In the last two years, the development and deployment of devices based on microelectromechanical systems for radio-frequency applications, or RF MEMS, have gained rapid ground. These new technologies have made possible discrete micro-scale mechanical circuits, which are capable of low-loss filtering, mixing, switching, and frequency generation. While the successful demonstration and implementation of RF MEMS devices is a cause for celebration, the real accomplishment is the beginning integration of RF MEMS into the module and IC. Devices such as RF switches and on-chip inductors are moving into production at several major semiconductor manufacturers. In this way, the industry is making measurable strides toward a major intermediate objective: the single-chip transceiver.
This goal is not as far away as one might think. Consider the existence of validated technologies that merge micromechanics with transistor circuits onto single silicon chips. Following the line of development, it seems that single-chip transceivers will be feasible in the next few years. The solution will not be found solely in the development of silicon-compatible MEMS devices, however. It also will require the use of alternative architectures that deploy large numbers of passive, high-Q MEMS circuits to reduce power consumption for portable applications.
This article looks closely at these micromechanical circuits, along with the associated technologies that will most likely play key roles in reducing size and power consumption for future communications transceivers. It also examines ways of integrating RF MEMS components for the creation of a complete RF front end.
The architecture of today's transceivers has enabled many advances. Yet the creation of communications devices continues to rely on evolutionary techniques and a hodgepodge of technologies. Engineers must combine a variety of building materials, such as silicon, gallium arsenide, quartz, and ceramic. They mix on-chip and off-chip circuits for high-Q inductors, filters, capacitors, and oscillators. Plus, they insist on hanging on to challenging packaging methods like LTCC, CSP, and modules. This lack of an integrated approach has created inherent design challenges.
Standard mechanical circuits, such as quartz-crystal resonators and surface-acoustic-wave (SAW) filters, provide essential functions in the majority of transceiver designs. But manufacturers tend to avoid using too many of them. For this trend, the circuits can fault their larger size and cost. The downside of this practice is that when designers minimize the use of high-Q components, they often trade power for selectivity (in other words, Q) and thus sacrifice transceiver performance.
Additionally, the use of current high-Q passive solutions prevents the tailoring of termination impedances, which is required by RF and IF filters. This factor can be an added advantage when designing power-hungry, low-noise amplifiers (LNAs) and mixers in CMOS technology. At the output of an LNA, higher impedance can enable a major power savings.
RF MEMS ACTIVITY
Currently, the industry is seeing a broad number of RF MEMS devices in development for strategic applications. In the high-Q micromechanical domain, for example, Agilent has already launched a successful MEMS-based, film bulk-acoustic wave resonator (FBAR). It is now shipping in cell phones worldwide. Similarly, companies such as Discera have conducted extensive work in micro-resonator technology to enable the replacement or integration of front-end components. The first of these components is a micro-oscillator that can replace crystal oscillators.
As far as research goes, one of the most popular areas has been the MEMS-based radio-frequency switch. Now, companies such as Teravicta and Microlab are commencing production of switches based on differently actuated devices.
Another hot area is on-chip inductors. These inductors have long been a focus for several companies, including PHS MEMS. This company helped to develop and industrialize a thick copper process that enables very high Q. Micromachined variable capacitors may soon reach the market in volume, thanks to intensive development by STMicroelectronics and other companies. On-chip micro-antennae are also the subject of intense ongoing research and development.
The simplicity of micromechanical devices, coupled with their small size, translates into a range of possible solutions. Such solutions can help address the disparate nature of transceiver architecture today, as well as the need for lower power consumption. In addition, MEMS can be integrated on-chip using silicon-style batch-fabrication techniques. This capability offers the industry a cost-effective way to move toward additional functionality and high-performance circuitries.
The following micromechanical-solution scenarios use vibrating micro-resonator beams. They offer examples of the kind of gains that can be achieved using the partial and full-scale deployment of micromachined technologies. The current generation of micro-resonators, freed at both ends of the beam, offers Q factors of more than 10,000 at frequencies relevant to CDMA cell-phone applications (FIG. 1).
