Microelectromechanical systems (MEMS) can be realized in many fashions. To date, however, pressure sensors and accelerometers have been the majority of high-volume applications. In fact, a number of these devices are currently produced in quantities greater than a million devices per month. As time goes on, other MEMS devices, including rate sensors/gyros, optical switches, RF MEMS, and chem/bio MEMS, will experience widespread application.
The early focus on the development of MEMS has been on device design. Now that MEMS are being commercialized at an ever increasing rate, the focus is on delivering a robust and cost-sensitive product. Packaging and testing have become major ways to differentiate between products. Furthermore, the cost of undertaking these activities is receiving great attention by manufacturers. Typically, the cost of testing can be as much as 33%, with packaging from 33% to 45%, of the solution's total cost.
Any company that wishes to be a market leader must pay attention to this fact. No longer will it be acceptable to design the test strategy after the design of the device and package. Testing issues must instead be included in the overall design of the device/package in the early phase of development.
Testing of MEMS devices can occur at various stages during the development of a MEMS solution. Figure 1 is a flow chart illustrating a standard MEMS development cycle. From early on in the determination of material properties by real life testing to final acceptance testing, MEMS are a major candidate for extensive testing. This article will focus on front end, or materials testing, and the back end, the reliability and acceptance testing.
The current rapid pace of materials development for MEMS rivals the similar development experienced by the steel industry about a century ago. New MEMS materials are continually developed to exploit specific material properties and process compatibilities. A decade from now, the structural material of choice for MEMS might not be any of the materials utilized in today's devices. A sample list of materials presently used in MEMS devices appears in Table 1. The exciting pace of material development in the MEMS industry requires the development of material testing techniques that can provide the properties of these materials as they evolve.
The unknown reliability of many MEMS devices limits their incorporation into commercial products. The long-term stability of these devices can only be ensured with greater knowledge of the basic material properties and failure mechanisms of the materials employed in MEMS designs. The motivation for this knowledge arises for numerous reasons.
Some "new" materials actually aren't new but rather just being employed for the first time in MEMS. These materials include thin films for actuation, such as shape-memory alloys and multilayers for optical reflectance. Other materials are truly new "alloys" like the new class of silicon-germanium (SiGe) materials. Those provide special processing advantages over the more mature polysilicon comprising the majority of MEMS devices. The obvious advantages provided by SiGe materials, including CMOS compatibility, ensures their continued development and incorporation into commercial products.
Still, there's only limited data available on the properties and long-term performance of these SiGe alloys, including elastic stiffness, corrosion resistance, fracture toughness, and creep resistance. But, opportunities for developing additional "alloys" haven't ended with SiGe. There's no question that more MEMS materials will arise, and their properties will need to be quantified.
There are substantial challenges in measuring the properties of MEMS materials—whether those materials are new or old. The materials making up MEMS are deposited as thin films. Although the electrical characterization of thin films is well established, the mechanical characterization of the same films is difficult. There are numerous properties that need to be measured for each material, including elastic modules, yield strength, fracture toughness, fatigue resistance, corrosion resistance, creep behavior, and residual stress. No single standard of testing for any of these properties exists. Numerous tests have been proposed and are being carried out by different manufacturers and institutions. But, the rigorous comparison and benchmarking of these test techniques have yet to be performed.
The challenge of testing materials is increased by the variation in material between manufacturing sites and even between manufacturing batches. Material properties of a common MEMS material, polysilicon for example, deposited by one manufacturer can vary substantially from that deposited by another. Manufacturers even see variations in film properties between wafer batches. Furthermore, Exponent has documented variations in properties across single wafers. A map of the variation in residual stress, measured by Exponent, in a silicon nitride film across a single silicon wafer is shown in Figure 2. The variation is sufficiently large to change device performance if the design depends on a given value of residual stress.
Therefore, the question of appropriate, validated testing remains an open issue. An important effort to coordinate MEMS testing is currently taking place within the American Society of Testing and Materials Task Group (ASTM) on Structural Films and Electronic Materials, E.08.05.03. This task group is meeting to compare the various test techniques used to evaluate MEMS materials with the final objective of setting qualified, tested standards through which MEMS designers, manufacturers, and end users can communicate. The next meeting of the task group is at the Fall ASTM meeting in Orlando, Florida (Nov. 10-12).
MEMS material testing also is essential because numerous MEMS devices operate under conditions unknown in the macro world. Certain characteristics of two classes of devices, RF and optical MEMS switches, can be examined in Table 2.
