Enabling MEMS Technologies for Communications Systems

Victor M. Lubecke*a, Bradley P. Barber**b, and Susanne Arneya

aBell Laboratories, Lucent Technologies, Murray Hill, NJ, 07974 USA

bAgere Systems***, Murray Hill, NJ, 07974 USA

ABSTRACT Modern communications demands have been steadily growing not only in size, but sophistication. Phone calls over copper wires have evolved into high definition video conferencing over optical fibers, and wireless internet browsing. The technology used to meet these demands is under constant pressure to provide increased capacity, speed, and efficiency, all with reduced size and cost. Various MEMS technologies have shown great promise for meeting these challenges by extending the performance of conventional circuitry and introducing radical new systems approaches. A variety of strategic MEMS structures including various cost-effective free-space optics and high-Q RF components are described, along with related practical implementation issues. These components are rapidly becoming essential for enabling the development of progressive new communications systems technologies including all-optical networks, and low cost multi-system wireless terminals and basestations. Keywords: MEMS, RFIC, BAW, acoustic, resonator, inductor, MARS, micro-mirror, cross-connect, modulator

1. INTRODUCTION The growing demand for larger more capable communications networks and smaller less expensive hardware has inspired many MEMS research efforts to bring to development new technologies that can enable revolutionary new approaches in systems and production. In this paper, various Bell Labs MEMS research efforts for addressing needs in Lucent Technologies, and now also Agere Systems***, wireless and optical network systems are reviewed. In wireless networks, there is a pressing need for extremely compact, low cost, and power efficient radio circuitry that can enable improved wireless handsets and other terminals, and new network infrastructures. While the size, performance, and cost benefits of silicon integrated circuit technology has been widely exploited in baseband circuitry for mobile phones and basestations, RF front-end circuitry has remained heavily dependent on large discrete passive components, particularly for resonant functions where inefficient integrated components would seriously degrade performance [1]. These circuit applications include resonators for low phase-noise voltage controlled oscillators (VCO’s), filter components, and reactive impedance matching elements. Many RF resonator functions can be performed with conventional inductor-capacitor (L-C) circuits. These are passive components that store energy in localized electromagnetic fields. However, integrated-circuit versions of these components, particularly inductors, are generally severely limited by parasitic losses. In order to minimize ohmic loss, thin-film spiral inductors are invariably large, and thus account for the majority of circuit area in typical RF integrated circuits (RFIC’s). Furthermore, when combined with the high conductivity of RFIC substrates, this broad, flat geometry results in a large parasitic capacitance that limits both quality factor (Q), and self-resonance frequency (SRF). Loss reduction techniques such as local substrate removal and vertical construction can help, but not without imposing fabrication and compatibility issues [2,3]. Through micro-electromechanical systems (MEMS) techniques, inductors can be made which minimize this loss mechanism through three-dimensional self-assembly [4]. The technique also allows for the creation of variable inductors that are not subject to the same constraints as those achieved through active circuitry [5,6]. Various MEMS inductors with Q values greater than 13 and inductance variations exceeding 18% are

* lubecke@lucent.com; phone +1 908 582-1587; fax +1 908 582-4941; http://www.lucent.com; Bell Laboratories, Lucent Technologies, Murray Hill, NJ, 07974 USA. ** bbarber@agere.com; phone +1 908 582-2751; fax +1 908 582-4228Agere Systems, Murray Hill, NJ, 07974 USA. *** Agere Systems was formerly known as the Microelectronics division of Bell Labs, Lucent technologies.

