Nanoscale probing of electronic band gap and ...

JOURNAL OF APPLIED PHYSICS 109, 024311 共2011兲

Nanoscale probing of electronic band gap and topography of VO2 thin film surfaces by scanning tunneling microscopy W. Yin,1 S. Wolf,1 C. Ko,2 S. Ramanathan,2 and P. Reinke1,a兲 1

Department of Materials Science and Engineering, University of Virginia, 395 McCormick Road, Charlottesville, Virginia 22904, USA 2 School of Engineering and Applied Science, Harvard University, 29 Oxford St., Cambridge, Massachusetts 02138, USA

共Received 28 September 2010; accepted 20 November 2010; published online 21 January 2011兲 The metal-insulator transition 共MIT兲 in vanadium dioxide in the vicinity of room temperature makes it one of the most interesting materials for novel switching device applications. It is therefore essential to have a fundamental understanding of the VO2 surface when it is incorporated into multilayer structures or nanodevices. This study focuses on the surface modification of VO2 in response to the thermal treatment during phase transition. Vacuum annealing at temperatures in the vicinity of the MIT triggers a partial reduction in the surface, and thus initiates a chemical phase transition. Scanning tunneling microscopy and spectroscopy are used to investigate the electronic properties and surface structure of the VO2 thin film on 共0001兲 sapphire substrates. Band gap maps with a high spatial resolution and single point spectroscopy I-V curves are measured as the sample is cycled through the MIT, and thus provide a direct observation of the surface phase transition at the nanoscale. The VO2 surface exhibits a homogeneous insulating behavior with a typical band gap of ⬃0.5 eV at room temperature, and the surface becomes more metallic and spatially inhomogeneous in conductivity during MIT, and wide range of surface oxides can be identified. The surface still remains partially metallic after cooling down from a long period anneal, and such irreversible surface electrical change is attributed to the loss of oxygen. The location of metallic islands after thermal cycling is strongly coupled to the topography of the film, and relaxation processes and continued modification of the spatial distribution of the metallic regions are recognized on a longer timescale. The impact of film morphology, strain, surface chemistry, and structural phase transition on the electronic characteristics of VO2 surfaces are discussed. © 2011 American Institute of Physics. 关doi:10.1063/1.3528167兴 I. INTRODUCTION

The metal-insulator transition 共MIT兲 in vanadium dioxide has been described more than 50 years ago but the nature of this phase transition is still debated. Vanadium dioxide VO2 is semiconducting at room temperature 共monoclinic lattice兲 and undergoes a first order phase transition at TC = 341 K to a metallic phase with a tetragonal lattice.1–15 This phase transition is of considerable interest for a wide range of applications such as, for example, switchable tunnel barriers, and sensors. However, for most applications the use of bulk single crystalline material is not feasible and VO2 thin films on a variety of substrates are used. The stability of the surface, and the reproducibility of the MIT play a particularly important role in sensing applications, or in the use of ultrathin VO2 layers and coatings.16,17 The MIT in thin films is in contrast to bulk single crystal samples, influenced by the complex interplay between strain fields imposed by intergrain interactions and the substratefilm interface. These local strain fields, which arise even in films exhibiting no net strain, are known to depress the MIT.18–21 The ability of vanadium to adopt a wide range of stable oxidation states further complicates our understanding of the MIT.22–24 Stable oxides are formed with V+5 to V+2 a兲

Electronic mail: [email protected].

0021-8979/2011/109共2兲/024311/8/$30.00

ions, and a wide range of mixed oxides such as the Wadley phases 共V+5 and V+4兲, and Magnéli phases 共V+4 and V+3兲 are known. The reduction of the V-ion has been observed previously and will be labeled here as a chemical phase transition in order to distinguish it from the structural phase transition described above. The interplay between these two types of phase transition plays a central role in the interpretation of our results. Our study focuses specifically on the surface changes in VO2 thin films, which are initiated during the annealing of the sample across the MIT. The electronic changes in the surface are observed with scanning tunneling microscopy 共STM兲 and scanning tunneling spectroscopy 共STS兲.25,26 We record the modification of surface topography, band gap, and conductivity distributions with nanometer resolution as a function of sample temperature. The spatial distribution and contributions of metallic and semiconducting regions at the surface is driven by the interplay between the MIT of VO2, and the chemical phase transitions, which are controlled by the loss of surface oxygen. The long-term relaxation 共one week兲 of the surface is recorded, and shows substantial condensation of metallic regions. II. EXPERIMENTAL

VO2 thin films were grown on 共0001兲 Al2O3 single crystal substrates by reactive dc sputtering from a V target. The 109, 024311-1

© 2011 American Institute of Physics

Downloaded 30 May 2011 to 137.54.26.233. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

024311-2

Yin et al.

