Thinking outside the chip: MEMS-based systems solutions
by Roger H. Grace, Roger Grace Associates
In the September/October issue of Small Times (p.32) I introduced a “MEMS Commercialization Report Card” which addressed 14 barriers needed to be overcome to realize a successful commercialization process for MEMS. Three of the “tier one” barriers are marketing, infrastructure, and design for manufacturing and test/calibration (DFM/DFTC). In this article, I will solely address the issues of DFM/DFTC as it relates to the concept of MEMS-based system solutions vs. the issue of individual MEMS devices.
Figure 1: Early version of ADXL accelerometer showing high degree of monolithic functional integration of signal conditioning sharing the same chip with the accelerometer.
MEMS devices and their development have historically been the major focus of interest and resources for MEMS providers. One of the purposes of this article is to provide MEMS designers as well as people who are specifying MEMS into an application some “food for thought” and encourage them to “think outside the chip” (i.e. the MEMS device). Since the “S” in MEMS stands for systems, I believe that we need to look carefully at all of the elements that provide functionalities and support to MEMS devices so that they can be optimally useful in a microelectronic-based solution.
MEMS-based system solution considerations
If one can “think outside the chip” when designing or specifying a system, one quickly realizes that the MEMS device plays a small (but important) role in the overall system solution. The raw physical, chemical, and optical signal with which a MEMS device frequently interacts requires a great deal of signal conditioning applied to it before it becomes a useful source of data/information to the application. Typically, the signal conditioning consists of functions including analog-to-digital conversion, amplification, comparators, programmable memory (E2PROM), filtering, and temperature sensor(s) for compensation. A big decision must be reached at the outset of the design process as to where these functions should reside–whether on the same chip as the MEMS, or on another totally separate chip that will be connected to the MEMS chip. In addition, the system designer needs to understand where the functions of ESD protection, control logic, embedded software, and power management should reside.
Figure 2: Chip stacking of an accelerometer die with signal conditioning ASIC.
Once this partitioning strategy has been established, the designer/system architect needs to look at the need for this “system” to communicate with the outside world either from a wired or wireless format. Finally, all of these system functionalities need to be interconnected, packaged, and tested.
The main point here is that if a MEMS-based system solution is to be commercially viable, the system electronic and mechanical architecture for its creation must be addressed from day one by all the members of the design/development team in order to optimize the price-performance of the resulting solution–which includes provisions for packaging, test, and calibration.
Signal conditioning/functionality partitioning tradeoffs
Looking at current industry examples of system partitioning strategies, it’s clear there exist numerous various opinions and approaches to accomplish this. The main criteria for selecting the optimum partitioning strategy is based on a number of factors including unit cost, NRE, process compatibility, production volume, and–most importantly–performance. Analog Devices and its ADXL line of accelerometers (Figure 1) have historically taken the approach of integrating the signal conditioning functions with the company’s capacitive accelerometer on the same piece of silicon (though not for the entire ADXL product line). This highly integrated monolithically integrated approach was also adopted by MEMSIC in its accelerometer.
Conversely, organizations including Freescale (Figure 2), Kionix, VTI, and STMicroelectronics have adopted a two-chip approach that separates the accelerometer sensor chip from the signal conditioning ASIC.
Figure 3: MEMS CMOS analog microphone (1mm × 1mm) showing high degree of monolithic functional integration of signal conditioning electronics sharing same chip with the microphone.
When we address the MEMS microphones partitioning strategy, we see that Akustica (Figure 3) has adopted the monolithic approach while the new market entrants, Analog Devices and Wolfson as well as Infineon, and market pioneer and leader Knowles Acoustics, have adopted the two-chip approach. Pulse (formerly Sonion) has a three-chip approach in which the microphone and ASIC are side-by-side mounted on a silicon substrate.
Eric Eisenhut, VP of sales and marketing at Kionix, notes that the digital content of ASICs has increased dramatically, evolving rapidly due to market demand. “We offer a number of ASIC options to our customers that provide a broad range of functionalities to best suit their specific application,” he says. “Our ASIC solutions provide the capability to optimize our MEMS device performance ...we do not need to have a perfect sensor, since the algorithms that we have developed take into account the tolerances of the manufacturing process.”
Figure 4: Chip-on-MEMS three axis accelerometer (2mm × 2mm) showing SOI accelerometer stacked with ASIC. (Courtesy: VTI)
On the microphone side, Marcie Weinstein, director of strategic marketing at Akustica, touts the company’s “intimate knowledge of the interface between our transducer and the surrounding electronics” as the key to successfully designing a CMOS MEMS microphone, enabling the company to innovate quickly and introduce new products. As we can see, the ASIC and its functionalities provide MEMS suppliers with a great deal of flexibility to satisfy customers’ varied needs.
Unlike many semiconductors, MEMS devices need to be in contact with the environment they measure. In the case of an accelerometer or rate sensor (gyro), these devices find themselves mounted on a PC board or substrate. With pressure sensors, however, many applications consist of harsh media (e.g., engine oil or human blood) and require mechanical attachments (e.g., screw threads). These conditions require packages that must be robust but at the same time low-cost, and must not influence the measurement of the sensor by imparting stresses to it.
