Graham Prophet, Editor, EDN Europe, from EDN, 2/7/2002

Micromachining and MEMS (microelectromechanical systems) are far from new. The concepts have been around in various forms since the early days of IC fabrication.

In some areas, micromachined devices have already achieved widespread use; the best-known application is probably the accelerometer, which is most famous for its use in autoairbag sensors. Still, the number of silicon devices deployed in airbag systems has yet to surpass the number of purely mechanical switches out there.

Sensors of various types represent the MEMS that have penetrated markets to the greatest extent, and pressure sensors lead the pack. Micromachining processes, with their ability to craft silicon diaphragms with precisely defined properties, have found a ready application in that area.

Better switches, better passives

This article, however, concerns MEMS for RF applications. Why go to the trouble of adding micromachined structures to the complexities of RF circuitry?

The answer lies almost entirely in the three letters that dominate RF design: dB and Q. Micromachined structures hold the promise of performing certain switching, routing, and signal-processing functions with lower losses or much better performance than conventional components.

You can build switches that give you lower transmission losses and higher isolation in their "off" state; inductors and capacitors to form tuned circuits with much higher Q factors than you can get using conventional integrated processes; bandpass filters and phase shifters that reach new levels of performance exceeding those possible with SAW structures; high-performance variable capacitors to perform tuning functions with characteristics much closer to the ideal than you achieve with varactor diodes; and much more.

Like the fundamentals of the technology, using MEMS for RF applications is not new; the military and aerospace sector has been using such devices in low-volume, cost-insensitive applications for some time. (And if you are working in such an environment and have access to the in-house facilities of that sector, this article is not for you!) The author listing at any of the conferences that cover MEMS is invariably a role call of that sector. But the mass-market applications that are demanding the performance that only MEMS can achieve are new, prompting an upsurge in efforts to transform MEMS from an effectively handcrafted technology to a volume-production technique.

As with so many other technologies today, a key driver is cellular communications, and 3G (third-generation) mobile phones, in particular. (Although, if the costs come down far enough, 2.5G or new designs of 2G phones could also benefit.) MEMS IP supplier MEMSCAP publishes a block diagram that identifies at least 10 functions in the RF section of a typical transceiver that could benefit from the use of MEMS devices; most are either bandpass filters depending on high-Q inductors, or frequency-setting functions in oscillators or PLLs that exploit high-Q filters and variable capacitors.

Switches will be first

The device that most suppliers expect to initially bring to market is, at first sight, a simple one: the RF signal switch. You might use such a device at various circuit nodes, the most obvious being transmit/receive switching. But newer portable, handheld designs may also require switching between antennas to support multiple air interfaces or to use antenna-diversity techniques.

Future multi-band products may reconfigure RF circuit blocks on the fly, an application for which MEMS switches would be all but essential. Almost certainly, 3G handsets will be the first volume application of MEMS switches. It is already clear that the available bandwidth, and hence the user experience on which the 3G concept is being sold, will depend critically on the quality of the RF link between the base station and the handset. Every decibel that the handset conserves in its receive path, by passing through a physical switch rather than a diode or FET switch, adds to that link quality. Every decibel saved in the transmit path represents power saved and battery life extended.

Manufacturers acknowledge that a new-generation product that appears to step back several generations in its time between recharges is unlikely to attract consumers.

The reality is that the RF MEMS market-in commercial terms-barely exists today but is set for a spectacular takeoff. Cahners' In-Stat Group is forecasting growth from around $1 million in 2001 (which accounts for little more than a few technology-transfer agreements and a handful of advanced samples) to nearly $350 million by 2006.

A relay-but so small

MEMS RF switches are, in principle, nothing more or less than relays. As with any other relay, you use a control or activation signal to physically make or break a separate signal path. Vendors will offer them as discrete devices for assembly to RF module substrates or as IP for incorporation into highly integrated system-on-chip designs. So, at the start of 2002, what is the switches' real position? And what can you expect to be able to source and design within the coming months?

Some of the following examples illustrate the various techniques that companies are proposing to build MEMS RF switches. Unlike diode or FET switches, all of them hold the promise of nearly ideal RF performance-that is, isolation in the "off" state of typically 40 dB or better and insertion loss in the "on" state of only a small fraction of a decibel. This performance is achievable because you can construct MEMS RF switches to connect directly to strip-line RF signal paths, with minimal losses to reflections. Switches proposed for introduction in early 2002 differ in their actuation mechanism and in the contact arrangements to make and break the circuit.

