The structural phase transition (SPT) in VO2 and its hysteretic behavior is used in the present work to program mechanical displacements of a micro-electro-mechanical memory through electro-thermal actuation (i.e., Joule heating). The memory element consists of a silicon dioxide cantilever with buried Ti/Pt heating electrodes and a nanocoating of VO2. The integrated heating electrode is used to actuate the structure, and its resistance is monitored to determine the phase transition of the material. A closed-loop system is implemented to ensure proper operation over a larger temperature window than previous work. The closed-loop control system is designed using the system parameters obtained from a static and transient response analysis. The advantages of the feedback configuration over an open-loop configuration are demonstrated by a larger difference between programmed states (i.e., higher noise margin). The micro-electro-mechanical memory shows no noticeable degradation after being operated for 12 000 cycles. Multiple mechanical states are programmed using oscillatory pulse sequences that produced bi-directional displacements.
In 1959, F. J. Morin at Bell Labs observed an insulator-to-metal-transition (IMT) along the c direction in bulk crystals of vanadium dioxide (VO2) that was induced upon heating at ∼68 °C . This observation was later confirmed by Verleur et al., who also noticed an abrupt (and hysteretic) change in the optical properties of VO2 for λ = 4 µm during the IMT .
Although VO2 is a strongly correlated material, which is still not well understood and very rich in new fundamental science [3, 4] its proximity of the IMT to room temperature has made this material the focus of many applied researchers who want to integrate it in practical devices. Other phase-change multifunctional material oxides also exhibit abrupt changes in their properties, but their transition temperatures (Ttr) are much further from room temperature than VO2, hindering their potential integration and use for most typical applications. Furthermore, the inherent hysteretic behavior that multiple properties of VO2 show across the phase transition has motivated several studies on the memory capabilities of the material, opening the possibility of using VO2 thin films for developing programmable multifunctional micro-electro-mechanical systems (MEMS) devices.
The memory aspect of VO2 associated with the change in conductivity was recently exploited by Driscoll et al. for the development of a memory metamaterial [5, 6]; work that was later expanded by the same research group . Later, the use of electric and optical reading/writing techniques was used by Coy et al. to program both: electronic and optical states in VO2 film . Subsequent work focused on the use of the electrical memory behavior of VO2 to demonstrate two terminal resistive memories written by electrical  and optical signals . All of this previous work has made use of the IMT of VO2.
The IMT in VO2 is immediately followed by a more energetically demanding structural phase transition (SPT) from a low temperature monoclinic phase to a high temperature rutile phase . Using the lattice parameters for both phases of VO2 (and noting that the unit cell in the monoclinic phase corresponds to two unit cells in the tetragonal phase), it can be calculated that the volume of the VO2 unit cell increases during this SPT.
Although the overall volume increases during the phase change (that is, during the heating cycle), the areas of some planes are reduced . When polycrystalline VO2 films are deposited by pulsed-laser-deposition (PLD) over amorphous SiO2 or single crystal silicon (SCS), the material tends to orient itself with its (011) and (110) planes parallel to the substrate for T < Ttr and T > Ttr, respectively . The area of the plane for the tetragonal phase is smaller than that of the monoclinic phase. Thus, when SiO2 or SCS microcantilever structures are coated with VO2 films, the structure bends concave toward the VO2 film coating during the phase change.
In 2010, it was observed that the SPT in VO2 can be used to generate curvatures over 2000 m−1 in 350 µm long SCS cantilevers . Such phenomenon was not a result of the difference in thermal expansion coefficients (Δα) between the SCS and VO2, since the observed strain was much larger than what could be achieved by this mechanism. Instead, the underlying mechanism was a result of the crystallographic changes that the VO2 film experiences during the phase transition. Subsequent relevant work was mainly focused on the use of VO2's SPT for microactuation purposes, which included the use of single crystal VO2 nanobeams [14-16], the dynamic behavior of VO2 coated microactuators , the potential of VO2 as a solid engine , and the demonstration of multiple programmable mechanical states on VO2-coated SCS cantilevers .