It is a well-known fact that MEMS packaging and test/calibration typically constitute over 65% of the total cost of the solution. Recently, the wafer-level packaging (WLP) technology popular in the semiconductor industry has migrated into the MEMS area, which is a common phenomenon. While WLP can effectively enclose or encapsulate a MEMS device and provide it with bump solder capability, quite frequently these devices find themselves inside of another “mechanical” package. Most recently, chip-stacking approaches have become popular with MEMS, as in the case of the Freescale line of accelerometers, as well as the “chip-on-MEMS” chip-stacking approach of VTI (Figure 4), which encapsulates a silicon-on-insulator three-axis accelerometer between two glass silicon capping chips. A “redistribution layer” connects the ASIC to the accelerometer via a flip-chip bonding approach and routes the signals from the accelerometer via the solder bumps to the system. As such, this provides a standalone to packaging requiring no additional external package, and associated manufacturing and package cost.
“Earlier versions of our accelerometers used a side-by-side approach of the accelerometer sensor and its signal conditioning ASIC,” said Ray Roop, member of Freescale’s technical staff. The current Freescale approach has a chip stack with bond wires connecting the accelerometer chip to the ASIC chip. It is interesting to note that Analog Devices selected a two-chip approach for its MEMS microphone, “because we determined that they were not able to achieve the requisite level of performance in dynamic range and frequency bandwidth associated with a single-chip approach because of the single chips’ coupling resonance effects with the package,” stated Harvey Weinstein, manager of ADI’s MEMS applications group. The goal, he added, was to achieve performance levels “much more demanding than those required by portable electronics, which would include high-quality audio applications.” [Stay tuned for upcoming special features in Small Times on MEMS packaging and MEMS testing.]
Testing MEMS devices requires a major set of unique considerations vs. what is typically seen in semiconductor test, according to Dan Popa of the U. of Texas at Arlington’s Automation and Robotics Research Institute (ARRI). “This is due to the fact that MEMS devices tend to operate in a multi-domain environment???e.g. electro-mechanical, electro-fluidic, electro-optic, electro-chemical, etc. Therefore, the MEMS device must have a stimulus representative of the similar stimulus of the intended application to be properly tested.” In the case of an accelerometer, a vibration table is required to “shake” the devices over their measurement range and over their operating temperature range. The creation of this test system is far more complex and tends to be a custom solution approach. These test system designs need to be addressed early in the MEMS development stage, and need to have scalability capability from R&D testing of the devices to full production. Typically, these systems are co-designed by the MEMS device manufacturer and their test systems integrator. Since no existing MEMS test foundries currently exist, facilities responsible for the packaging of these devices use the test systems provided by the MEMS manufactures to conduct final testing and calibration.
It is noteworthy that a recently created organization, MEMUNITY (www.memunity.org), has assumed the role of educating the MEMS industry about the intricacies of wafer-level packaging and how this approach can enable the automated, high-throughput, low-cost testing of MEMS devices. “VTI has extended the wafer-level test strategy to the active calibration of our three-axis acceler-ometer product line,” noted Scott Smyser, VTI’s VP and GM. “We believe that this approach of active calibration is unique in the industry, and helps us to dramatically reduce the cost of calibration and test of our devices while providing a 100% level of testing.”
I have proposed an approach of “thinking outside the chip” in the creation of a MEMS-based system solution, a.k.a. “MEMS modules.” This approach requires a broad-based interdisciplinary team which tends not to exist where the product needs to be developed, and requires competencies in MEMS device design, wafer processing, signal conditioning/ASIC design, packaging, testing/calibration, supply chain management, and electronics system design. With the availability of over 60 MEMS foundries worldwide, this takes the pressure off the MEMS design and wafer process. Many MEMS companies (Analog Devices, Freescale, Kionix, etc.) have internal ASIC device designers. Companies including Si-Ware Systems and Sensor Platforms provide signal conditioning ASIC design. Austria Microsystems provides both ASIC design and ASIC manufacturing capabilities. Backend network chips are available from a broad selection of suppliers.
In my opinion, the real challenge is creating MEMS-based system solutions that exactly address customers’ needs vis-à-vis classical system integration and packaging/testing approaches. Many organizations–including Axept, Crossbow, IMEC, Infotonics, LV Sensors, Tronics, U. of Texas-Arlington/ARRI, and most recently Acuity–have demonstrated the capability to undertake such a systems-based solutions approach. In addition, organizations including Fraunhofer’s Einrichtung Elektronische Nanosysteme Institute (Chemnitz, Germany), the University of Michigan’s Wireless Integrated Microsystems Research Center (WIMS), and ARRI are focusing on this approach. Axept (Figure 5) has integrated two separate pressure sensors along with a number of discrete signal conditioning ICs to create a dual-redundant pressure sensor module. It provides a virtual data acquisition platform providing operational status assessment, sensor-drift compensation, trend analysis, calibration setting by poling and voting, service history recording, smart sensor recalibration algorithm, and sensor status alerting.
UT Arlington’s ARRI (Figure 6) has created a micro-opticalelectromechanical (MOEMS) switch module that includes a MEMS die, four optical fibers, and wire bonds for interconnects inside of a (hermetically sealed) Kovar metal carrier. Tire pressure monitoring systems currently being produced in large volumes by suppliers including Bosch, Freescale, and TRW are excellent examples of MEMS-based systems solutions.
These systems (a.k.a. “modules”) embrace pressure sensors, motion sensors, temperature sensors, signal conditioning, battery and battery management, software algorithms, wireless communications, and package integration and test–and are projected to be a major “killer application” for MEMS in the very near future, approaching 100 million units worldwide in the early next decade. I believe that there exists many other major opportunities for MEMS-based system solutions in the near future. Carpe diem!
Roger H. Grace, president of Roger Grace Associates (Naples, FL), is a technology marketing consultant to the MEMS and nano industries with more than 35 years’ experience. Contact him at firstname.lastname@example.org or www.rgrace.com.