MEMS: New Materials, Markets And Packaging

Challenges and opportunities in manufacturing, testing, and deploying increasingly complex microelectromechanical systems.


Semiconductor Engineering sat down to talk about future developments and challenges for microelectromechanical systems (MEMS) with Gerold Schropfer, director of MEMS products and European operations in Lam Research’s Computational Products group, and Michelle Bourke, senior director of strategic marketing for Lam’s Customer Support Business Group. What follows are excerpts of that conversation.

SE: In the past, MEMS devices were either high-volume commodities, or low-volume, highly specialized devices. Is that changing?

Schropfer: During the early years of the MEMS industry, the automotive sector was the main driver for new MEMS technology. For example, Ford acquired a fab to build their own early MEMS accelerometers. With the entrance of smart phones in our lives, the MEMS market shifted to become dominated by consumer-driven applications. While the cost pressure was high in automotive, it became much higher in consumer-focused markets like smart phones. Beyond these high-volume consumer and automotive applications, there were always a large number of specialized applications that used a low to medium volume of MEMS devices. What we see today is that some safety-critical automotive applications are being pushed toward consumer cost models, while at the same time consumer products are requiring greater and greater reliability. For example, an RF MEMS switch used in antenna tuning for a consumer-level smart phone can’t experience early failure, particularly if it is being used by millions or billions of users.

SE: What does that mean for the economics of MEMS devices?

Schropfer: New MEMS products that can leverage existing process platforms are much easier to bring to market than next-generation products that require a completely new manufacturing technology. (This does not include chip redesigns or adaptation of existing fabrication processes). Typically, MEMS actuators are more difficult to commercialize than MEMS sensors, especially if you need to include mechanical contacts in the product. These contacts require special attention during the design, process development, and material selection phases to guarantee reliable operation. A holistic engineering approach is required to accelerate development of MEMS actuators and sensors, which enables optimization of both the design and fabrication technology at the same time.

SE: What new markets are you seeing for MEMS, and what’s changing in the existing markets?

Bourke: The markets that MEMS devices address are primarily the same as they have always been — consumer electronics and automotive. The consumer electronics market is still the key driver from a growth perspective, with new and ever-increasing uses emerging. Future technologies, such as micro-speakers and gas sensing, have a smaller market today but large growth potential when adopted by mobile devices. Gas sensing and environmental sensors currently are used in engine management and monitoring the environment within the car, but a larger market would be the adoption of gas sensors within the smart home environment, and potentially mobile phones. MEMS-based oscillators would be one device that has migrated from the fab to volume production. Other new markets include:

  • Single-use lab-on-a-chip for point of care, using microfluidics, which has significant growth potential;
  • Thermal imaging, which includes microbolometers for monitoring people’s temperature, and
  • Piezoelectric micro-machined ultrasound transducers (PMUT) for medical imaging, which are low volume.

SE: In the past, MEMS devices were either high-volume commodities, or low-volume, highly specialized devices. Is that changing?

Bourke: When we think about low volume, highly specialized we think about aerospace and space applications. The high-volume MEMS devices are generally the inertial MEMS devices and RF filters that are very technology-driven. The technology to manufacture these devices continues to evolve, and with this evolution the technical challenges in manufacturing are more demanding.

SE: What makes it harder to develop one MEMS device than another?

Bourke: End markets can make some devices harder to gain approval, such as biomedical and automotive. It depends on what you need to do to qualify the device. What are the reliability requirements? How complex are the process requirements to meet the device specification? How easy is it to test the device in order to iterate the design? As an example, inertial MEMS devices need to go through the entire manufacturing process and be taken under vacuum to understand the device performance.

SE: Now that MEMS devices are being used in safety-critical markets, such as automotive, there is more attention required for safety, reliability, and potentially even security. What’s the impact of that?

Schropfer: MEMS products have a long tradition in safety-critical applications. For more than 20 years, MEMS has proven to be a very reliable technology. I started my industrial career in 1998 at SensoNor, one of the first companies to develop and produce MEMS for automotive applications, including accelerometers for airbag crash detection and pressure sensors for tire pressure monitoring. Both of these uses are truly safety-critical applications. Silicon is a fantastic material for MEMS devices, offering a high degree of mechanical reliability.

SE: How about issues like variation, packaging, die shift and test?

Schropfer: The question is less about whether MEMS can be made to be safe and reliable, but rather about how to prove and characterize this safety and reliability without spending years of development effort. We see an increasing demand for what we call “digital MEMS product qualification.” Variability in material properties and manufacturing processes can be considered in the early MEMS product development cycle, if these properties are well-characterized in advance. We can de-risk MEMS market introduction by predicting and qualifying MEMS device performance, including variability concerns, early in the development cycle using virtual experiments. Safety and reliability analysis is definitely a part of this experimentation, including aspects such as failure mode investigation and the application of virtual shock tests.

Bourke: The first MEMS device that we all typically reference is the air bag sensor in cars, so it’s safe to say they have always been used in safety-critical applications. In automotive there is a drive to zero defects associated with the devices incorporated within the vehicle. This will continue as the level of autonomy increases, which in some cases means a level of 10-8 (1 in 100 million) performance in terms of failure. This challenge means the reduction of latent defects attributable to the MEMS fabrication process that could cause in-service failures.

