The Trouble With MEMS

The advent of the Internet of Things will open up a slew of new opportunities for MEMS-based sensors, but chipmakers are proceeding cautiously.

There are a number of reasons for that restraint. Microelectromechanical systems are difficult to design, manufacture and test, which initially fueled optimism in the MEMS ecosystem that this market would command the same kinds of premiums that analog designs have been able to maintain. MEMS chips are custom-designed for specific applications, and many are customized engineering marvels with micro-machined moving parts and advanced electronic controls.

Pricing has not always been proportionate with the importance or difficulty of developing these devices, however. Accelerometers, gyroscopes and capacitive touch sensors, for example, have become so pervasive in smartphones that big systems companies have been able to play one small vendor off against another to whittle down the average selling prices. While volumes are huge, average selling prices have plummeted.

“If you look at gyroscopes and accelerometers, price erosion has been 3% to 5% per quarter,” said Mike Rosa, director of technical marketing for Applied Materials‘ 200mm Group. “Serial entrepreneurs in the MEMS space are talking about a stall. Systems vendors have been able to drive growth in high volumes, but they’ve driven the price down so far that it’s difficult to sustain business. And it’s not getting easier to manufacture these devices with every new generation.”

That view is echoed by the foundries. “These are challenging times for MEMS,” said Walter Ng, vice president of business management at UMC. “The systems guys want to drive MEMS to the commodity level even though it’s actually custom development. It’s a difficult business to support if we push it to the commodity level, which is why there is an effort to bring standardization. But along with that, there are also a lot of requests for customized processes.”

Those pricing pressures have whittled down the amount of innovation in the most popular markets because there isn’t enough margin to support it, while triggering a search for new markets where there is less cost pressure. “New entrants in the MEMS market are microphones, where they use multiple microphones for higher-quality audio input,” said Stephen Breit, vice president of engineering at Coventor. “We’re also seeing new markets for micro-ultrasound transducers for fingerprint ID and gesture detection. These have lower noise sensitivity than capacitive MEMS. There also is a lot of new functionality in cars, such as safety features.”

What are MEMS devices?
MEMS, as the name implies, are systems built through a combination of mechanical and electrical engineering. They can include some moving parts, or none at all, and they can range from relatively simple to extremely complex. Until 2006, they were largely unheard of outside of narrow markets such as airbags and inkjet cartridges. That was the year Nintendo introduced the Wii with an accelerometer built into the controller. The MEMS market exploded after that, expanding out in every direction, including smart phones and other portable devices.

MEMS chips today are used in everything from microvalves, micromirrors, pressure sensors for microphones, to labs-on-a-chip, which can test a drop of blood, for example, within minutes instead of hours.

Gears on a MEMS chip. Source: Mentor Graphics

Making these devices generally takes advantage of existing manufacturing technology, but some of those technologies are drawn from the leading edges of design. That includes a whole range of advanced techniques, including multiple types of deposition (CVD, PVD, low-pressure CVD, ALD) They also utilize a range of lithography techniques, a number of etching approaches, including deep reactive ion etching, and materials ranging from SOI substrates to scandium thin films.

Source: MEMS & Nanotechnology Exchange

MEMs chips typically fall into four categories:

• Capacitive. This technology can be used to detect anything that is conductive. They can be found in touchscreens and fingerprint sensors.
• Gyroscopic. These devices use an oscillation component to detect acceleration in any direction.
• Piezoelectric. These thin-film-based devices are still in the early rollout phase, but are expected to be used in a variety of applications, including energy harvesting. They produce electrical signals in response to mechanical stress.
• Laser-based. These devices also are in the developmental phase. The basic premise is that they can fine-tune lasers for a variety of purposes, from advanced automotive headlights to acousto-optic filters.

MEMS also typically are comprised of four basic components, according to the MEMS and Nanotechnology Exchange: microsensors, microactuators, microelectronics and microarchitectures.

What gets used where depends upon a combination of pricing pressure, which can affect R&D as well as manufacturing, as well as the maturity of the technology and how much performance is needed for a particular application.

New approaches
While MEMS really gained popularity in the mobile phone market, the sector’s role in the Internet of Things could prove much bigger. For the IoT to be successful, sensors will be required everywhere. Many of those sensors will be MEMS chips. Some will be standardized and commoditized, while others will command premiums because MEMS customers won’t have as much of a lock on the market—or at least not initially.

“Microphones are one of five devices with volumes in the billions and revenues of more than $1 billion,” said Applied’s Rosa. “What’s different is that microphones hold their gross margin. A phone has lots of cool features, but if the microphone doesn’t work it’s basically a brick. And as performance improves, there will be new applications. There are multiple applications just in a car with noise cancellation.”

He noted that MEMS for microphones are migrating from capacitive to piezo, with an improvement in performance coming from new materials, as well.

