A self-excited MEMS electro-quasi-static induction turbine generator The MIT Faculty has made this article openly available. Please share how this access benefits you. Your story matters. Citation Steyn, J.L. et al. “A Self-Excited MEMS Electro-Quasi-Static Induction Turbine Generator.” Microelectromechanical Systems, Journal of 18.2 (2009): 424-432. © 2009, IEEE As Published http://dx.doi.org/10.1109/JMEMS.2008.2011692 Publisher Institute of Electrical and Electronics Engineers / American Society of Mechanical Engineers Version Final published version Accessed Thu May 26 06:22:56 EDT 2016 Citable Link http://hdl.handle.net/1721.1/61619 Terms of Use Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. Detailed Terms 424 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 18, NO. 2, APRIL 2009 A Self-Excited MEMS Electro-Quasi-Static Induction Turbine Generator J. Lodewyk Steyn, Sam H. Kendig, Ravi Khanna, Stephen D. Umans, Fellow, IEEE, Jeffrey H. Lang, Fellow, IEEE, and Carol Livermore Abstract—This paper presents a microfabricated electro-quasistatic (EQS) induction turbine generator that has generated net electric power. A maximum power output of 192 μW was achieved under driven excitation. We believe that this is the first report of net-electric-power generation by an EQS induction machine of any scale found in the open literature. This paper also presents self-excited operation in which the generator resonates with an inductor and generates power without the use of external activedrive electronics. The generator comprises five silicon layers, fusion-bonded together at 700 ◦ C. The stator is a platinumelectrode structure formed on a thick (approximately 20 μm) recessed oxide island. The rotor is a thin film of lightly doped polysilicon also residing on a 10-μm-thick oxide island. Carrier depletion in the rotor conductor film limited the performance of the generator. This paper also presents a generalized state-space model for an EQS induction machine that takes into account the machine and its external electronics and parasitics. This model correlates well with measured performance and was used to find the optimal drive conditions for all driven experiments. [2007-0253] Index Terms—Electro-quasi-static (EQS) induction generator, electrostatic machine, power MEMS, self-excitation, turbine generator. I. I NTRODUCTION B ATTERIES have been, and still remain, the energy-storage medium of choice for many portable electric and electronic applications. Most hydrocarbon fuels have energy denManuscript received October 24, 2007; revised August 26, 2008. Current version published April 1, 2009. The work of J. L. Steyn was supported in part by an Applied Materials Graduate Fellowship. The work of S. H. Kendig was supported by the Reed Fund. The work at Lincoln Laboratory was supported by the Defense Advanced Research Projects Agency under Air Force Contract F19628-00-C-0002. Funding for the development of microengine technology was provided by the Army Research Laboratory (DAAD19-01-2-0010) under the Collaborative Technology Alliance in Power and Energy Program, managed by J. Hopkins (ARL) and Dr. M. Acharya (Honeywell) and by the Army Research Office (DAAG55-98-1-0292) managed by Dr. T. Doligalski. This work was presented in part at the 18th International Conference on Micro Electro Mechanical Systems, Miami, FL, January 30– February 3, 2005. Subject Editor C.-J. Kim. J. L. Steyn is with Pixtronix Inc., Andover, MA 01810 USA (e-mail: lodewyk@alum.mit.edu). S. H. Kendig is with Bose Corporation, Framingham, MA 01701 USA. R. Khanna is with BAE Systems, Lexington, MA 02421 USA. S. D. Umans is an independent Consultant residing in Belmont, MA 02178 USA. J. H. Lang is with the Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA 02139-4307 USA. C. Livermore is with the Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139-4307 USA. Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JMEMS.2008.2011692 sities that are approximately 20–30 times greater than those of most batteries and, therefore, present an attractive alternative, provided that a means can be found to convert the enthalpy of combustion of a hydrocarbon fuel into electric power. On the macroscale, gas-turbine (Brayton) cycles can have high efficiencies and coupled turbomachines and electric generators lend themselves to reliable continuous operation for long periods of time. Such gas turbine generators have been very successful at the kilowatt level and above. Research at MIT has focused on the development of a miniaturized gas turbine generator to deliver 1–50 W of electric power [1]. In this device, a small gas turbine engine provides the shaft power needed to drive a small electric generator. Presented in this paper is an electro-quasi-static (EQS) induction generator for that system. Although magnetic machines are preferred at large scales, EQS machines become attractive at small scales, primarily because very small air gaps between the rotor and stator allow higher electric fields before breakdown, on the order of 108 V/m. Higher electric fields make it possible to build a high-power-density EQS machine. Macroscale EQS motors have been reported previously, but even relatively small conventionally fabricated devices (e.g., [2], [3]) perform rather poorly as compared to their magnetic counterparts. Previously at MIT, Nagle et al. [4], [5], Fréchette et al. [6], and Livermore et al. [7] presented microfabricated EQS induction micromotors. The device in [7] attained a maximum speed of 55 krpm, a maximum torque of 3.5 μN · m, and a maximum air-gap power of 20 mW, the highest air-gap power of any MEMS electric micromotor. Attempts to produce an EQS induction generator have, until now, not been successful [8]. This paper describes the first successful creation and operation of an EQS generator that outputs net electrical