SE: What does the move from 200mm to 300mm accomplish? Is it better process control and higher volume, or are there other benefits? How does it affect yield? And does this require entirely new equipment for fabs?

Bourke: It allows the integration with an advanced ASIC, where they are only available at 300mm. There also is improved equipment performance of 300mm tools, as they are typically developed for a more advanced (smaller) technology node. In addition, you get the economics or reduced cost for high-volume or large-die applications. Examples would be microphones, lab-on-a-chip, and PMUTs. Another potential benefit would be the upgradeability of equipment within the fab to 300mm, so not all fabs will require new equipment. But it’s important to realize that many 300mm foundries are now manufacturing MEMS devices, so the toolsets are already in place.

SE: The MEMS market has been dealing with heterogeneous packaging issues since its inception. How does this affect stress, potential thermal mismatch, and material purity?

Schropfer: Packaging has been, and continues to be, a well-recognized commercialization barrier for MEMS-enabled products. The packaging of MEMS components differs significantly from the packaging of ICs, primarily because the functionality of the MEMS device is much more impacted by the package. MEMS devices typically contain moving parts that are very sensitive to package effects, and may often need to interact with the environment in some way. This makes MEMS devices highly susceptible to package-based mechanical and temperature stress.

Fig. 1: Heterogeneous packaging with MEMS device. Source: Lam Research

SE: So how does this affect the package design?

Schropfer: MEMS developers must consider package effects as part of their MEMS design, or risk multiple costly design-and-fabrication cycles to resolve packaging problems. One solution is to incorporate package effects in a compact or behavioral model during the design of a MEMS device, instead of simply performing finite element analysis on the device alone. Not only is simulation time reduced, but much more sophisticated simulations (such as Monte Carlo analysis, virtual testing or transient simulations) can be included in these packaging effects. MEMS engineers can use these compact or behavioral models to investigate a variety of complex MEMS design interdependencies, such as temperature dependent drift, frequency shifts, and the response of the packaged device to external stimuli such as noise sources, vibration, or shock loads.

SE: There is work underway to improve the precision of MEMS devices, particularly inertial and gyro sensors. How is this done? How do you deal with issues like drift, interference from other devices (especially with sensor fusion), and aging (longer lifetimes)?

Schropfer: To improve the precision of MEMS inertial devices, one needs to consider the complete R&D chain, starting from concept and ending in production. Focusing only on design before fabrication, or only on the manufacturing process after the design is completed, is not enough. What is required is an early exploration of the design space and process development options to optimize the design and process stages together, and to improve precision, electrical performance, reliability, and yield.

SE: Can you provide an example of this?

Schropfer: Most MEMS inertial sensors, such as MEMS gyroscopes, use comb-shaped structures for sensing and actuation, and include high-aspect ratio trenches fabricated with deep etch processing, such as deep reactive ion. The detailed sidewall profiles of these trenches have a significant impact on device performance and yield. Asymmetry or variation in the sidewall profile of a MEMS gyroscope can cause quadrature error, and result in unacceptable yield loss due to mechanical cross coupling. Variation of a few nanometers in the sidewall profile is critical, but difficult to measure by metrology. The relationship between quadrature error and structural (fabricated) profiles is difficult to understand without deep design knowledge, leading to uncertainty about etch recipe adjustments. The sidewall profile depends not only on the etch process recipe, but is strongly influenced by the geometry of the design — the so-called pattern dependency. At the end, the electrical performance of the MEMS device is the ultimate measure of success.

SE: This potentially adds to the time and ultimately the cost. What’s the solution?

Schropfer: A holistic and concurrent engineering approach is required that considers process and design optimization at the same time. Predicting electrical performance of a MEMS inertial sensor requires an understanding of both the device design (geometry) and material and process parameters. You need to co-optimize the design and the manufacturing process if you want to improve the performance, yield, and reliability of the MEMS inertial sensor. This involves a concurrent engineering approach. You need partners with strong design and manufacturing expertise if you want faster time to market, improved precision and performance, and acceptable yield. One key enabler to success is to use MEMS device modeling to link electrical measurement results to physical parameters in order to allow for unit process optimization. A predictive device model can translate between electrical results — for example, offset or compensation voltage — and physical geometries, such as sidewall angles or CD loss. These predictions can be used to select the right experiments to carry out in the fab and to reduce overall cycle-time.

Bourke: As inertial MEMS devices evolve, there is a drive to improved performance that requires tool/ process performance close to what we see from a leading semiconductor device performance. Quadrature error and resonant frequency performance of inertial MEMS devices is directly linked to the precision of the deep reactive ion etch process. As the cycle time for the devices is long, as the ultimate result requires the device being taken under vacuum, Lam and Coventor have been working with our customers to reduce the cycle time and constantly improve the precision of these processes.

SE: The RF MEMS switches are a new area. Why do we need them? What kinds of issues will this solve? And what kinds of new issues will arise?

