MEMS' Rocky Road

FEATURE FOCUS: MEMS
ON THE COVER

MEMS' rocky road

Taking a design from concept to high-volume production can be a bumpy ride.

By John DeGaspari, Associate Editor

Few people will argue that there is potential demand for MEMS devices across a range of industries. In 2002, the five leading applications of microelectromechanical systems will use 21.5 million disposable blood pressure sensors, 28.7 million manifold absolute pressure sensors for engines, 85 million packaged airbag accelerometers, 425 million inkjet printer heads, and a whopping 1.58 billion read/write magnetic heads for computer hard drives.

Those are "killer applications," according to Roger Grace, a MEMS technology consultant based in San Francisco. There are others that have broken the one-million-parts-per-month threshold that, he said, qualifies them as high-volume products.

Yet the numbers tend to obscure another reality: Making the transition from concept to high-volume production is both expensive and risky. While engineering resources exist for taking a good idea for a micro device through design, prototyping, and testing, and putting it into high-volume production, the infrastructure for doing so is still evolving. There are no guarantees of success and no standards to follow.

Analog Devices' optical MEMS device steers signals by micromirrors. Electronics and MEMS structure are integrated on a single monolithic IC.

Companies seeking to commercialize microelectromechanical devices can take a number of routes. For instance, they can draw on university laboratories for early development and prototyping, or turn to foundries specializing in MEMS, or invest in their own manufacturing equipment to develop devices in-house.

Companies that have successfully commercialized MEMS have done so despite of a lack of standards and having to work with a technology that is fragmented in terms of engineering resources. Some of the biggest players, such as Analog Devices, a $2.5 billion semiconductor company in Norwood, Mass., have the deep pockets to invest in a dedicated MEMS facility and then hang in for the long haul.

The operating results of Analog Devices' MEMS business, which produces more than three million single-chip accelerometers a month, were in the black some nine years after producing its first part in 1991. Others, without the in-house manufacturing capability, face tougher hurdles, not the least of which is finding a partner to take a concept into high-volume production. It is a tough journey, according to those who have made it successfully, with plenty of pitfalls along the way.

Choosing Partners

In MEMS, process is driven by design, so it's important for companies seeking to commercialize a micro device to evaluate the capabilities of a foundry, Grace said. "You need a company that understands what it's doing and understands your design, and has a process that is compatible with what needs to be used," he said.

The decision of where to build a device after proof of concept is a big problem for companies getting into the MEMS business, said Demitrios Papageorgiou, principal design engineer with Memsic Inc., a supplier of accelerometers in North Andover, Mass. Many times, porting to a high-volume fabrication line requires expensive tweaking, he said.

Michael Huff is founder and director of the MEMS Exchange in Reston, Va., a clearinghouse that puts developers in touch with foundries. He believes that current technology has some manufacturing issues that must be addressed before it will be widely adopted. Huff, formerly a technical fellow at Baxter Healthcare, said he recognized a need for a network of foundries with a single portal when Baxter decided to outsource fabrication for a MEMS device it was developing.

What he discovered was a mixed bag of foundries that differed widely in capabilities and range. Some specialized only in one process, such as laser sintering or a certain type of film deposition; others had much broader capabilities, from electronics to integrated MEMS. No single foundry can possibly handle everything, he said.

So far, the market is not large enough to support its manufacturing capability, according to Grace and other industry sources, who say that some foundries are operating at 15 percent capacity or less.

Automated pick-and-place minimizes die handling on a MEMS fabrication line. MEMS devices are highly specific to their applications.

The MEMS Exchange has identified a network of 22 foundries, which, according to Huff, account for around 1,000 process technologies. Its purpose, he said, is to create an environment in which MEMS developers can pick and choose to customize a process sequence. Since its inception in 2000, the MEMS Exchange has been used for early development and prototyping. Huff envisions that, in the future, it can help users transfer their prototypes to full-scale manufacturing environments, although that is still untested.

