Controlling the resistive switching hysteresis in VO2 thin films via application of pulsed voltage

We investigate the origin of the variation in resistive switching hysteresis of VO2 thin films. Using pulsed electrical measurements in textured VO2 thin film devices, we show that the hysteresis observed in I-V curves results from Joule heating effects, particularly in the low-resistance state. The hysteresis is reduced by increasing the cooling time between pulses. Based on a mechanism of Joule heating-induced metal-insulator transition, numerical simulations are performed which agree with the experimental variation in the hysteresis. Finally, a framework for engineering the I-V curves of VO2 devices is proposed.

We investigate the origin of the variation in resistive switching hysteresis of VO 2 thin films. Using pulsed electrical measurements in textured VO 2 thin film devices, we show that the hysteresis observed in I-V curves results from Joule heating effects, particularly in the low-resistance state. The hysteresis is reduced by increasing the cooling time between pulses. Based on a mechanism of Joule heating-induced metal-insulator transition, numerical simulations are performed which agree with the experimental variation in the hysteresis. Finally, a framework for engineering the I-V curves of VO 2 devices is proposed.
The metal-insulator transition (MIT) is a topic of long standing interest in condensed matter physics and materials sciences. 1,2 This transition deals with the increase in conductivity of a material with external stimulus such as heat, 1 light, 3 strain 4 or electric field. 5 Systems exhibiting MIT behaviour include organic compounds 6 and transition metal oxides. 7,8 Along with the resistivity; optical 9 and magnetic properties 10 can also undergo a sharp change through these transitions. Generally, the control parameters for the MIT can be classified into three categories: temperature control; bandwidth control 4 (e.g. strain); and band-filling control 11 (e.g. doping or applied fields), although more specific classifications can be found in this article by Imada et al. 12 The use of an electric field to induce a change in resistivity is referred to as resistive switching (RS), and has attracted much attention due to the potential applications in electronics such as oscillators, neuromorphic devices and memory. 13,14 For many RS systems, their I-V curve shows some level of hysteresis, i.e. the current does not retrace itself in voltage sweeps up and down. 6,15 The V-O system is a family of compounds with a large number of phases, many of which undergo a MIT. 16 Of particular focus in this study is vanadium dioxide (VO 2 ), with a MIT transition temperature (T MIT ) of 340K. Over a range of 0.1K, single crystal VO 2 undergoes a structural transition, accompanied by a 5 order of magnitude change in conductivity. 17 There is also a intrinsic thermal hysteresis of ∼1K. 18 VO 2 has also been shown to exhibit RS when external electric fields of 6.5×10 7 V m −1 are applied across it. 19 For thin films of VO 2 , the amplitude of the transition decreases, while the hysteresis and range over which the transition occurs increases, as described by structural effects such as phase propagation and grain size. 20 VO 2 RS takes place at either voltage polarity with well-pronounced hysteresis. 21 In many examples, the power required to trigger the MIT in the insulating state is equal to the power required to maintain the metallic state before it returns to the insulating phase. 22 The mechanism for the RS is debated, with arguments made for temperature control via Joule heating, and banda) Electronic mail: murtagho@tcd.ie filling control via carrier injection. 23 Experiments using different methodologies have isolated these mechanisms, 24,25 but in most cases both simultaneously contribute to the RS. Understanding the relative contributions of the mechanism in any given system is important. Furthermore, understanding the dynamics of each mechanism -for example temperature change induced by Joule heating -can provide insight into morphology, structure and phase changes, which is vital for device optimisation. Ideally, in electronic applications, heat fluctuations and power dissipation should be kept to a minimum to reduce energy consumption and mitigate adverse heating effects, while the ideal profile of the hysteresis varies between applications. 14 The ability to tune this I-V profile is therefore an important consideration in RS devices.
In this study, we determine the origin of hysteresis in RS of our VO 2 thin films. The influence of pulsing on hysteretic behaviour and the temperature profile is discussed. Through both continuous and pulsed voltage measurements we show that the hysteretic behavior originates from local Joule heat- ing of the channel once the system is switched to the lowresistance state. We further demonstrate how pulsing -in particular the time between pulses -can change the hysteresis. A thermal model, predicting channel temperature using only Joule heating and environmental cooling, agrees with the data and predicts substantial temperature fluctuations between the metallic and insulating states. The model demonstrates that one needs to consider heating and thermal conductivity beyond the VO 2 material itself.
