Employing a MEMS plasma switch for conditioning high-voltage kinetic energy harvesters

Triboelectric nanogenerators have attracted wide attention due to their promising capabilities of scavenging the ambient environmental mechanical energy. However, efficient energy management of the generated high-voltage for practical low-voltage applications is still under investigation. Autonomous switches are key elements for improving the harvested energy per mechanical cycle, but they are complicated to implement at such voltages higher than several hundreds of volts. This paper proposes a self-sustained and automatic hysteresis plasma switch made from silicon micromachining, and implemented in a two-stage efficient conditioning circuit for powering low-voltage devices using triboelectric nanogenerators. The hysteresis of this microelectromechanical switch is controllable by topological design and the actuation of the switch combines the principles of micro-discharge and electrostatic pulling, without the need of any power-consuming control electronic circuits. The experimental results indicate that the energy harvesting efficiency is improved by two orders of magnitude compared to the conventional full-wave rectifying circuit.

and key recent work (focused on agile manufacturing, but still relevant to the microplasma "big picture") 6-Somewhere in the text, the authors should clearly state that solid-state switches are not good for this application becasue they are inherently leaky. You need a mechanical switch that physically disconnects the circuit. 7-Please comment on reliability issues. Your devices are made of Si. Are you concerned about lifetime? would it help to make the devices in other materials, e.g. tungsten? please help us understand the trade-offs, e.g. energy function, fatigue, stability of physical properties (e.g. single crystal vs, multi-crystal). This is a significant issue for your technology to be adopted as mainstream.
8-Based on your data, the choice of go = 9 um seems to be a fortunate coincidence; what would it take to optimize the design? In other words, maybe there is a better value for go, with even better performance. What would it take to find it? 9-Page 11, you mention that 80 pairs of tips are working in the fixed-fixed switch. I respectfully disagree. The microplasma is a non-linear phenomenon, it is very sensitive to the tip radii (tip electric fields depend on their tip radius), my guess is that only a few of them are working during the discharge due to the tip radii spread (it is unavoidable when you make arrays of anything, and when one makes arrays of very small features, the spread tends to have long tails).
10-Talking about the multi-tip switch, why do you have so many tips? are you concerned about lifetime (so as soon as the sharper tips get damaged, the duller tips start working)? please comment 11-Supplementary video 2. Maybe is my browser, but your video is upside down! can you please check you uploaded the video in the right orientation?
12-Thinking some more about your multi-tip device, the authors should point out that you might need current regulators in each emitter to uniformize the current in the device, to protect the tips from burning/damaging, to efficiently use the array of tips. In a nutshell, you can take care of the tip radii spread if you integrate negative feedback, i..e, having some ballast in series with each tip, so there is a maximum current per emitter, so the sharper tips can work at the same time the duller tips work. An ideal ballast has high impedance and high saturation current, e.g. an ungated field effect transistor (FET) operating in saturation, which is not hard to achieve given the voltage involved (you can get the FETs to saturate with volts). Given that you are making your structures in silicon, you can easily make long and narrow fingers ending in tips, which would make the fingers act as ungated FETs monolithically integrated to the sharp tips. Researchers  please add such discussion. Again, this is a significant issue for your approach to be adopted mainstream.

Response:
The environmental changes like the humidity surely have impacts on the approach, by influencing the breakdown voltage of the plasma switch. Learning from the previous research [67], we know that water vapor has a higher breakdown strength than air, so a mixture of the water vapor and air (i.e. higher humidity) results in a higher breakdown voltage. Fortunately, previous researches have confirmed that this kind of breakdown voltage fluctuation was normally not higher than 10% when the relative humidity is lower than 70% [68,69]. For our approach, a little breakdown voltage (ON actuation voltage) increase is good for the performance improvement as we are always pursuing a high operation voltage. The important point that needs to be paid attention to is that the switch breakdown voltage should not be higher than the reverse voltage of the diodes. In our design, the switch breakdown voltage was between ~300V and ~420V, which was far away from the diode limit of ~570V. Thus we believe that reasonable humidity change will not have a big impact on the suitability of the approach.
Compared to humidity, temperature has much less impact on the breakdown voltage if the temperature of operation is within the industrial temperature [70] [71].
Contrariwise, if there is water condensation between the two electrodes of the switch, conduction could occur at a much lower voltage, without any dielectric breakdown. In that case the system won't work correctly. Then, for a practical application, a hermetic package is preferable, also to protect the tips from dust pollution. This can even be obtained at the wafer scale (batch process) by performing an anodic bonding with a glass wafer. Combined with getter material, it also can also provide a good long-term vacuum [74] [75]. We added related comments to the manuscript.
In the "The MEMS plasma switch" section (p. 8): "Therefore, the ON-actuation voltage of the switch can be controlled by designing the gap properly. However, the breakdown voltage cannot always be kept precisely constant but will be influenced by the environmental changes, i.e. humidity and temperature. Fortunately, the breakdown voltage fluctuation resulting from the humidity change is normally not larger than 10% [67][68] [69], while the effect of the temperature on the breakdown voltage is ignorable within the industrial temperature range [70] [71]. Concerning the device reliability and for avoiding function failure, the ON-actuation voltage should be designed with enough margin over the reverse voltage of the diodes (~570V)." In the "Discussion" section (p.14): "The MEMS device used in this paper is exposed in air. It can be easily polluted by the dusts and become invalid if there is water condensation between the two electrodes of the switch, causing conduction between electrodes at a much lower voltage, without any dielectric breakdown. Therefore, for a practical application, a hermetic package is preferable. This can even be obtained at the wafer scale (batch process) by performing an anodic bonding with a glass wafer. Combined with getter material, it can also provide a good long-term vacuum [74] [75]." Q3-Page 7, Fig. 2 (c) and (d). Please comment on why your modeling underestimates the voltage and overestimates the current (the opposite effects are good, they try to compensate each other when calculating power). Also, it is good that modeling and experiment agree on the lower bound (what you call "bottom envelope"), but there are features in the experimental data that are not captured by your model, e.g. the current ripples that seem to have a slower timescale than the natural oscillations Fig. 2(d). Where does this come from? Is this an artifact in the experiment of is inherent to your approach? please explain.