While MEMS technology does not necessarily mean working with silicon, in the United States, at least, silicon micromachining dominates. MEMS fabrication offers a new lease on life to older integrated circuit foundries, which are becoming obsolete as the semiconductor indus-try drives geometries smaller and smaller. "Some get into MEMS, hoping that as their integrated circuit business starts to wane, they can backfill with MEMS," Huff said. Yet there are also important distinctions between MEMS and integrated circuits, which point to thorny issues in manufacturing.

For one thing, MEMS are largely mechanical rather than electrical, Huff said. Integrated circuits use generic, well-defined processes working with a limited set of electronic devices—transistors, capacitors, and resistors—that can be configured into many variations on chips. "That simplifies things," he said. The diversity of microelectromechanical systems is enormous, and the implementation of the devices requires very different process sequences. Materials and fabrication processes are also of a much wider variety than those in the integrated circuit world.

Analog Devices' MEMS process is based on standard IC process equipment, used for both product development and production, such as this photolithography coat and develop track.

University laboratories are flexible and comprehensive resources for making prototypes, and generally are less expensive than foundries run for profit. By their charter as nonprofits, however, university labs cannot be used to manufacture large volumes, so a developer still must seek out a manufacturing partner or invest in a foundry of its own.

A look at the experiences of companies that have successfully commercialized microelectromechanical systems points out some key issues involved in bringing MEMS devices from concept to high-volume production.

Stand by Your Niche

One of the problems with MEMS in general is the diversity of the niche environments in which the devices operate, according to Dale Gee, senior director of business development and strategic marketing for NovaSensor Inc., a subsidiary of the TRW Co. in Fremont, Calif. The company produces micro sensors used in medical and automotive applications.

The company entered the blood pressure monitoring market in the early 1990s, with a sensor designed specifically for that application. "We had to design not only the silicon, but the package itself," Gee recalled. The device had to withstand radiation sterilization, ETO sterilization, and defibrillation voltages, and it had to meet medical specifications.

"For that particular application, there were very specific things we had to design into that product," Gee said. "It's not a good product for anything else, but if you can identify a good market, it makes sense to put the effort into it." Development of the blood pressure sensor took a couple of years, about a year each for the packaging and the silicon, he said.

According to Huff, small volumes are a problem for companies that develop micro devices. The relatively low volumes of devices tied to very specific applications limit return on investment, he said.

MEMSIC's dual-axis thermal accelerometer combines the sensor and electronics on a single chip that is manufactured on a standard CMOS process.

Semiconductor sales passed 293 billion units worldwide last year, according to the Semiconductor Industry Association in San Jose, Calif. Airbag accelerometers, running at approximately 85 million pieces worldwide this year, barely register on the radar screens of typical semiconductor foundries, Huff said.

That is significant for return on investment. Companies that invest millions of dollars in fabrication facilities and specific process lines can take years to recoup their costs, he said.
The biggest problem in MEMS manufacturing is process development, according to Joe Giachino, industry liaison at the Engineering Research Center for Wireless Integrated Microsystems at the University of Michigan in Ann Arbor.

The design of the device and the process go hand in hand, so that the process is being developed with the design. In Giachino's view, it's important that the process is robust. In many cases, people tweak the process while they are doing designs. If the process is not robust, it can lead to problems in commercialization, he said.

According to Bob Sulouff, director of business development and marketing in the micromachined products division of Analog Devices, the company has a 10-year history in MEMS manufacturing, which went from zero to 3 million accelerometers a month. One thing the company discovered, said Sulouff, was that MEMS is highly process dependent. MEMS are very sensitive to variations in the process because of their small mass and abundant surface area, he said. The surface roughness or chemical absorbency of a material has a big effect on the part and its performance.

Accelerometers are a well-established MEMS application. Some 85 million packaged accelerometers will be manufactured this year worldwide.

Analog Devices does its MEMS development work on the actual manufacturing line, using the same equipment and even the same people who produce in volume. The reason, said Sulouff, is that there are so many effects that are detected on the equipment in the process.

"The real tools that you use to make the part have an effect on the performance," he said. A process that is surface-dependent and chemistry-dependent, for example, may not show variations when a device is produced in limited quantities in a lab, but may turn out differently in the actual commercial manufacturing using production tools.