J. Appl. Phys. 109, 024311 共2011兲

FIG. 1. 共Color online兲 Topography images of a surface section, which is representative of the VO2 thin film morphology. A bias voltage of 0.2 V is used to display the metallic regions at room temperature 关296 K, 共a兲兴, and on hot samples at 325 K 共b兲 and 360 K 共c兲. The dramatic change in the apparent brightness of the images is due to the modification of the LDOS as the VO2 surface undergoes a phase transition from insulating to metallic.

dc source power was set at 250 W and the chamber pressure was maintained at 10 mTorr by flowing 9.5 SCCM 共SCCM denotes cubic centimeter per minute at STP兲 oxygen and 90.5 SCCM argon gas mixture during synthesis. The substrates were preheated at 550 ° C for 30 min. The film thickness was estimated to be ⬃400 nm by cross-sectional scanning electron microscopy. Dc-sputtered VO2 films show a sharp conductivity jump across the insulator-to-metal transition with four-order magnitude change in resistivity. The ratio of electrical resistance at 25 ° C to that at 100 ° C was measured to be 4.75⫻ 104共⍀ / ⍀兲 on as-grown VO2 films. The details of electrical transport measurement can be found elsewhere.7,27 The STM analysis was conducted in an Omicron variable temperature ultrahigh vacuum 共UHV兲 scanning probe microscope with a base pressure of 10−10 mbar. The VO2 samples are introduced from air into the UHV chamber and heated to 373 K for an hour to remove the surface contaminants.28 The samples were mounted on a sample holder with an integrated resistive heater unit, which was used for the in situ STM-studies at elevated temperatures. All measurements were performed with a chemically etched tungsten tip. Bias voltages between 0.2 and 1.0 V and a feedback current 0.3 nA was used for topography imaging. The topography images and spectroscopy data were analyzed using the Scala 共Omicron Nanotechnology SPM兲 and WSXM software.29 The VO2 samples were heated from room temperature, and stabilized for about 1.5 h at 325 and 360 K 共first sample兲, and at 339 and 349 K 共second sample兲 for a total of 6 h, respectively. The complete heating process requires a substantial amount of time in order to minimize thermal drift during imaging. The determination of a complete band gap map is time consuming, and in order to limit sample degradation during measurement, we recorded a smaller number of complete band gap maps during the annealing but were able to probe several positions on the sample after cooldown. The samples were stored in UHV and measured again at room temperature 12 h and one week after finalizing the annealing experiment. III. RESULTS

The transformation of the VO2 surface can already be observed in the topography images, which are shown in Figs. 1共a兲–1共c兲 as a function of temperature. The thermal drift dur-

FIG. 2. 共Color online兲 Representative STS curves for metallic to large band gap 共⬎0.4 eV兲 regions. The I-V characteristics for the metallic and large band gap regions are shown on the right hand side. The four dI/dV characteristics, which are representative of the LDOS, correspond to the four band gap regimes subsequently used for the band gap maps shown in Figs. 3–7.

ing the heating of the sample was corrected manually in order to image the same group of crystallites repeatedly. The images are recorded with a bias voltage of 0.2 V and the electronic states close to the Fermi energy dominate the tunneling current, which can be seen in the spectroscopy data summarized in Fig. 2. The metallic regions of the surface, where the local density of states 共LDOS兲 close to the Fermi energy, EF, dominate, therefore appear brighter in the topography images, and the progression from a predominantly insulating to a metallic surface is clearly seen in the evolution of the images in Fig. 1. The changes in the topography images, which were recorded as a function of temperature with a bias voltage of 1.0 V are far less prominent due to the only small differences in LDOS between metallic and insulating regions deeper in the band. These images then indeed reflect the topography of the thin film. The voltage dependent topography images gave a first impression of the insulator-metal transition at the surface, and a more quantitative assessment of the contribution from different vanadium oxide phases was obtained with STS. The surface density of states and their spatial distribution was studied by measuring the I-V characteristics in the voltage interval from ⫺1 to 1 V by single point STS 共averaging over 20 times of I-V scan in each spot兲 and grid measurements. The grid measurement consisted of a matrix of 100 ⫻ 100 pixels and I-V curves were taken at each pixel. The band gap maps shown here have a pixel size of 2.52 nm2, with the exception of Fig. 4共b兲 where the pixel size is 102 nm2. Figure 2 shows four representative examples of I-V and dI/dV characteristics, which are associated with different band gap materials/regions at the surface. The color-code indicated in the next sentence is used in the presentation of the band gap maps in Figs. 4–6. The I-V characteristics are measured at selected points within an image; 共a兲 metallic 共blue兲: showing an Ohmic behavior with an almost linear shape, and a relatively large tunneling current at V = 0 共dI/ dV⬎ 0.05 nA/ V兲; 共b兲 very small band gap 共yellow兲: less than 0.2 eV or with a very small current at V = 0, and the shape of the I-V curves deviates slightly from the Ohmic behavior of the metal; 共c兲 small band gap 共white兲: between


024311-3

Yin et al.