Perhaps the most direct way to harness micromechanical circuits is via the direct replacement of the off-chip ceramic, SAW, and crystal resonators. Such devices are used in RF pre-select and image-reject filters, IF channel-select filters, and crystal-oscillator references in conventional superheterodyne architectures (FIG. 2). Micromechanical switches also can be used to replace FET T/R switches. The micromechanical switches greatly reduce wasted power in transmit mode. For example, they can cut wasted power by as much as 280 mW if the desired output power is 500 mW. For further miniaturization, medium-Q micromachined inductors and tunable capacitors can be used in VCOs and matching networks.
Although they are beneficial, the performance gains afforded by the mere direct replacement by MEMS are quite limited—especially when compared to the more aggressive uses of the technology. To fully harness the advantages of micromechanical circuits, one must first recognize that these circuits offer the same system complexity advantages over off-chip discrete components that planar IC circuits provide over discrete transistor circuits. This is due to the MEMS circuits' microscale size and zero-dc power consumption. Because of this fact, micromechanical circuits should be utilized in large numbers in order to maximize performance gains. Even with banks of micromechanical circuits (100 high-Q resonators on a 2-mm2 die), the area and height consumed by the devices is a fraction of the area consumed by their off-chip counterparts.
Figure 3 presents a system-level block diagram for a possible transceiver front-end architecture. It takes full advantage of the complexity that is achievable via micromechanical circuits. The main driving force behind this architecture is power reduction. In several of the blocks, power reduction is attained by replacing active components with low-loss passive micromechanical ones. It is further reduced by trading power for high selectivity (i.e., high Q).
Among the key performance-enhancing features are:
- An RF channel selector, which is comprised of a bank of switchable micromechanical filters. This selector offers multi-band reconfigurability as well as receive-mode power savings via relaxed dynamic-range requirements. By allowing the use of a more efficient power amplifier, it also enables transmit-mode power savings.
- Use of a passive micromechanical mixer-filter to replace the active mixer normally used, thereby providing obvious power savings
- A VCO that is referenced to a switchable bank of micromechanical resonators. This VCO can operate without locking to a lower frequency reference. It therefore operates with orders-of-magnitude-lower power consumption compared to present-day synthesizers.
- Use of a micromechanical T/R switch, with already described power savings in transmit mode
- Use of a micromechanical resonator and switch components around the power amplifier to enhance efficiency
Although it is already quite aggressive, the architecture of Figure 3 may still not represent the best power savings afforded by MEMS. In fact, even more power savings are possible. To achieve them, the high-Q micromechanical circuits in the signal path must post losses that are so low that the RF LNA is no longer needed.
Normally, the RF LNA is required to boost the received signal against losses and noise from subsequent stages. In this case, the RF LNA can actually be removed. The needed gain to the baseband can then be provided by an IF LNA. The IF LNA consumes much less power, as it operates at the much lower IF frequency. Without the RF LNA or transistor mixer, the receiver front-end architecture reduces to an all-MEMS topology, like the example shown in Figure 4.
With the absence of RF transistor circuits, the dynamic-range concerns have been removed. Subsequently, the channel-selecting filter bank of Figure 3 has been converted to a mixer-filter bank. It has moved down to the IF frequency, where it might be easier to implement. Now, a single-frequency RF local oscillator (LO) can be used to downconvert from RF to IF.
Because the RF LO is now a single-frequency oscillator, power-hungry phase-locking and pre-scaling electronics are no longer necessary. This change invites power advantages that are similar to what the VCO experienced in the architecture of Figure 3. In fact, the architecture of Figure 4 attains all of the power advantages of Figure 3. In addition, it boasts power savings due to the lack of an LNA. It does so, however, at the cost of a slightly higher overall noise figure and decreased robustness against hostile (i.e., jamming) interferers.