The approach takes longer, Sulouff admitted. "There is a natural tension between people who want to get products out on time and consistent, and people who want something new and different," he said. It's also expensive, given the high cost of sophisticated production machinery. A good MEMS fabrication line can cost anywhere between $60 million and $100 million.

Christos Monovoukas, director of business development for Corning IntelliSense, a MEMS foundry in Wilmington, Mass., recommends a multidisciplinary approach to building micro devices. The company designs, develops, and manufactures MEMS for telecommunications, life sciences, and microinstrumentation applications for outside customers. The design of a biosensor, for instance, may call on the disciplines of physics, chemistry, and electronics, he said.

There also has to be sufficient communication between project stages, from design to development, and then to manufacturing. People have to work closely together in moving a product from one stage to the next, he said.

The High Cost of Packaging

Packaging remains the biggest bottleneck in MEMS, accounting for 80 percent of the cost, according to Amir Mirza, director of advanced technology of NovaSensor. Unlike semiconductors, MEMS have no well-established committees that ensure exactly what a package will look like, he said. Standard MEMS packages don't exist, because there are no large oversight bodies.

"Packaging has been a big problem for a lot of companies and it has kept some players out," Mirza said. That could change if volumes get high enough in certain applications, he added. Until then, he expects packaging to remain customized.

Sulouff of Analog Devices does not expect to see standardized MEMS packaging any time soon. "There have been people meeting for a long time trying to standardize, and the best they can do is find some basic voltage or nominal parameters," he said.

Customized packaging comes with the territory, in Sulouff's opinion. The nature of MEMS tends to be sensors, and there is no standard sensor, he said. Applying an enabling technology like MEMS to make sensors allows the use of semiconductor tools and the creation of controls to get good performance, he said. But sensors by their nature have different sizes and connections. The best Analog Devices has been able to do is put a package around an element that is very small, and then put the packaged part into the next assembly, he said. "You make a few variations of a simple package for the basic element, and do all of the fitting into the system by this module," he said.

In general, people tend to underestimate the difficulty of packaging, according to Monovoukas of Corning IntelliSense. "We've seen a lack of consideration for packaging up front," he said.

Testing of MEMS devices is often developed by foundries in-house. A technician performs customized testing for a gas detector produced at Corning IntelliSense, a comprehensive MEMS foundry.

Some industry groups are starting to recognize standardization as an issue, and are focusing on fabrication. The MEMS Industry Group, a trade association based in Pittsburgh, identified standardization as a key challenge in its 2001 annual report. The group plans to issue a report on foundries and fabrication next year, according to special projects director Karen Lightman.

Groups that have addressed the issue of MEMS fabrication standards also include the Semiconductor Equipment and Materials International, a computer-oriented trade association based in San Jose, Calif. Standards will be one of the issues under discussion at the 2002 Commercialization of Microsystems Conference (www. coms2002.org), co-chaired by Joe Giachino at the University of Michigan and sponsored by the Michigan Economic Development Corp. this September in Ypsilanti, Mich.

Giachino acknowledges that, given the lack of universal established processes, MEMS fabrication standardization will never approach that of the semiconductor industry. But people are pushing for standards, which can make it easier to transfer MEMS between foundries.

Put to the Test

Although the last few years have seen the appearance of some testing companies, such as ETEC Inc. in West Peabody, Mass., most MEMS manufacturers must develop their own methods to test their parts for performance and durability, according to Mirza.

Analog Devices, for instance, has developed its tests in-house. Reliability is an uncharted area when a company starts up a new process or tries out new designs, Sulouff said. Although a lot of people know about semiconductor reliability, few know about micromachine or sensor reliability because it is very specific, he said.

Sulouff said that the only way to gain confidence that micro devices work is to put them through their paces, by changing temperature quickly, applying high voltages, or going to other extremes. The company does an analysis to determine why a part failed and what could go wrong, trying to make the process more robust.

Overall, the troubles in the MEMS landscape can be attributed to growing pains, many say. They expect that, as the infrastructure matures, the technology is likely to grow friendlier to more players. Right now, as Giachino put it, MEMS is entering its "teenage years."