presented, with clear potential for even better performance. These inductors are well suited to integration in RFIC VCO’s and low noise amplifiers (LNA’s) [7]. Single element RF resonators are attractive for wireless handset front-end duplex filters. Often discrete ceramic resonators are used for high Q values, with the caveat of large size. Surface acoustic wave (SAW) filters offer a more compact option yet typically provide reduced quality performance. Alternatively, bulk acoustic wave (BAW) resonators can be used for many RF filter and oscillator applications. In bulk-mode acoustic filters, energy is stored as both electromagnetic and mechanical energy through piezoelectric coupling. Because the velocity of sound in the piezoelectric layer is much smaller than that of light, these devices can be much smaller than conventional high performance cavity resonators while maintaining improved power handling and frequency capability over SAW devices. Various BAW resonators and filters are described, demonstrating excellent Q values up to 1000. These BAW resonator technologies are also well suited for integration with RFIC’s [8]. In optical networks, there is a need for faster traffic routing and load balancing systems, and low cost home and central office hardware that can enable higher capacity networks, with improved bandwidth and data delivery to a myriad of users. Optical fiber systems are the backbone of modern communication systems. However, in most cases it is not fiber that brings high performance services to the end user. Typically, some handoff takes place between the optical network and conventional cost-effective systems already in place in homes and businesses, like telephone lines and coaxial cable links. Another major bottleneck occurs every time an optical path needs to be rerouted. Typically, data carried by optical fibers must be converted to electrical signals for switching functions, and then converted back. This greatly slows and limits traffic. One important element needed to realize the promise of fiber to the home, is a cost effective end user optical interface that can provide upstream data to the network. Passive modulators offer an attractive alternative to active laser or LED sources for reasons of maintenance of wavelength coherence, optical device reliability, temperature insensitivity, and above all, cost [9]. In such a system, the same light transmitted downstream from the central office to the home optical network unit, is over-modulated and looped back upstream to the central office. These systems have the advantage of keeping expensive high-performance optical hardware at the central office where the cost can be shared by many users. For these systems to be practical, the home modulator must be kept affordable. A MEMS based optical modulator capable of high data rates (>10 Mbit/sec) and good optical contrast (>20db) is described here. The heart of this unit is a voltage controlled reflective component known as a Mechanical Anti-Reflection Switch (MARS). This component has also been demonstrated as a highly effective variable attenuator and spectral power equalizer. Another MEMS reflecting device has also shown excellent promise for high capacity traffic routing. Arrays of such devices have been demonstrated for direct switching of optical signals from fiber to fiber in a manner that is independent of bit rate, protocol, and format of the data [10].

2. MEMS RF INDUCTORS When placed on a low-resistivity substrate, planar inductor geometries present both a desired inductance and a parasitic capacitance. Reducing the size of the conducting elements brings about a reduction in parasitic capacitance, but also increases the resistive loss. The approach presented here for improving this situation involves lifting the structure off the substrate plane to reduce the capacitance without increasing the resistance. The inductors assemble by means of an interlayer stress that causes portions of the inductor to bend away from the substrate in a controllable manner [11,12], as shown in Fig. 1(a). The fabrication process involves conventional silicon surface micromachining techniques that allow batch processing, as shown in Fig. 1(b), and can potentially be integrated with electronic circuits. A high performance variable inductor can be formed by creating a structure that remains sufficiently isolated from the substrate at all operating temperatures, yet incorporates mutually coupled current-carrying members that move with respect to each other with varying temperature, thus affecting the mutual component of the total inductance. The temperature variations needed to actuate such a structure can be environmental, or localized joule heating effects induced by an applied DC current [4]. One such variable inductor is shown in Fig. 2(a). The inductor consists of two loops that assemble themselves above the substrate, with a relative angle between them that can be thermally controlled.