RF MEMS emerge as a market at a time when product-delivery mechanisms in their parent semiconductor industry are in flux. The embryonic MEMS market reflects this state by proposing to offer, from the outset, each of the formats in which you can buy semiconductor products. Vendors will deliver devices as discrete components (in effect, as RF relays that offer massive advantages in size and weight over conventional components), as subsystems, or as IP. Today's cell phones go some way toward consolidating the passive components ancillary to IC chip sets in one or a few packages.

A feasible way forward for some MEMS applications might be to incorporate them into an integrated device at that level. The ultimate objective of IP offerings is to enable you to integrate MEMS IP into a system-level ASIC along with every other functional block your system demands. This ambition represents a formidable challenge at a time when IC fabrication and its associated EDA are already struggling to meet the challenges of system-level IC design (see box below "Making MEMS mainstream").

At present, any MEMS-switches deliveries that have taken place have been strictly of advanced-prototype status, so it is unclear which of these methods will become the preferred way to design in MEMS switches.

So, how do you build an RF MEMS switch? If you want a microminiature relay, why not build exactly that? Microlab uses this approach. James Valenzuela, Microlab's vice president of marketing, says the company expects to release samples of its MagLatch switch to a "select group of customers" early in 2002. Microlab has built, in microfabrication, a switch that closely resembles a conventional magnetic relay (Figure 1). A planar coil, acting on a pivoted beam of a magnetic alloy, provides the activation energy; micromachined torsional supports provide the pivots. The beam carries a highly conductive metallization-of gold, for example-to complete the circuit when it is brought down to the substrate.

There is also a latching structure; an underlying bias magnet "captures" the beam with either end down, so overcoming the bias to change the relay's state requires only a short pulse. The switch comes in spst/spdt and other standard relay formats. The company quotes activation energy as a pulse of 60 mA from 2 to 5V, and the energy required is less than 40 µJ. No power is required in the latched state. Insertion loss is 0.1 dB, and isolation in the off state is more than 40 dB. Microlab says it can build the device onto a variety of substrates, including a silicon device, after fabrication of the RF circuitry.

Electrostatic activation

The more common approach to activating a MEMS switch, however, uses electrostatic forces to change the position of the moving elements. Motorola takes this approach with its packaged MM7500-4P RF switch. Like many other designs, the switch is based on a cantilever bar. As the name suggests, the moving element is anchored at one end and makes contact when the "free" end is drawn down to the substrate (Figure 2). The switch targets cell-phone antenna-switching applications, and Motorola expects sampling to begin early in 2002. The four-way switch has one common terminal for the antenna connection. It is intended to provide low- and high-frequency switching in transmit and receive paths in a single device. The contact sets are all carried on a single mechanical element and "make " simultaneously. So you might use one path in series (closing the signal path) and one in shunt (close-to-ground) modes. Motorola presents the device in a hermetic package with an option to include passive matching networks; it also includes a high-voltage-CMOS charge pump and control chip. Motorola has opted for a relatively high electrostatic voltage for activation (around 60V) to achieve fast switching and long lifetimes. (A single device can reach 6×109 operations.)

Motorola's Cliff Vaughn, a MEMS manager, notes that more complex switch configurations stretch the capabilities of PIN diode and GaAs FET switches; you may need to cascade devices to get the off-state isolation you need, leading to increased losses in conduction. Vaughn says that the capacitance of the MM7500 switch element is only 10 fF (open) yielding isolation of 60 to 70 dB, or 45 to 50 dB once packaged. He adds that for volume applications, the standard part will function as a development vehicle, anticipating customized control interfaces and other details of production items.

Although the device comes in a hermetic-style package, Vaughn says that the switch is not atmosphere-dependent, and you can operate it in free air on the bench. He also notes that you will see benefits from the linearity of any mechanical switch compared with their semiconductor alternatives, especially when it's operating at more than 2 GHz. Vaughn expects volume production to begin in 2003 or 2004 and anticipates that, at first, only top-of-the-range cell phones will support the likely two-to-one price premium of MEMS over diode or FET switches.

Watch those reflections

As an aside, Vaughn points out that when you use a MEMS switch, you must be sure of the operating state of your transmitter power amplifier at the instant of switching. With a metallic switch, you get a very high isolation and, therefore, a very high VSWR and reflection of energy back into the transmitter output, if it is still running. Whereas the imperfect isolation and presence of various parasitics in a diode or FET network might serve to "mop up" such energy, you are likely to run into trouble if you do not make sure that the power amplifier is off before you switch the antenna connection.

A number of companies continue developing switch mechanisms that are similar in their broad operating outlines, if not in the detail of their execution. For example, at Analog Devices, a program separate from the company's well-known accelerometer and sensor product line is working to produce the µmRelay; also a cantilevered-beam, electrostatically operated switch, µmRelay has processing options to provide dielectric isolation of the control-voltage path from the signal path (Figure 3).

