Keep Bugs Out of MEMSs
Prototype evaluation tools and production test schemes can help you get high-volume MEMS sensors out the door now.
Rick Nelson, Senior Technical Editor -- Test & Measurement World, 4/1/1999
| Microelectromechanical systems (MEMSs) are proliferating like ants at a picnic, and engineers might find them almost as annoying. Designers of MEMS fabrication facilities must prove their processes yield reliable, long-lived devices that can withstand environmental extremes. Engineers integrating MEMSs into products must develop appropriate support circuitry and verify that their designs yield the accuracy and bandwidth their applications demand. Production test engineers must adapt their ATE so it can evaluate moving microengines, as well as solid-state semiconductors, at high throughput rates. Micromachines are more than just troublemakers for test engineers; they’re poised to help out as well. For example, a MEMS microrelay holds promise for probe-card-reconfiguration applications in ATE (see “Are Micromachines in Your Future?’’ below). Near term, though, you can expect more difficulty than assistance. Today, MEMS-based accelerometers and pressure transducers are presenting test challenges in automotive applications, where the devices are used in high volumes. In university and government labs, researchers are building rotating microengines that you might need to evaluate with a tiny tachometer, instead of a DMM. Some of these contraptions look like they belong in Charlie Chaplin’s Modern Times. At least the Little Tramp is too big to get entangled in these contraptions’ tiny gearwheels, whose diameters might not exceed 300 microns. Spider mites, though, should beware. All MEMSs present complex test problems that traditional benchtop instruments and ATE don’t address. Your test needs will differ depending on whether your firm builds MEMS-based products or the micromachines themselves. If you are integrating a commercial MEMS device into a product, the device’s vendor can help, in the form of application assistance, evaluation boards, and built-in device-test capabilities. If you are responsible for performing production test on MEMS devices or of boards on which they reside, you will face challenges obtaining a test system, because off-the-shelf ATE for micromachines does not yet exist. You can, however, enlist the aid of specialists offering MEMS test services. If you are designing and evaluating novel microstructures, you can benefit from research at university and government labs. Testing Your MEMS Prototypes Analog Devices suggests that you employ a microcontroller running software polling loops to decode the ADXL210’s duty-cycle output (see “Measure a Changing Duty Cycle,’’ Test & Measurement World, September 1998, p. 25). Designing-in and testing this device is straightforward if you follow the vendor’s cookbook procedure(1). Some subtleties, though, might escape your notice without the cookbook. And keep in mind that you’ll have to write your own cookbook if you develop your own MEMS rather than buy a commercial part. The subtleties range from bandwidth to noise performance. When applying the ADXL210, you’ll have to set (via an external resistor) the square-wave frequency in accordance with your application’s bandwidth—more than twice the bandwidth of the acceleration signal you wish to measure to satisfy the Nyquist criterion. You should select (via external capacitors) the lowest analog bandwidth your application permits to minimize the effects of the ADXL210’s Gaussian white-noise characteristic. The ADXL210 has a handy self-test pin that, when asserted, alters the duty cycle of a good device by 10%. But you’ll need to perform more than this simple test before shipping your ADXL210-based product. For instance, you will have to measure and compensate for the device’s initial 0-g offset, which you can accomplish by burning a calibration factor in EEPROM. Further, you may wish to characterize your product’s accuracy and linearity over its full operating range. Testing Sensors in Production Called STeP (for Sensor Test Platform), ETEC’s platform is based on a Windows NT-based distributed client/server network architecture. Test programs written in C++ and Visual Basic control VXIbus instruments. STeP integrates the instrumentation with pressure or vibration stimulus, an optional burn-in function, and—for automated testing—a device handler. Based on STeP, ETEC can develop a MEMS sensor test system for you, beginning with specification generation. ETEC president Henry Klim says generating a concise spec is critical: The spec should cover throughput rate and other performance issues as well as cost, delivery date, and clear benchmarks for system acceptance at ETEC’s facility and, finally, at your site. Klim estimates that most applications could be served by one of the test configurations represented in Table 1.
