VOLUME , NUMBER
P HYS ICAL REVIEW LETTER S
Femtosecond Structural Dynamics in VO2during an Ultrafast Solid-Solid Phase Transition
A. Cavalleri,* Cs. Tóth, C.W. Siders, and J.A. Squier
University of California San Diego, La Jolla, California 92093-0339
Opt-X, Inc., Lake Forest, California 92630
P. Forget and J.C. Kieffer
Université du Québec, INRS énergie et matériaux, 1650 Lionel-Boulet, Varennes, Québec, Canada
(Received 1 March 2001; published )
Femtosecond x-ray and visible pulses were used to probe structural and electronic dynamics during
an optically driven, solid-solid phase transition in VO2. For high interband electronic excitation ??5 3
1021cm23?, a subpicosecond transformation into the high-T, rutile phase of the material is observed,
simultaneous with an insulator-to-metal transition. The fast time scale observed suggests that, in this
regime, the structural transition may not be thermally initiated.
DOI:PACS numbers: 78.47.+p, 71.30.+h, 78.70.Ck
A number of oxides of vanadium exhibit insulator-to-
metal transitions upon heating, most notably V2O3?TM?
150 K? and VO2?TM? 340 K? . In the latter case,
changes in the electronic band structure are associated with
atomic rearrangement between a low-T monoclinic and
a high-T rutile phase. The intriguing nature of this pro-
cess [2,3] and the importance of transition metal oxides in
modern condensed matter physics  make this problem
amenable to dynamic studies using ultrafast techniques.
However, time resolved measurements of phase transitions
in highly correlated materials are generally very demand-
ing, because they require simultaneous access to the fem-
tosecond dynamics of more than 1 degree of freedom of
the system (e.g., electronic, structural, magnetic).
In this Letter, we report on ultrafast optical and x-ray
diffraction  measurementsof aphotoinducedphase tran-
sition in VO2. We demonstrate the first direct measure-
ment of a femtosecond solid-solid phase transition and the
combined measurement of electronic and structural dy-
namics in a correlated solid. For intense interband exci-
tation ??5 3 1021carrierscm23?, we directly observe the
formation of the rutile phase, simultaneous with an insula-
tor to metal transition, on a time scale that is comparable
to or shorter than internal thermalization times of the pho-
toexcited system. Thus, our results challenge the thermal
model for a solid-solid transition in this excitation regime.
This experiment raises several additional issues for future
studies, including the questions of what microscopic pro-
cess is responsible for initiating the structural distortion
and what causes the disappearance of the band gap after
Optical measurements were performed combining
the commonly used pump-probe technique with visible
microscopy, i.e., spatially resolving the pumped/probed
area of the VO2 surface for different time delays.
50 fs, 800-nm, p-polarized optical pump pulse, impinging
at an angle of 60±, was used to excite 200-nm VO2
crystalline films on glass substrates, while a variably
delayed probe pulse provided reflectivity snapshots of the
surface imaged onto a charge-coupled device (CCD). The
peak pump fluence was set to 25 mJ?cm2, significantly
lower than the measured single-shot damage threshold
of 63 mJ?cm2.Figure 1 shows spatially dependent
reflectivity curves, measured along the vertical direction
on the indivsidual images. The laser pulses were spatially
filtered before interaction with the sample, resulting in a
single-transverse-mode, Gaussian spatial profile.
different positions corresponded to different local excita-
tion fluences that were precisely known. As immediately
evident in the plot, the reflectivity of the center drops most
rapidly toward the equilibrium reflectivity of the metallic
phase , reached well within hundreds of femtoseconds.
At a time delay of about 10 ps, a larger area exhibits
the same reflectivity, as parts of the sample that are
pumped at lower fluence switch at a slower rate. The low-
reflectivity area progressively enlarges, reaching a maxi-
mum after approximately 5 ns and slowly returning
toward the optical properties of the low-temperature phase
in several tens of nanoseconds.
