Continuing chip scaling has contributed a lot to this goal, but today a situation has been reached where the presence of the expensive, off-chip passive RF components plays a limiting role. Despite many years of research, on-chip passive components based on electronic solutions have not resulted in components with the high quality offered by discrete passive components and required by most wireless applications. MEMS technology is based on IC fabrication technology that can yield small, low weight, and high performance components to replace (some of) the bulky, expensive, and unwanted discrete passive RF components such as switches, varactor diodes, high-Q resonators, and filters (Larson, 1999).

The main performance characteristics of an RF switch are insertion loss, isolation, power consumption, and linearity. Radio-frequency switching is presently realized for the greater part with PIN diode- and GaAs MESFET-based semiconductor switches. Existing and future applications, however, are asking for improved RF performance of switches based on semiconductor solutions.

For RF applications, semiconductor switches provide the desired performance in terms of switching speed, but present power-consumption constraints and introduce significant insertion loss. Moreover, the non-linear characteristics of semiconductor-based switches become an important problem for the non-constant envelope modulation scheme used in present CDMA technology and which will also be implemented in the future third-generation UMTS standard for mobile phones.

A possible route to overcome these problems can be found in the development of RF-MEMS switches. RF-MEMS switches are in essence mechanical switches, allowing capacitive contact (AC) switching as well as ohmic contact (DC) switching. Research work has already demonstrated that RF-MEMS switches implementing electrostatic actuation offer very low power consumption during switching with no standby power consumption, low insertion loss, and high linearity as compared with semiconductor-based switches (Goldsmith et al, 1998, and Brown, 1998).

All these features make RF-MEMS switches very attractive for use in a number of systems, including radio front-ends, capacitor banks and time-delay networks (Brown, 1998). In particular for space applications, another very interesting feature of RF-MEMS switches (and of RF-MEMS components in general) is that these devices are expected to be less prone to radiation-induced degradation as compared to semiconductor switches.

Lumped-element LC bandpass filters can be built using planar micromachining or MEMS technology and can be used for various wireless applications in the low GHz range. In those cases where high performance filters are not required, these filters offer a simple, cheap, and compact alternative to thin-film resonator (TFR), dielectric resonator (DR), strip line, active, and surface acoustic wave (SAW) filters. Moreover, the implementation of MEMS technology offers the potential for building tunable bandpass filters (Kim et al, 1999) that can vary their center frequency and bandwidth, a feature not readily available with the aforementioned filter types. This makes them attractive for modern multi-band communication systems.

Although an integrated design approach may result in a slow start, it is believed that in the long run this approach will lead to more rapid, cost-effective industrialization and commercialization. In particular, packaging and testability are very often neglected in an early design stage, a fact already pointed out by Senturia and Smith in 1988 for the development of microsensors (Senturia et al, 1988).

This is surprising in that one of the purposes of the package is signal distribution. Therefore, it is fairly obvious that the RF performance may very well be (adversely) affected due to interference of the package. Testing the RF performance during processing is important to improve yield and lower cost. For this reason the design and technology have been chosen in a way to establish not only acceptable off-chip RF performance but at the same time an easily testable and properly packaged device while maintaining good device performance.

An example of a successful realization of a MEMS device following the above elucidated integrated design approach is the MIRS micro relay, which is considered the first fully packaged operational micro relay for DC and low frequency applications (Fullin et al, 1998, and Tilmans et al, 1999). An integral design and fabrication approach incorporating all the key elements of a microrelay—actuator, electrical contacts, housing of the electrical contacts, structural design, micromachining fabrication process, and packaging—has resulted in the functional micro relay in Figures 1 and 2.

Figure 1: The MIRS micro relay mounted in a ceramic package (Fullin et al, 1998, and Tilmans et al, 1999). The flip-chip assembly of the two chips is clearly seen.

Figure 2: The MIRS micro relay mounted in a plastic SOIC-16 package (body size: 10.2- x 7.5- x 1.98-mm³) (Fullin et al, 1998, and Tilmans et al, 1999).

The research described here is designated as preliminary or screening work. The integrated design approach previously described has only marginally been implemented. Current and future work, however, is and will be based more and more on this approach.

The CPW is defined by the first-level metal layer deposited directly onto the substrate consisting of 3-µm-thick aluminum. The CPW signal line is 100-µm wide and the slots to ground are 25-µm. On a glass substrate (AF45) this results in a characteristic impedance of 50 (Pieters et al, 1999). The loss in the CPW consists of three main contributions: conductive loss, substrate loss, and loss due to mismatch. The conductive loss of a 3-µm thick Al CPW is on the order of 0.04-dB/mm at 5-GHz. The substrate adds an additional loss of 0.001-dB/mm (for a glass substrate). The losses caused by mismatching can be minimized by a proper design.