The present work presents a significant improvement in the previous advances on the memory applications of VO2 [9, 19]. We demonstrate a monolithically integrated micro-electro-mechanical memory device, where electrical signals are used to program deflection of a microcantilever, and unlike the relevant prior work mentioned above, the present system is operated in a closed loop configuration that compensates for external temperature variations, allowing for a very robust system that can operate in a much larger temperature window. MEMS have been found to be useful in a broad range of fields, including RF circuits, biomedical devices, sensors, actuators, imagers, and global positioning systems; and their operation is fundamentally based on the displacement of a suspended mechanical structure. The programmable mechanical states demonstrated in this work allows for multiple frequency reconfiguration of single RF circuits (e.g., antennas and transceivers), and repeated precise and controlled large displacements in microactuators used in manufacturing or surgery.
The design of the VO2 MEMS memory device is discussed first. Then, the response of the device for static and dynamic excitation is characterized. Such characterization is required, since it will determine the necessary system parameters for the design of the closed-loop control system, as well as the operating current value. The discussion of the designed closed-loop system follows. It is demonstrated that a feedback system is essential if the memory capabilities want to be exploited in applications where fluctuations in ambient temperature are possible. Finally, the repeatability of the device and the multiple state capability is demonstrated.
2.1 Thin film deposition
The VO2 thin film was deposited by PLD using a KrF excimer laser (LambdaPhysik LPX 200, λ = 248 nm) at 10 Hz repetition rate and 350 mJ pulse energy incident on a rotating vanadium target for 30 min. The sample was kept at 470 °C in a 15 mTorr Oxygen atmosphere. After deposition the sample was annealed for 30 min under the same deposition conditions.
2.2 Device fabrication
The fabrication process is outlined in Fig. 1. First, a 1 µm thick layer of SiO2 is deposited by LTO. Then the metal layer that will make the heater and the traces connecting to it, is deposited using thermal evaporation and lift-off. This metallization consists of an adhesion layer (50 nm of Ti) and 150 nm of Pt. Then a second layer of LTO SiO2 (1 µm thick) is deposited. The two oxide layers are then patterned using plasma etching, this will define the geometry, and will serve as a hard mask for the final step. Finally, the structure is released using XeF2 dry etching. After dicing, a VO2 film is deposited over the entire DIE, and the DIE is connected to an IC package for measurement.
2.3 Measurement setup
The measurement setup used was the same as used in . A temperature sensor and Peltier heater was attached to the IC package and the temperature of the DIE was controlled using a closed loop PID temperature controller. In order to measure the deflection of the cantilever, a low intensity IR laser (CW λ = 808 nm) was focused on the tip of the cantilever. The intensity of the laser was adjusted so that it had little effect in heating the cantilever, but that the reflection had enough intensity for detection. The reflected light was then aimed at a position sensitive detector (PSD, Hamamatsu S3270) which would output a voltage proportional to deflection. The signal was then calibrated by using a CCD camera aimed at a side view of the cantilever. All signal acquisition and control was done using an embedded real time controller (NI cRIO-9075) programed in LabView.
3 Results and discussion
3.1 Design of the memory device
The memory cell in this work consists of a VO2/SiO2 bimorph cantilever with an embedded titanium–platinum heater, where the titanium is only used for adhesion purposes. The thickness of the heater and its “on-chip” electrical connections (hereby referred to as “traces”) to the metal contact pads is 200 nm. A top view of the device can be seen in Fig. 2. The fabrication process is discussed in the experimental section. The thickness of the cantilever was 2 µm and the thickness of the VO2 film was approximately 200 nm. The chip was cemented to an IC package using silver paint and a gold wire was bonded from the Ti/Pt metal pads on the chip to the connections of the IC package, through which all the actuation and feedback electrical signals are transmitted.
Because of the large linear sensitivity of the resistivity of Pt to variations in temperature, it is possible to use the resistance of the heater as a feedback measurement to stabilize the temperature of the cantilever. Figure 3 shows how the resistance of the device varies as the temperature is raised from 30 to 90 °C. This change in resistance with temperature was used on later experiments to monitor the temperature of the microcantilever. The entire device can be actuated and controlled through this single terminal, simplifying the complete system.