observed in a number of optical experiments investigating
phase transformations , above a distinct threshold the
reflectivity settles to the value characteristic of the new
phase, independent of the local excitation fluence. Thus,
a true phase change can be hypothesized, as opposed to
simple carrier excitation, observed at the earliest time de-
lays when the optical properties follow the spatial profile
of the pump laser. A 7-mJ?cm2threshold was obtained
by comparing the measured maximum transformed area
and the pump-fluence profile . Reflectivity evolutions
sampled at three different positions of the photopumped
spot are shown in the respective insets. Exponential fits
to these curves indicate that for increasing pump fluence
the insulator-to-metal transition time decreases from
more than 50 ps to about 100 fs. A sharp decrease in
Similar to what was
-10031-9007?01?()?(4)$15.00© 2001 The American Physical Society-1
VOLUME , NUMBER
P HYS ICAL REVIEW LETTERS
Middle plot: Plots of the reflectivity during the phase trans-
formation. The probe light illuminates a region that is several
millimeters in diameter, in order to provide homogeneous in-
tensity conditions over the photopumped region. The spatially
dependent reflectivity is measured along the vertical direction,
where the nonzero angle of incidence of the pump pulse does
not affect the pump-probe time delay. Lower plots: Time re-
solved evolutions of the reflectivity for three different positions
on the sample, corresponding to local fluences of (a) 7 mJ?cm2,
(b) 15 mJ?cm2, and (c) 25 mJ?cm2. The time scale is loga-
rithmic.The dashed line is the calculated reflectivity for a
200-nm-thick metallic phase on a glass substrate. The continu-
ous curve along the experimental points is a one-parameter, ex-
ponential curve fit for the transition time constants, obtained by
assuming exponential behavior.
Upper plot: Gaussian spatial profile of excitation.
the transition time, from 10 ps to 900 fs, was observed
between 10 and 12 mJ?cm2. We repeated our measure-
ment using 0.5 ps pump pulses of the same fluence and
found no significant variation in the threshold, as well as
a longer minimum transformation time of about 2 ps, in
good agreement with that reported in the literature .
To extract the transformed depth from the optical data,
we calculated the reflectivity of a three-layer structure,
composed by a metallic film of thickness Dx (high T
phase at the surface), a 200 nm-Dx thick insulating film
beneath (low T phase) and a semi-infinite glass substrate.
Comparison of the calculated reflectivity with the mea-
sured value did not provide a unique value for Dx, which
was found to be compatible with 80-nm, 150-nm, or
200-nm transformed depths. Thus, optical measurements
provide no direct information on the structural dynamics
and ambiguous indication on the transformed depth.
Ultrafast structural probing was achieved by using x-ray
bursts of spin-orbit-split 8-keV (1.54-Å) Cu-Ka1 and
Cu-Ka2 line radiation at 20 Hz, generated by focusing
terawatt femtosecond laser pulses onto a moving copper
wire . The probing x rays, emitted into 4p steradians
by the plasma point source, were focused using a pair of
ellipsoidal grazing incidence reflectors in a Kirkpatrick-
Baetz configuration, providing about 2000 Cu-Ka pho-
tons/pulse in a 0.3±cone angle onto a 50-mm spot. The
experiments were conducted on a bulk VO2 sample,
exhibiting the best crystalline quality and maximizing
the diffracted signal. Angle and time dependent diffrac-
tion were measured in an optical pump, x-ray probe
configuration, with excitation limited to the intermediate
fluence range ?15 mJ?cm2?, where no cumulative damage
was observed . The static temperature of the crystal
was held below the transition temperature. The sample
was excited over an area of several millimeters in diameter,
resulting in an x-ray probed region that was homoge-
neously pumped. The diffracted signal was measured
in the (110) direction of the low-T phase using an x-ray
CCD, with the Bragg angle being 13.9±. For calibration
purposes, the unpumped sample was reversibly heated
across the transition temperature (340 K), where a shift
of the Bragg angle to that of the high temperature rutile
phase ?13.78±? was observed.
The measured diffraction profiles were, at negative time
delays, identical to those measured from the unperturbed
sample, thus evidencing no significant cumulative or pre-
pulse effects. At positive time delays, a shoulder origi-
nating from the new crystallographic phase appeared at
about0.1±degrees fromthe centerofthe unperturbed curve
(see Fig. 2a). Because the high-T phase is initially formed
over a small fraction of the depth probed by the x rays
??3 mm?, the peak of the shoulder was observed to be
only a few percent of the main lines. Figure 2b displays
normalized diffraction curves at early times, obtained by
dividing the measured time resolved signals by those from
the unperturbed sample. As pointed out before, the curves
at negative time delays resemble those from the unpumped
case, whereas after a few hundred femtoseconds a signifi-
cant should is visible at the diffraction angle of the equi-
librium, high-temperature rutile phase. At time delays of
several picoseconds (not shown), a corresponding feature
was observed at higher diffraction angles, due to a com-
pressive response ofthe low-temperature monoclinic phase
to the expansive transformation taking place at the surface.
