First in a series: Challenges increase across the board, but for different reasons. Low profits temper investments.
The MEMS market is growing rapidly, profits not so much.
In most market segments, this would be a signal that more automation and standardization are required. But in the microelectromechanical systems world, fixes aren’t so simple. And even where something can be automated, that automation doesn’t work all the time. In fact, while MEMS devices are extremely difficult to design, build and manufacture, the business side of the market is arguably even tougher to manage.
MEMS devices are created through a combination of electrical and mechanical engineering. Included in this category are inertial sensors, such as gyroscopes and accelerometers, both of which are present in almost every mobile device to detect motion. The combination of MEMS devices is how Google’s Waze, a popular peer-to-peer GPS application, can tell if you’re stuck in traffic or moving, how fast you’re moving, and when you’re likely to arrive at your destination given current traffic conditions. There also are compasses, vibration sensors, as well as capacitive touch sensors for security.
Billions of these tiny units are sold each year, with no end in sight for increases in volume. But every MEMS maker complains that ASPs have fallen faster than sales volumes have increased. Downward price pressure has dominated this market for nearly a decade, since smartphones began replacing feature phones.
Meanwhile, there are other MEMS devices that are sold in much lower volume, such as pressure-sensitive diaphragms for microphones, tiny speakers, and ultrasound fingerprint sensors based on piezoelectric materials. Chips for these market segments provide significantly higher returns on investment, but they lack sufficient demand to achieve economies of scale. So while this portion of the market is potentially lucrative for some companies, scalability is limited and competition is growing (see chart below).
For the commodity group, the goal is to create a more standard approach to manufacturing. For the latter, the key is staying far enough ahead of the curve to maximize profits because some of this technology is still evolving.
Process and standardization
One of the missing pieces in the MEMS market is a standard PDK. It’s a good idea on paper, but it has proven harder than most people anticipated.
“The diversity of physical device structures and process architectures used for MEMS does pose a challenge to the standardization of MEMS manufacturing,” said Rakesh Kumar, senior director of MEMs at GlobalFoundries. “However, there can be some level of standardization built within a family of sensors. For example, inertial MEMS such as accelerometers and gyroscopes can be manufactured using a common platform technology with some level of customization to create differentiated functionalities. Such a platform approach will help to reduce development costs and time to market.”
So far, though, no one has been able to utilize a single process development kit for accelerometers or gyroscopes the way they have for ASICs or SoCs. Numerous industry experts say that, in general, foundry offerings look more like a menu of items than the kind of well-vetted and highly refined processes available for ASICs and SoCs. That menu can include, for example, an etch step that is out of order, or a CVD step or aluminum deposition that is out of sequence. But the menu is more about changing the order of the process steps than establishing a set of rules that can be used to trim development costs.
“Though the concept of using a standard PDK for inertial sensors is possible and has been used by some foundries, it is mostly an effort to establish a set of design criteria, such as design rules, process flow, TLR (topological layout rules) and DRC (design rule checking) for fabless companies to follow,” said Yan Qu, senior regional marketing manager at UMC. “Even though there is some merit to a foundry offering a standardized process, a fabless customer probably would view standardization as a negative offering for creating differentiated advantages.”
Process development isn’t quick, either. MEMS is complicated technology from a manufacturing standpoint, and it can vary significantly from one foundry to the next.
“On average, development of successful mass production requires about four to five years,” said UMC’s Qu. “There are many tool dependencies and equipment limitations or variations. For example, the working principle of a gyroscope requires a vacuum, special tools and eutectic bonding. A foundry has to consider carefully how much return they could generate from those added equipment costs and tools.”
The foundries provide IP libraries for those MEMS devices, which do speed time to market the same way IP does for ASICs or SoCs. And all of the foundries utilize similar equipment, which includes deep reactive ion etch tools and wafer bonders. Beyond that, however, this is largely a semi-custom or fully custom development process.
“The foundries can provide the IP library, so that you pick a sensor to match the interface specifications,” said Stephen Breit, vice president of engineering at Coventor. “That works for gyroscopes and accelerometers. And even for sensors that are still under development, the foundries can enable those to some extent. But this area still requires a fair level of expertise. With microphones, that’s a bit more commonplace. But with things like artificial noses and spectrometers, it’s a different matter.”
New complex MEMS devices
These new applications are where the MEMS market diverges sharply from the more commoditized accelerometers and gyroscopes.