After deposition of the VO2 film, the deflection of the cantilever was characterized as the substrate temperature was varied across the SPT window. The results are shown in Fig. 4. The transition temperature in this work is defined as the points in the heating and cooling curves where the rate of change in deflection is the largest. The transition temperature for the heating and cooling curves are marked in green, and the corresponding device resistance is also marked in Fig. 3. This curve shows the classical curve for the SPT found in previous work, showing that we have a highly oriented film of VO2. Since the temperature sensor used to measure the temperature of the IC package was placed far from the chip, conductive losses from the device to the sensor resulted in a lower temperature reading. On a separate experiment, the resistance of the VO2 film was measured as its temperature was varied across the phase transition, and the curve was found to be shifted approximately 10 °C higher, which matches the typical transition temperature for VO2. Because of residual stress during fabrication, there is an initial out-of-plane deflection of the cantilever at room temperature. All deflection measurements obtained were relative to that initial deflection. The static and transient response analysis discussed next is done using electro-thermal actuation (i.e., Joule heating).
3.2 Static and transient response
The deflection of the cantilever was measured as the current through the device was cycled from 0 to 7 mA. Figure 5 shows the resulting deflection curve. It can be readily noticed that the shape of the deflection curve in Figs. 4 and 5 are different. This is due a change in temperature distribution when heating with the integrated heater in the cantilever, vs. heating the whole chip with the Peltier heater. In conductive actuation, the Peltier heater covers the entire bottom surface of the chip and therefore, the temperature is very uniform across the entire chip. However, in the case of Joule heating, only the suspended structure is being heated and the cantilever's anchor acts as a heat sink.
The total measured resistance of the heating electrode can be considered as two electrical resistances connected in series: (i) the resistance of the heating element, and (ii) the resistance of the traces. It is necessary to estimate the values of these two resistance components (since they can not be physically measured directly). The data from previous curves and the dimensions of the heater were used for this calculation, and heater resistance was found to be 216 Ω whereas the trace resistance was found to be 84 Ω (both values at room temperature). The thermal coefficient of resistance (TCR) of the Ti/Pt metallization was measured to be 3.17 × 10−3 K−1.
As in all the previous work that studied memory in VO2, the presented VO2-based MEMS memory device is required to operate within the hysteretic region. In order to maximize programmable states, it is best if the device is preheated with a monotonic increase in temperature to a value at which the distance between the outermost hysteretic loops (heating and cooling) is largest [19, 21]. In Fig. 5 the outermost loops are the red and blue lines, and the optimum operating resistance is shown by the green line, 317 Ω in this case. The vertical separation of the loops at this resistance is the theoretical maximum difference between the memory “states”, 44 µm in this case. The current needed to maintain this operating temperature when the device is at 30 °C is 3.6 mA and the resistance of the heater is 233 Ω which gives us the average power needed to operate the device at room temperature, 3.02 mW.
Before discussing the memory experiments, where mechanical states were programmed using current pulses super-imposed to the operating current discussed above, a closed-loop controller for the system was designed. The main purpose of this controller is to maintain the operating resistance of the device under fluctuating ambient conditions. In order to design this controller, it becomes necessary to know the speed of the mechanical structure. All the control design parameters (e.g., sampling time, duration of programming pulses, etc.) will depend on this speed. Previous work has demonstrated that thermally induced VO2-based actuators are only as fast as the speed at which temperature can be changed in the structure [17, 22]. A full-scale open-loop current step response is shown in Fig. 6.
The measured time constant was calculated from the average curve to be 4.4 ms. In order to guarantee that the structure achieves a steady-state temperature after a step input, the duration of the programming pulses were 40 ms which is several times larger than the times constant. To have a more robust system that can compensate for temperature fluctuations, a feedback system was designed, where the device's temperature is constantly monitored and used to control the current supplied to the heater. The closed-loop system should be independent on the programming pulses; and thus, it was designed to be 10 times slower than the programming pulse (400 ms). The control system was implemented using a field programmable gate array (FPGA).
3.3 Open-loop versus closed-loop behavior
Figure 7 shows the schematic of the circuit used and the block diagram of the control system. The programming pulses are added to the output of the controller (which is limited to voltage amplitudes between 1 and 10 V). The device, represented by RD in the schematic, was connected in series with a known resistance (R1) of 1.2 kΩ. The value of R1 was chosen to protect the resistive heater in the actuator by limiting the current due to the maximum applied voltage of 10 V. Using the measured value for the voltage across R1 and simple circuit analysis, the current through the device and the value for RD was calculated. The controller setpoint RSet is set to the previously determined operating resistance 317 Ω.