0.2 and 0.4 eV; 共d兲 large band gap 共red兲: larger than 0.4 eV. The room temperature, semiconducting phase of VO2 is the only oxide with an oxidation state of +4 which possesses a band gap larger than 0.4 eV and it is depicted as “red” in the band gap maps. All dI/dV curves are slightly asymmetric, with a Fermi energy 共V = 0兲 positioned above midgap. A defect-doping of the VO2 and contributions from a tipinduced band bending can be the origin for this asymmetry.30,31 The band gaps are determined from the extension of the flat section at the center of the dI/dV characteristics, and the spatial distribution of band gaps can be visualized by setting a threshold tunneling current at a set voltage in the dI/dV curves from the grid measurement. Band gaps determined from the flat section are less prone to measurement errors in noisy data sets but are also systematically smaller by 0.1–0.3 eV than if one uses the inflection point method. The tunneling threshold current at a set voltage is uniquely linked to the band gap, which has been confirmed in a large number of I-V characteristics. Setting the correct threshold current then yields images, which only show the respective band gap regime. These images are then overlaid to visualize the entirety of the band gap map across all band gap regimes. The error in this analysis is about 0.2 eV from the noise in the I-V curves and a conservative estimate of the reproducibility of the measurement in each spot. The spatial distribution of band gaps and conductivity maps 共dI/dV at V = 0兲 are obtained from the grid measurements, and the band gap maps are generated using the four band gap regimes defined in the single point I-V characteristics shown in Fig. 2. Figures 3共a兲–3共c兲 show the spatial distribution of band gaps, topography and conductivity map of the same VO2 surface region at room temperature before the in situ heating measurement. The 250⫻ 250 nm2 topographic image of the VO2 surface shows small roundish grains with an average size of about 15 nm. In the band gap map 关Fig. 3共a兲兴 the dark gray pixels mark the large band gap areas 共BG⬎ 0.4 eV兲 which corresponds to the insulating phase of VO2. The metallic phase is denoted by white pixels, and band gap values in between these two extremes are shown in medium gray. The large band gap contributions clearly dominate and amount to 78% of the total area of the image. The conductivity map is presented in a continuous color scale, where the darkest gray denotes regions of lowest conductivity, which dominate the image in agreement with the band gap map. Increasingly lighter shades of gray correspond to increased conductivity and the metallic areas 共white兲 are few, and dispersed within a poorly conducting matrix. The metallic regions in the band gap map are reflected as bright spots of enhanced conductivity in the conductivity map. The pixels where no high quality I-V curve could be obtained are marked in black. These areas are often regions with a steep slope in the topography. The band gap for a representative dI/dV characteristic 关Fig. 3共d兲兴 is about 0.5 eV, and thus consistent with reported values for the VO2 band gap.10,32 The surface is dominated by VO2 and contributions from metallic regions are small; a percolation path is not present. The band gap distribution is only weakly correlated with the topography, a slight increase in small-band gap material is seen

J. Appl. Phys. 109, 024311 共2011兲

FIG. 3. 共Color online兲 This figure shows the VO2 surface before the annealing experiment but after the short annealing cycle used for sample cleaning. 共a兲 Band gap map: dark gray - band gap ⬎0.4 eV, white - metallic, and intermediate gray for all other band gaps. 共b兲 Conductivity map: derivative of the I-V curves at a bias voltage of 0 V. The darkest gray corresponds to the lowest conductivity, white denotes metallic conductivity. 共c兲 STM topography image 共1 V, 0.3 nA兲, and 共d兲 a representative dI/dV characteristic for a VO2 surface spot with a band gap of ⬃0.5 eV. For image pixels marked in black the quality of the I-V curves was insufficient to determine the band gap.