The differential motion results from a cross-member corrugation structure in the inner loop that causes it to bend with temperature at a different rate than the outer loop. The conductors were ~50µm wide, separated by a ~20-µm gap, and the longer loop was about 1200µm long. The pitch between anchor pads was 150µm. The same self-assembly technique has been used to make fixed value inductors with high Q and SRF values [4]. Limiting parasitics for these structures occur near the anchor pads where the loop structure remains close and roughly parallel to the substrate, and the performance for these inductors is subject to change when exposed to physical shock and thermal variations that cause the structure to flatten towards the substrate. An improved variation on this design shown in Fig. 3(a). This hairpin inductor is attached to the substrate by hinges rather than anchor pads and warping elements are used to assemble the inductor into a locking semi-vertical position that changes very little with subsequent temperature changes. The hinges can be further fixed in place by electroplating the structure, which can also create a dependable ohmic connection to a circuit. The conductors were ~50µm wide, and the loops about 1200µm long. The pitch between attachment points was 150µm. Fabrication of the fixed and variable inductors was performed through photolithographic techniques. The structures were formed as a Cr-Au layer (0.5µm+) over a polysilicon layer (1.5µm), patterned on a sacrificial oxide layer (2µm) over the substrate, with a final etch-release/self-assembly step to achieve the desired three-dimensional structures [4]. The resulting (non-hinged) structures were resilient, springing back after mechanical probing, and the designs were repeatable.

Sacrificial oxide

Metal Polysilicon

High-σ Si substrate

Parasitic C

(a) (b)

Fig. 1. Self-assembling inductor. When released, interlayer stress causes the inductors to bend away from the substrate and reduce parasitic capacitance (a). SEM micrograph shows various inductors simultaneously assembled (b).

2 3 4 5 6 7 8f, GHz

0

1

2

3

4

5

6

Q

0.6

0.7

0.8

0.9

L, n

H25C, Q25C, L75C, Q75C, L150C, Q150C, L200C, Q200C, L

∆∆∆∆L >18%

∆∆∆∆Q <10%

(a) (b)

Fig. 2. Self-assembling variable inductor. When heated, mutually coupled loops (about 1200µm long) bend at different rates to allow controlled variation of inductance as loops deform with temperature (a). Q values remained fairly stable around 5 (<10%variation), while inductance values varied over an 18% (22% referenced to the heated extreme) range (b).

Fabrication was carried out using the Cronos Multi-User MEMS process (MUMPS) [13]. While this process was suitable for demonstrating the inductor concept, it imposes unnecessary limits on inductor performance. The substrate has higher conductivity (1 W-cm) than needed for an RFIC (up to 10 W-cm), which results in lower Q. Another significant limitation is the single metal layer in this process, which is too thin to minimize ohmic losses (less than one skin depth), and limits inductance to low values by restricting designs to a single turn (no bridge layer). The process however, is convenient, widely used, and allows for the demonstration of an effective technique for fabricating MEMS-first integrated RF inductors. The process provides a reasonable representation of RFIC demands, provides the basic features of promising embedded MEMS approaches [14], and can be readily modified to remove the aforementioned limitations. Various self-assembling inductors were fabricated and their scattering parameters measured using a Cascade-Microtech Microchamber probe station with ambient temperature control, and an HP 8510B network analyzer. The performance of a variable inductor is shown in Fig. 4. Frequency-swept measurements were made for temperatures ranging from 25˚C to 200˚C. At room temperature, the outer loop of the inductor stood at an angle of about 45 degrees, and the inner loop was bent even further (see Fig. 2(b)). As temperature was increased, both loops began to straighten out and flatten towards the substrate at different rates. In this case, neither loop went totally flat against the substrate, even at 200˚C. In the vicinity of 4 GHz, Q values remained fairly stable around 5 (<10% variation), while inductance values varied from 0.67 nH to 0.82 nH, greater than 18% (22% referenced to the heated extreme). The values were repeatable as the temperature was cycled. The ambient temperature for the inductors was varied here, but it is also possible to achieve this effect by directly heating only the inductor structure by applying a DC current. A comparison of two similar fixed-value inductors is shown in Fig. 3(b). Both were hairpin-shaped, but one was a simple warping structure attached to the substrate with flat anchor pads (similar to the inner loop in Fig. 2(a)), while the other was attached with hinges as previously described (see Fig. 3(a)). Measured inductance values were very similar for both (~1 nH), with a slightly higher value for the simple inductor due to the extra length of the anchor pads. At very low frequencies, Q values were similar, as the thin metal was the limiting factor in both cases. At higher frequencies though, the Q value was greatly improved for the hinged inductor, in excess of 13. Furthermore, very little change in inductance was observed for the hinged inductor at varying temperatures. While a simple hairpin design was useful for demonstrating the variations in Q and inductance described above, the absolute performance for this structure was not ideal. Even with only thin metal, wider inductors with a more circular or triangular shape have demonstrated Q values that rise more quickly with frequency and peak at levels greater than twice