power. The fundamental challenge of EQS generators (as compared to EQS motors) is their sensitivity to parasitic capacitances between the point at which power is generated and the point at which that power is used. If losses caused by the presence of parasitic capacitance within an EQS generator exceed its generating capacity, its power output falls to zero. In contrast, those same losses in an EQS motor would take place before the point at which power is converted to the mechanical domain. Thus, they would reduce motoring efficiency but not the ultimate mechanical output power. The generator demonstrated in this paper is therefore similar but not identical to the motors described in [6] and [7]. Key differences in the structure and operational protocol that enable generation in the present device are presented in this paper along with the power-generation results that they enable. 1057-7157/$25.00 © 2009 IEEE STEYN et al.: SELF-EXCITED MEMS ELECTRO-QUASI-STATIC INDUCTION TURBINE GENERATOR 425 Fig. 2. Three-dimensional section view of the turbine-generator device. The generator stator electrodes are shown arranged like spokes on a wheel. The connections of three of the six phases to the external bonding pads (stator leads) are shown. Fig. 1. Essence of an EQS induction machine. Basic six-phase machine consists of a stator with a set of electrodes arranged such that every sixth electrode is connected. Sinusoidal voltages on the six electrode sets, phased 60◦ apart, produce a traveling wave. This in turn induces a traveling potential wave on the rotor—a high-resistivity polysilicon film in this case. II. D EVICE L AYOUT AND O PERATION Fig. 1 shows a description of the essence of the EQS induction machine. The machine comprises a spinning rotor and a fixed stator that interacts with the rotor. On the surface of the stator is an array of radial electrodes; every sixth stator electrode is connected to form one phase in a six-phase machine. The rotor in the present machine is coated with a highresistivity polysilicon film. Sinusoidal voltages on the six phases, phased 60◦ apart, produce a traveling stator potential wave. The traveling potential wave on the stator induces a traveling potential wave on the rotor. If the rotor spins slower than the stator potential wave, the machine operates as a motor. If the rotor spins faster, as shown in Fig. 1, it operates as a generator. Correspondingly, the combination of the rotor film’s high resistivity and its stray capacitance causes the rotor potential to lag/lead the stator potential during motoring/generating operation. In the present machine, the backside of the rotor disk (away from the stator) is occupied by a set of turbine blades that spin the rotor when they are supplied with pressurized air [1]. During operation, the rotor is supported on air bearings [9], [10]. The actual device is shown in schematic 3-D cross section in Fig. 2. The device structure is similar to devices presented earlier [4], [6], [7]. In Fig. 2, the first layer L1 provides connections for the main air supply of the turbine and for the air supply to the front thrust bearing. Layer L2 includes the turbine main air-distribution manifold and the air-injection ports for the front thrust bearing. The turbine rotor and stator blades are formed on the topside of L3; the rotor itself and the journal bearing that supports it in the plane of the wafer are also defined in L3. The bottom of L3 has the rotor film for the induction machine on top of a thick (10 μm) oxide island. Layer L4 is the stator, with 786 platinum electrodes arranged in 131 interleaved groups of six electrodes on top of a thick (approximately 20 μm) oxide island. L4 also includes the air-injection ports for the rear thrust bearing. L5 provides air connections for the rear thrust bearing Fig. 3. Schematic cross-sectional view of a section of the generator, showing the stator electrodes fabricated on the oxide-coated stator substrate along with one of the six concentric interconnection rings that groups the electrodes together into six phases. Also shown is the rotor film on its oxide-coated rotor substrate. The potentials Vi are indicated on the six stator phases; the phase interconnection for phase four is shown. Fig. 4. Photograph of the generator device. and the journal bearing of the turbine. Fig. 3 shows a schematic cross section of one section of the rotor/stator structure, and Fig. 4 shows a photograph of the actual device. III. F ABRICATION The generator is fabricated using standard integrated-circuit fabrication techniques, deep reactive-ion etching (DRIE), and multiwafer bonding. Its fabrication is similar to that described in [6] and is identical to that described in [7], except for the high-speed bearing fabrication and the preparation for and execution of the full-stack wafer bond, which are described as follows. Layers L1, L2, and L5 are fabricated primarily through double-sided DRIE. Layer L4 combines those processes with the fabrication of the electrical features of the stator; L3 is 426 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 18, NO. 2, APRIL 2009 similar but includes the additional challenge of etching the deep high-aspect-ratio journal-bearing feature as in [11]. A. High-Speed Bearing Fabrication The journal bearing in L3 is designed to be 20 ± 1 μm wide and 300 μm deep. The results presented in this paper are from a device with a 21 ± 1 μm wide and approximately 300-μmdeep bearing. Good control of the journal-bearing fabrication enables rotational speeds of up to 850 krpm to be reached. The speed limit is attributed to destabilizing forces resulting from air-gap nonuniformities caused by deviations from a flat stator topography. B. Rotor and Stator Fabrication With Thick PECVD Oxide Films Thick silicon-dioxide layers are required in Layers L3 and L4 to minimize parasitic