around some grain boundaries. It should be noted that this surface has already undergone a short heating cycle to temperatures above the MIT in order to remove surface contamination. The insulating VO2 surface was recovered very well, and the surface has not experienced any irreversible changes. Figure 4共a兲 shows the band gap map after heating the sample to 349 K, which is just above the MIT transition. The characteristic I/V curves for each of the band gap regimes are summarized in Fig. 2. The colors in the band gap maps are assigned as following: a兲 metallic - blue, b兲 very small band gap - yellow, c兲 small band gap - white, and d兲 large band gap - red. Examples for the dI/dV characteristics, which correspond to each of these band gap regimes are shown in Fig. 2. A substantial section of the sample has transformed to a metallic phase, and the contribution from the wide band gap VO2 phase is greatly diminished. The contributions from different phases measured after each annealing and cooling step is summarized in Fig. 7. However, sizeable contributions from intermediate band gap materials are readily apparent. The metallic phase in some areas accumulates to larger islands, with few nonmetallic contributions, and is in other parts of the sample randomly distributed, which is reminiscent of the “random metallic clusters” described by Chang et al.32 The surface is electronically highly inhomogeneous on a short length scale, and a comparison with the topographic image shows that some insulating regions are located preferentially close to grain boundaries. The overall surface conductivity at 349 K has therefore changed dramatically com-


024311-4

Yin et al.

J. Appl. Phys. 109, 024311 共2011兲

FIG. 4. 共Color兲 Band gap maps recorded for hot samples at temperatures close to 共339 K兲 and above 共349 K兲 the structural phase transition of VO2. 共a兲 shows a high resolution band gap map at 349 K, and 共b兲 are two band gap maps of the same sample area measured at 339 and 349 K. The image size and therefore also pixel size is larger than in 共a兲, which leads to an increased number of poor or highly asymmetric I-V characteristics 共black pixels兲.

pared to room temperature 关Fig. 3共b兲兴. The growth of the metallic phase is illustrated in more detail in Fig. 4共b兲; two band gap maps for 339 and 349 K are shown, which cover the same area of 300⫻ 600 nm with an STS pixel size of 10⫻ 10 nm2. The 339 K measurement was performed before the sample temperature was raised to 349 K. A metallic island has nucleated in the upper right corner of the image at 339 K, and it expands as the temperature is increased to 349 K. Subsequent to the annealing experiment, where the sample was held at elevated temperatures for several hours, we performed more extensive studies on the recovery of the VO2 insulating surface after the cool down. Figure 5 shows the band gap maps recorded at two different locations on the sample surface, which were imaged 12 h after cooling-down 关in total three different regions of the sample were investigated, two of them correspond in appearance to Fig. 5共a兲兴. The sample surface was quite homogenous prior to heating, with VO2 being the dominant contribution but during annealing and after cooldown this homogeneity is lost.

FIG. 5. 共Color兲 Band gap maps and topography images recorded 12 h after cooldown of the sample. The band gap maps were taken at two different areas of the sample, and the metallic contribution in 共a兲 is 17%, and in 共b兲 61%. The topography image for band gap map 共b兲 illustrates the correlation between surface morphology and the extension of the metallic islands.

The surface does not completely revert back to the semiconducting phase of VO2 but contains sizeable amounts of metallic and small band gap regions. In some areas of the sample surface the different phases are arranged nearly randomly as seen in Fig. 5共a兲, in other areas the metallic regions form well defined islands whose extension is closely related to the surface topography. This can be seen quite well in Fig. 5共b兲: some of the crystallites exhibit a metallic surface while others are dominated by semiconductor phases. The correlation between topography and electronic characteristics of the surface is even more pronounced for band gap maps recorded a week after the annealing experiment 共Fig. 6兲. The metallic phase has condensed into islands, which are tied to specific crystallites. The somewhat random distribution of different phases, which still dominated the images recorded shortly after the cooldown, is only seen in small sections of the images. The long-term relaxation of the surface was one of the most surprising results of our study, and will impact strategies in the use of VO2 thin films. The original, predominantly semiconducting nature of the surface is not recovered after an extended annealing cycle, and the surface continues to change on a relatively long time scale. A previous study28 of VO2 thin films showed that the surface is modified early in the annealing process, which is partially due to the fact that these processes are performed far away from thermodynamic equilibrium and a chemical transformation and reduction of the oxide is likely. The repeated low-temperature cycling through the MIT does not initiate significant changes in the resistivity 共R兲 as a function of temperature, which represents an integral measure of resistivity for the entire film. The R-T curves for the VO2 thin films, which are equivalent to the ones used in the present work, are discussed in detail in a previous publication by Ko and Ramanathan.7 However, annealing at temperature exceeding 100 ° C indeed modifies the bulk material


024311-5

J. Appl. Phys. 109, 024311 共2011兲

Yin et al.