0 5 10 15 20f, GHz

0

3

6

9

12

15

Q

0.0

0.5

1.0

1.5

2.0

2.5

L, n

H

Anchor pad, QHinge, QAnchor pad, LHinge, L

(a) (b)

Fig. 3 Self-assembling fixed-value inductors. Interlayer stress causes the legs to bend, raising the hinged hairpin-shaped structure (about 1200µm long) to a locking position for improved Q, SRF, and thermal/mechanical stability (a). Hinged structure showsimproved Q over warped structure, and can be further improved with thicker metal (b).

that of the hairpin. These structures could also be adapted in the same manner as the hairpin inductors described here. Previous analysis indicates that increasing the metal thickness for these inductors to about two skin depths (3µm at 2 GHz) should result in peak Q values greater than 20 [4]. An additional set of metal and sacrificial layers would also be useful for creating a conducting bridge or underpass, which would allow for the creation of multiple turn inductors with much higher inductance. These improvements could be easily achieved through custom MEMS processing, or post-processing depositions on a foundry processed wafer. Wire loop model simulations for variable inductors using Fasthenry [15] field solver software suggest the range of inductance variation for the two-loop configuration can be enhanced, by increasing the coupling between the loops. Reducing the gap between inner and outer loops (Fig. 2(a)) from 25 µm to 5µm results in the maximum inductance variation changing from of 27% (37% referred to the alternate extreme) to about 34% (51%), when the angle between the loops varies from 0 to 90 degrees. Substrate coupling was neglected for these simulations as both loops maintain a position above the substrate for the majority of travel.

3. BULK ACOUSTIC WAVE RESONATORS One of the most promising technologies for high-Q integrated wireless filters is based on the micromachining of piezoelectric films to form bulk acoustic wave resonators. A combination of sharp cut-off characteristics, compact size, and practical fabrication makes these resonators attractive for demanding applications, such as wireless PCS duplex filters used to separate closely spaced transmit and receive frequencies [16]. While ceramic resonator technology can meet challenging PCS specifications, smaller integrated solutions would allow for smaller and less expensive mobile handsets. Acoustic wave resonator technology can be used to this end. One key advantage in fabricating BAW resonators over SAW devices is that the critical frequency dependent dimensions are the thickness of the films, rather than lateral dimensions of planar lithography. Fig. 4(a) illustrates this difference [17].

Two methods for the construction of such resonators are illustrated in Fig. 4(b). The figure shows a membrane based resonator, with the piezoelectric film suspended over a cavity etched in the substrate [18]. The air-piezoelectric film boundary forms an optimum impedance discontinuity. An SEM photograph of an array of front-side etched membrane

b =λ/2a ~ 0.5b

+

-

Bragg Stack

SiO2

SixNy

SAW Device

d = λ/2 Etched Membrane

BAW Device (a) (b)

Fig. 4. Acoustic wave resonators. SAW devices depend on planar lithographic dimensions that become small with increasingfrequency and are thus more difficult to control than the film thickness parameters of a BAW device (a). Concept and side viewSEM photographs are shown of acoustic mirror type BAW resonator, using air discontinuity, and Bragg stack type, formed byalternating layers of different dielectrics to create a large impedance discontinuity below the piezoelectric film at the designfrequency (b).

