Vanadium dioxide is one of many correlated-electron transition metal oxides that undergo an insulator-metal transition (IMT) driven by various stimuli including temperature, strain, doping and optical excitation. More specifically, VO2 undergoes both an IMT and a structural phase transition (SPT) from a semiconducting monoclinic to a metallic rutile phase at a temperature of roughly 340 K.[2, 3] IMT materials interposed between two metallic electrodes are known to exhibit volatile conductance switching, a property which has applications ranging from non-volatile memory to neuromorphic computing. In such two-terminal devices, the driving mechanism behind the IMT-induced conductance-switching has been under debate, with competing claims that an electric-field or Joule-heating is the dominant process. There is also debate about whether or not the SPT is triggered concurrently with the electrically induced IMT.[10, 11] Here we use a combination of optical, infrared and X-ray spectro-microscopy techniques to show that driving the device beyond the switching threshold results in an abrupt redistribution of local temperature as well as the emergence of metallic filaments with rutile phase ordering. We additionally show that the thermal redistribution and the SPT occur at the same applied current even if the SPT occurs at a different temperature than the IMT, consistent with a Joule-heating driven phase transition model. These results demonstrate that spatially resolved techniques are critical for unambiguously characterizing and understanding this inhomogeneous phenomenon.
From both technological and scientific standpoints it is important to unambiguously understand the driving mechanism behind conductance switching in VO2, especially since the material exhibits both an IMT and an SPT which can be triggered by a range of extrinsic forces.[2, 10, 12, 13] When switching was first observed in VO2, it was believed that Joule-heating caused a temperature-driven IMT simultaneously with the SPT, giving rise to the current-controlled negative differential resistance (CC-NDR) and conductance-switching. This conclusion has been challenged by recent claims of electric-field-driven switching in which Joule-heating was negligible.[6, 17] To add additional complexity to the understanding, it remained unclear as to whether or not SPT occurs, even under a Joule-heating model, since TSPT can be higher than TIMT.[10, 11] Here we employ spatially-resolved observations of the local temperature and electronic structure of planar VO2 devices in order to directly characterize the inhomogeneous behavior of the device during switching. Using these techniques we observed that the sharp conductance-switching phenomenon coincided with the emergence of filamentary optical contrast between electrodes, a sharp redistribution of the thermal profile concentrated around the filament, and the emergence of filamentary rutile phase-ordering in electronic structure. The coincidence of these phenomena at the same applied current is described well by a Joule-heating driven thermal phase transition model in which both the IMT and SPT occur at or near the same temperature. Some of the individual techniques discussed here have been applied to similar devices, but without the spatial resolution required to observe redistributions in both the local temperature and electronic structure of a functioning device.[8, 12, 16, 18]
In order to characterize the conductance switching phenomenon in VO2, we fabricated planar devices by evaporating two platinum electrodes of width W separated by a gap of length L onto a thin VO2 film (Figure 1a). Data reported in Figures 1, 2, and 3 were obtained from the same device, with L,W = 16,8 μm. These devices exhibited a sharp drop in low voltage (<50 mV) resistance, R0, as the ambient temperature, Tamb, was raised above insulator to metal transition temperature (TIMT = 340 K) and exhibited hysteresis upon cooling (Figure 1a). When larger currents (>1 mA) were applied to devices at Tamb = 300 K they exhibited switching to a low-resistance state, or turned ON, at threshold point (ION, VON) (Figure 1b). When the current was subsequently ramped down below a second threshold point (IOFF, VOFF) the device reverted to its original high resistance state, or turned OFF. The switching process results in a two-terminal current-voltage curve which is dynamical, hysteretic, purely dissipative, and reversible and can be mathematically described under a memristive system model.[19, 20] The threshold positions depended on ambient temperature, as evidenced by measurements of VON which monotonically decreased with increasing temperature and approached zero as Tamb approached TIMT (∼340 K) (Figure 1c). The temperature dependence of R0 and VON are consistent with previous reports for similar devices.[15, 21]