capacitive coupling to the rotor and stator backplanes. To minimize wafer bow, the insulator oxide is limited to buried islands where it is needed as in [12]. The rotor oxide is a 10-μm-thick plasma-enhanced chemicalvapor-deposition (PECVD) film, and the stator oxide is an approximately 20-μm-thick PECVD tetraethyoxysilane film. The 0.5-μm-thick polysilicon rotor film is deposited by lowpressure CVD and doped with boron by ion implantation. The stator fabrication process is described further in [7]. The bond between L3 and L4 is a fusion bond between PECVD oxide and silicon. The PECVD oxide is smoothed by repeated oxide deposition and chemical–mechanical polishing until a bondable surface is obtained as demonstrated by successful contacting to a test wafer. After smoothing, L4 is annealed at 750 ◦ C for 3 h to outgas the PECVD oxide, while ensuring that the platinumstator-electrode structure does not develop hillocks. Only then does the fabrication of L4 proceed to the etching of the throughwafer flow paths and to bonding. The bondability-verification procedure with known good blank wafers, combined with rework where needed, enables the bonding of the five-layer wafer stack with 100% bond yield. The final room-temperaturebonded stack is room-temperature pressed with a uniform load of 3 kN for 24 h, then placed under a uniform thermal press of 3 kN for 4 h at 500 ◦ C, and finally, annealed at 700 ◦ C for 22.5 h. IV. C ONCEPTS FOR S YSTEM A RCHITECTURE , M ODELING , AND O PERATION When the rotor of the EQS device spins faster than the traveling speed of the stator’s potential wave, power is transferred from the rotor to the stator. However, if the power lost in the rest of the generator system (including its parasitics and external electronics) exceeds the power sourced by the rotor–stator interaction, the generator system as a whole will fail to generate net power. Therefore, to demonstrate a complete generator system, one must first choose the system’s mode of operation, design the external electronics that enable that operation, and model the complete system in detail to ensure that it can be tuned to operate as intended. Two different system architectures and modes of operation were used in the course of the experiments described here. Fig. 5. strays. Machine model that includes the resonant electronics and the external The two operational modes have related (but not identical) challenges and require related (but not identical) models to understand and address those challenges. In the first operating mode, the six generating phases of the machine were each driven externally at a common frequency but with phase-specific driving voltages and phase shifts. Driven excitation could be implemented by driving each of the generator phases directly with electronics based on MOSFET inverters, but the power consumption of the external electronics would exceed the power produced by the EQS generator and prevent net generation of electric power. Instead, electronics were created to drive the generator phases through a combination of low-level drive voltages and resonant circuits to amplify the voltages onto the generator phases. A single phase of the EQS machine, its parasitics, and the resonant external electronics are shown in Fig. 5. The inductor resonates at the design electrical frequency with the combined machine capacitance and external stray capacitance, and RSi represents the load. The parasitics shown in Fig. 5 are described further in Section V. The second operating mode employed a self-excited architecture. The electronics for this approach are similar to those shown in Fig. 5, except that the driving voltages VDi are each replaced by a connection to ground. The resonant circuits formed by the generator, the stray capacitance, and the inductors self-excite the phase voltages by amplifying small fluctuations. As long as the rotor speed is set to an appropriate value above the speed of the stator’s traveling wave and the stator’s parasitics are sufficiently small and well balanced, the resonance automatically selects the optimal phase voltages, phase shifts, and electrical frequency. Both driven- and self-excitation require that the parasitic capacitance and the wiring resistance be minimized to reduce ohmic losses. Both operating modes also require that the imbalance among the generator phases be minimized, so that the capacitance of each generator phase to ground and to the other generator phases is as uniform as possible. Finally, each mode requires that some set of electronic parameters be optimized in order to ensure that net power is generated. These requirements were met through a combination of the following features: 1) complete system modeling; 2) careful design and layout of the external electronics; and 3) detailed parameter estimation to ensure the accuracy of the models. The externally driven approach requires a steady-state sinusoidal model that predicts power output; parameters can be chosen by optimizing for maximum power output. The self-excited approach requires an eigenvalue analysis; parameters can be chosen by maximizing STEYN et al.: SELF-EXCITED MEMS ELECTRO-QUASI-STATIC INDUCTION TURBINE GENERATOR 427 the real part of the eigenvalues. A detailed state-space model was used to meet the requirements of both sets of experiments; for the experiments under driven excitation, the system was analyzed with sinusoidal driving voltages. The model is described in Section V. Detailed parameter-estimation experiments were performed to determine the values of the internal and external stray capacitances and also to characterize the external inductors. These experiments are described in Section VI. The driven and self-excited experiments are described in Sections VII and VIII, respectively. (RL ), and wire-insulation-dielectric (RC ) losses. This model accurately and successfully