resonators is also shown in Fig. 4(b). The figure also shows an alternative technique, with a piezoelectric film over an acoustic mirror made of alternating dielectrics layers [19]. The dielectric Bragg stack forms a large impedance discontinuity at the design frequency. An SEM photograph of an acoustic mirror type resonator made using layers of silicon dioxide and silicon nitride is also shown. The electrical response of a BAW resonator has a sharp impedance minimum (zero) near a high impedance peak (pole). The speed of sound and thickness of the piezoelectric film (AlN here) determine the zero’s frequency, and the zero-pole spacing is set by the electromechanical coupling k2. The impedance characteristics for a classic “T” cell structure is shown in Fig. 5(a), illustrating how series and shunt elements of slightly differing frequencies can be used to form a band pass filter.

From Fig. 5(a) it is evident that a wider separation between zero and pole allows for wider bandwidth (BW) filters. However, material constraints limit the available resonator range and thus additional circuit elements are sometimes required to achieve a satisfactory filter. For example, to cover the PCS band with a filter using an acoustic-mirror based filter, external series inductors can be required. (SiO2 & AlN Bragg stack) under an AlN piezoelectric film, external inductors would be required in series with the BAW resonators for full coverage of the frequency range [8]. One such filter was constructed using an SiO2/AlN Bragg stack under an AlN piezoelectric film and external series inductors. The performance is shown in Fig.5(b). The external inductors for such filters are on the order of 1 nH, which means that integration and size advantages can be lost unless high-Q integrated inductors can be used. While more difficult to construct, membrane based resonators can achieve a typical 50% increase in pole zero separation without inductors. Filters for GSM applications, with their large required bandwidth, can thus be constructed without additional passive elements. Other RF bands, e.g. 5.2 GHz LAN, may require both membranes and inductors to cover the bandwidth.

4. OPTICAL MEMS REFLECTORS MEMS techniques are well suited for realizing low cost optomechanical components that can solve many of the problems associated with modern dynamic optical networks. In particular, electrostatically controllable micromechanical reflective membrane structures can be used to provide a wide range of modulation, attenuation, and switching functions purely in the optical domain, and promise to radically improve the capacity and cost limitations once expected in these networks.

(a) (b) Fig. 5. Resonator impedance and transmission from a T-cell configuration. Series and shunt elements of similar frequency arecombined to form a steep cut-off band pass filter (a). PCS BAW filter made on an acoustic mirror using series inductors to widenthe resonators’ bandwidth (b).

Data modulators, variable attenuators, and equalizers based on a micromechanical antireflection coating have been demonstrated for various Wavelength Division Multiplexing (WDM) network applications. The basic component used for these functions is known as a Mechanical Anti-Reflection Switch (MARS), which operates as a multi-region dielectric mirror for which the thickness of one of the regions is voltage controlled [9]. The controllable region is essentially an air-gap membrane capacitor, which gradually collapses due to electrostatic force as increased voltage is applied. The structure and operating principle are illustrated in Fig. 6(a). A λ0/4-thick film of silicon nitride is suspended above a silicon substrate with a 3λ0/4-thick air gap between (λ0 being the center-operating wavelength). In this configuration the nitride/air/substrate film structure provides a high reflection (>70%) condition at λ0. When a voltage is applied between a gold electrode on top of the nitride and the silicon substrate, the film is electrostatically attracted toward the substrate, reducing the air-gap thickness. When the air-gap becomes λ0/2, a low reflection (<1%) condition is achieved. Such devices can be readily achieved through silicon surface micromachining techniques, where a sacrificial layer is etched away to produce the free-standing membrane [9]. The geometry for a typical device is shown in Fig. 6(b), consisting of a central electrode plate suspended by crossed support beams. At the center of the plate is an optical window defined by the absence of metal. Modulators with central plates typically ranging from 10µm to 100µm on an edge and support beams ranging from 4µm to 20µm wide and 5µm to 100µm long have been demonstrated with about a 25V-50V drive voltage range [20].