Response:
Thank you for pointing out this disparity between simulations and experiments. Indeed, additional information is needed for a good understanding of these apparent mismatches: actually lower voltage bound, upper voltage bound and current ripples can be correctly caught by simulations. a) Concerning the higher amplitude of the TENG's voltage in Fig 2c: If we do a zoom in of a few cycles of the experimental curve, we observe a peak at each front edge followed by a damped oscillation at 50Hz:

Figure R1
Raw data of the measured voltage of the TENG output after a 1GΩ/70GΩ voltage divider showing a peaks at the front edges and 50Hz ripples.
The voltage peak is directly induced by the setup we used for measuring the high-voltage, i.e. a 1/71 resistive divider made of 2 resistors of 70GΩ and 1GΩ. Indeed, it comes from the parasitic capacitance of the 70GΩ resistor, which shows a low impedance compared to the resistance for this voltage having such a high slope. We can simulate this peak by adding a 100fF capacitor in parallel to the 70GΩ resistor, as shown below: Figure R2 (a) Simulation of the voltage across the TENG (b) Simulation of the TENG output voltage after the 1G/70G resistive divider including a parasitic capacitor of 100fF across the 70G resistor.
As the peak is quite sharp, it doesn't contribute significantly on the output average current nor the harvested energy. For the new experimental curve in Fig. 2c, we have added a low-pass filter at 40Hz for attenuating the electromagnetic noise, followed by a 50 th percentile filter of 65 points for smoothing the peak induced by the parasitic capacitance. The raw data and the post-process operations are presented in the Supplementary Materials Fig. S2. In the revision, we replaced the experimental data in Fig. 2c with the smoothed data and added the corresponding instructions in the main manuscript in page 6 marked in red. "The raw data of the measured as well as the signal processing are given in Supplementary Materials Fig. S2." b) Concerning the ripples in the TENG's current in Fig. 2d The low frequency ripple of the measured current comes from the low sampling rate (50Hz) of the used picoamperemeter, which is far below the Nyquist rate associated to this signal. So, the full signal is not captured and a low frequency ripple is introduced.
We can simulate this ripple by resampling at 50Hz the simulated current through the TENG, whose sampling rate is 10kHz. In the revision, we added both simulated curves with different sampling rates in Fig. 2d, and added the corresponding descriptions in the main manuscript in page 7 marked in red.
"The experimentally measured current peaks were actually underestimated and showed some lowfrequency ripples, because of the low sampling rate of the picoamperemeter, which can be derived by comparing the simulated current curves with sampling rates of 10kHz and 50Hz in Fig. 2d." Q4-page 8. You are missing a paragraph on microplasmas to provide context. In nutshell, by miniaturizing the inter-electrode separation, plasma sources can operate stably at less vacuum, even at atmospheric pressure, which opens very exciting opportunities, e.g. creating excited species that are otherwise only possible to create at extreme conditions. You should also some key references on micro plasmas. I suggest to include these: Please refer to page 8, paragraph 1, marked in red.
"As the switch in the conditioning circuit plays the role of controlling charge transfer, plasma discharge becomes a promising solution by providing reliable physical disconnection between electrodes, as well as high operation voltages. At a high-voltage threshold, a current flow will pass through two conductive electrodes due to the electrical breakdown in a specific gas [57][58] [59][60] (air or some rare gases  The descriptions of the MEMS plasma switch in the following paragraph "Fixed-electrode switch" were also revised as a logical consequence, marked in blue in the revised manuscript. Q5-Page 8. Spring softening (the spring constant is reduced) due to gap reduction is a well known phenomenon in MEMS mechanical switches. How spring softening affects (or improves?) your approach? please explain.
Response: Thank you for this question. Regarding the softening effect, we believe that the electrostatic softening effect for our device is not very obvious as comb-drives are used and the movement direction is longitudinal. The device works more like an electrostatic actuator rather than a resonator. The simulated resonant frequency of the device is ~13k while the actual movement frequency of the device actuated by the electrostatic force is lower than 2Hz. The frequencyamplitude effect caused by the spring softening can thus be ignored.
However, it is still valuable to explain clearly the dynamics and working principles of the movable switch, whose breakdown voltage is always varying because of the movement of the anode, i.e. dynamic changes of the gap between anode and cathode. In this revision, we added more simulations and theoretical discussions to explain why we have a full-hysteresis loop when 6 and a narrow hysteresis loop when 9 and 12μm. Please refer to the following description that is added in page 10 and 11, marked in red. where is the designed initial gap between anode and cathode, x the movement of the anode due to the applied voltage (see Supplementary Materials eq. S11), the operating pressure, , A and B are a constants related to the gas composition, the excitation-ionization energies and the saturation ionization respectively. The calculated dynamic breakdown voltage with different initial gaps between anode and cathode, as well as the relation between the voltage applied to the anode i.e.
(eq. S11) and the anode displacement are shown in Fig. 4c. A predicted breakdown voltage curve at x~0 (x=0.02μm) is shown in Fig. 4d, which corresponds to the normal Paschen's law in air. Detailed theoretical analysis of the electrostatic pulling and Paschen's law can be found in section III of Supplementary Materials. Seeing from Fig. 4c, there are crossing points between the breakdown voltage curves of 9μm, 12μm and the related green curve of versus , while no crossing when 6μm occurs. If there is no crossing, it means that the air breakdown voltage is always higher than the voltage leading to a physical contact, then the breakdown will never happen. The charges stored in the buffer will be fully released ( drops to 0V) during the contact, thus a full-hysteresis loop is expected. In contrast, air breakdown happens at the crossing points (for example when 2.3 ) before the anode touches the cathode. During the breakdown, will slightly drop as the plasma current is quite low, resulting in a narrow-hysteresis loop." Q6-Somewhere in the text, the authors should clearly state that solid-state switches are not good for this application because they are inherently leaky. You need a mechanical switch that physically disconnects the circuit.