guided the experiments discussed in this paper and helped to determine the optimal drive voltages and phase angles VDi , i ∈ {1, 2, 3, 4, 5, 6}, whenever a driven excitation was used. Combining the lumped-parameter representation of the fields with the model representation of the strays yields a state-space model of the form V. D EVICE M ODELING In (1), the state vector X is comprised as [V p ψ0 ξ w VC iL ]T . The entries in the state matrices A, B, and C are obtained from the Fourier decomposition of the fields problem; they are given with derivation in [15]. VD is the drive-voltage vector. The rotor potential p has dimension 2n × 1, where n is the number of harmonics used to approximate the stator wave; we need twice the number of harmonics to account for the sine and cosine terms or, equivalently, the amplitude and spatial phase of each harmonic. We used 13 harmonics for all subsequent modeling of the generator, and therefore, p has dimension 26 × 1. The vector V represents the potentials on the six statorelectrode sets. ψ0 and ξ are scalars and represent the zeroth harmonic of the rotor conductor potential and the potential of the rotor backside. VC and iL are a potential in and the current through the inductor, respectively, as shown in Fig. 5. With V, w, VC , and iL all being of dimension 6 × 1 and p being of dimension 26 × 1, (1) represents a 52-state system. The machine casing is assumed to be at zero potential. The concepts described in Section IV are implemented in a state-space model of the EQS machine. This model improves on the work by Nagel [4] and Bart and Lang [13] that addressed mainly the rotor- and stator-field analyses and assumed balanced operation in which all the phases of the electric machine can be treated as being identical. The model presented here treats each of the six machine phases as unique and also accounts for unique coupling between the various phases. The first step in developing the machine model is to represent the fields within the rotor–stator structure in a lumpedparameter form. The field and circuit theories developed in this paper then feeds into the state-space model. We use the same starting point that was used in [4] and [13] and approximate the polar machine, shown in Fig. 1, as a wrapped linear induction machine. The Cartesian representation of the wrapped linear induction machine is shown in Fig. 3, where φ(t) represents the potential on the stator and ψ(t) represents the potential on the rotor. The turbine disk is assumed to be grounded through a capacitor CB . A Fourier decomposition of the stator potential, in turn, provides a Fourier representation of the rotor potential and the air-gap electric fields as well as the fields in the rotor and stator half-spaces. Once the fields on both sides of the stator electrodes are obtained, one can use Gauss’s law to solve for the charge on the stator electrodes, thereby obtaining a current–voltage relationship for the machine in lumped-parameter format. The next step is to add the relevant internal and external parasitic stray components, as shown in Fig. 5. The model includes the inductors that make this a resonant system and the coupling to the other phases. In Fig. 5, CSij are components of the general nonsymmetric stator capacitance matrix and include contributions from the six stator interconnect rings, the six stator leads, and the six bond pads. The subscripts i and j refer to phases one through six. The contributions of the interconnect rings toward CSij were obtained by simulating a portion of the stator structure in FastCap, a boundary-elementmethod capacitance solver [14]. The contributions of the stator leads were computed using parallel-plate theory. Rmi are the combined resistances of the leads, interconnect wiring, vias, and electrodes of the stator itself. CEij are components of the general nonsymmetric external stray-capacitance matrix. CEij were modeled using FastCap. To obtain an accurate model, it was necessary to reestimate this matrix using experimental techniques. Each of the six inductors attached to the machine was independently modeled to include core (RLp ), wiring AẊ = BX + CVD . (1) VI. C HARACTERIZATION AND P ARAMETER E STIMATION To excite the six-phase generator, six voltage amplitudes (VDi in Fig. 5), five phases shifts, the frequency, and the speed must all be specified. The generated power will depend on all 13 of these parameters. Finding the optimal operating point experimentally is impractical, so a model-guided computer-optimized approach is preferred. This requires an accurate model. Therefore, the model described in Section V must be calibrated to the actual machine and its experimental setup. The machine was characterized using a frequency-domain method. The generator was spun at 50 krpm. Each phase (VDi ) of the machine was in turn swept in frequency while the other phases were grounded. The resulting voltages wi (magnitude and phase) were recorded for all six phases. This experiment was repeated six times to give six data sets, each with a magnitude and phase response for all six phases, and therefore, 36 frequency-response functions (FRFs). The model was then compared to the experimental data, and a nonlinear least squares algorithm was used to adjust the external capacitance matrix CEij , the gap G, and the rotor conductivity σs to best fit the data. Fig. 6 shows a description of this parameter-estimation procedure. Fig. 7 shows the FRF for the case in which Phase 1 was the driven phase and all the others were grounded, together with the model result after fitting. Typical values for CEij , i = j and CEij , i = j were 4.5 and 0.5 pF, respectively. For the gap and the rotor conductivity, values of G = 4.2 μm and σs = 0.75 nS were obtained. 