Arranged as a “loop back” modulator, the MARS device can be switched on and off to provide a low cost consumer unit for fiber to the home networks. Measured 25V response characteristics for a 30µm square plate suspended by 5µm-wide, 20 µm-long beams are shown in Fig. 7(a) [9]. Device response to a square wave pulse train is shown, where rise and fall times for the device are 132 and 125 ns, respectively. For a data communications system this translates to a bitrate greater than 2Mbits/s [20]. Fig. 7(b) shows the response of such a modulator to a pseudo random bit stream of 215-1 word length at 3.5 Mbit/sec. The modulator achieves this data rate with error rates of less than 10-8 with less than 2dB power penalty from the system base band.

(a) (b) Fig. 6. MARS device. The Mars device operates as a multi-layer dielectric mirror stack for which the thickness of one of the layers is controlled by varying an electrostatic potential to change its size by a quarter wavelength (a). SEM of a typical device shows the optical window at the center of a cross-shaped support structure.

(a) (b) Fig. 7. MARS Modulator performance. The response of a Mars modulator device to a 25 V square wave pulse (a), and bit errorrate and eye pattern for a fully packaged modulator.

If the displacement of the MARS is gradually controlled in an analog manner, rather than switching it on and off with a square wave, it can be used as a variable attenuator for a particular wavelength [21]. An advanced implementation of this function is illustrated in Fig. 8(a), where an array of many such attenuators is used to form a WDM equalizer. Here a free space optics wavelength-multiplexing package is used, with refractive optics imaging an input fiber through a planar diffraction grating to disperse the input light by wavelength across the array and then direct the reflected light back through the grating into an output fiber. With a MARS array at the device plane, separate wavelengths can be individually attenuated as needed to compensate for gain variations resulting from other parts of the network system. The flattening of the emission spectrum from an amplified spontaneous emission (ASE) source is shown in Fig. 8(b), demonstrating the capability of a MARS equalizer. The input spectrum is shown along with the output spectrum with no control signal applied, and with MARS devices in operation. Here, a spectrum with over 15dB of dynamic range is flatted to less than 0.25 dB over a 22-nm wide spectrum, although with roughly an 8-dB insertion loss [21].

An exciting new application of MEMS technologies to optical networks is through the realization of transparent optical crossconnects. In these systems, arrays of electrostatically controlled micromirrors are used to intelligently route traffic from incoming fibers to outgoing fibers, without the conventional bottleneck of converting the light beams to intermediate electronic signals for switching operations. The active mirrors can be micromachined through batch processes, allowing large arrays to be realized in a cost effective manner. An example of one such mirror which uses electrostatically actuated gimbal structures to produce 2-axis tilt motion is shown in Fig. 9(a) [10].

(a) (b) Fig. 8. MARS spectral power equalizer. A grating is used to disperse separate wavelengths to independent MARS attenuators which can be varied in an analog manner to adjust the power level for each (a). Flattening of the emission spectrum from anASE source using a MARS equalizer is shown (b).

Fiber/microlens assembly

Fiber/microlens assembly

Micromirrorarray

Micromirrorarray

(a) (b) Fig. 9. Optical MEMS crossconnect. A two-axis beam steering micromirror is shown (a), along with a crossconnect configurationin which arrays of mirrors are used to steer beams from one fiber assembly to another.

In these optical crossconnects it is desirable to have the number of active components scale by port count, N, rather than by the number of connections, N2, to effectively keep up with the rapidly expanding size and capacity of networks (N >1000). One such system is illustrated in Fig. 9(b), where a set of 2-dimensional MEMS mirror arrays is used to provide a crossconnect path from one fiber/collimating-microlens array to another. In this design, a crossconnect path consists of light leaving an input fiber and being collimated and projected onto one mirror in the first MEMS array by a microlens [22]. The tilt of this first mirror directs the reflected beam toward a mirror in the second array, which tilts to direct the beam to a second microlens where it is couples to the output single-mode fiber. An optical crossconnect of this kind is transparent, with performance that is independent of bit rate, protocol, or format of the data. Typical switching times of <10ms, crosstalk below –50dB, and low insertion loss have been demonstrated. Bit error rate curves at hundreds of gigabits per second compared with back-to-back performance show no penalty from chromatic dispersion, polarization dependent loss, or polarization mode dispersion [10].