Textured VO 2 thin films were grown on c-plane sapphire substrates via pulsed laser deposition to 30nm thickness. A KrF laser (248nm wavelength) and a pure vanadium target were used. Distance from target to substrate was 7cm. During growth the substrate was held at 600 • C, base pressure was approximately 1 × 10 −5 mbar, O 2 pressure was held at 15 × 10 −3 mbar and the laser energy used was 170mJcm −2 . The films were characterised by x-ray diffraction and resistance measurements, showing the VO 2 <010> ( Figure S1) family of reflections 26,27 and an approximately 3 order of magnitude resistance change across the MIT. Films grown by this method have been shown to be granular. 28 Sheet resistance measurements were performed using a 4 point probe and a ceramic heating stage. The sheet resistance was measured to be 120kΩ at 330K, 16kΩ at 342K and 150Ω at 360K. Due to hysteresis, upward and downward switching temperatures were 350K (T on MIT ) and 345K (T o f f MIT ), characteristic of thin film VO 2 . 29 The wider hysteresis is evidence of a granular film of slightly mixed phase causing a percolative effect, while the high switching temperature suggests a slight oxygen excess as compared to single crystal VO 2 ( Figure S2). 20,30 Devices for two-terminal measurements were prepared using photolithography. VO 2 channels of length 4µm and width 8µm were fabricated using a CF 4 RIE etch followed by deposition of layered Ti/Au contacts of thickness 5nm/30nm via electron beam evaporation. An image of the device can be seen in Figure S3. Different devices showed variations in threshold switching voltages and resistance due to variations in local film structure and lithography accuracy.
Contacting the samples was performed using a JANIS probe station with a heated stage. All measurements were made in the local Dublin atmosphere. For electrical characterisation, input waveforms were created using the LeCroy Arb-Studio 1102 arbitrary signal generator. A resistor (560 Ω) was placed in series with the selected VO 2 device and the Digilent Analog Discovery 2 oscilloscope was used to collect voltage data from the signal generator and across the resistor. The circuit is described in Figure S3. This arrangement allowed a statistically relevant number of measurement to be taken quickly. Two different measurements were performed: first, a continuous sweep of the voltage up and down at; and second, a pulsed sweep implemented with controlled pulse parameters. These measurements were performed on different devices.
At stage temperatures below the transition temperature, the upward continuous sweeping I-V curve of the VO 2 device shows a discontinuity at a critical voltage (V on sw ) as the material enters the metallic state and the current jumps. The closer the stage temperature is to the transition temperature, the less voltage is required ( Figure 1). As the voltage is lowered, the device will return to an insulating state at some V o f f sw . For a continuous voltage sweep, V o f f sw is always lower than V on sw resulting in a well-defined hysteresis.
In the pulse experiments voltage pulses 10µs in length were applied with varying gaps of 0V between pulses, from 100µs to 32ns. The input waveform was generated by multiplying this pulsed wave with a triangular wave. The beginning of each pulse occurs at voltage intervals of 0.01V ( Figure S4). For this device, the high and low resistance states of the VO 2 device were found to be between 10-15 kΩ and 60-260 Ω respectively measured via the I-V curves. This variance is believed to be due to conduction filaments 31 within the VO 2 channels changing or rerouting as the experiment progressed.
Two measured quantities from the experiment are total voltage dropped over the circuit (the source voltage, V s ) and the voltage drop over the resistor (V R ). The current through the circuit is calculated via I = VR R where R is the resistance of the resistor, while the voltage drop across the device V D = V s − V R . Figure 2 shows I-V curves of the VO 2 device using V D . This data is used to calculate the potential and power dissipated across the isolated VO 2 device. Figure 3 shows the relationship between gap length, V o f f sw and V on sw for the VO 2 device (using V D ). 100 measurements were taken for each gap length. Each data point represents the average and standard deviation of these measurements. Measurements were take with a stage temperature of 342K. Two trends are clear from This is the author's peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI: 10.1063/5.0017784
Controlling the resistive switching hysteresis in VO 2 thin films via application of pulsed voltage 3 The sample begins at the set temperature (T set : the stage temperature). In the continuously applied voltage regime, power I × V is applied to the channel, heating it. Any heat added will be conducted away by the channel, contacts and the highly thermally conductive sapphire substrate. 32 At V on sw , input power will be enough to bring the channel temperature (T ch ) to T on MIT (power = P on ). The resistance of the channel will drop and the current will increase. In this state the power applied is much higher, heating the channel. The channel will now stay in a metallic state until the applied power is low enough (P o f f ) to allow the channel to cool to T o f f MIT . Because the device is in a low resistance state, this will be at a much lower voltage (V o f f sw ). Note, in this case P on ≈ P o f f . However, if a pulsed voltage is used, the channel is able to cool between consecutive pulses. Therefore, more power will be required to heat the channel to T on MIT , and for longer gaps the V on sw will increase as observed. Similarly, the device will switch back to an insulating state once T ch reaches T o f f MIT . For a very large gap between pulses, this would occur at a voltage very close to V on sw as no heat would be retained in the channel between pulses. However as the gap decreases, ∆V will grow as heat is retained in the device, lowering the voltage required for T ch to reach T o f f MIT . This will be the trend until the gap disappears and the system is in the continuous voltage regime again. Thus, the experimentally observed changes in switching voltage in relation to the gap between pulses ( Figure  3) is consistent with the heating picture.