In the low resistance state, e.g. when currents above ION were applied, a visible filament emerged which bridged the electrodes. This filament was present only while the device was held in the low resistance state, grew in width with increasing current (Figures 2a-2b), and disappeared when the applied current was reduced below IOFF. Optically visible filaments have been previously reported in similar planar VO2 devices and were previously attributed to a local temperature increase within the filaments. To confirm the nature of the optical contrast in our devices, we measured the thermo-reflectance of the VO2 film, without applying current, as a function of ambient temperature (Figure 2c). The reflectance at green wavelengths (∼530 nm) decreased with temperature until it saturated at around 353 K, consistent with previously reported data, indicating that the optical contrast is likely related to localized temperature changes. A spatially localized reflectance spectrum was measured by focusing the detector near the edge of the filament and revealed a spectrum similar to the spectrum expected for the material at Tamb = TIMT (black curve in Figure 2c). This result qualitatively suggests that the high contrast region is at a temperature above TIMT and that the halo around the filament reflects a thermal gradient to ambient temperature outside of which the material remains insulating. Since the thermoreflectance saturates just above TIMT, it is not well-suited to perform quantitative thermometry and hence we used black-body spectro-microscopy.
In order to quantitatively map local temperature of the device in situ throughout the ON and OFF switching processes we employed an InfraScope black-body spectro-microscope. A series of temperature maps, emissivity-calibrated to a temperature resolution of ∼1 K and with a diffraction-limited spatial resolution of ∼2 μm, were acquired at current steps of 10 μA. Figure 3a shows four of the temperature maps in which the current was ramped beyond the on-switching threshold, revealing the emergence and growth of a filament-shaped hot spot between the electrodes. We characterized this changing temperature profile by plotting a line scan of the thermal map taken across the central position between the electrodes (hash marks in Figure 3a) as a function of applied current for the ramp up and ramp down (Figure 3b). This figure reveals a discrete reorientation of the thermal profile at the switching currents, as shown in more detail in Figure 3c for the line profiles just before and just after ON and OFF switching. The thermal profile is relatively more concentrated while the device is in the low resistance state and relatively more diffuse while the device is in the high resistance state. We note also that just before ON switching the maximum temperature was measured to be just below the heating IMT temperature while just before OFF switching the maximum temperature was measured to be just above the cooling IMT temperature (both of which were measured previously from Figure 1a). These two observations are displayed quantitatively in Figure 3d in which the distribution width (defined as the width of the region which contains the middle 30% of the area under the temperature distribution) and maximum temperature are plotted as a function of current.
The coincidence of conductance switching with the local crossing of the IMT temperatures clearly demonstrates the thermal nature of switching in these planar devices and the detailed thermal distribution change provides a good description of the switching process. Since the current density, and resulting Joule-heating thermal profile, is inhomogeneous in this device geometry it is only the maximum temperature portion of the device that dictates switching. Once the switching process is initiated at a point the local resistance drops, causing an unstable positive feedback effect which causes more current to flow into the local low-resistance filamentary pathway until some significantly different steady state is reached. Thus the end result of the switching process is a discontinuous jump in maximum temperature and a coincident focusing of the current into a filament. Interestingly, this redistribution results in a drop in local temperature of around 15 K at some points (e.g. xc = -2 μm in Figure 3c) despite an increase in current. The maximum temperature displayed in Figure 3d changes more rapidly with current in the on-state compared to the off-state. This is because incremental change in current has a bigger effect on local temperature in a confined filament in the on-state than in a uniform current distribution in the off-state. This observation also counters prior predictions of temperature distributions within a metallic filament. Due to the inhomogeneity of the transition, it is critical to perform high-spatial-resolution thermal measurements since focused single-point or spatially averaged temperature measurements could grossly underestimate the switching temperature leading to an incorrect conclusion on the nature of switching.