5. CONCLUSION Two important wireless resonator and optical reflector MEMS technologies have been presented. MEMS RF inductors can provide improved Q and SRF over conventional RFIC inductors, without sacrificing the benefits of integration. Fixed value thin metal inductors were demonstrated with Q values better than 13, with the potential for Q’s in the 20’s. Variable inductors were demonstrated with a near 20% change in inductance, with the potential for variations on the order of 50%. Bulk acoustic wave resonators were also examined for use in wireless filter applications, where they offer reduced size over ceramic resonators and improved high-power and high-frequency response over SAW devices. Simple acoustic mirror BAW devices offer good resonator characteristics with bandwidth that can be improved with external inductors. Membrane devices offer improved stand-alone bandwidth, with the potential to extend the technology to higher frequency wireless applications. Electrostatically controlled MEMS reflector elements have been demonstrated as a modulator for low cost loop back systems, spectral power equalization, and all-optical traffic switching. While providing the high performance of fundamentally mechanical optics, these components can be produced with the cost efficiency of electronic integrated circuits and scaled to address the growing capacity needs of modern dynamic communications networks.

ACKNOWLEDGEMENTS The authors wish to acknowledge the support of Lucent Technologies and Agere Systems (formerly the Microelectronics division of Lucent Technologies), and to thank J. Lin, O. Boric-Lubecke, V. Aksyuk, and F. Pardo, Lucent Technologies, Murray Hill, NJ, and K. Grenier, D. Lopez, and L. Fetter of Agere Systems, Murray Hill, NJ, for useful discussions on IC’s and MEMS.

REFERENCES 1. V.M. Lubecke, B. Barber, and L. Fetter, “High-Q MEMS for Wireless Integrated Circuits,” TELSIKS' 2001,

Proceedings of the Fifth Conference on Telecommunications in Modern Satelite and Cable Services, Niš, Yugoslavia, October, 2001.

2. C.-Y. Chi and G.M. Rebiez, “Planar Microwave and Millimeter-wave Lumped Elements and Coupled-line Filters Using Micromachining Techniques”, IEEE Trans. Microwave Theory. Tech., vol. MTT-43, pp. 730-738, April 1995.

3. D. J. Young, V. Malba, J.-J. Ou, A.F. Bernhardt, and B.E. Boser, “Monolithic High-performance Three-dimensional Coil Inductors for Wireless Communication Applications,” Technical Digest, IEEE Intl. Electron Devices Meeting, Washington, D. C., pp. 67-70, Dec. 8-11, 1997.

4. P.L. Gammel, B.P. Barber, V.M. Lubecke, N. Belk, M.R. Frei, “Design, Test and Simulation of Self-Assemblied, Micromachined RF Inductors”, SPIE Symposium on Design, Test, and Microfabrication of MEMS and MOEMS, pp 582-591, March 1999.

5. Shifang Zhou, Xi-Qing Sun and William N Carr, “A Monolithic Variable Inductor Network Using Microrelays with Combined Thermal and Electrostatic Actuation”, Jrnl. of Micromechanics and Microengineering, vol.9, no.1, pp.45-50, March 1999.

6. S. Hara and T. Tokumitsu, “Monolithic Microwave Active Inductors and their Applications”, IEEE International Symposium on Circuits and Systems, pp. 1857-1860, June 1991.

7. V. Lubecke, B. Barber, E. Chan, D. Lopez, and P. Gammel, “Self-Assembling MEMS Variable and Fixed RF Inductors,” Proc. of the Asia Pacific Microwave Conference (APMC), Sydney Australia, Dec. 2000.