If you’re hell-bent on designing, building, and testing your own MEMS, good luck! For help, you can turn to the MCNC MEMS Technology Applications Center, whose Multi-User MEMS Processes (MUMPs) program offers surface micromachining for prototype development. MUMPs does not support integrated electronics; in conjunction with Defense Advanced Research Project Agency (DARPA) funding, however, MCNC provides access to Analog Devices’ BiCMOS electronics technology for development of experimental devices that integrate bipolar and CMOS transistors along with moving microstructures on one substrate. You can also seek help from the MEMS Exchange, hosted by the Corporation For National Research Initiatives (CNRI). The MEMS Exchange, also with DARPA support, aims to connect MEMS designers with fabrication facilities. The organization acts as a “trusted intermediary,’’ a disinterested referral service that will neither compete with you nor solicit your business. (CNRI, a not-for-profit organization, has no fab capability and does not compete with commercial ventures.) Currently, the MEMS Exchange lists only university fab sites—Cornell, Stanford, and University of California at Berkeley—but project manager Michael Huff says several commercial sites have expressed interest in offering MEMS fab services through the MEMS Exchange and that a contract may be in place by the time this article appears. CNRI envisions a distributed network for prototyping MEMS devices, in which you could send your designs to one or more remote fabrication sites. To enable you to evaluate devices throughout the manufacturing process, CNRI is implementing a client/server system in which a Java applet remotely controls an optical microscope, which would allow you to investigate, for example, whether one manufacturing step damages microstructures built in a previous step. A demonstration (Fig. 1) went online in December.2 CNRI plans to add remote control of other instrumentation as well—including wafer probers, profilometers, and scanning electron microscopes.(3)
A commercial firm that independently offers MEMS foundry service is Advanced Custom Sensors Inc. (ACSI), which also makes off-the-shelf silicon pressure sensors. ACSI will design a MEMS package for you and build prototypes as well as production units. The firm provides manual wafer probing as well as leak, humidity, temperature, pressure, vibration, mechanical-shock, thermal-shock, and light-sensitivity tests. Convincing Customers To make TI’s “no’’ answer to these questions convincing, Mike Douglass, distinguished member of the technical staff at TI’s Digital Imaging Division, published a paper describing DMD failure mechanisms and design steps taken to compensate for them.4 One failure mechanism—hinge memory—can prevent, for instance, a mirror that spends 95% of its time in the off state from fully assuming the on state when required. TI researchers minimized hinge memory effects by controlling metal creep in the hinge material and dynamically controlling voltages during mirror transitions. TI continues to test the devices described in Douglass’s paper, and he reports that as of January they had experienced 1.7 trillion cycles with no failures related to hinge memory. A similar MEMS reliability initiative is underway at Sandia National Laboratories’ Reliability Physics Department. While TI’s MEMS technology is proprietary, you can take advantage of Sandia’s efforts by licensing its technology or by entering into a CRADA (Cooperative Research and Development Agreement) or User Facility agreement. Under a CRADA, you can work with Sandia researchers on projects that benefit you as well as the Department of Energy, (Sandia’s parent organization). Under a User Facility agreement, you perform work at a Sandia facility, with Sandia providing only the support staff required for maintenance or safety. Sandia Researchers have built a test system called SHiMMeR (Sandia High-volume Micromachine Measurement of Reliability) that can exercise 256 MEMS parts at a time while subjecting them to varying environmental conditions (Fig. 2).
Although MEMS researchers at Sandia and elsewhere had expected polysilicon fracture to be a major failure mechanism, the Sandia researchers found that MEMSs fail as a result of normal wear and tear (Fig. 3), “much like a car engine fails without oil,’’ according to department manager Bill Miller. Sandia’s devices already seem more reliable than cars—one microengine completed 7 billion revolutions without an oil change.
Miller says current research focuses on using SHiMMeR to investigate the effect of humidity on reliability. Humidity exacerbates the clumping of tiny particles that tear away from micromachines’ polysilicon at moving joints; the particle clumps in turn jam the joints, causing failures. In addition, Miller’s team is working to optimize the waveform of the electrostatic drive signals that control device motion. Optimized waveforms enhance the reliability of mechanisms such as the comb drive in a microtransmission. Pragmatic Instruments, a manufacturer of arbitrary waveform generators, has licensed one set of Sandia’s waveforms—called the Sandia Super mDriver waveforms—and makes it available in a $5000 waveform-creation software package that runs on a PC. According to Pragmatic president Henry Reinecke, in a typical application, an AWG generates a waveform based on the PC representation; the AWG output serves as input to an electrostatic amplifier, which in turn drives the MEMS device. Although MEMS sensors represent a mature technology, actuators such as Sandia’s microtransmissions and MCNC’s microrelays represent early steps on the road to cost-effective volume production. Roger Grace, president of Roger Grace Associates, a market-research firm that follows MEMS technology, expects that devices like MEMS microrelays won’t reach full commercialization until 2004. Roadblocks include not only the perfection of high-volume fabrication techniques and optimization of drive waveforms, but also development of adequate modeling of MEMS devices’ mechanical faults. In pursuit of this goal, researchers at Carnegie Mellon University have developed a tool called CARAMEL (Contamination and Reliability Analysis of Microelectromechanical Layout), which combines MEMS layout data and a contaminant description to help predict aspects of a MEMS design susceptible to contamination-induced failures.5 This approach coupled with techniques that model MEMS mechanical faults as electrical analogs will pave the way for commercial MEMS actuators to take their place beside commercial MEMS sensors in the next decade. T&MW FOOTNOTES 1. “Preliminary Technical Data: ADXL210, REV Pr.A,’’ Analog Devices, Norwood, MA, 617-329-4700, www.analog.com
FOOTNOTES 1. “Preliminary Technical Data: ADXL210, REV Pr.A,’’ Analog Devices, Norwood, MA, 617-329-4700, www.analog.com
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
