MEMS Integration Challenges, Today and Tomorrow
As MEMS component suppliers compete to offer more functionality in their components, companies will be driven to adopt a MEMS design-automation platform that can most efficiently integrate multiple technologies.
MEMS integration means different things to different audiences. To pioneers in the MEMS industry, integration may imply a monolithic fabrication process, in which the MEMS and accompanying CMOS electronics are fabricated on the same die. MEMS design automation software suppliers have a broader view of integration, one the includes combining MEMS with CMOS electronics, whether in a monolithic process, as separate die in the same package, or even in separate packages. And integration encompasses packaging effects on the MEMS as well. The electronics are analog/mixed-signal (A/MS) circuits that provide electrical input to the MEMS and perform A/D conversion on its output. Such circuits are designed and simulated at a high level of abstraction with tools such as MATLAB and Simulink, and at lower levels of abstraction with EDA software such as the Cadence Virtuoso suite. MEMS designers, therefore, must deliver models of their designs that are compatible with the tools of choice for electronics design. The integration challenge is a major focus throughout the MEMS ecosystem.
Of course, integration can be a matter of perspective: there are even higher levels of MEMS integration if you view things from a system company’s point of view. To put it simply, one supplier’s system is another’s component. Qualcomm, for example, views the integration challenge from a different perspective. Recently, Len Sheynblat, VP of technology at Qualcomm, spoke at an industry conference about the challenges of integrating sensors into mobile handsets. Today’s smartphones have as many as 14 sensor types, many of them based on MEMS technology. Sheynblat pointed out there are currently more than 18 sensor vendors offering more than 26 sensor product lines. With no standards for sensor I/O, even across products from the same vendor, the job of selecting and integrating sensors is unnecessarily time consuming. Even the data sheet specs for sensors of the same type are not standardized. The difficulty of sensor selection and integration flies in the face of increasing pressure on handset makers to bring new generations of handsets to market in ever-shorter cycles (now as short as six months). In fact, Sheynblat stated that with prices dropping for the sensors themselves, the integration costs comprise a growing proportion of the end-product cost. If sensor integration is challenging for a company with Qualcomm’s technical and financial resources, one can easily imagine that other companies face similar if not greater challenges.
Easier Integration Through Standards
Most industry observers suggest that the way forward is through standardization of hardware and software interfaces. On the hardware side, the system companies would like to see standardization at the digital interface level, such as compatibility with I2C bus. To address this requirement, MEMS-based sensors such as microphones, accelerometers and gyroscopes are increasingly being labeled as digital,meaning they have a digital rather than an analog hardware interface. Thus a MEMS component may actually be a sub-system that combines a microcontroller, an A/MS ASIC and the MEMS sensing element in a single package. A 10 degree-of-freedom motion sensor component, for example, includes a 3-axis accelerometer, 3-axis gyro (usually two or three- separate MEMS), a 3-axis magnetometer (compass) and a barometric pressure sensor, all packaged together with A/MS circuits and a microcontroller. But there are opportunities for standardization of higher-level software layers as well. For instance, combining the output from 10 sensors into the position and orientation information that’s needed by application developers is a nontrivial task that is now handled by “sensor fusion” software. Leading vendors of motion sensors such as ST Microelectronics and InvenSense, as well as a number of software startups, are providing sensor-fusion software to make it easier for application developers to utilize input from motion sensors.
The standardization of software and hardware interfaces to MEMS components will surely happen, either as a result of industry standardization efforts or de facto standards imposed by a dominant component supplier. At the physical level, perhaps interfacing with I2C bus will become mandatory for market success. At the application level, the programming interfaces to sensor fusion software will no doubt converge. The crucial question, for both system integrators and MEMS component suppliers is: When? The standardization and/or convergence of interfaces will undoubtedly take a number of years. In the interim, both MEMS component suppliers and system integrators will continue to face substantial integration costs, both in engineering effort and time to market.