428 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 18, NO. 2, APRIL 2009 Fig. 6. Parameter-estimation flow diagram. Fig. 8. Individual per phase powers versus mechanical speed. Negative power indicates power generated, according to the sign convention that power entering the device is positive. The uncertainty in power is ±5 μW per phase. Fig. 7. Frequency response of all the phases under Phase-1 excitation. The measurement uncertainty in log(V ) is ±0.05. VII. E XPERIMENTS U NDER D RIVEN E XCITATION The first experiments performed on the EQS generator were driven-excitation experiments. Under these conditions, ’s in Fig. 5 were applied with six function generators the VDi (Agilent 33220A), one for each phase. A 1-kΩ current-sense resistor was placed in series with the voltage sources. The perphase real and reactive powers were measured with six Analog . The inductors for Devices AD835 multipliers at points VDi this application were JW Miller Model 4671 inductors with a nominal inductance of L = 8.2 mH, chosen to set the operating frequency to about 400 kHz. For these inductors, the following were measured to be the nominal model parameters: CL = 1.7 pF, RL = 40 Ω, RLp = 1 MΩ, and RC = 1 kΩ (see Fig. 5). To further reduce the capacitive loading on the device and make net generation possible, the voltage probes (Agilent 10440B 100:1 probes, 2.5 pF10 MΩ) used to measure wi during device characterization were removed for this experiment. Therefore, CEij , i = j was reduced from approximately 4.5–2 pF per phase. With the new CEij matrix, a nonlinear optimization was performed on the model to find the power-optimal drive-voltage amplitudes and phases, as well as the optimal speed and electrical frequency. All voltage magnitudes converged to their upper bounds for most optimizations. Fig. 9. Sum of the per-phase powers versus mechanical speed. Negative power indicates power generated, according to the sign convention that power entering the device is positive. The uncertainty in power is ±30 μW. A. Experiments at Moderate Levels of Excitation We found that the linear state-space model correlated well with measured power output when we used moderate-excitation levels on the generator phases. Fig. 8 shows the per-phase power–speed relationship obtained when exciting the machine with VDi = 1.6 Vpp and phase angles 0◦ , −30.4◦ , −110.1◦ , −195.7◦ , −217.7◦ , −254.6◦ at 402.9 kHz. The curve shown in Fig. 9 is the real power sum of all the phases. In the moderateexcitation regime, the calculated sensitivity of peak power to gap is about 400 μW per 1 μm of change in air gap. The good model agreement suggests that the gap is close to the previously fitted 4.15 μm. Negative power indicates generated power, which in this setup was returned to the function generators. A maximum of 108 μW ± 30 μW at 245 krpm was generated by the system—the generator and its external electronics—in this experiment. The corresponding voltages Vi on the electrodes and wi at the device’s external electrical connections were ∼ 30 Vpp . STEYN et al.: SELF-EXCITED MEMS ELECTRO-QUASI-STATIC INDUCTION TURBINE GENERATOR 429 B. Device Efficiency The electromechanical model described earlier can be used to estimate the power lost in various parts of the system and to calculate the mechanical to net electrical efficiency of the device. To determine the overall (airflow to electric) efficiency of the device, the losses in the turbine and air gap should also be taken into account; these are estimated from the measured pressures and mass flow rates, assuming that the flow through the turbine is adiabatic. At 245 krpm, a total of approximately 300 mW enters the turbine in the airstream. Turbine inefficiencies dissipate 219 mW. This is understandable, given that the design speed of the turbine is 2.4 Mrpm. A further 54 mW is dissipated in air-gap viscous losses, with 26.7 mW accounting for journal- and thrust-bearing viscous losses. The electromechanical model indicates that a negative torque of 0.03 μN · m is produced by the air-gap electric fields, corresponding to a mechanical power input to the generator of 790 μW at 245 krpm. Of this shaft power, 255 μW is lost in the rotor conductor, and 535 μW enters the external circuit shown in Fig. 5. Internal machine interconnect and wiring resistances account for 75 μW of the losses. The bulk of the power from the machine, 280 μW, is dissipated in the inductor core, with inductor wiring and the combination of dielectric and proximity losses accounting for 50 and 22 μW, respectively. These calculations yield an overall (airflow to electric) efficiency at peak power of 0.036%. The low efficiency as compared with macroscale systems reflects several factors, including operation of the turbine far from its design point and the power-limiting effects of carrier depletion in the rotor. The mechanical to net electrical efficiency at peak power is 14%, not including viscous losses in the machine’s air gap, or 0.2%, including the air-gap viscous losses. C. Results at Higher Levels of Excitation The power generated under driven excitation was also measured as a function of excitation voltage for excitations as high as 3 Vp−p . Fig. 10 shows the sum of the per-phase powers versus the driving voltage squared for four different excitation frequencies and the same set of phase angles as was employed previously and a speed of 235 krpm. As the excitation voltage increases, the generated power first increases and then decreases. For two of the excitation frequencies examined, the machine ceases to generate power. In contrast, the model from Section V predicts that the generated power should increase as the square of the excitation voltage. This trend is predicted to increase until the electric field across the air gap reaches the onset of breakdown at about an order of magnitude greater voltage than is experienced in these experiments. The discrepancy between predicted and measured performance