FIG. 7. 共Color online兲 Summary of the average contributions of different band gap regions in all band gap maps measured during the annealing experiment and after the cooldown. After the cooldown, three band gap maps were measured at different positions on the sample surface and the metallic contributions were 15%, 17%, and 61% 共12 h after cooldown兲, and 29%, 36%, and 32% 共one week after cooldown兲. The variability in all other contributions was below 5% with the exception of the 0.2–0.4 eV region 共12 h after cooldown兲, which shows a disproportionally small value of 14% and mirrors the high metallic contributions in the same band gap map.

IV. DISCUSSION

FIG. 6. 共Color兲 Band gap maps and topography images recorded one week after the annealing experiment was finalized.

and consequently leads to a change in the R-T behavior.28 The modification in the material properties begins at the surface and propagates throughout the entire film if the annealing process is continued. Figure 7 summarizes the overall contributions from the different phases for all four treatment steps. The rapid reduction in the semiconducting VO2 after the first heating step is evident, and the overall surface composition remains fairly stable in the hot sample, and after the cool down. The metallic contributions are between 30% and 40%, the largest band gap contributions drop to around 20%. The variability in the metallic contribution after the cool down is relatively large between images, and the metallic contribution can change quite dramatically if a large metallic island dominates the image frame 关Fig. 5共b兲兴. In summary, the relative contributions of the different band gap regions remain fairly constant after the initial annealing step but the spatial distribution is modified considerably during the extended timeframe of the experiment after cooldown.

The interpretation of the surface transformation requires the assignment of vanadium oxide phases to the band gap regimes, which were defined at the beginning of the results section 共Fig. 2兲. The vanadium-oxygen phase diagram is complex and includes four oxides with a single oxidation state, from V2O5共+5兲, VO2共+4兲, V2O3共+3兲, and VO共+2兲, and a large number of mixed oxides including the Wadley and Magnéli phases. VO is the least stable of these oxides and can sustain a large concentration of metal or oxygen vacancies, which leads to substantial variations in its conductivity and presumably also band gap. The few available data indicate a band gap around 0.2 eV for VOx 共x is close to 1兲.14,24,33–35 Nearly all vanadium oxide compounds undergo metal-insulator transitions, with VO2 having the highest transition temperature. The electronic state, metallic, or insulating, of the oxides of relevance in our study, and the respective transition temperatures are summarized in Table I. V2O5 is not present in our samples but was included to complete the data set. At room temperature only two oxide phases, VO2 and VOx, are insulating. All other oxide phases have already undergone a phase transition and are metallic. The surface indeed shows a dominant contribution from the insulating phase of VO2 prior to heating, which is the only wide-band gap phase at room temperature. After heating across the MIT for VO2, the surface should be completely metallic if VO2 remained the dominant oxide phase. However, while the metallic contributions are considerable and amount to nearly half of the surface, a wide range of different band gap regions can now be identified. Even a small percentage of semiconducting VO2 is still present, probably due to an incomplete transition. The MIT follows a hysteresislike behav-


024311-6

J. Appl. Phys. 109, 024311 共2011兲

Yin et al.

TABLE I. Summary of vanadium oxide phases and their respective MIT temperatures 共TC兲. Columns 3 and 4 list whether these phases are insulating 共I兲 or metallic 共M兲 at room temperature, and above the critical temperature for VO2.

Composition V 2O 5 VO2 V6O11 V 4O 7 V 5O 9 V 2O 3 VOx V

Oxidation state +5 +4 +3.6 +3.5 +3.3 +3 ⱕ+2 0

T ⬎ 341 K 共TC of VO2兲

Room temperature

M M M M

I I 共TC177 共TC250 共TC139 共TC168 Ib M

K兲 K兲 K兲 K兲

a a a a

I M M M M M I M

a

All these transitions proceed from insulator to metal at the critical temperatures indicated in the table. These oxides are metallic at the temperature relevant to our experiment. b The electronic properties of VOx depends strongly on the stoichiometry with a concurrent variation in conductivity and band gap.