Abstraction Shields Complexity
As usually occurs when systems get more complex, higher-level layers of MEMS-enabled systems are being shielded from lower levels by interfaces that abstract details of the lower layers. The emerging stack of software and hardware layers for sensor-enabled smartphones is illustrated in Figure 1. At the highest level are sensor-enabled apps such as augmented reality apps that use knowledge of the phone’s position and orientation to annotate the camera output with useful information. The augmented reality app in turn makes use of the sensor-fusion software, which runs as a service provided by the phone’s operating system. The operating system requires generic or vendor-specific software drivers to interface with the sensor components via a hardware bus such as I2C. And the sensor components include their own stack, from firmware, to microcontroller to A/MS circuitry and finally the MEMS sensing elements themselves.
 |
| Figure 1: The stack of software and hardware layers in a sensor-enabled smart phone or tablet |
We are in a transitional period in which the dominant MEMS suppliers are racing to provide vertical integration across the stack. For instance, a MEMS supplier that also supplies sensor-fusion software (with a proprietary interface) hopes to gain competitive advantage by locking in apps developers. Once an app is developed to one vendor’s motion interface, porting to another vendor’s interface may require a considerable effort. Meanwhile, system integrators, app developers and operating system vendors would prefer a sensor-fusion layer that’s independent of the underlying sensing components, allowing them more choices of MEMS component suppliers. Recognizing this need, startups such as Movea and Sensor Platforms offer sensor-fusion interfaces that are hardware independent. Under the covers, these startups have to do the integration with different hardware, and that’s part of their value-add. The competition between vertically integrated suppliers and new entrants who offer value by isolating systems integrators from vendor-specific interfaces will continue at all levels of the sensor stack. Over time, the market momentum could shift from vertically integrated MEMS suppliers to companies that specialize in horizontal layers, as it has in other, more mature sectors of the technology industry.
Integrating MEMS Requires Multiple Levels of Abstraction
Horizontal stratification is also occurring at the lower levels of the sensor stack. Fabless MEMS companies such as Knowles Acoustics and Invensense are using MEMS foundry services to make inroads in a market previously dominated by IDMs with captive fabs. They are delivering components that offer digital interfaces and combine ASICS with MEMS sensing elements in a single package. But the race is on to integrate more and more sensing capabilities into the same package. To remain competitive, these companies must either develop a full suite of sensors themselves or partner with specialized suppliers of complementary sensors. For example, a company that specializes in MEMS motion sensors (accelerometers and gyros) might partner with a company that specializes in magnetometers. Either way, that word “integration” comes up again. And that’s where MEMS design-automation solutions enter the picture.
 |
| Figure 2: Coventor’s vision for a MEMS design automation platform and MEMS development kits (MDKs) that complement ASIC design tools and methodology |
System architects and ASIC designers who are responsible for designing a MEMS component (microcontroller plus ASIC plus multiple MEMS sensing elements) need to work at multiple levels of abstraction. At the highest level of abstraction, the algorithmic level, they’ll likely use MATLAB and Simulink to simulate the sensor(s) in combination with electronics and even digital signal processing. In order to simulate the whole system (again, a component to a handset maker), they’ll need a schematic symbol and underlying model for each MEMS device. At a lower level, the ASIC designers will use EDA tools like Cadence Virtuoso to create and verify a circuit design. They, too, will need a schematic symbol and underlying model to place in their circuit schematic. The design of the MEMS devices themselves will likely remain in the hands of specialists who we call MEMS designers. Those specialists must hand over models of their MEMS devices that can be simulated in the MATLAB/Simulink and Cadence Virtuoso. Today, the handoff from MEMS designers to system architects and ASIC designers is almost entirely manual. Besides being labor-intensive and a source of human errors, the manually generated MEMS models do not capture all of the physical behavior, leaving the possibility of unexpected interactions between the ASIC and the MEMS that can be difficult to diagnose further down the line.
The challenges of integrating ASICs and MEMS are now being addressed through sophisticated design automation systems, such as the Coventor MEMS+ platform, that enable MEMS designers to automatically generate models of their device that run in MATLAB, Simulink or the Virtuoso environment. Ideally, this platform would provide automatic links to layout and the ASIC verification flow as well. As demand for MEMS-enabled systems grows, and MEMS component suppliers compete to offer ever more functionality in their components, we anticipate that most MEMS companies will be driven to adopt a MEMS design-automation platform. At the MEMS component level and all higher levels of the sensor stack, the spoils will go to the companies that can most efficiently integrate multiple technologies.
Dr. Stephen R. Breit is vice president of engineering at Coventor, Inc. He has been responsible for overseeing R&D on all products since joining Coventor in 2000. Prior to that, he was director of embedded technology for Dragon Systems and held management positions at Kendall Square Research and at BBN Systems and Technologies. Dr. Breit has 25 years of experience in developing technical and scientific applications software. He holds a Ph.D. in engineering from Massachusetts Institute of Technology.