“There are incremental changes in the existing sensors, but there also are new fingerprint sensors based on ultrasound technology instead of capacitive sensors, ” said Mike Rosa, director of technical marketing at Applied Materials. “So instead of just doing a capacitive measurement when you push the button, it will take an ultrasound of your finger. It will read the ridges on the epidermal layer, and it can read the ridges in the dermis (second layer of skin). That’s a high-security addition. And because it’s a piezo-electric material, you’re not prone to the moisture or dirt or particles that leave you frustrated when you try to scan your finger and it doesn’t work. That’s in the works.”
Piezo materials are coming into use in microphones, as well, which provide a higher signal-to-noise ratio. In a microphone noise is never good, so a high signal-to-noise ratio is the goal—particularly if the noise is beyond the threshold of human hearing.
“The first bump is the actual architecture,” said Rosa. “The second bump is purely related to the materials. And these are different piezo materials. In the case of microphones, it might be aluminum nitride. It might also be aluminum nitride with scandium doping in the range of 20% to 30%. In the case of microphones, companies want to do scandium doping as high as 43%. You have the potential to increase the signal-to-noise ratio with ever-higher doses of scandium.”
The list doesn’t stop there, either. Some Lidar chips, which are considered a safety requirement in autonomous vehicles, are being developed using MEMS technology, incorporating a laser diode and a mirror. Lidar scans an area using a laser beam, then reconstructs an image based upon the reflection of that beam, or pulses of a beam. While the initial mobile versions of this technology cost as much as $70,000, a MEMS-based version is expected to sell in the several-hundred-dollar range.
MEMS technology also is finding its way into some 3D scanners and retinal scanners, based upon MEMS scanning mirrors. Germany’s Fraunhofer Institute reports that MEMS scanning mirrors can be miniaturized enough to be used in mobile applications. While not as good as a scanning laser ophthalmoscope, a MEMS-based solution is much smaller, cheaper and more secure than other biometric technology because retinal blood vessels are “almost impossible to fake,” according to the research house.
Fig 5: MEMS scanning mirrors. Source: Fraunhofer Institute.
These are all interesting technologies, to be sure, but economics plays a pivotal role in this part of the MEMS market, too. For engineers working on conventional CMOS designs, there are glaring differences between the tooling for those chips and MEMS devices.
“A lot of this is done without a real flow,” said Coventor’s Breit. “Even with inertial sensors and microphones, those require complex membranes and structures, and then you need to make the connections. All of this has been ad hoc. PDKs can help automate the flow so you can produce piezo MEMS and the interconnects, anchors, and ways to connect the die. But the DRCs (design rule checks) that everyone is used to in CMOS are not there or they’re incomplete. And there is no LVS (layout versus schematic).”
Breit said another problem involves insufficient characterization of these devices, because the parameters are not as well defined as in a digital part that is mass manufactured. The result is that small changes can occur that are not fully anticipated. And that, in turn, can have an impact on sensor fusion designs, which can include multiple MEMS chips in a single device.
“This is going to create new trends,” said Babak Jamshidi, deputy director of product technology marketing at STATS ChipPAC. “An inertial sensor could house a pressure sensor and maybe an optical sensor. You’re also going to see a lot more MEMS devices in biomedical, which at this point is a very fragmented market. The ASPs are high and volumes are lower in this market, so within a decade there will be a consolidation of sensors in that market.”
The industrial sector is another area where MEMs is gaining traction. “We’re starting to see custom solutions with integrated motion and pressure sensors,” said Christophe Zinck, senior application engineering manager at ASE. “With the industrial IoT, there has been a really big push to integrate new functionality without taking up more space. So based on this, people are creating their own solutions and seeing how all the different sensors can work together. For the design team, they are developing a full module.”
GlobalFoundries’ Kumar agreed. “There is a rapid migration of MEMS sensors from discrete devices into multi-functional integrated devices such as a combination of accelerometer and gyroscope at the wafer level, together with a magnetometer, pressure sensor and ASIC in a system-in-package. A wider adoption of MEMS products into the Internet of Things (IoT) will continue to drive integration of MEMS and low-power ASICs for processing and connectivity.”
The MEMS market has experienced both explosive growth and crushing reductions in average selling prices. But as the technology finds new uses, both supplementing and replacing much more expensive devices, the economics of this market could change significantly.
Part two of this series will examine those changes in detail, including how to improve yield, thermal and test issues, packaging options, and what’s missing from the tool chain—all pieces that could help improve the cost equation for companies working in this market and their return on investment.
The Trouble With MEMS
Severe price erosion is putting this whole sector under pressure at a time when demand is growing.