As relaxation time, τ. 34 τ is defined by the temperature change over time with no applied heating: Where T set is the stage temperature. τ is dependent on the thermal conductivity of the device and substrate. The heat flux into the sample is given by dQ in /dt = V × I. We further assume that the heat dissipation can be described as: Note this model assumes heat capacity C v and τ are independent of temperature. 35 The temperature variation with respect to time t considering total heat flux (heating and cooling) of the system is given by: Solving Eq. (3) an expression for channel temperature is given by: There are two fitting parameters in this model that predict the temperature; heat capacity C v and relaxation time τ. Equation (4) was used in a python script to model the channel temperature and calculate the fitting parameter values. V D and I values for each time step were taken from the measured data. The fitting parameters were adjusted so that the temperatures reached at V on sw and V o f f sw corresponded to T on MIT and T o f f MIT respectively.
An example of the temperature calculated by Eq. (4) is shown in Fig. 4. The orange line traces the maximum temperatures reached during pulses (voltage on), while the red line traces the minimum temperatures reached between pulses (voltage off). At V on sw the power jumps as the device becomes metallic and this leads to an increase in the temperature. The opposite occurs at V o f f sw . It is worth noting from Figure 4 how erratically the channel temperature changes between pulses, rapidly heating and cooling while reaching a maximum temperature of 430K. This is similar but lower in magnitude to temperatures calculated in similar experiments, 36 with difference attributed to the lower powers used here. The adjusted optimum values of C v and τ are 3.6 × 10 −11 J/K and 1.8 × 10 −6 s respectively. All pulse measurement simulations match V on sw and V o f f sw with respective temperatures T on MIT and T o f f MIT within 1K. Using the Eq. (4) the voltage hysteresis was simulated as a function of gap length, and is depicted in Figure 5. The dependence of the hysteresis magnitude on the gap time is This is the author's peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI: 10.1063/5.0017784
Controlling the resistive switching hysteresis in VO 2 thin films via application of pulsed voltage This value of C v can be used to approximate the size of the channel. The specific heat capacity C s v of VO 2 is 530-600 JK −1 kg −1 . 37 Using the equation; and using the estimated heat capacity, this gives a switching volume of 1.5 ×10 −17 m 3 . Given that the volume of the VO 2 channel is 9.6×10 −19 m 3 , about 16 times smaller, it can be assumed that the substrate plays a substantial role in absorbing heat in this experiment, an expected result due to the high thermal conductivity of the Al 2 O 3 substrate. In this experiment the applied field over the device never exceeds 1 × 10 7 V m −1 , lower than the field of 6.5 × 10 7 V m −1 required to cause a purely electronically driven switch. 19 Because of this, it can be assumed that the dominant mechanism for the MIT is self-heating. Further evidence is seen by the shift of V on sw in Fig. 3, as a higher power is required to heat the channel to T on MIT . A purely field driven switch would see a constant V on sw at the critical field. Despite this, it is maintained that even in channels which solely switch via electric field, this self-heating effect is still what dominates their substantial hysteresis. For this reason, the temperature of VO 2 in the on state needs to be considered in the use of VO 2 electronics, and cooling methods to prevent the material from reach- ing unstable temperatures or use of voltage pulsing should be implemented to reduce achieved temperatures.
We have shown that the hysteretic behaviour seen in resistive switching of VO 2 thin films can be altered via pulsing. Increasing the gap time, or allowed cooling between pulses, raises the V o f f sw value, reducing hysteresis. This effect is shown to be thermal, agreeing with the work of Lee et al. 36 We have also shown that V on sw increases with increasing gap time, with the cooling adding to the power requirement. Increasing the period of the applied waveform, or the pulse width, would allow more heating at low voltages, decreasing V on sw . This agrees with the results of Lee et al. That work has also shown that by decreasing the resistive load in series with the device, V o f f sw and V on sw can be lowered without changing the hysteresis. From our model, this can be attributed to increasing the power applied to the channel, reducing the required voltage.
With this information a picture of tuning the RS curve of VO 2 can be built. V on sw may be raised or lowered by the period of applied voltage waveforms. The hysteresis of the device may be altered via pulsing of this waveform, and the magnitude of the switching voltages can be altered by changing the power applied to the channel.
Of course, this doesn't take into account the cooling effects of substrates or the intrinsic material properties of the VO 2 film itself. Our model shows the effect of substrate heat capacity plays a substantial role, while the effect of structural properties will change how the film reacts to changing temperatures and electric fields. Strict environmental temperature control will also be a requirement. However, this is proposed as a useful framework for engineering electronic devices using VO 2 films.
The data that support the findings of this study are available from the corresponding author upon reasonable request.
See supplementary material for the resistivity vs temperature graph of the VO 2 film, the XRD spectrum of the film and diagrams of the circuit and input waveform used. This is the author's peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI: 10.1063/5.0017784
Controlling the resistive switching hysteresis in VO 2 thin films via application of pulsed voltage