Top Takeaways From SEMI-MSIG MEMS & Sensors Executive Congress 2017

Emerging low power technologies and machine learning could shake up the sensor industry.


The MEMS and sensors sector has been talking about smarter, lower power devices forever, but this year’s recent SEMI-MSIG Executive Congress stressed the market drivers and the emerging technologies that look to bring those changes to the market. Ubiquitous sensing now demands lower power for its always-aware sensors to be useful, while acoustic wave and piezoelectric technologies are emerging to make wake-on-demand practical. Applications are also starting to find ways to use maturing machine learning technology to convert simple sensor data to mine useful information. Though who’s in and who’s out in the dynamic mobile phone business still drives the ups and downs in revenue this year and next, with time-of-flight light sensors and MEMS timing devices winning new slots, last year’s fast growth sector of fingerprint sensors sees revenues fall on increasing price pressure.

Low power may be the next disruptor
“While sensing is ubiquitous, we haven’t yet capitalized on the real potential of it,” suggested Intel Fellow and Director of its Anticipatory Computing Lab, Lama Nachman. To be really useful, technology will need to be adjusted individually for each person’s preferences, and will need to be able to offer the right information at the right time, she noted. That means systems will have to know the full context of what you are doing to figure out what you might want, with constant context information from the sensors, and with deep learning to figure out how to adjust to that context. But with the sensors always on, the battery will become the major pain point. To reduce power usage going forward, sensors will increasingly need to wake up on demand, and turn on the system only when needed, Nachman urged.

Amit Shah, Partner, Artiman Ventures, concurred that low power sensors could be a disruptive opportunity, though suppliers would be smart to capture more of the value from the component by aiming at more of the vertical stack. “Low power sensors will be big – but the enabling sensor needs to be hidden in a must-have application,” he noted.

Acoustic wave devices and piezoelectric thin films offer low-power solutions
Sensors using acoustic waves and piezoelectric thin films look like the most likely technologies for low power sensors that can wake on demand. From their review of the academic papers at the most recent Transducers event, Alissa Fitzgerald and Keith Jackson, of AMFitzgerald & Associates, highlighted the potential of emerging technologies from the research community based on common SAW filters to wake the sensor with the sonic wave from RF interrogation, to take, say, a temperature reading, eliminating the need for a battery. The technology could also work for pressure or gas sensors. Another emerging low-power approach bent a structure by differential absorption of IR on different materials on different sides, to close a circuit switch, for a sentinel IR detector of tailpipe emissions that operates with zero leakage in the off state.

Fitzgerald predicted that by 2020 sensors made of piezoelectric thin films, which can be awakened by sound waves or other motion to generate an electrical signal, and are generally lower power, more sensitive, and likely easier to manufacture, will largely replace capacitive sensing devices now made with DRIE processes, including microphones and inertial sensors. One such emerging commercial application is Vesper’s piezoelectric MEMS microphone, which is awakened by sound waves from the voice converted to an electrical signal. CEO Matt Crowley noted that other voice recognition devices now typically used 2-3W continually just while listening for something to happen, which rapidly drains their batteries. “Zero power is probably extensible to other sensors as well,” he suggested, noting that could make energy harvesting more practical. “Energy harvesting would be necessary in a trillion sensor world,” he quipped. “How many batteries will a trillion sensors need?” Finland’s VTT Lab’s James Dekker, Sr. Scientist, also reported a more efficient piezoelectric MEMS- mirror time-of-flight LIDAR solution, developed with Murata.

Fitzgerald also noted that research papers on sensors on paper and plastic have increased significantly in recent years, growing from 12 percent of Hilton Head papers in 2004 to ~50 percent in 2017, as those on silicon correspondingly declined, suggesting that innovation on silicon is slowing down in academia. She predicted demand for ubiquitous, ultra-low-cost sensors could drive development of sensors on paper and plastic by 2030, particularly for point-of-care diagnostics. Sensors for RF, temperature, humidity, pressure, and gas on these alternative substrates are also possible. Note, however, that it is also easier to do novel research for publication on a new material than on a mature and equipment-intensive technology like MEMS, and that the paper and polymer sensors would likely not replace MEMS sensors, but be for new applications.

Machine learning starts to wring useful information from basic sensor data
Sensor makers are starting to talk about real applications of deep learning. Machine learning enabled Netherlands startup Soilcares analyzes soil fertility from an image from Si-Ware’s near infrared MEMS spectrometer module taken in the field, instead of the traditional wet-chemical test in expensive laboratory equipment. Infineon reported using deep learning to extract a good blood pressure measure from simple photo-diode recordings of heart rate and pulse oximetry. Intel assumes future smart systems will use adaptive learning to match content to specific user needs.

Soilcares collected a database of some 50,000 soil samples worldwide, analyzed them with traditional wet chemistry, then used deep learning to map those results to the IR spectrum readings. With this system, the farmer can shoot an image of his soil sample with a lunch-box- sized near-IR spectrometer system, then send the result by smart phone to get the analysis and suggestions for the needed nutrients to add. With 60 percent of crop yield typically controlled by soil fertility, the easily accessible analysis could significantly improve yields anywhere in the world.

CEO and President Henri Hekman reported that sophisticated farmers in the Netherlands are seeing 2-5 percent improvements in yield from the system, while some in Kenya with more depleted soil have seen 1.5-2X improvement.

Strong automotive growth, and changes in dynamic mobile phone market, drive sensor company results this year and next
The sensor market remains driven by strong growth in automotive applications, and continued fast turnover in the dynamic mobile phone market. STMicroelectronics’ win of a slot for its laser-based proximity sensor on the front of the iPhone, to turn the screen off when the phone is next to the head for a call, will likely expand the total addressable market for such time-of-flight sensors by 6X, said Jérémie Bouchaud, Director MEMS & Sensors, IHS Markit. The iPhone use also draws other phone makers’ attention to the time-of-flight sensor, enables other applications of the sensor for range finding, and grows the volume to bring down the cost of the light sensors for autofocus as well. ST currently has 99 percent of this fast growing market, but there is room for a quick entrant to become the second source, although there’s a small window before the market saturates in the next few years.

Who’s doing best in MEMS and sensors this year? Automotive and RF device suppliers see strong growth. Bosch won this round of musical players for the IMU in the iPhone, while STM won a slot for its time-of-flight sensor. Arrows represent estimated growth in revenue in 2017 over 2016. Source: IHS

The iPhone 7 and 8 also for the first time replace the quartz timing device with SiTime’s MEMS unit, raising the credibility of the MEMS timing technology, and likely giving a major boost to the MEMS timing market, said Bouchaud. Bosch won the latest round of musical chairs for the IMU slot for the iPhone 8 to shake up that sector. Meanwhile, last year’s fast growing fingerprint sensor makers now face increasing completion that’s driving down prices and revenues. RF filters and microphones continue strong, with four MEMS mics in each iPhone and 7 in each of Amazon’s various Alexa devices, and speech recognition systems seeing expanding use.

Next year, disruption many come to the microphone sector, as InvenSense will reportedly introduce the first performance improvement for some years, with 70-74dB in a size small enough for a phone. Look also for breakthroughs in ultrasonic fingerprint sensors, the launch of the first MEMS speakers, the re-launch of MEMS autofocus, growing applications for MEMS scanners, and designs wins for automotive LIDAR, predicted Bouchaud. Little uptick is expected, however, in the disappointing industrial IoT market. “Industrial IoT has seen slower growth than we anticipated,” he noted. “For the next 5-7 years, we see few big markets here.” He suggested that adoption was likely delayed by the complexity of fusing and interpreting the collected sensor data for actionable information.

Leave a Reply