MEMS Market Shifting

The MEMS sector is beginning to look more promising, bolstered by new end-market demand and different packaging options that require more advanced engineering, processes and new materials. All of this points to higher selling prices, which are long overdue in this space.

For years, the market for microelectromechanical systems was populated by too many companies vying for too few opportunities. Some devices became commoditized to the point where costs failed to keep up with selling price reductions. Even in the more specialized and higher-margin fringes of this segment, such as MEMS-based microphones and speakers, market sizes were too small to support more than a handful of smaller companies.

But over the past year, the overall landscape has changed significantly. As with many other parts of the semiconductor industry, this sector has experienced some high-profile consolidation. Broadcom merged with Avago ($37B), TDK acquired InvenSense ($1.3B), and Qualcomm has signed a deal to acquire NXP/Freescale ($47B). That leaves fewer big companies in direct competition, and the impact could be sizable for the highest volume portion of this market, which also is where the steepest price erosion has occurred (See Fig. 1 below). While commoditization is expected to continue, vendors predict price drops will be slower than in the past.

Fig. 1: MEMS market 2016 vs. 2015, in US$ million. Source: Yole Développement, May 2017

Concurrently, though, new opportunities are opening up in markets such as automotive, drones, robotics and IoT/IIoT, all of which require sophisticated MEMS designs.

Hubbub over hubs 
MEMS always has been an umbrella term for two separate markets. On one side are gyroscopes, accelerometers and magnetometers, which are present in billions of smart phones and tablets. These chips are difficult to develop, package and test, but the lure of huge volumes has drawn a lot of companies and produced some intensive competition. Despite the volumes, few companies in this space showed any significant growth between 2015 and 2016.

The latest strategy to circumvent that trend is by building sensor hubs, which MEMS vendors are developing for markets such as automotive and IoT. This goes a step beyond sensor fusion. The goal here is to package sensors together into a more standardized format, which is basically a plug-and-play platform. That reduces the cost to design, manufacture and sell these devices on the chipmaker side, while making it easier to customize and integrate on the system vendor side. And it adds predictability for both sides.

“The trend is to implement that capability into the sensor so it can be used in a sensor hub,” said Jay Esfandyari, senior manager of MEMS product marketing at STMicroelectronics. “That doesn’t mean you have to put it into a hub, but you do have the capability. So you may want an accelerator and a gyroscope in a hub, and you may want to add a magnetometer or a pressure sensor, or you may want all of them. The customer can select which sensors they want and connect it to the module.”

That sounds straightforward enough, but it gets more complicated. The question is how these hubs will being integrated into larger systems. And that feeds into a discussion that is ongoing in a number of markets, which is how to split up the processing of data in these systems. As more sensors are added, it becomes too expensive and far too slow to process all of the data centrally. Some of that data needs to be processed locally, but how much isn’t clear.

“The big question is how you weight input to decide what’s more important,” said Steven Woo, distinguished inventor at Rambus. “How are you going to make sense of all of this data? There is a lot of exploration going on right now and a lot of tools are available to do everything in one place.”

That viewpoint is being echoed across the industry. “We’re seeing sensor fusion in the MEMS market, particularly as these companies try to put more value into the system,” said Jeff Miller, product marketing manager for electronic design systems in the Deep Submicron Division of Mentor, a Siemens Business: “An integrated sensor with multiple degrees of freedom and a multi-access GPU, versus a standard MEMS sensor, provides a lot more value. But it requires a lot more integration to do that sensor fusion. For example, if you were to take apart one of the Amazon Echos, you’d find seven microphones carefully arranged in order to do that far-field voice detection. That requires quite a bit of sensor fusion in order to isolate the voice that’s speaking. Combining IMUs (inertial measurement units) is a great way to add value to what otherwise would be undifferentiated sensors. That typically involves a processor, as well.”

Integration concerns
Along with tighter integration, systems vendors also need to understand how these devices potentially can interact with each other. Like most chips, MEMS devices are sensitive to heat, which can affect performance in extreme environments such as cars. But because they are mechanical as well as electrical, they also are sensitive to vibration and other types of noise. That requires more extensive characterization.

“Characterization gets harder when they’re on the same die or packaged together, and it’s different for each type sensor,” said Stephen Breit, senior director for MEMS at Coventor, a Lam Research Company. “You can get cross-coupling effects with gyroscopes and accelerators. A gyroscope has a drive mode that can couple with sensors in close proximity.”

There are ways to mitigate these issues. Vibration in a gyroscope can be dealt with like any other noise. It has a predictable frequency, so it can be filtered out. But that noise also can be interpreted as a signal to other nearby devices, so it has to be accounted for beyond just a single MEMS device.

“It’s very situational,” Breit said. “With inertial sensors, a big concern is quadrature effects, which is the coupling between sensing axes. That can happen for various reasons. There are manufacturing variables, so you have to tune designs generation by generation.”