Response:
We thank the reviewer for this suggestion. The corresponding comments are added in the introduction, page 3, paragraph 2, marked in red.
"Besides, additional energy dissipation will be brought in because this kind of solid-state electronic switch is inherently leaky. Superior electrical insulation between the two stages is mandatory, therefore an acceptable switch must have good physical disconnection properties when the switch is OFF." Correspondingly, when stating the advantages of our micro-plasma switch, we have added the property of "no ohmic contact" as marked in red in the following paragraph in page 4. "Compared to the electronic switches, the proposed micro-plasma switch has the advantages of no electronic control, no ohmic contact and no need to be supplied with some external energy. The proposed switch is a fully "stand-alone" device and does not require direct integration with the TENG." Q7-Please comment on reliability issues. Your devices are made of Si. Are you concerned about lifetime? would it help to make the devices in other materials, e.g. tungsten? please help us understand the trade-offs, e.g. energy function, fatigue, stability of physical properties (e.g. single crystal vs, multi-crystal). This is a significant issue for your technology to be adopted as mainstream.

Response:
The dominant point we concern in this paper is to verify the effectiveness of the MEMS plasma switch used in energy harvesters to improve the energy harvesting efficiency. It is wellknown that silicon-on-insulator (SOI) fabrication has the advantages of simple process, low-cost, and batch fabrication. Thus we choose this simplest and fastest fabrication solution that we are good at, i.e. etching tips with single-crystal silicon, to demonstrate a basic energy-harvesting system.
Regarding the reliability of the silicon switch, our major concern when we did the design was avoiding having a current so high it could damage the silicon tips. Micro-discharges could vaporize or sputter the electrode material after a long-term running at high current [57], which increases the pressure in the gap spacing if device is packaged and changes the gap itself, thus decreases the breakdown voltage. According to [59], a continuous current as high as tens of mA will cause physical damage and function failure for a silicon microplasma device. Fortunately, in our experiments, we calculated the average emission current in orders of µA. For example, in Fig. 3f, the average current can be calculated as i=Q/t=ΔV•Cbuf/Δt=70V•4.7nF/0.1s=3.3uA. In addition, in some devices, the charge transfer is distributed on multi pairs of tips (possibly up to 80 pairs). If we consider only 10 pairs of tips working simultaneously, the current through each pair of tips is about 0.3uA. Therefore, in terms of the current limit, our silicon MEMS switch can fully fulfill the requirements of our application.
However, the shortages of using silicon are also obvious, including the poor electrical conductivities of the material and the thin oxide layer that grows on the surface that can limit the current. For our application in the energy harvesting field, it may decrease the energy transfer efficiency because of the energy consumption caused by the poor electrical conductivity if a highly doped substrate is not used.
Employing tough metal materials will help to improve the emission efficiency and stability. For example, tungsten coating on silicon-based gated emitters have been used [76] [77]. However, we have to notice that such kind of tungsten coating or metal deposition will decrease the turnon/breakdown voltage [57], which is on the contrary with our purpose of having the operation voltage as high as possible. Therefore, the system can benefit from the good conductivity and low leakage of the metal materials, only on the basis of not decreasing to much the breakdown voltage.