While no detail of the actual atomic dynamics during the
growth of the metallic phase can be retrieved at this stage,
the formation time at the very surface unequivocally ap-
pears to be of the order of a few hundred femtoseconds.
Figure 3 shows the integrated x-ray reflectivity from
the new phase, normalized to the signal from the low-
temperature monoclinic crystal. During the first 10 ps,
the step in the integrated diffraction from the rutile phase
VOLUME , NUMBER
P HYS ICAL REVIEW LETTERS
nal measured for negative and positive time delays. The diffrac-
tion profiles originate from the convolution between the rocking
curve of the dynamically strained/transforming crystal and the
spectrum of the x-ray plasma source. The dashed and continuous
curves correspond to delays of 2300 fs and 11 ps, respectively.
The two crystalline structures (monoclinic insulating phase and
tetragonal metallic phase) are sketched by displaying only the
vanadium atoms, which dominate the diffraction signal. Lower
plot (b): Ratio between the x-ray reflectivity of the excited and
Upper plot (a): Angle dependent x-ray diffraction sig-
reaches about 8% of the signal from the main peak. Since
the scattering factors for these two phases are approxi-
mately equal, the ratio of the two diffraction signals di-
rectly yieldsthe thickness ofthetransformed layerfrom the
3-mm x-ray penetration depth. This resolves the ambiguity
on the transformation depth, indicating that for this inter-
mediate fluence range ?15 mJ?cm2?, a 250-nm thick layer
of material undergoes transformation in about 10 ps. The
inset displays the signal measured during the first few pi-
coseconds, demonstratingthatthe first step in the structural
phase transition occurs on this time scale over a depth of
approximately 40–60 nm (consistent with an 80-nm-thick
layer, as extracted from the observed transition in the opti-
cal properties). At long time delays (hundreds of picosec-
onds, not shown), slow oscillations were observed around
the 8% value, probably due to coherent acoustic response
[12,13] initiated by the ultrafast transformation. However,
analysis of the response at longer times requires a full ther-
moelastic treatment and is beyond the scope of this paper.
While we were not able to perform ultrafast x-ray mea-
nal from the metallic rutile phase normalized to the integrated
diffraction from the monoclinic phase. The integrals are calcu-
lated over a region of 0.1±around the center of the respective
lines. Inset: Region near zero time delay. Continuous curve:
exponential fit of the rise time.
Main plot: Time-dependent, integrated diffraction sig-
surements in the highest fluence regime (nonreversible),
we believe that our data cover the essential features of the
structural phase transition. Higher excitation energies are
likely to result in similar dynamics and in a thicker layer
being transformed in the subpicosecond time domain.
The measurements reported here point toward two dif-
ferent physical mechanisms for the phase transition. Close
to threshold, the process requires tens to hundreds of pi-
coseconds before completion. A thermal pathway governs
the transformation, with the lattice being heated in sev-
eral picoseconds above the critical temperature (340 K)
and growth of the new phase proceeding incoherently and
at spatially separated sites by statistical transitions across
an activation barrier (nucleation and growth). This exci-
tation regime corresponds to the conventional pathway for
the first order phase transition and is not the main focus of
At higher degree of electronic excitation ??5 3
1021cm23?, optical and x-ray data point toward a transi-
tion occurring over a macroscopic volume within 500 fs or
less . This time scale is comparable to that for energy
transfer to the lattice, and it is shorter than the typical
internal thermalization time of a nonequilibrium phonon
distribution [15,16]. Excitation of a dense carrier popula-
tion across the 600-meV band gap  may significantly
perturb the potential energy surface of the electronic
ground state and depress the barrier separating the two
phases.This process could, for example, result from
weakeningof the covalent bonds, similar to the mechanism
VOLUME , NUMBER Download full-text
P HYS ICAL REVIEW LETTERS
governing ultrafast melting [8,18]. Alternatively, direct
laser excitation of coherent lattice displacements [19,20]
or excitation of hot, nonequilibrium phonons may drive
the phase transition along an unperturbed potential energy
surface .The current experimental data cannot
resolve this issue. Further, it is not clear whether the
system becomes metallic from disruption of electronic
correlations or from structural distortion. This issue is
intimately related to the nature of the insulating phase
of VO2 and represents an important question for future
experimental studies [2,3]. In these respects, theoretical
work is required to address the nature of the highly excited
state of insulating VO2 as well as the ensuing dynamics.
Experimentally, future advancements in the ultrafast x-ray
measurements will allow for improved signal to noise
ratio, as well as for the simultaneous detection of several
diffraction orders  and retrieval of atomic positions at
the early stages of the transition.