There is little history or experience to leverage here. This is partly due to the fact that the devices themselves are evolving and the potential interactions are unknown. It’s also partly because these are being used in new markets such as drones and robotics, or in evolving markets such as automotive. In all of these segments, there are no roadmaps for how technology will be used in the future, and what ultimately will be integrated into the same system or package.

“Companies are looking for a whole solution, especially in the packaging space,” said Mike Rosa, managing director of strategic and technical marketing for Applied Materials’ 200mm Equipment Group. “You see some of these companies integrating MEMS with an ASIC and TSVs, but they want the whole process flow with end-to-end integration on a package, and then they want help to farm that out to an OSAT. The challenge is how to support all of that.”

Precision becomes critical
Within a number of MEMS end-market segments, there also is a need for greater precision. An automotive accelerometer used in an ADAS system, for example, needs to be much more accurate than one used in a smart phone. While this translates into added value and higher selling prices, the amount of engineering work and the complexity of that work goes up significantly. In some cases, it will require new skill sets.

“Precision requirements for autonomous vehicles is much higher than for an airbag or rollover, which typically measure when you exceed a threshold,” said Coventor’s Breit. “Now, inertial sensors are being used for dead reckoning for a period of time. There is a race going on to achieve much higher specs for these devices, and whoever gets there first will have a real competitive edge.”

This isn’t a trivial amount of work. “You need to account for noise biasing and stability, temperature and sensitive,” said ST’s Esfandyari. “You need to reduce noise, increase bias stability and improve resolution. So a gyroscope that was 1,000 degrees/second now needs to be 4,000 degrees/second. And the noise floor now needs to be very low. Some of that is done on the ASIC side, where you deal with averaging and remove the noise. You have to make sure the IC is low-noise, too.”

Some of these devices require new materials, as well, to step up the accuracy.

“The architecture has not changed, but the specs are tightening,” said Applied’s Rosa. “Fingerprint sensors and microphones are moving from capacitative to piezo-based materials. The previous spec for a microphone’s cone fingers were plus/minus 0.5 degree tilt on the sidewalls, and in practice the center of the wafer had straight sidewalls and the edge of the wafer would tilt out of spec. That created a quadrature error, which would say, for example, that a phone was tilted. So the spec is now between plus/minus 0.5 to 0.3 degree tilt.”

Fingerprint sensors are changing to add more security into devices, as well. The newer devices can read fingerprints into the grooves of the finger using a laser. “Our customers are looking at new optical films for this, the integration of III-V materials on a silicon wafer, and new optical coatings,” Rosa said.

Fig. 2: USound’s MEMS-based speakers.

IoT/IIoT
Even with less-expensive devices, there is more engineering involved. The barrier for entry is rising, and so is the base level of expertise that is required to develop these devices.

“The answer to putting more intelligence into the MCU is to use cheaper sensors and then do more processing on the signals that come back from those sensors,” said John Tinson, vice president of sales at Sondrel Ltd. “If you reduce the component counts, you are reducing the total number of units. That is one way to go. It does have design complexities, because nothing is for free. But sensor hubs are a design direction, and they will get more popular.”

That raises some questions about whether they will supplant other devices or coexist alongside them. Mike Eftimakis, IoT product manager in ARM’s Systems and Software Group, believes it will be the latter, particularly with the use of more sensors and the need for processing more data generated by those sensors.

“If you think about IoT, in general, it’s about getting data from the field and streaming that to the cloud,” Eftimakis said. “If you do that with billions or trillions of devices, it’s not possible to get all of that data to the cloud. So you need some kind of up-front processing. If it’s on the same chip, it’s sensor fusion. But it also can be done with gateway processors or some other interim states in the RTF (rich text format) of the devices that go to the cloud. So there will be more and more requirements for processing at the edge, being able to process more and more data locally.”

In either case, this is good news for the MEMS world, because many of these sensors are based upon the integration of mechanical and electrical components. And that extends from the front end design all the way through to manufacturing and materials.

“The fundamental expansion of the 28/22nm RF nodes and the development of MEMS sensor process skills are two critical links to support the wide variety of applications being deployed for autonomous driving, while cloud-based infrastructure will handle the massive machine-to-machine (M2M) data processing requirements,” said Wenchi Ting, associate vice president in charge of UMC’s Specialty Technology Division.

Conclusion
As more connected devices hit the market, the need to connect the physical world to the digital world will require enormous numbers of sensors. Many of those will be based on MEMS technologies, and an increasing percentage of those will require highly specialized, precision devices. For the MEMS market, this looks promising on a number of fronts, from new markets to higher-value silicon. But now it’s also much harder work, and  companies that can get out ahead in this space may be poised to take a much bigger cut of the field.

Related Stories
MEMS: Improving Cost And Yield
Second in a series: New packaging options could help boost profitability, but testing and thermal issues remain problematic.
MEMS: A Tale Of Two Tough Markets
First in a series: Challenges increase across the board, but for different reasons. Low profits temper investments.
How Testing MEMS, Sensors Is Different
These devices require more than an electrical input and output.
What’s Up With MEMS?
New technologies and different ways of integrating more intelligence are generating new opportunities in the MEMS market.