The corresponding comments were added to the discussion part (p.15, marked in red): "The single-crystal silicon devices used in this paper can contain the current limits in most of the energy harvesting applications, because the calculated average ON-current is in orders of μA far below the safe current of tens of mA [59]. The MEMS silicon fabrication process holds the wellknown advantages of simple process, low-cost, and batch fabrication, however, the tradeoff is the poor electrical conductivities of the material and the thin oxide layer that grows on the surface which limits the current. For our application in energy harvesting fields, it may decrease the energy transfer efficiency because of the energy consumption caused by the poor electrical conductivity if a highly doped substrate is not used. Micro-discharges could also vaporize or sputter the electrode material after a long-term running at high current [57], which increases the pressure in the gap spacing if device is packaged and changes the gap itself thus decreases the breakdown voltage. Employing tough metal materials will help to improve the emission efficiency and stability. For example, tungsten coating on silicon-based gated emitters have been used [76] [77]. However, we have to notice that such kind of tungsten coating or metal deposition will decrease the turnon/breakdown voltage [57], which is on the contrary with our purpose of increasing the operation voltage as high as possible. Therefore, the system can benefit from the good conductivity and low leakage of the metal materials, only on the basis of proper design for not decreasing the breakdown voltage." Q8-Based on your data, the choice of go = 9 um seems to be a fortunate coincidence; what would it take to optimize the design? In other words, maybe there is a better value for go, with even better performance. What would it take to find it?
Response: We thank the reviewer for this good question. Actually, with unstable charge pump like our circuit, there no existing optimized design because we always have better performance by approaching a higher and a narrower hysteresis loop. However, and the width of the hysteresis loop are limited by three factors: (1) the inverse breakdown voltage of the diodes (it is ~570V in this paper), which defines the upper limit voltage of ; (2) the ON voltage of the switch (from ~300V to ~450V in this paper), which defines the working voltage of ; and (3) the passing current when the switch is ON, which defines the charge releasing time, i.e. the width of the hysteresis loop.
We added some descriptions in the discussion part (p.14, marked in red): "However, there is still a large space to further improve the performances of the circuit. For example, we can have more diodes in series to get a higher inverse breakdown voltage, which will increase the upper limited working voltage for the buffer capacitor. According to the Paschen's law, we can increase the gap to tens of or even one-hundred micrometers to approach a ~kV ON voltage. At the same time, the number of pairs of tips can be reduced to narrow the hysteresis. However, keeping a proper redundancy of the tips ensures the robustness and sustainability of the switch." Q9-Page 11, you mention that 80 pairs of tips are working in the fixed-fixed switch. I respectfully disagree. The microplasma is a non-linear phenomenon, it is very sensitive to the tip radii (tip electric fields depend on their tip radius), my guess is that only a few of them are working during the discharge due to the tip radii spread (it is unavoidable when you make arrays of anything, and when one makes arrays of very small features, the spread tends to have long tails).

Response:
We agree that not all of the 80 pairs of tips work at the same discharge cycle. Although it cannot be precisely localized, we can still find proof from the Supplementary Video 1: the appearance of the flare is not always fully distributed over the 80 tips when the switch is ON and there is a slight difference in the flash positions for each time. Another proof is that the ON/OFF voltage at each switch operation cycle is not exactly the same, seen from Fig. 3f (previous Fig. 3c).
We added the discussions as follows (p. 12, marked in red): "The narrower hysteresis of the movable switch ( 9μm) compared to the fixed switch can be explained by the fact that only one pair of triangular tips is working for the movable switch, whereas multi pairs of tips are working in the fixed switch. But it has to be noted that not all of the 80 pairs of tips are working at the same time for a specific discharge cycle. Indeed, we can see from the Supplementary Video 1 that the flare did not appear all of the positions of the tips and there is a slight difference time to time. The disparity of the working tips results in the slight ON/OFF voltage variations as shown in Fig. 3e."