In summary, we have reported the first conjunct op-
tical and x-ray measurements during an optically driven
solid-solid phase transition in the correlated oxide VO2.
While a variety of characteristic time scales are found,
laser pulses of sufficiently high fluence can trigger a subpi-
cosecond structural transition from the low-T monoclinic
phase to the high-T rutile phase. Our results are suggestive
of a nonequilibrium pathway between the two equilibrium
Because the reported solid-solid transformation lasts
only a few hundred femtoseconds and is reversible, it may
be useful for an ultrafast Bragg switch , capable of
selecting a subpicosecond portion of a longer x-ray pulse
(e.g., from a synchrotron). Finally, as opposed to previous
studies of nonthermal order-disorder phase transitions, this
is the first experiment where the product state of a trans-
formation is structurally identified, as opposed to indirect
signatures such as loss of scattering efficiency. This ca-
pability may have ramifications in femtosecond dynamics
of condensed matter, such as the possibility of achieving
coherent control of phase transitions by adjusting the ex-
citation conditions to maximize a specific product state.
The authors are indebted to K. Sokolowski-Tinten and
R. Merlin for enlightening discussions and suggestions.
This work was partially supported by the National Science
Foundation through Grant No. INT-9981720.
Email address: email@example.com
 F. Morin, Phys. Rev. Lett. 3, 34 (1959).
 R.M. Wentzcovic, W.W. Schultz, and P.B. Allen, Phys.
Rev. Lett. 72, 3389 (1994).
 A. Zylberstein and N. Mott, Phys. Rev. B 11, 4383 (1975).
 M. Imada, A. Fujimori, and Y. Tokura, Rev. Mod. Phys.
70, 1034 (1998).
 C. Rischel et al., Nature (London) 390, 490 (1997); A.H.
Chin et al., Phys. Rev. Lett. 83, 336 (1999).
 The equilibrium indexes of refraction n 1 ik are, at
800 nm wavelength, 2.9 1 i0.5 and 2.3 1 i0.72, for the
semiconducting and metallic phases, respectively.
 K. Sokolowski-Tinten et al., Phys. Rev. Lett. 81, 224
(1998); K. Sokolowski-Tinten et al., Phys. Rev. B 58,
 For the sake of convenience, the threshold is here defined as
the lowest fluence that is necessary to induce the transition
over the whole 200-nm layer. The threshold for a thinner
layer of material is likely to be lower than 7 mJ?cm2.
 M.F. Becker, A.B. Buckman, R.M. Walsh, T. Lépine,
P. Georges, and A. Brun, Appl. Phys. Lett. 65, 1507 (1994);
J. Appl. Phys. 79, 2404 (1996).
 C. Rose Petruck et al., Nature (London) 398, 310 (1999).
 The size of the VO2bulk crystals prevented sample transla-
tion and constrained the pump fluence to values that did not
cause cumulative alteration, either visible or in the diffrac-
 A.M. Lindenberg et al., Phys. Rev. Lett. 84, 111 (2000).
 A. Cavalleri et al., Phys. Rev. Lett. 85, 586 (2000).
 Because the x-ray pulse duration has not been measured
independently, the observed 500 fs rise time could be in-
 D. von der Linde et al., Phys. Rev. Lett. 44, 1505 (1980).
 Subsequent thermalization of the system raises the equilib-
rium temperature above the critical temperature and stabi-
lizes the phase transition into the metallic phase. Retrans-
formation into the low-T phase follows thermally within
tens of nanoseconds, caused by heat diffusion into the un-
 S. Shin et al., Phys. Rev. B 41, 4993 (1990).
 C.V. Shank, R. Yen, and C. Hirliman, Phys. Rev. Lett.
50, 454 (1983); P. Stampfli and K.H. Bennemann, Phys.
Rev. B 49, 7299 (1994); P. Saeta et al., Phys. Rev. Lett.
67, 1023 (1991); K. Sokolowski-Tinten, J. Bialkowski, and
D. von der Linde, Phys. Rev. B 51, 14186 (1995).
 R. Merlin, Solid State Commun. 102, 207 (1997).
 H. Zeiger et al., Phys. Rev. B 45, 768 (1992).
 P. Daugherty, G.P. Wiederrecht, K.A. Nelson, M.H. Gar-
ret, H.P. Jensen, and C. Warde, Science 258, 770 (1992).
 S. Techert, F. Schotte, and M. Wulff, Phys. Rev. Lett. 86,
 P.H. Bucksbaum and R.D. Merlin, Solid State Commun.
111, 535 (1999).