Figure 3: Schematic cross section of a metal bridge capacitive switch.

The switch consists of a suspended movable metal bridge or membrane, which is mechanically anchored and electrically connected to the ground of the CPW. When the bridge is up, the capacitance of the signal line to ground is low and the switch is in the RF-ON-state. Upon activation, the bridge pulls down onto a dielectric layer placed locally on top of the signal line. The switch thus changes its state, the capacitance becomes high, and the switch is in the RF-OFF-state. The DC actuation voltage at which the switch changes state is called the pull-in voltage. In operation, the DC control voltage and the RF signal are superimposed and applied to the signal line.

It is clear that a RF signal traversing the CPW will always experience a capacitive reactance due to the nearby presence of the grounded metal bridge. When the bridge is up, this capacitance is (or better, must be) very small (e.g., on the order of 10-100fF). In the down state the capacitance can easily increase by one or two orders of magnitude. By virtue of efficiently transmitting and effectively rejecting the RF signal, this micro mechanical variable capacitor operates as a microwave switch.

Figure 4: Shunt MEMS capacitive switch implemented on a CPW. (a) Top view of a fabricated switch (the bridge is 300-µm long and 50-µm wide). (b) Equivalent circuit (lumped elements).

The area underneath the air bridge is covered with a thin film of tantalum pentoxide (Ta₂O₅) with a dielectric constant of 25. A high dielectric constant is advantageous as this leads to low impedance from the signal line to ground when the bridge is pulled down. In terms of RF characteristics this means a high isolation. The dielectric further serves as a decoupler for the DC control signal. The metallic bridge consists of a 1-µm-thick aluminum layer. The stiffness (bending stiffness and built-in stress) of the beam should be high enough to ensure "pull-up" after removing the DC control voltage.

Figure 5: First-order filter. (a) Top view (L=1-nH, R=1.9, C=3.6-pF). (b) Equivalent circuit.

Figure 6: Second-order filter with series capacitive coupling. (a) Top view (L=0.8-nH, R=1.9, C=3.1-pF, C_c=2.2-pF). (b) Equivalent circuit.

For the LC resonators, spiral inductors in combination with an air gap capacitor are used. In the examples described, the inductor (L) consists of a half-turn loop implemented in the first-level metalµthe same metal as used for the CPW. Multi-turn coils are possible by using the second metal layer (i.e., the switch bridge metal) as an air bridge for interconnecting the inner part of the inductor to the rest of the circuit.

An example of a multi-turn inductor is shown in Figure 7. Due to ohmic losses in the inductor, an additional series resistance (R) occurs. The inductor designs are readily derived from the available library of the inductors as developed for the microwave MCM-D technology (Pieters et al, 1996). The capacitance (C) is identical to the switch capacitance when the bridge is in the up-position. Applying a DC voltage across the capacitor plates results in a decrease of the gap spacing. This allows tuning of the capacitor and therefore also of the center frequency of the BPF. The LC resonators of the filter in Figure 6 are series coupled by means of a capacitor (C_c). The capacitor C_c is simply formed by an interruption in the CPW signal line.

Figure 7: Top view of a multi-turn inductor.

1. The process starts with alkali-free glass wafers (AF45 from Schott)
2. A 3-µm-thick layer of aluminum alloy (1% silicon) is sputtered and patterned to define the CPW transmission lines (plus the bottom electrode of the tunable capacitors and the lower metal of the spiral inductors)
3. A 0.3-µm-thick layer of tantalum is next deposited, anodized, and patterned to form the switch dielectric
4. A polymer sacrificial layer (Shipley S1828) is spin-coated and patterned
5. A 1-µm-thick aluminum bridge layer is deposited and patterned to define the switch bridge
6. The sacrificial layer is removed by a plasma etch to release the bridge. A number of 5-µm holes are patterned throughout the bridge (see Figures 3 and 4). These holes enhance the sacrificial layer etch by providing additional access points for the sacrificial layer etchant.

After removal of the sacrificial layer, the bridge is mechanically released, allowing it to move up and down in response to an applied DC control voltage.

Figure 8 shows the capacitance of the switch of Figure 4 as a function of the applied DC control voltage. Around 35V, pull-in occurs and the capacitance abruptly changes from 0.13- to 3.6-pF. This is the point at which the switching occurs from the RF transmit (or ON-) state to the RF reject (or OFF-) state. The measured capacitance values are in close agreement with the predicted values.

Figure 8: Switch capacitance as a function of the applied DC voltage.

The IL (transmit state) and the I (reject state) of the switch are shown in Figure 9. The insertion loss is approximately 0.15-dB at 5-GHz. Note that the measured insertion loss includes both the intrinsic loss of the switch and the loss introduced by the CPW. Switch isolation is close to 14-dB at 5-GHz. Both IL and I are close to expected values and can still be further optimized.