is attributed to carrier depletion in the polysilicon rotor-conductor film, which effectively increases the air gap; the electrode/air-gap/rotor conductor structure is analogous to the metal/oxide/semiconductor structure in a MOS capacitor, in which the same phenomenon occurs [16]. The increase in effective air gap has two effects. First, since the effective air gap increases faster than the voltage for sufficiently large voltages [17], an increase in the electrode voltage can decrease the electric field in the gap and decrease Fig. 10. Comparison of fitted model and experimental results for large drive voltages. The measurement uncertainty is ±30 μW. the power output. Second, the increase in effective air gap decreases the device capacitance and modifies the system’s resonance. This is consistent with the measured dependence of power output on frequency and voltage. At low voltages, the power output is highest at 411 kHz, near the system’s resonant frequency, as expected. The higher voltage deviations from the model in Section V are also most pronounced at 411 kHz, because a given driving voltage produces the largest electrode voltage when it is applied at the resonant frequency. At lower excitation frequencies, the maximum power output occurs at progressively higher values of the driving voltage, consistent with the fact that the amplification of the driving voltage onto the stator electrodes is reduced off-resonance. The actual rotor gap remains constant to within 0.2 μm, as evidenced by the nearly constant air flows through the thrust-bearing gaps [10]; the depletion of the carriers in the rotor conductor was incorporated into the model described in Section V by replacing the actual air gap with an effective gap that is a function of electrode voltage. The voltage-dependent effective gap was used as a fitting parameter to fit the model to the measured data at increased drive voltages. A strongly saturating (tanh) fitting function was then fit to the effective gap to determine the saturation voltage. At higher voltages, as much as 0.7 μm of air-gap increase was needed to match the data. Because this air-gap increase is greater than the rotor conductor-film thickness, this indicates a fully depleted film and a change in rotor conductivity at higher voltages. The change in rotor conductivity was not accounted for in the model. The smooth curves in Fig. 10 represent the revised model, including the nonlinear rotor conductor. The maximum power output of the generator was 192 μW ± 30 μW, recorded at an excitation voltage of VDi = 2.4 Vp−p , which corresponds to voltages Vi and wi of about 50 Vp−p . VIII. E XPERIMENTS U NDER S ELF -E XCITATION The performance of the machine was also measured under self-excitation. All external driving voltages VDi were set to zero, and the mechanical speed was increased until the phase 430 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 18, NO. 2, APRIL 2009 Fig. 12. Self-excited EQS generator produced enough power to light up four high-intensity LEDs on two of the phases, phases 4 and 5. The LEDs on phase 6 remain unlit. Picture taken with a Canon Powershot A80 camera in a dimly lit room. Exposure time: 15 s, ISO speed 100, and f-stop 3.2. Fig. 11. Eigenfrequency and average voltage magnitude versus speed for the generator as measured at points wi defined in Fig. 5. The scatter in the frequency data is attributed to small variation in the airgap due to presumed nonidentical thrust-bearing pressures at the same operating speed. The uncertainty in frequency is ±1 kHz and in voltage is ±0.1 V. voltages were self-excited through the amplification of random fluctuations; in this mode of operation, the generator and its load are electrically unstable. The model was first used to predict the conditions (rotational speed and presence of various measurement probes) under which self-excitation was expected to occur. This was determined by examining the real parts of the eigenvalues of the matrix A−1 B from (1). At speeds greater than 215 krpm, the machine self-excited. Fig. 11 shows the measured and predicted eigenfrequency over the range of rotational speeds at which the frequency was measured. The results show good agreement with the model, with the eigenfrequencies very close to the predicted values. Self-excitation begins close to the predicted rotational speed for onset of selfexcitation; its end at about 300 krpm reflects the fact that no measurements were made at higher rotational speeds because of bearing-stability concerns. The voltages wi were measured using low-capacitance probes, and the average voltage magnitude is plotted versus speed in Fig. 11. The maximum voltage wi attained under self-excitation was approximately 29 Vp−p and is limited by rotor-conductor depletion as described earlier; the models indicate that the value of Vi is essentially the same as that of wi . All of the power generated by the EQS generator in these experiments was dissipated in the wiring resistances and the inductors. In separate experiments, the self-excited generator also provided power to an external load in the form of four highbrightness LEDs. Two LEDs were coupled to the points wi on each phase through adjustable coupling capacitors. The coupling capacitances were adjusted until the LEDs on phases four and five lit up; the LEDs on the other phases remained unlit. Fig. 12 shows a photograph of the LEDs lit by the generator. With the visibility threshold of each LED at approximately 35 μW, the power delivered to the four LED load is estimated at 140 μW total. IX. S UMMARY AND C ONCLUSION The results reported here demonstrate the feasibility of an EQS induction micromachine as a generator. The amount of power it can generate is limited by its external and internal stray capacitances. It is therefore necessary to minimize all the strays and accurately model