SOI is one such material. The advantage is that wafers are thicker, which makes them easier to work with and therefore potentially less expensive. Some of the early MEMS chips were based on wafers where the build-up was only two to four microns.

“There is still a lot of room to improve performance, even for inertial sensors—and especially for industrial and military applications,” said Coventor’s Breit. “For consumer markets, the existing technology is ‘good enough.’ But for IoT and wearables, they will need smaller form factors, so they will have to be integrated with other technologies. This is basically an IP play, but not on the same silicon as the rest of the chip, so it will require wafer bonding or some type of packaging. These devices will be in the tens or hundreds of microns, so using that area on a chip does not make sense for MEMS.”

Testing
Testing MEMS requires a spectrum of approaches. Because some of these chips involve mechanical devices with moving parts, tests can involve physically shaking them and measuring the range of motion or structural integrity of the parts. But this slice of the test market has its own set of issues, including cost.

“Big iron ATE is overkill and a misfit for cost-sensitive MEMS parts,” said Joey Tun, principal market development manager at National Instruments. “But there’s also very little standardization. Every semiconductor company treats MEMS as important IP and they get involved in the test. It’s a whole different world in physical stimulus. There is a lot of vertical knowledge required.”

Tun said the package itself may require a linear accelerator or rotational test. But there also are a number of do-it-yourself approaches based on open platforms, which helps in terms of customizability.

“The end customer determines how the device will be tested,” he said. “So with automotive, for example, testing will be more rigorous. Right now, about 30% to 50% of the cost for MEMS is test-related. That percentage is coming down, but the physical world still throws a wrench into testing. Physical stimulus has uncertainties of its own. It requires fine-tuning and calibration and can take a lot of time if you need to really fine-tune it.”

Technology improvements
Fine tuning and testing will become much more prevalent as the MEMS industry begins to incorporate new materials such as scandium and doped aluminum nitride for next-generation RF filters and piezo-based MEMS in inkjet printers.

“Inkjet started as solvent-based inks, but today they’re wax-based,” said Applied’s Rosa. “To jet it, you need to melt the wax to 100° C. But to jet a more viscous material, you also need an actuator that can deliver higher force. Piezo-actuated devices can deliver much higher force.”

Rosa said that scandium is making its way into RF filters and fingerprint sensors these days, as well, in concentrations of between 5% and 40%. Every 5% is the equivalent of one decibel of signal to noise ratio. “If you look at RF filters, that’s basically a bell curve that responds to frequency. What scandium does is tighten the sidewalls of the curve so there is less interference with other signals.”

So far, almost all MEMS manufacturing happens using 200mm wafers. But as 200mm fab utilization rates increase due to rising demand for automotive components and IoT devices, foundries are starting to look at 300mm for MEMS. The key question there is whether it will generate enough profitability to make the investments worthwhile.

“The scaling and re-use of flows in much of this sector is impossible, because there is no standardization,” said UMC’s Ng. “We’ve had some success focusing on specific applications to get economies of scale. But it’s also challenging for some devices. They’re physically quite small. This isn’t like a huge application processor. They have a small area, so the aggregated wafer volume is not huge.”

Nevertheless, 200mm is restrictive. A gyroscope developed at 200mm doesn’t have enough room for a large sidewall tilt. That problem disappears with 300mm tools. In addition, Rosa said that because of the increasing interplay between ASICs, many of which are manufactured using 300mm processes, and MEMS chips, it’s much simpler and more cost-effective to bond together two 300mm wafers.

“The ASIC is getting more sophisticated and there are more outputs, and customers are working on integrating more on a single chip,” Rosa said. “The three things we continue to see in MEMS are new devices to drive devices, new applications for existing devices, and continued price erosion that is forcing companies to think creatively about pricing strategies.”

Those creative strategies include wafer bonding, stacking ASICs and MEMS in an advanced package, or a fully integrated monolithic ASIC and MEMS, where the MEMS is stacked directly on CMOS. And it includes adding more materials into a PVD chamber to create new thin films.

Conclusions
There is no question that MEMS will play an increasingly important role in a variety of markets. Pressure and motion sensors are critical components of the IoT and connected devices, and MEMS are a core technology.

But price erosion will continue to haunt this market until there is either enough consolidation to create bigger players—or enough demand to overshadow the supply of these chips and warrant both a higher return on investment as well as increased investments in new technologies, methodologies and manufacturing processes.

At least for the moment, many companies are pausing to weigh their options. There are still plenty of MEMS chips being designed and manufactured, but what happens over the next couple years is not clear.

Related Stories
Flexible Sensors Begin Ramping
Technology opens up new possibilities for the Internet of Everything.
Sensors Enable ADAS
Advanced driver assistance requires a careful and complex balance of hardware, software and security.
Wearable Sensors Reveal Health Data From Sweat