A two-chip solution that combines a MEMS chip with a transistor chip can certainly be used to interface micromechanical circuits with transistor circuits. But such an approach becomes less practical as the number of micromechanical components increases. For instance, practical implementations of the switchable filter bank in Figure 3 require multiplexing support electronics. These electronics must interconnect with each micromechanical device. If they were implemented using a two-chip approach, the number of chip-to-chip bonds required would become quite cumbersome. A single-chip solution therefore seems all the more desirable.
In the pursuit of single-chip systems, several technologies that have been developed and implemented merge micromachining processes with those processes used for integrated circuits. Much of this has taken place in the past several years. One such technology, for example, combines CMOS transistor circuits and polysilicon surface-micromachined structures in a fully planar, modular fashion. Here, transistor and MEMS fabrication steps are separated into modules. No intermixing of process steps from each process is permitted. These kinds of modular processes are advantageous. They allow the greatest flexibility when changes are made to either the transistor or MEMS process steps.
It should be noted that none of the existing fully planar approaches are truly modular. Each requires some degree of sacrifice in the MEMS modules, the transistor modules, or both.
In addition to fully planar integration methods, the industry is witnessing a resurgence in bonding processes. This type of process merges circuits and micromechanics by bonding one onto the wafer of the other. In particular, the advent of more sophisticated aligner-bonder instruments is making possible much smaller bond-pad sizes. Soon, wafer-level bonding may have bond-pad sizes that are small enough to compete with fully planar-processed merging strategies in interface capacitance values.
If the bond capacitance can indeed be lowered to this level with acceptable bonding yields, this technology may well be the ultimate in modularity. It will, in effect, allow the combination of virtually any micromechanical device (e.g., even those made with diamond) with any transistor integrated-circuit technology.
A recently introduced process combines micromechanics with transistor circuits using a microplatform bond and transfer approach (FIG. 5). In this process, micromechanics are first fabricated onto microplatforms. The microplatforms themselves are released and suspended over their "MEMS-carrier" wafer by temporary tethers. Next, they are flipped and bonded to receiving bond pads on a transistor wafer. They are then physically torn from the MEMS-carrier wafer by breaking the suspending tethers.
This bonded platform technology allows the low-capacitance, "single-chip" merging of MEMS and transistors. It offers several key advantages:
- Being truly modular, it requires no compromises in either the MEMS or transistor modules.
- It attempts to minimize the Q-degrading anchor losses experienced by previous bonding-based methods. It does so by bonding the platforms' housing resonators instead of directly bonding the anchors of resonators.
- It constitutes not only a wafer-scale batch approach, but also a repeatable one. A step-and-repeat procedure can be used to allow a single MEMS wafer to service several transistor wafers.
From a broader perspective, the integration techniques discussed previously are really methods for achieving the low-capacitance packaging of microelectromechanical systems. As mentioned, another level of packaging is required to attain high Q vibrating micromechanical resonators—vacuum encapsulation.
The requirement for a vacuum is unique to vibrating micromechanical resonators. In contrast, the requirement for encapsulation is nearly universal for all of the micromechanical devices discussed in this article. In fact, it is universal for virtually all micromechanical devices in general. Some protection from the environment is necessary, even if it is only to prevent contamination by particles or even by molecules. Or such protection could serve to isolate the device from electric fields or feed-through currents. Needless to say, wafer-level encapsulation is presently the subject of intense research.
RF MEMS devices, such as micro-resonators, are well positioned as the building blocks for a new integrated mechanical circuit technology. In this technology, high Q will serve as a principal design parameter, thereby enabling more complex circuits. Wireless designers can combine the strengths of integrated micromechanical and transistor circuits. They can then use both in massive quantities. In doing so, they can attain functions that were previously impossible. The industry can soon expect to see new transceiver architectures offering orders-of-magnitude performance gains. In particular, high-Q micromechanical circuits can enable paradigm-shifting transceiver architectures that trade power for selectivity (i.e., Q), with the potential for substantial power savings and multiband reconfigurability.