8. K. Grenier, B. Barber, V. Lubecke, M. Zierdt, H. Safar, P. Pons, and P. Gammel, “Integrated RF MEMS for Single Chip Radio,” Transducers digest, Munich, Germany, May 2001.

9. J. Walker, K. Goossen, and S. Arney, “Fabrication of a mechanical antireflection switch for fiber-to-the-home systems,” Journal of Microelectromechanical Systems, Vol. 5, No. 1, pp. 45-51, March 1996

10. D. T. Neilson, et al., “Fully provisioned 112x112 micro-mechanical optical crossconnect with 35.8 Tb/s demonstrated capacity,” Optical Fiber Communication Conference, 2000, Vol. 4 , pp. 202-204, 2000.

11. W. Cowan, V. Bright, V, A. Elvin, D. Koester, “Modeling of Stress induced curvature in surface micromachined devices,” SPIE Symp & Education Prog. on Micromachining and Microfabrication, pp.56-67, 1997.

12. V. Aksyuk, B.P. Barber, C.R. Giles, R. Ruel, L. Stultz, D.J. Bishop, “Low Insertion Loss Packaged and Fiber-Connectorized Si Surface-Micromachined Reflective Optical Switch,” Proc. 1998 Solid State Sensor and Actuator Workshop, Hilton Head pp. 79-82.

13. D. Koester, R. Majedevan, A. Shishkoff, K. Marcus, “Multi-user MEMS Process (MUMPS) Introduction and Design Rules,” Rev. 4 7/15/96 MCNC (CRONOS) MEMS Technology Application Center, Research Triangle Park, NC 27709.

14. J. Smith, "Embedded Micromechanical devices for the Monolithic Integration of MEMS with CMOS", Proc. IEDM ‘ 95, pp. 609-612, 1995.

15. Fasthenry User’s Guide, version 3.0, M. Kamon, L. M. Silveira, C. Smithhisler, J. White, Massachusetts Institute of Technology, Cambridge, MA, 11 November 1996. Fasthenry is an inductance extraction program that computes the complex frequency-dependent admittance matrix under the magnetoquasistatic approximation.

16. P. Bradley, R. Ruby, J. Larson, Y. Oshmyansky, and D. Figuerdo, “A Film Bulk Acoustic Resonator (FBAR) Duplexer for USPCS Handset Applications,” IEEE MTT-S Int. Microwave Sym. Dig., May 2000.

17. B. Barber, “Thin-Film Bulk Acoustic Wave Devices for RF Applications,” presented at Sarnoff 2000, NJ March 2000.

18. R. Ruby et al. “PCS 1900MHz duplexer using thin film bulk acoustic resonators (FBARs)”, Elec. Lett. Vol. 35, No 10, pp794-795, May 1999

19. K. M. Lakin et al. “Solidly Mounted Resonators and Filters” IEEE Ultrasonics Symp, pp 905-908, 1995 20. J. Walker, K. Goossen, and S. Arney, “Mechanical anti-reflection switch (MARS) device for fiber-in-the-loop

applications,” Advanced Applications of Lasers in Materials Processing/Broadband Optical Networks/Smart Pixels/Optical MEMs and Their Applications. IEEE/LEOS 1996 Summer Topical Meetings, pp. 59-60, 1996.

21. J. Walker, J. Ford, K. Goosen, D. Bishop, D. greywall, and V. Aksyuk, “Surface Normal Optical MEMS in Dynamic WDM Transport Networks,” SPIE Symposium on Design, Test, and Microfabrication of MEMS and MOEMS, pp 41-47, March 1999.

22. R. Ryf, et al., “1296-port MEMS transparent optical crossconnect with 2.07petabitis switch capacity,” Optical Fiber Communication Conference and Exhibit, 2001. OFC 2001, pp. PD28_1-PD28_3, 2001