ior and small pockets of the insulating phase can still be preserved even after transition through the MIT. The prevalence of regions with a band gap below 0.4 eV can be attributed to reduced vanadium oxides, which are formed due to loss of oxygen from the surface and the buildup of a large concentration of oxygen vacancies. The band gap of VOx agrees well with the band gap regimes below 0.4 eV but intermediate oxides and Magnéli phases would contribute to the metallic surface regions. However, it is likely that highly nonstoichiometric compounds with a large density of oxygen vacancies cannot be described with the properties known for bulk phases 共Table I兲. It has also been shown with STS measurements that the surface of a V2O3 thin film exhibits a variation in the local electronic structure on the nanoscale.36,37 These studies illustrates that the well-known electronic characteristics of the bulk material are not always reflected in the surface properties, and that local defects can play a critical role. The defect concentrations, which are required to suppress the MITs for different vanadium oxide phases are not known, and this would indeed be a very interesting area of research. We can assume that the MIT will not occur in highly nonstoichiometric, defect rich compounds, which contribute then to the regions with a band gap below 0.4 eV rather than the metallic ones. However, only metallic VO2 regions can switch back to the insulating VO2 phase with a band gap ⬎0.4 eV. The concentration of VO2 increases after cooldown by about 10%, therefore up to 20% of the metallic region 共above MIT兲 might be attributed to other oxides 共Fig. 7兲. All regions with band gaps below 0.4 eV are neither VO2 nor do they correspond to one of the well-defined bulk oxides included in Table I with the exception of VOx. A chemical phase transition, which is characterized by the loss of oxygen, therefore coexists with the structural phase transition 共MIT兲 of VO2. The relative contribution from the chemical phase transition as compared to the structural MIT depends on the annealing time of the sample. The spatial distribution of the different phases can be

discussed using Fig. 4, which shows the sample surface at temperatures close to or above the VO2 transition temperature. Even at 339 K 关Fig. 4共b兲兴 contributions from the insulating phase of VO2 are small and the structural transformation is likely finalized leaving only few, small pockets of insulating VO2 共red pixels兲. The correlation with topography is weak, and metallic regions appear not to be tied to specific crystallites. The spatial distribution of the metallic phase agrees with the model of a first order phase transition, which proceeds through the nucleation of the new phase. The larger metallic islands of rutile VO2 might therefore form in the initial stages of the transition, and continue to expand into the low-temperature VO2 phase 共MIT transition from the rutile to monoclinic phase兲. Simultaneously some sections of the sample surface begin to loose oxygen, and the smallband gap phases begin to dominate the surface 共chemical phase transition兲. Overall, these reactions lead to the somewhat random distribution of metallic pixels, which serve as percolation pathways. Chang et al.32 observed the island nucleation and percolation of the low temperature phase of VO2 by performing STS during the sample cooling. The progression of the surface transformation is similar to our observations, and their failure to use the two-dimensional percolation model to describe surface conductivity might be due to an admixture of a chemical phase transition at the surface. The progression of the chemical phase transition through the loss of oxygen and reduction in vanadium can also be seen in Fig. 4共b兲, where a metallic island grows at the expense of a small-band gap region. The transformation is much more extensive in the areas adjacent to the island than in regions with a random distribution of band gaps like in the lower left corner. The metallic islands might nucleate during the structural phase transition but they also appear to contribute to the rapid chemical transformation of the surface which progresses more rapidly along the island perimeter. This analysis clearly illustrates the degradation of the surface during heating, and confirms the detrimental impact of oxygen loss and surface reduction. The annealing time is a critical parameter since the short anneal used to clean the sample did not destroy the surface but allowed a nearly complete recovery of the insulating VO2 surface. The rapid surface degradation, and chemical phase transition is in contrast to the stability of the bulk, which can be cycled through the MIT repeatedly without detriment to performance. It is expected that the stability of ultrathin films and coatings however is limited by the chemical phase transition, and recovery of VO2 in well-defined oxygen environments might be required. The spatial distribution of band gap changes considerably after the sample is cooled down: the correlation of metallic phase islands with the topography is greatly enhanced, most metallic islands visible in Figs. 5 and 6 correspond to single crystallites in the topography image. This trend is more pronounced one week after cooldown and points to a considerable modification of the surface properties on a long timescale. The segregation of the metallic and non-metallic regions can at room temperature only be driven by the redistribution of surface oxygen. It should be noted that only the


024311-7

J. Appl. Phys. 109, 024311 共2011兲

Yin et al.

metallic regions aggregate to continuous islands while the other band gap regimes remain dispersed. Since the VO2 is in the insulating state, only oxide included in Table I or fully reduced V can contribute to metallic islands, all other phases are in the small-band gap regime. The driving force for oxygen diffusion is related to a local gradient in chemical potential created by the coexistence of a wide range of oxide phases. However, the segregation of the metallic phase is also tied to the topography, which indicates that not only the chemical diversity of the surface but also local strain fields might play a substantial role in the phase separation. The complexity of the strain fields does unfortunately not allow for a quantitative assessment of their impact on oxygen diffusion at present. Using a VO2 bar with a well-defined strain field distribution instead of a polycrystalline thin film would enable one to elucidate the role of the strain field in the phase segregation. Recent studies21,38 demonstrate the formation of a domain structure with metallic, insulating, and strain-stabilized intermediate phases18 on a VO2 nanobeam in response to local variations in the strain field. However, in contrast to our observation, the whole nanobeam reverts to the insulating VO2 after cooldown. The long-term relaxation processes, which have been observed in polycrystalline films, and are often related to the width of the hysteresis loop in the R-T curve, are likewise attributed to the local strain fields and intercrystallite interactions in these films.5,39,40 The strain field might indeed influence the transition temperatures in the individual crystallites in our samples but the irreversible transition to a partially metallic surface can only be explained by a substantial contribution from a chemical phase transition. However, the long-term relaxation might indeed have a strain-related component.

