers and rapid bandgap collapse, but whether electronic changes precede or follow structural transitions remains debated. Transient states are expected to be nanoscale heterogeneous due to inhomogeneous excitation and intrinsic material heterogeneity, complicating interpretation by spatially averaged probes. The purpose is to directly image and spectroscopically characterize the transient phase with nanoscale spatial and femtosecond temporal resolution to resolve whether multiple local processes or metastable phases arise, and to clarify the role of heterogeneity in early-time dynamics.

Past work using ultrafast X-ray diffraction and absorption established the basic timeline of VO₂’s photoinduced transition, including sub-100 fs dimer disordering and bandgap collapse. Terahertz conductivity and electron diffraction suggested nucleation-and-growth dynamics on tens to hundreds of picoseconds and proposed a long-lived metastable monoclinic metallic phase persisting up to microseconds. However, this interpretation is contentious, with reports both supporting and refuting a long-lived monoclinic metallic state. Resonant coherent diffraction has inferred domain statistics but lacked real-space images. Grain boundaries are known to pin domains and influence phase coexistence. These prior findings motivate a real-space, spectrally resolved, ultrafast imaging approach to directly test heterogeneity and phase identity in VO₂.

The authors employ time- and energy-resolved coherent resonant soft X-ray imaging at the PAL-XFEL to perform wide-field nanoscale imaging with ~150 fs temporal resolution. Two imaging modes are used: Fourier transform holography (FTH) for rapid single-exposure inversions (but without absolute complex transmission due to a beam block removing the d.c. component) and coherent diffractive imaging (CDI) using multiple exposures and iterative phase retrieval to recover quantitative amplitude/phase and hyperspectral information. Samples are 75-nm VO₂ films on Si₃N₄ membranes; a Cr/Au multilayer mask defines a 2 μm field of view with multiple reference apertures. The pump is 800 nm with fluences up to ~24 mJ/cm² (saturation regime without irreversible domain change at the chosen conditions), and the probe is tuned across the V L₃ and O K edges (notably 517, 529.5, 531.25 eV; hyperspectral CDI at 31 energies across the O K edge at 20 ps). Spatial resolution is sub-50 nm; the XFEL timing jitter limits time resolution to ~150 fs. Imaging sequences alternate positive/negative delays to monitor initial domain stability. Principal component analysis (PCA) decomposes dynamic image stacks to identify the minimal set of spatial-temporal components; significant PCs are identified above a noise threshold defined from negative-delay data. Early-time dynamics (up to 20 ps) are modeled by a double exponential convolved with the instrument response, with time constants shared across datasets but amplitudes allowed to vary. CDI reconstructions of transient states at 20 ps yield quantitative transmission spectra for regions initially M1 vs initially R, enabling differential spectroscopy. Complementary DFT-based XAS simulations (supercell core-hole method in VASP; PAW pseudopotentials; LDA/PBE/PBE+U; polarization ⟂ c_R) model spectral signatures and strain effects, comparing undistorted rutile and an orthorhombically strained rutile structure representing the transient state. Acoustic/thermal propagation estimates connect observed picosecond dynamics with strain wave travel times.

- Early-time dynamics are spatially uniform in time: PCA shows a single principal component suffices up to ~20 ps, indicating all regions share the same temporal response despite initial domain heterogeneity. - Two characteristic timescales describe the transition: a near-resolution-limited drop of 203 ± 18 fs and a slower evolution of 4.98 ± 0.04 ps, both occurring in the same spatial regions. The initial ~200 fs change corresponds to the ultrafast M1→R transition; the picosecond timescale reflects strain/acoustic dynamics rather than distinct phase growth or a separate monoclinic metallic phase. - FTH artifacts can mimic changes in metallic regions due to the missing d.c. component; PCA and CDI clarify that genuine early-time changes originate where the sample was initially M1. - Hyperspectral CDI at 20 ps reveals that regions initially M1 and initially R exhibit very similar metallic spectra; their difference is <1% at 530.5 eV, far smaller than the ~10% M1→R spectral change, confirming complete switching of initially M1 regions to a metallic state. - The transient metallic state is identified as a highly orthorhombically strained rutile metallic phase. Spectral differences relative to equilibrium R are attributed to strain generated during ultrafast transformation without immediate volume expansion; out-of-plane strain relaxes on picosecond timescales via strain waves, consistent with the observed ~5 ps component and estimated ~7 ps acoustic traversal. - Nanoscale heterogeneity and spatially dependent dynamics emerge primarily at hundreds of picoseconds, likely associated with slower in-plane strain relaxation and mesoscopic effects.

The results demonstrate that the photoinduced VO₂ transition proceeds with an ultrafast, spatially synchronous electronic-structural switch to a metallic state across pre-existing domains, contradicting interpretations that invoke concurrent monoclinic metallic phase formation or nucleation-limited growth dominating early times. The single-PC description up to ~20 ps indicates that both the ~200 fs and ~5 ps dynamics occur in the same regions, incompatible with scenarios where different regions host different phases or where nucleation and growth would necessitate multiple PCs. Hyperspectral CDI shows the transient state is metallic and spectrally uniform across regions, with small differences attributable to strain rather than temperature or distinct phases. A strain-wave picture accounts for the picosecond dynamics and their fluence dependence via changes in acoustic velocities and penetration depth. These findings underscore the necessity of spatially and spectrally resolved ultrafast probes to correctly interpret non-equilibrium phases in quantum materials and avoid artifacts from spatial averaging or imaging limitations (e.g., FTH d.c. loss).

This work introduces time- and energy-resolved coherent soft X-ray imaging to capture nanometer-resolved, femtosecond-resolved ‘movies’ of a light-induced phase transition in VO₂. The transition exhibits a rapid (~200 fs) switch to a metallic state followed by a picosecond (~5 ps) evolution explained by strain-wave dynamics, with early-time behavior being spatially uniform despite initial heterogeneity. Hyperspectral CDI identifies the transient metallic state as an orthorhombically strained rutile metal, countering claims of a long-lived monoclinic metallic phase in the early-time window. The approach highlights how combined spatial and spectral resolution is critical for disentangling genuine transient phases from artifacts. Future research could extend these methods to other correlated materials, probe longer times to map in-plane strain relaxation and domain coarsening, explore fluence and thickness dependence to tune acoustic dynamics, and correlate with complementary lattice-sensitive ultrafast probes to fully map coupled electronic-structural evolution.

- FTH imaging removes the d.c. component, introducing correlated artifacts that can mimic changes in metallic regions; CDI mitigates this but requires multiple exposures and reconstructions. - Temporal resolution (~150 fs) is limited by XFEL–laser timing jitter; uncertainties in timing, pump fluctuations, and penetration depth mismatch are not fully included in early-time error estimates. - At higher fluences (>~24 mJ/cm²), irreversible domain pattern changes occur (likely due to occasional high-flux XFEL shots), constraining the usable fluence range and complicating comparisons. - Sample inhomogeneity prevented conventional clustering of hyperspectral data; differential spectra rely on carefully selected regions and noise correlations. - Interpretation of the ~5 ps component as strain/acoustic dynamics is supported by modeling but not by direct lattice parameter measurements during the transient; DFT XAS modeling carries usual approximations (functional choice, broadening, alignment).