In order to directly map structural changes that are associated with inhomogeneous heating, we used scanning transmission X-ray microscopy (STXM) to study the localized electronic structure changes in situ during conductance-switching. Since STXM is a transmission technique we fabricated a set of devices on 20 nm thick silicon nitride membranes using an otherwise identical process to the one used for optical and thermal measurements above. Due to the constraints of the two measurement techniques, we fabricated two varieties of devices for characterization: one for black-body spectro-microscopy and another for X-ray measurements. STXM enables X-ray spectroscopic mapping with a 30 nm spatial resolution and 70 meV energy resolution, giving local electronic structure measurements throughout the device. Figure 4a shows two X-ray absorption spectra (XAS) of the oxygen K-edge, collected at a single point midway between the electrodes with no applied current and with current significantly above IT. XAS has been studied extensively in the VO2 system, and the O K-edge was shown to be particularly sensitive to modifications in the structural ordering. The differences observed in the electronic structure of the two spectra, particularly the emergence of the d‖* band, are attributed to known differences between semiconducting monoclinic (M1) and metallic rutile (R) phase-ordering, showing that there is indeed a SPT.[10, 27] We used these point spectra to identify particular absorption energies that gave the maximum contrast between the two phases in STXM, 528.95 eV and 530.2 eV, shown at the bottom of Figure 4a.
Using these maximum contrast energies we collected spatially resolved absorption maps at both energies for a set of different applied currents and constructed contrast maps, calculated as the normalized ratio of absorption at the two energies, for increasing current (Figure 4b). These maps show regions of high contrast, signaling the presence of metallic rutile phase-ordering, between the electrodes with a similar shape to the thermal hotspots of Figure 3a. As shown in Figures 4c and 4d, rutile phase ordering was observed only above the current threshold ION and grew in width with a similar trend as the thermal hotspots and optical contrast regions. These results show that, although TSPT could be higher than TIMT by several degrees in some cases,[10, 11] the discrete jump in local temperature of up to 45 K associated with conductance-switching in this configuration is sufficient to cause a SPT whether or not they actually occur at the same temperature.
We used a 3D finite element COMSOL simulation of our model for thermal switching in order to validate its mathematical feasibility based on fundamental physics. The simulation included basic electrical and thermal physics, including electron drift and heat diffusion, a three-dimensional model of the planar device, and literature values for the materials parameters. To implicitly account for the phase transition of the VO2 without introducing any explicit physics for phase transition into the model, we used simple sigmoid functions for the dependence of its electrical conductivity on temperature. We roughly calibrated this functional form to the data of Figure 1a. Figure 5a shows simulated temperature maps for different currents. For comparison to the optical results of Figure 2a, we clamped the color scale of the simulations to coincide with the saturation temperature of thermo-reflectance, 353 K. The I–V curve and the current dependence of the width of the filaments obtained from simulations are also shown (Figures 5b,5c). Data from simulation qualitatively agree with the experimental data while the quantitative differences could be attributed to several computational parameters that may not account for a real-world scenario, discussed in the supporting information.
In conclusion, we have utilized a complementary set of techniques to study the electrical and thermal behavior, and electronic structure of fabricated VO2 devices. Blackbody spectro-microscopy provided a comprehensive description of abrupt redistributions in local temperature during Joule-heating-driven IMT. Filamentary conduction was accounted for in numerical simulations using a simplified electro-thermal model. Using STXM maps we observed that the SPT occurred beyond the same current threshold that induced the IMT, regardless of whether or not they occur at precisely the same temperature. The occurrence of IMT and SPT at the threshold current was justified by the jump in local temperature upon on-switching. The techniques employed in this study provide a spatially-resolved, direct and unambiguous means to study local temperature and local structure correlated with electronic transport. These results present a clear understanding of the driving mechanism in metal-insulator transition devices, which have gained recent interest due to their promising applications.