V. CONCLUSION

We observed the surface transformation of a VO2 thin film with STM and STS during a thermal cycle. The spatially resolved band gap maps show that upon heating across the MIT the insulating VO2 surface transforms into a surface, which is defined by metallic islands, and a mixture of small band gap regions. The electronic structure of the surface is highly inhomogeneous, and is controlled by the structural MIT for VO2 and a chemical phase transition. The latter is attributed to the loss of oxygen and the presence of nonstoichiometric compounds with a high defect concentration. The extent of the chemical phase transition is directly related to the duration of the annealing process. The original VO2 surface cannot be recovered after cooldown, and metallic islands still cover about 35% of the surface, and coexist with VO2 and regions composed of a mixture of small band gap surface structures 共reduced V-oxides兲. The surface conductivity after cooldown is strongly coupled to the topography of the sample, and this effect is even more pronounced a week after cooldown. The long-term segregation of different surface phases is driven by oxygen diffusion and presumably closely connected to the intricate, local strain fields within the film.

ACKNOWLEDGMENTS

The authors gratefully acknowledge the support from the following awards and contracts: Office of Naval Research MURI Grant No. CHIP-N0014-06010-428, the Defense Microelectronics Agency under Contract No. DMEA2-H9400308-2-0803, the Army Research Office under Contract No. W911NF-07-1-0477, the Defense Advanced Research Projects Agency 共DARPA兲 under Grant No. N00014-05-10901, and Air Force Office of Scientific Research under Grant No. FA9550-08-1-0203. 1