The first memory experiment was designed to test the robustness of the system. The temperature of the chip was maintained at 30 °C using the Peltier, while the closed-loop control system maintained the device resistance at 317 Ω. Then, programming current pulses that maximized the difference in states were applied. In order to maximize the difference in the programmed mechanical states, the pulse had to be large enough to transition the phase change region completely during heating and cooling. The amplitude of the pulses required were +1.52 V (for heating) and −3 V (for cooling). A train of pulses that cycled between the two states was applied, and the result is shown in Fig. 8. This experiment was repeated when the chip is at a temperature of 40 °C (result also shown in Fig. 8).
In memory devices it is important to be able to differentiate between the different states, for example, by using a comparator with a threshold somewhere between the states. This means that at all times the states of a memory device must remain inside certain “windows” that must never overlap. Figure 8 shows that as the temperature is raised the states shift down by a small amount, but they remain inside windows that are separated by a sufficiently large amount that permits reading of the states independently from temperature.
An important detail to notice is that for the higher background temperature of 40 °C, the steady-state deflection is lower. However, from Figs. 4 and 5, the steady-state deflection should be larger for a higher temperature. This can be explained by considering that the resistance of the device, RD, is actually composed of two parts as explained before,
where RD is the measured resistance of the device, T the ambient temperature in °C, V the magnitude of the applied voltage pulse, R the resistance of the heater in the device, and Rtrace is the resistance of the Ti/Pt traces, given by
Notice that Rtrace has been assumed to be only a function of the ambient (or background) temperature (i.e., the traces are assumed to be perfect heat sinks). From these two equations, it can be noticed that when the ambient temperature rises by +10 °C, Rtrace increases by 2.6 Ω and since the controller forces RD to remain at 317 Ω, the value of Rheater has to decrease by the same amount of 2.6 Ω. From the data plotted in Fig. 5 it can be calculated that, a drop of 2.6 Ω from the operating resistance of 317 Ω corresponds to a drop in deflection of approximately 14 µm; which matches the difference in deflection between the steady- state deflection values for the two different background temperatures shown in Fig. 8.
Therefore, the difference between memory states at different background temperatures can be reduced by reducing the ratio of Rtrace/Rheater.
It is important to demonstrate the advantages of the designed closed-loop system over the simpler open-loop version. To this end, the same tests were performed on the device in open loop, with a constant average operating voltage that only keep the device resistance at 317 Ω in room temperature. This voltage was 5.65 V. The device was exposed to the same sequence of pulses. The results are shown in Fig. 9.
At 30 °C the resulting memory states were very similar to the closed loop response, but when the temperature was raised by +10 °C the deflection curve shifted up by more than 40 µm. The separation between states for the 40 °C background temperature is half of that for the 30 °C. Furthermore, with these new curves it is impossible to set a threshold to differentiate between the two states of the device. Therefore, due to the abruptness of the changes in the properties of VO2 with temperature, some method for temperature compensation (like the closed-loop system used here) is necessary for practical VO2-based memory applications.
It should be noted that, since the main purpose of the experiment in this section is to validate the robustness of the closed-loop system, only two states are shown. The performance and the demonstration of multiple states on the presented MEMS memory device follows.
3.4 Performance of the VO2-based MEMS memory
The next set of experiments was designed to characterize the reliability of the device, and its multiple state capability. The device was continuously cycled between its two states for 10 min (approximately 12 000 cycles). The values of the two states before and after the cycling were measured and are presented in Fig. 10. After cycling, no noticeable change in the deflection for each state, nor in the shape of the device's response to the programming pulses was noticed.
Finally, multiple memory states were demonstrated by controlling the amplitude of the programing pulse. A sequence of positive and negative pulses were used, and the results are shown in Fig. 11.
A VO2-based micro-electro-mechanical memory is implemented with a closed-loop system, which used the resistance of the integrated heater as a feedback variable and expanded the usable temperature window of the device. The various system parameters that affect the design and tuning of the control loop were studied by determining the static and transient response of the device, and the necessity to operate the device in close loop was demonstrated. A large difference between the programed deflection states was shown (30 µm) as well as the capability to demonstrate multiple programmable states. The final memory cell showed high reliability by showing no degradation after 12 000 cycles.
This work was supported by the National Science Foundation (NSF ECCS Award ECCS-1306311 monitored by Anupama Kaul). E. Merced acknowledges the support from the National Science Foundation, award no. DGE-0802267 (GRFP Program). This work was performed in part at the Lurie Nanofabrication Facility, a member of the National Nanotechnology Infrastructure Network, which is supported in part by the National Science Foundation.