Schropfer: We are seeing rising demand for RF MEMS switches in next-generation telecommunication systems and smart phones. Due to their mechanical nature, RF MEMS switches have several advantages over competitive technologies, including a very low resistance when closed, and very high resistance when open. Applications are numerous, and include tunable filters, MIMO antennas, tactile radio, and RF ID. This is not a new technology. The development of RF MEMS switches started more than 20 years earlier, but market success was limited at the time and only a few companies succeeded in commercializing RF MEMS switches. One main barrier to commercialization is related to reliability. RF switches need to survive billions of actuation cycles. Many first-generation MEMS switches were based on ohmic contacts. It is a huge challenge to find materials that are hard enough to sustain a large number of switching cycles, while at the same time soft enough to make good ohmic contact when closed. RF MEMS switches require a completely new fabrication technology, very different from inertial sensors and based on composite layers of mechanical materials. Reliability issues include electrical and mechanical stress in these composite materials, along with temperature dependencies and sensitivity to shocks and vibrations.

SE: What’s changed?

Schropfer: RF MEMS switches can exhibit dynamic nonlinear behavior, and understanding this behavior and the associated design/manufacturing constraints is crucial to commercializing these devices. Pull-in (closing), lift-off (opening), frequency hysteresis and transient behavior are very sensitive to device dimensions and process variations, and critical to performance and yield. You need to understand the dynamic behavior of your MEMS-based switch before you can design the system around it, including not only the MEMS chip but other IC and RF components. This needs to be done using realistic and not idealistic device models. I see an exciting future for RF MEMS switches. While development remains challenging, many of the barriers that existed 20 years ago have been removed. A lot of progress has been made in developing, stabilizing, and characterizing new manufacturing processes and materials for RF MEMS. The latest generation of switches are mostly capacitive-based devices. When a capacitive-based switch is closed, direct contact between electrodes can be avoided through mechanical stoppers, improving long-term reliability. Simulation platforms are available that can model the dynamic behavior of an RF switch, and can be used to co-design the RF switch together with the control circuitry. In addition, fabrication processes are now available to integrate an RF MEMS device on a standard CMOS process, which allows the MEMS device and its surrounding circuitry to be manufactured on the same die.

Bourke: Today the biggest challenge that RF MEMS switches face is reliability. RF MEMS switches offer potential benefits in terms of performance, but they face strong competition from ever-improving RF-SOI solutions, which are dominant in the market today.

SE: MEMS microphones have been discussed for some time. What kinds of response do these mics provide, and why is this such a small part of the market? Is it purely volume, or is this just really hard to design and manufacture? And what kinds of electronics will these be paired with? Does it require machine learning to be able to optimize sound?

Schropfer: MEMS microphones are rising in popularity. 2020 was a good year for MEMS microphone companies. The strong demand for MEMS microphones is being driven by multiple end-user markets that address requirements in lifestyle, health care, automotive and industrial applications. MEMS microphones are being used extensively in smart phones, and provide excellent voice quality with little to no background noise, enabling smart voice recognition applications. Artificial intelligence can now be used to accurately interpret voice commands, and provide innovative value-added services such as virtual assistants and instant language translation capabilities.

SE: What are the key metrics for these devices?

Schropfer: MEMS microphones are membrane-based devices. There are a number of challenges in MEMS microphone development. In every application that uses a MEMS microphone, key performance parameters are sensitivity and signal-to-noise-ratio (SNR). Increasing sensitivity by increasing membrane size is usually not cost effective. We are seeing more complex mechanical structures, such as multiple membranes and backplates within a single microphone. Membranes are thin and often composed of multi-stacked layers of different materials. Material layer stress is an issue that must be addressed in both the design and manufacturing of these microphones. Optimizing the signal-to-noise ratio requires concurrent optimization of the mechanical structure and the surrounding electronics, and these optimization requirements are all interrelated and must be addressed concurrently. The design of the MEMS and surrounding circuitry must be done at the same time, since they both influence sensitivity and SNR. Both the MEMS device and the CMOS electronics are subject to operating and process variations that must also be considered in the design. Virtual design of experiments can be used to look at the co-dependencies in the MEMS and CMOS design components, but the designer must bridge a gap between the high model accuracy required by the mechanical portion of the MEMS microphone and fast simulation speed required for the surrounding circuitry.

Bourke: MEMS microphones have been deployed for a considerable length of time. Excluding RF MEMS devices, MEMS microphones are the most established in terms of unit shipments. They are well established in mobile handsets, and with the increased use of the home assistant, such as Alexa and Google, the market continues to grow. MEMS microphones today have a similar performance to studio-quality microphones. They continue to evolve to improve the signal-to-noise ratio. In the future, we will see a move from capacitive to piezoelectric-based microphones.

SemiEngineering’s MEMS Knowledge Center
Top stories, blogs, white papers on MEMS


Mike Cummings says:

Amazing article, slightly of subjest, I have seen assembers using vacuum reflow (a feeble attempt at reducing voids in the solder of bottom terminated devices), I am conerned that MEMs will be damaged by the use of vacuum @ 250 to 260 C, pretty sure the MEMs will be degraded by this process.


Leave a Reply

(Note: This name will be displayed publicly)