S. Biermann, A. Poteryaev, A. J. Lichtenstein, and A. Georges, Phys. Rev. Lett. 94, 026404 共2005兲. 2 B.-G. Chae, Y. T. Kim, D.-H. Youn, and K.-Y. Kang, Physica B 369, 76 共2005兲. 3 Y. J. Chang, C. H. Koo, J. S. Yang, Y. S. Kim, D. H. Kim, J. S. Lee, T. W. Noh, H.-T. Kim, and B. G. Chae, Thin Solid Films 486, 46 共2005兲. 4 H. S. Choi, J. S. Ahn, J. H. Jung, and T. W. Noh, Phys. Rev. B 54, 4621 共1996兲. 5 J. H. Claassen, J. W. Lu, K. G. West, and S. A. Wolf, Appl. Phys. Lett. 96, 132102 共2010兲. 6 E. U. Donev, R. Lopez, L. C. Feldman, and J. R. F. Haglund, Nano Lett. 9, 702 共2009兲. 7 C. Ko and S. Ramanathan, J. Appl. Phys. 104, 086105 共2008兲. 8 E. Z. Kurmaev, V. M. Herkashenko, Y. M. Yarmoshenko, S. Bartkowski, A. V. Postnikov, M. Neumann, L. C. Duda, J. H. Guo, J. Nordgren, V. A. Perelyaev, and W. Reichelt, J. Phys.: Condens. Matter 10, 4081 共1998兲. 9 F. J. Morin, Phys. Rev. Lett. 3, 34 共1959兲. 10 M. M. Qazilbash, M. Brehm, B.-G. Chae, P.-C. Ho, G. O. Andreev, B. J. Kim, S. J. Yun, A. V. Balatsky, M. B. Maple, F. Keilmann, H. T. Kim, and D. N. Basov, Science 318, 1750 共2007兲. 11 D. Ruzmetov, S. D. Senanayake, and S. Ramanthan, Phys. Rev. B 75, 195102 共2007兲. 12 D. Ruzmetov, K. T. Zawilski, S. D. Senanayake, V. Narayanamurti, and S. Ramanathan, J. Phys.: Condens. Matter 20, 465204 共2008兲. 13 A. Sharoni, J. G. Ramirez, and I. K. Schuller, Phys. Rev. Lett. 101, 026404 共2008兲. 14 S. Surnev, M. G. Ramsey, and F. P. Netzer, Prog. Surf. Sci. 73, 117 共2003兲. 15 A. Zylbersztejn and N. F. Mott, Phys. Rev. B 11, 4383 共1975兲. 16 M.-G. Willinger, G. Neri, E. Rauwei, A. Bonavita, G. Micalli, and N. Pinna, Nano Lett. 8, 4201 共2008兲. 17 E. Strelcov, Y. Lilach, and A. Komakov, Nano Lett. 9, 2322 共2009兲. 18 A. C. Jones, S. Berweger, J. Wei, D. Cobden, and M. B. Raschke, Nano Lett. 10, 1574 共2010兲. 19 K. Nagashima, T. Yanagida, H. Tanaka, and T. Kawai, J. Appl. Phys. 101, 026103 共2007兲. 20 A. Tselev, E. Strelcov, I. A. Luk’yanchuk, J. D. Budai, J. Z. Tischler, I. N. Ivanov, K. Jones, R. Proksch, S. V. Kalinin, and A. Kolmakov, Nano Lett. 10, 2003 共2010兲. 21 S. Zhang, J. Y. Chou, and L. J. Lauhon, Nano Lett. 9, 4527 共2009兲. 22 R.-P. Blum, H. Niehus, C. Hucho, R. Fortirie, M. V. Ganduglia-Pirovano, J. Sauer, S. Shaikhutdinov, and H. J. Freund, Phys. Rev. Lett. 99, 226103 共2007兲. 23 A. Irizawa, A. Higashiya, M. Stunekawa, A. Sekiyama, S. Imada, S. Suga, T. Yamauchi, T. Ueda, M. Arita, Y. Takeda, H. Namatame, M. Tanigushi, and T. Namba, J. Electron Spectrosc. Relat. Phenom. 144–147, 345 共2005兲. 24 M. Demeter, M. Neumann, and W. Reichelt, Surf. Sci. 454–456, 41 共2000兲. 25 T. Becker, C. Streng, Y. Luo, V. Moshnyaga, B. Damaschke, N. Shannon, and K. Samwer, Phys. Rev. Lett. 89, 237203 共2002兲. 26 M. Fäth, S. Freisem, A. A. Menovsky, Y. Tomioka, J. Aarts, and J. A. Mydosh, Science 285, 1540 共1999兲. 27 D. Ruzmetov, K. T. Zawilski, V. Narayanamurti, and S. Ramanathan, J. Appl. Phys. 102, 113715 共2007兲. 28 W. Yin, K. G. West, J. W. Lu, Y. Pei, S. A. Wolf, P. Reinke, and Y. Sun, J. Appl. Phys. 105, 114322 共2009兲. 29 I. Horcas, R. Fernandez, J. M. Gomez-Rodriguez, J. Colchero, J. GonezHerrero, and A. M. Baro, Rev. Sci. Instrum. 78, 013705 共2007兲.


024311-8

R. M. Feenstra, Phys. Rev. B 50, 4561 共1994兲. G. J. de Raad, D. M. Bruls, P. M. Koenraad, and J. H. Wolter, Phys. Rev. B 66, 195306 共2002兲. 32 Y. J. Chang, J. S. Yang, Y. S. Kim, D. H. Kim, T. W. Nob, S. W. Kim, E. Oh, B. Kahng, and J. S. Chung, Phys. Rev. B 76, 075118 共2007兲. 33 A. D. Rata, University of Groningen, 2004. 34 A. D. Rata, A. R. Chezan, M. W. Haverkort, H. H. Hsieh, H.-J. Lin, C. T. Chen, L. H. Tjieng, and T. Hibma, Phys. Rev. B 69, 075404 共2004兲. 35 J. B. Goodenough, Prog. Solid State Chem. 5, 145 共1971兲. 30 31

J. Appl. Phys. 109, 024311 共2011兲

Yin et al. 36

M. Preisinger, J. Moosburger-Will, M. Klemm, S. Klimm, and S. Horn, Phys. Rev. B 69, 075423 共2004兲. 37 V. Simic-Milosevic, N. Nilius, H.-P. Rust, and H.-J. Freund, Phys. Rev. B 77, 125112 共2008兲. 38 J. Wu, Q. Gu, B. S. Guiton, N. P. d. Leon, L. Ouyang, and H. Park, Nano Lett. 6, 2313 共2006兲. 39 J. Narayan and B. M. Bhosle, J. Appl. Phys. 100, 103524 共2006兲. 40 J.-G. Ramírez, A. Sharoni, Y. Dubi, M. E. Gomez, and I. K. Schuller, Phys. Rev. B 79, 235110 共2009兲.