Other actuation mechanisms also exist. At start-up, XCom Wireless CTO Dan Hyman is focusing the company's attention on the problems of ensuring process portability (see sidebar "Making MEMS mainstream"). XCom is working on a relay mechanism that, according to the company, is unlike any other on the market. It will be low power and latching and will address the problems of process portability and packaging cost. At present, the actuation mechanism is undisclosed, although the company is prepared to disclose that the device is based on a "bimorph element," perhaps indicating the use of a resistive heating/thermal effect.

Other technologies, like XCom's, that have their roots in the proprietary military/aerospace sector, are probably further away from commercialization, but you can expect some to be licensed out over time. An example that may prove to model this process is the technology agreement between Rockwell and Cronos. One agreement that has resulted in detailed product plans involves the Summit IV technology from Sandia Labs (Sandia Ultra Planar Multilevel MEMS Technology)-a library and a process specification that Fairchild Semiconductor will make available to the market.

You will be able to design both RF and optical MEMS through the program. Vice president and general manager of Fairchild's interface and logic group, WT Greer, positions the offering as appropriate to high-volume applications and describes Summit IV as, "a technology, not a product."

He says that Summit IV allows you to make a product that you cannot easily make in any other technology. He adds that as part of the introduction program, Fairchild's device-modeling group will develop fully characterized simulation models of the devices in the library, so system and circuit simulations that run on standard EDA tools can correctly represent them. Greer also expects to see switching devices available on the market before the middle of 2002.

IP/EDA combinations

MEMSCAP is one of a new breed of companies that is offering both MEMS IP and the EDA tools to design it into a more highly integrated context. A fundamental part of MEMSCAP's approach is that it can add the MEMS devices as extra processed layers on top of an IC whose conventional process steps are already complete.

The process deposits further layers (films) of metal and dielectric and then performs micromachining on those layers. MEMSCAP describes a wide range of components, but the switch technology that it is investing most of its effort into developing is based on a membrane and is electrostatically activated (Figure 4). An activation voltage deflects the membrane (which is the moving-contact part of the switch), drawing it down into contact with the signal lines that have been laid down beneath it.

MEMSCAP's product-development manager, Bertrand Guillon, notes that you can use the mechanism as either a capacitive or resistive switch using two metal layers and an appropriate choice of dielectric. (At gigahertz frequencies, you do not need an ohmic contact for good signal propagation through a switch, if a capacitive one suits your purposes better.) The activation voltage is normally 20 to 30V; MEMSCAP is also working on a lower voltage version with an activation voltage within the range of a mobile phone's battery voltage, to avoid the need for charge pumps. Guillon acknowledges that this version means fabricating a much more delicate membrane; early work has indicated that such a device has a lifetime in the order of 106 operations, whereas the higher voltage version can achieve 109.

The lighter membrane, all other things being equal, will also have a lower limit on the power it can transmit; a typical transmit path requires 2W. Guillon says that MEMSCAP will be in a position to go to production with an RF switch in mid-2002, and limited samples are available now. The company's business model accommodates various options-from manufacturing customized devices in its own wafer fab, adding MEMS structures to customers' prediffused wafers in the same facility, to a technology transfer of IP and design knowledge. The ultimate objective of the process is to integrate all passives and switches on top of a system-level IC.

IP supplier Coventor also has an extensive library of RF MEMS parts, with switches that will be made available for integration, as discrete parts for evaluation, or for assembly into substrate or circuit board-level designs. The switches, which are ohmic-contact designs, operate from dc to 40 GHz. The vendor will supply them as dice, for chip-and-wire assembly, or as packaged devices. Initially, Coventor will supply spst and spdt versions and quotes performance as offering isolation of greater than 35 dB and insertion loss less than 0.2 dB. Coventor offers a supporting design-tool package including the Architect, Builder, Designer and Analyzer programs. These programs provide design and analysis of MEMS structures through finite-element analysis of the microstructures, and they link to other software tools, such as Ansys's Ansoft and Agilent's HFSS high-frequency circuit simulator.

The IMEC research center in Belgium, working with Alcatel, has developed an electrostatically activated switch targeting shunt-mode operation over a coplanar waveguide. A bridge structure is electrostatically attracted to contact the waveguide and capacitively connect it to ground at a suitable point (Figure 5).