those that remain along with their imbalance. This also motivated the use of thick (10–20 μm) PECVD oxide films for electrical isolation in the present device. Power generation could only be attained at high operating speeds (> 200 krpm) and with a metal stator for lowresistive losses. Our generator addressed these requirements by being the first fully bonded MEMS EQS induction machine with a metal stator structure. This stator structure had a PECVD oxide surface as a bond surface, where surface bondability was achieved after the fabrication of that layer. A precise 20-μm-wide 300-μm-deep journal bearing allowed for stable operation beyond the synchronous speed. These or similar techniques to control the parasitic capacitances, to limit resistive losses, and to ensure high-speed operation are expected to be key components of creating effective EQS machines in the future. Creating an inherently phase-balanced machine design and employing accurate detailed models of its performance were also shown to be key aspects of successful operation of the EQS machine as a generator. A well-balanced machine is important because it minimizes electric losses and operational challenges of understanding and correcting for the machine’s fundamental imbalances. Accurate modeling, including inductor models that are accurate over a wide frequency range, enables the total model to be used to guide generator operation. This paper demonstrates that it is possible to estimate the external strays and other parameters—in this case, a total of 23—from a suitable set of frequency-domain experiments using a nonlinear least squares technique. The calibrated models are in agreement with the measured results, and similar models with parameters calculated from first principles can be used to design future devices of this kind. The device presented here was operated for tens of hours over a speed range of 200–300 krpm and a drive-voltage range of VDi = 0−3 Vpp . No degradation in performance was observed. The maximum power recorded was 192 μW. At VDi = 1.6 Vpp , 108 μW was produced, and the machine terminal voltages wi were ∼ 30 Vpp . At the terminals, 535 μW was generated. At both high-driven voltages (wi ∼ 50 Vp−p ) and under selfexcited operation, a significant power-limiting carrier-depletion effect was identified. Scaling to the design conditions, using the linear model and assuming that a better rotor conductor can be STEYN et al.: SELF-EXCITED MEMS ELECTRO-QUASI-STATIC INDUCTION TURBINE GENERATOR found, it is predicted that, with a terminal voltage of 600 Vpp , at 900 krpm, this device could produce 0.5 W at the terminals at an overall device efficiency of 15%. In addition, the possibility of self-excitation, without the need for external drive electronics, greatly simplifies its practical implementation for powergeneration applications. The results obtained in this paper suggest that EQS generators have significant potential for integration into MEMS gas turbine generators that output mechanical power at the several-Watt scale. The present EQS generator’s power output, while small, has the potential for a three orders of magnitude increase with the introduction of a second-generation linearrotor conductor, making it consistent with the power scales of MEMS engines. Although MEMS magnetic generators are expected to offer somewhat higher powers and somewhat higher efficiencies than future EQS generators (of order 60% rather than 45% for the best projected EQS generators) [18], the EQS generator offers important advantages. The fabrication of the EQS generator is fundamentally CMOS compatible, and integration of the generator into a silicon engine will not require the creation of additional technologies beyond a secondgeneration rotor conductor and the technologies demonstrated in this paper. Additional advantages include the small size of the EQS device, the use of a light rotor that is readily supported by air bearings, and compatibility with the high temperatures in a fuel-burning engine. [7] C. Livermore, A. R. Forte, T. Lyszczarz, S. D. Umans, A. A. Ayon, and J. H. Lang, “A high power MEMS electric induction motor,” J. Microelectromech. Syst., vol. 13, no. 3, pp. 465–471, Jun. 2004. [8] T. L. Willke, “Self-excited electrostatic generator,” M.S. thesis, MIT, Cambridge, MA, 1968. [9] L. G. Fréchette, S. A. Jacobson, K. S. Breuer, F. F. Ehrich, R. Ghodssi, R. Khanna, C. W. Wong, X. Zhang, M. A. Schmidt, and A. H. Epstein, “High-speed microfabricated silicon turbomachinery and fluid film bearings,” J. Microelectromech. Syst., vol. 14, no. 1, pp. 141–152, Feb. 2005. [10] C. J. Teo, “Bearings for microturbomachinery,” Ph.D. dissertation, MIT, Cambridge, MA, 2005. [11] H. Q. Li, N. Savoulides, L. Ho, S. A. Jacobson, R. Khanna, C.-J. Teo, L. Wang, D. Ward, A. H. Epstein, and M. A. Schmidt, “Fabrication of a high speed microscale turbocharger,” in Proc. Hilton Head Solid State Sens., Actuator Microsyst. Workshop, Hilton Head Island, SC, Jun. 2004, pp. 258–261. [12] R. Ghodssi, L. G. Fréchette, S. F. Nagle, X. Zhang, A. A. Ayon, S. D. Senturia, and M. A. Schmidt, “Thick buried oxide in silicon (TBOS): An integrated fabrication technology for multi-stack waferbonded MEMS processes,” in Proc. 10th Int. Conf. Solid-State Sens., Sendai, Japan, Jun. 1999, pp. 1456–1459. [13] S. F. Bart and J. H. Lang, “An analysis of electroquasistatic induction micromotors,” Sens. Actuators, vol. 20, no. 1/2, pp. 97–106, Nov. 1989. [14] K. Nabors, S. Kim, J. White, and S. D. Senturia, FastCap User’s Guide. Cambridge, MA: Res. Lab. Electron., MIT, 1992. [15] J. L. Steyn, “A microfabricated ElectroQuasiStatic induction turbinegenerator,” Ph.D. dissertation, MIT, Cambridge, MA, 2005. [16] S. M. Sze, Semiconductor Devices, Physics and Technology. Hoboken, NJ: Wiley, 1986. [17] R. F. Pierret, Semiconductor Device Fundamentals. Reading, MA: Addison-Wesley, 1996. [18] S. Das, “Magnetic machines and power electronics for power MEMS applications,” Ph.D. dissertation, MIT, Cambridge, MA, 2005. J. Lodewyk Steyn received the B.Eng. degree in mechanical engineering from the University of Pretoria, Pretoria, South Africa, in 1998, and the S.M. and Ph.D. degrees from Massachusetts Institute of Technology, Cambridge, in 2002 and 2005, respectively. He is currently the Lead MEMS Design Engineer with Pixtronix Inc., Andover, MA, where he is currently developing low-power MEMS-based directview displays for portable electronic devices. He has a keen interest in engineering in general and, in particular, in MEMS design, fabrication, and ACKNOWLEDGMENT The authors would like to thank T. Lyszczarz, J. Yoon, and the late T. Forte of the MIT Lincoln Laboratory for their contributions to the fabrication of the stator. They would also like to thank C. J. Teo for operating the spinning generator at speeds up to 850 krpm. They would also like to thank G. Donahue for the journal-bearing process development. They would also like to thank C. Law for providing the 3-D solid models and rendering under the MIT UROP program. All microfabrication was performed at the MIT Microsystems Technology Laboratories (MTL) and at the Micro Electronics Laboratory (MEL) at MIT Lincoln Laboratory. Opinions, interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by the United States Government. R EFERENCES [1] A. H. Epstein, “Millimeter-scale, micro-electro-mechanical systems gas turbine engines,” J. Eng. Gas Turbines Power, vol. 126, no. 2, pp. 205– 226, Apr. 2004. [2] E. R. Mognaschi and J. H. Calderwood, “Asynchronous dielectric induction motor,” Proc. Inst. Elect. Eng.—Sci., Meas. Technol., vol. 137, no. 6, pp. 331–338, Nov. 1990. [3] B. Bollée, “Electrostatic motors,” Philips Tech. Rev., vol. 30, no. 6/7, pp. 178–194, 1969. [4] S. F. Nagle, “Analysis, design and fabrication of an electric induction micromotor for a micro gas-turbine generator,” Ph.D. dissertation, MIT, Cambridge, MA, 2000. [5] S. F. Nagle, C. Livermore, L. G. Fréchette, R. Ghodssi, and J. H. Lang, “An electric induction micromotor,” J. Microelectromech. Syst., vol. 14, no. 5, pp. 1127–1143, Oct. 2005. [6] L. G. Fréchette, S. F. Nagle, R. Ghodssi, S. D. Umans, M. A. Schmidt, and J. H. Lang, “An electrostatic micromotor supported on gas-lubricated bearings,” in Proc. MEMS, Interlaken, Switzerland, Jan. 2001, pp. 290–293. 431 characterization. Sam H. Kendig received the S.B. degree in electrical engineering from Massachusetts Institute of Technology, Cambridge, in 2006. He is currently with Bose Corporation, Framingham, MA, where he is developing homeentertainment products. His interests include design of MEMS and analog electronics. Ravi Khanna received the B.S. degree in physics in 1983 and the M.S. and Ph.D. degrees in electrical engineering in 1992 from Pennsylvania State University, University Park. From 1992 to 1998, he was a Senior Device and Process Engineer with Skyworks Solutions, Woburn, MA, where he oversaw the design, fabrication, and testing of microwave-semiconductor components. From 1998 to 2004, he enjoyed the challenges of MEMS process development and implementation on the MIT Micro-Engine and MicroRocket programs. He is currently with BAE Systems, Lexington, MA, where he is working in both production and development of infrared imaging systems. 432 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 18, NO. 2, APRIL 2009 Stephen D. Umans (S’67–M’75–SM’82–F’95) was born in Cleveland, OH, on November 10, 1948. He received the S.B. and S.M. degrees in electrical engineering and the Sc.D. degree in electrical engineering from Massachusetts Institute of Technology (MIT), Cambridge, in 1972 and 1976, respectively. Recently, he was a Principal Research Engineer with the Electromechanical Systems Laboratory and the Electrical Engineering and Computer Science Department, MIT. He is currently an independent Consultant, with specific focus in the areas of electromechanics, electric power systems, and electric machinery. He is the coauthor of the textbook Electric Machinery (McGraw-Hill, 1990). Jeffrey H. Lang (F’98) received the S.B., S.M., and Ph.D. degrees from the Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology (MIT), Cambridge, in 1975, 1977, and 1980, respectively. Since 1980, he has been with the faculty of MIT, where he is currently a Professor of electrical engineering and computer science. From 1991 to 2003, he was the Associate Director of the MIT Laboratory for Electromagnetic and Electronic Systems. His research and teaching interests focus on the analysis, design, and control of electromechanical systems with an emphasis on rotating machinery, microscale (MEMS) sensors, actuators and energy converters, and flexible structures. He has written over 200 papers and holds 12 patents in the areas of electromechanics, MEMS, power electronics, and applied control. He is the coauthor of Foundations of Analog and Digital Electronic Circuits (Morgan Kaufman, 2005). Dr. Lang was an Associate Editor of Sensors and Actuators from 1991 to 1994. He was the recipient of four Best Paper Prizes from IEEE Societies. He is a former Hertz Foundation Fellow. Carol Livermore received the B.S. degree from the University of Massachusetts, Amherst, in 1993 and the A.M. and Ph.D. degrees from Harvard University, Cambridge, MA, in 1995 and 1998, respectively, all in physics. From 1998 to 2002, she was a Postdoctoral Associate and then a Research Scientist with Massachusetts Institute of Technology (MIT), Cambridge. Since 2003, she has been on the faculty of MIT, where she is currently an Associate Professor with the Department of Mechanical Engineering. Her research interests include power MEMS and the development of techniques and applications for nano- and microscale self-assembly.