Figure 9: Insertion loss (a) and isolation (b) of the MEMS shunt switch in Figure 4 as a function of frequency (log-scale). The black line represents the measurement (for a 50 source impedance and a 50 load). The red line represents the simulation from the equivalent circuit in Figure 4(b), thereby substituting C= 80-fF and R_line= 0.8 (transmit state), and C=5-pF and R_line= 5 (reject state). The simulation includes the losses from a 0.6-mm long CPW.

Due to the switches' extremely low loss in the RF transmit state, direct measurements of the actual or intrinsic IL of the switch tends to be inaccurate. To overcome this problem, comparative loss measurements must be performed using CPWs identical in size, but without a bridge. The difference between the measured loss responses of a CPW with and without a switch gives the intrinsic loss of the switch. This illustrates the need for a testable design. However, at this moment such test structures are not available and therefore only appropriate calibration has been performed prior to every measurement. This still gives valuable results, but as already indicated only reveals the overall loss.

The transfer function of the first order LC filter in Figure 5 is presented in Figure 10. The filter-design (mask layout) has been used to estimate (by analysis and simulation) the lumped-element values of the constituent elements of the filter (see also Pieters et al, 1996). Likewise, the measured S-parameter-data are used to extract the lumped-element values of the equivalent circuit in Figure 5(b).

Figure 10: Transfer function of the first-order lumped-element LC filter in Figure 5(a). The black line represents the measured data (for a 50 source impedance and a 50 load). The red line represents the simulation result from the equivalent circuit in Figure 5(b), thereby substituting L=1-nH, R=1.9, and C=3.6-pF.

Both methods lead to the same element value: R=1.5, L=1-nH, and C=3.6-pF. The resonance frequency can readily be obtained as 1/(LC), yielding 2.7-GHz, a value that can also be found from the peak in the curve of Figure 10. The circuit impedance at resonance is found to be ZL/(RC)= 208, resulting in a predicted insertion loss of 1db at resonance.

The transfer function of the second-order LC filter of Figure 6(a) is presented in Figure 11. Again both methods agree with respect to the lumped element value needed for correspondence: R=1.9, L=0.8-nH, C=3.1-pF, and C_c=2.2-pF. The resonance frequency can readily be obtained as 1/(LC), yielding 3.2-GHz, a value that can also be found from the peak in the curve, and an insertion loss of 2.5dB that is also confirmed by Figure 11.

Figure 11: Transfer function of the second-order lumped-element LC filter of Figure 6(a). The black line represents the measured data (for a 50 source impedance and 50 load). The red line represents the simulation result from the equivalent circuit in Figure 6(b), thereby substituting L=0.8-nH, R=1.9, C=3.1-pf, and C_c=2.2-pF.

The filter shown in Figure 6(a) should be tunable since the capacitors are MEMS variable capacitors. Due to a process-flow flaw the MEMS capacitors have not been fully released, thus preventing tuning of the gap spacing. This is currently under investigation and will be corrected in the next process run.

A device currently under development is a single-pole double-throw (SPDT) switch. More specifically, a demonstrator SPDT switch—a T/R switch for wireless local area networks (WLAN) in the 5- to 6-GHz bandµis being prototyped. To establish an integrated system maintaining maximum RF performance, including packaging and testing algorithms, it has been decided to make use of the microwave MCM-D technology already available at IMEC (Pieters et al, 1996) to allow for high-density packaging.

For instance, the chip with the heart of the MEMS device and the chip with the signal processing electronic circuitry are fabricated using separate and distinct manufacturing processes. After chip manufacturing, the individual chips are assembled and interconnected onto a single MCM-D carrier substrate, which already contains the necessary RF components such as transmission lines and quarter-wave stubs. This approach, referred to as the system-on-a-package (SoP) approach (Wambacq et al, 2000), allows great flexibility, is convenient, and opens up the possibility to independently design and moreover optimize the constituent components.

Figure 12: Illustration of the system-on-a-package (SoP) approach, based on the MCM-D technology.

Henri Jansen started his career in 1979 as an equipment engineer in the Royal Dutch Navy where he was responsible for the maintenance and repair of communication, radar, and sonar equipment. After obtaining a computer science degree, he received his Ph.D. from the University of Twente in Enschede, the Netherlands, in 1994 on the subject of plasma etching in microsystem technology. He continued working for one year as a plasma researcher at in the university’s micromechanical group. He currently works at IMEC in Leuven, Belgium, where he is responsible for RF-MEMS-based technology.

Originally published in the conference proceedings of the Third Round Table on Micro/Nano-Technologies for Space, ESTEC (Noordwijk, NL), May 15-17, 2000