But as Kris Baert, group leader of IMEC's MEMS microsystems, components, and packaging division, outlines, most of the work has been directed toward implementing an economical packaging technology to seal the device in a suitable protective environment. This goal has been achieved with a so-called zero- or wafer-level scheme, in which you adapt wafer-bonding techniques to apply lids or caps to each of the devices on a complete wafer in a single operation. Baert says that IMEC and Alcatel have achieved this goal without significantly degrading the RF characteristics of the switch itself. The team believes that the combination of a switch fabricated in a five-step CMOS- or BiCMOS-compatible process plus wafer-scale packaging goes a long way toward meeting necessary cost goals.

Graham Prophet, Editor, EDN Europe

Making MEMS mainstream

Researching this article brought to light many similarities between the evolving MEMS (microelectromechanical-systems) sector and the historical behavior of the hybrid market. For many years before the concept of "systems on silicon" emerged, companies promoted the hybrid as the vehicle of choice for implementing a complex system in a single package. Hybrids always struggled with manufacturing complexity (and hence cost) and the burden of expensive materials and packaging. The hybrid was - and remains - primarily a technology for the sector from which it emerged. With a few specialized exceptions, only the military/aerospace markets could support its materials costs and the inherently handcrafted, low-volume nature of its manufacture. MEMS show many of the same characteristics; many attractive functions and concepts have been demonstrated only at a laboratory or prototype scale, but transferring them to volume production remains a formidable challenge.

When contemplating integration at silicon level with other IC functions, stating that a product is compatible with a given semiconductor-manufacturing process is not the same as being able to routinely produce that product on the fabrication line. It is one thing to say that you can build a device with all of the same materials and process steps, but MEMS operation depends on attributes of those materials that you may be unable to control or optimize in a conventional process flow. Process variations that may not affect the usual parameters of RFIC operation may render a MEMS device inoperative. According to XCom Wireless CTO Dan Hyman, this situation reflects the tight coupling that design and production require in areas such as residual stresses in metal films. He also notes that the same model of sputtering machine, used in two fabs, by two operators performing the same process step, can yield different results.

There is not enough space here to describe the fabrication techniques, but following any of the links in the sidebar "For more information" will reveal detailed descriptions of processes such as surface and bulk micromachining, which add a third dimension to engineering at the micron level. The ability to create 3-D structures, freeing beams, membranes, pivots, and hinges to move, underlies the whole technology. It is also the basis of the high-performance inductor, freeing integrated inductances from their planar limitations and allowing you to build 3-D "coils" that are freestanding above the silicon wafer's surface and that can have air cores. It should be no surprise that in this early phase of rapid development, the technology still requires a high degree of handcrafting and process-control skill to yield functional devices. It mirrors the same state of maturity that IC fabrication went through decades ago.

Adding to the integration challenge is the fact that not all the materials are native to the IC-manufacturing process; in particular, switch contacts may use gold-one of process engineers' least favorite materials and usually regarded as a serious contamination risk. This article does not delve into the many areas of ongoing study to fully characterize the properties of materials that enable MEMS operation, but it is no exaggeration to say that researchers are developing a new physics, describing the behavior of materials at small dimensions and masses.

Some of the problems to overcome include the property of "stiction," in which a moving part unintentionally adheres or latches to a mating surface. This phenomenon can be due to surface effects (on an atomic or molecular scale) at the interface between the two surfaces; accumulated charge, either on the surfaces or buried in the materials concerned; or a combination of such effects that is sufficient to overcome the forces that should separate the parts. There is also the question of fatigue life, as many of the MEMS structures with moving parts employ flexure or torsion mechanisms. A Web search for any of these topics, plus the term "MEMS," will reveal a growing literature of detailed studies of these and other effects. As has been noted, many MEMS concepts emerged from military programs in which longevity may not have been a requirement; but if they are to transfer to consumer electronics, prospective vendors will have to characterize the MEMS' behavior over extended lifetimes.

Companies need to demonstrate the viability of agreements to mount a given range of MEMS devices on a particular foundry's process over time and volume production. However, the move to MEMS devices will be won or lost on the basis of cost. Some MEMS devices, again more relevant to their military/aerospace roots, require expensive hermetic packages enclosing either a protective atmosphere or, in the worst case, a vacuum. Clearly, this requirement is the opposite of what high-volume consumer classes of products need. First-generation products will, therefore, represent a more expensive option than established technologies; following the usual migration path, expect to see MEMS switches first in premium products. In the case of cell phones, MEMS will first appear in top-price models, for which the extra bill-of-materials cost will support more sophisticated services and longer battery life.

Can MEMS break out of the "hybrid trap" and become mainstream? Only time will tell, but you may note that several promised deadlines for product availability have already passed. Current market conditions and the slowdown in deployment of 3G cellular systems may have given the sector the much-needed breathing space to tackle some of these cost and manufacturing issues.