The actuation frequencies provided in the table result in a number of total accumulated actuation cycles that extends far beyond what has been required in "macro" applications. Both of these classes of devices can experience fatigue and wear from contacting surfaces during individual actuation cycles. There's little or no information about fatigue or wear for virtually any engineering material under these conditions—for neither macro nor MEMS devices. As a result, lifetime predictions are device specific. Plus, given the lack of fundamental material testing, predictions aren't always largely validated by statistics.
Therefore, the MEMS industry is employing new materials under conditions that macroscopically have never been characterized. New sets of material data will have to be developed in order to assist designers and manufacturers.
One example of MEMS testing that showed an unexpected reliability issue was crack growth in silicon. Exponent has developed a MEMS fatigue and crack-growth test specimen. The specimen is a perforated, triangular plate that's both suspended above and connected to a substrate by a single beam with a stress concentration (Fig. 3). The specimen oscillates in the plane, applying bending loads to the connecting beam. The specimen is resonated until failure, and multiple specimens can be tested to evaluate the fatigue resistance and crack growth under different levels of stress. Stress versus time and the number of cycles to failure for a set of tests on single-crystal silicon is provided (Fig. 4).
These tests showed that silicon MEMS could fail under conditions of relative humidity. This was unanticipated by the MEMS community and has promoted changes in packaging techniques to prevent failure of MEMS due to absorbed moisture.
The challenges of MEMS material testing and reliability permeate all stages of MEMS commercial development. The initial concept of a MEMS product has to include possible failure modes associated with different designs and the intended application. For instance, silicon has demonstrated a sensitivity to moisture. A sensor based on silicon may need to be isolated from a moist environment, or alternative materials might have to be considered. For the last decade, Exponent employees have worked within MEMS reliability testing and have provided reliability assistance throughout the MEMS product development cycle.
The good news about MEMS material and reliability testing is that MEMS devices are operating successfully in numerous applications where the stresses and demands exceed those frequently imposed on "macro" devices. The challenge in MEMS material testing is to understand how to measure the performance of those materials so that future devices can exploit those material properties and operate at comparable levels of reliability.
There are distinct differences in the test requirements for MEMS devices as a function of the MEMS product life cycle. In each phase, the meaning of test performance also is different.
The R&D phase requires that the test system perform comprehensive laboratory functions. The test system must have the flexibility to perform a broad set of measurements over a wide range of operating parameters. The typical users are either device designers or highly skilled people with the technical background to devise their own measurement requirements. The user also needs to employ the equipment in an interactive mode. System accuracy, flexibility, and the ability to transition the current test and measurement methodology to the pilot phase determine test performance.
The pilot phase requires that the test system perform the same tests that eventually will be implemented during full production. This system will provide the dual function of establishing product validation as well as production test validation. In addition to providing low-volume production capability, the test system provides data that will more closely indicate the costs of testing for the production phase. Test performance is determined by measurement throughput. It's advantageous to replicate some of the features of the R&D phase system.
The production-phase test system represents a scaled-up version of the pilot-phase test system. It usually must handle more devices at one time. Typically, it includes an automated material handling system to load and unload devices from the test system. The users may either be highly skilled production engineers or unskilled equipment operators. Test performance is determined by device throughput. Reliability and maintenance are important issues too.
Regardless of the MEMS product life-cycle stage, there's a generalized definition of a MEMS test system by functional categories. One, the computer, controls the entire system and provides communication interfaces to the outside world. Another, the instrumentation, includes all of the source and measurement electronics used during the testing process.
The stimulus is a separate assembly that actuates the MEMS device in the domain of its sensitivity. The fixturing category is an electromechanical assembly that provides a controlled path from the stimulus to the device undergoing testing while maintaining electrical contact to that same device.
The environmental control typically controls the ambient temperature of the device during testing. The material-handling/automation functional category is a robotic system that loads devices to and from the fixture assembly. It's only required during the full-production phase.
Yet another category, the application software, is a device-specific test program. Finally, the system software includes all of the nonapplication software modules which support the development of test programs and their operational use.
The Need For High Volume
The first consideration is to justify the need for high-volume equipment. Although there's some latitude in the definition of high-volume equipment, we will assume that the definition includes the ability to operate the production test process with minimal operator intervention. Furthermore, it's assumed that minimizing the skill level of the operator is desirable. Lastly, and obviously, the raw throughput rate must satisfy operational and financial objectives.
In order to define the detailed requirements, we must begin with a top-level view of the major components that comprise the high-volume test system. These are the electronic measurement system, the MEMS stimulus system, the device fixture mechanism, the environmental control system, and the material handling system.
The capabilities of each of these must be matched to the overall objectives of the MEMS device manufacturer. A typical list of objectives categories includes production test and calibration; test development; calibration and algorithm development; data collection and analysis; factory integration; and product engineering. These lists provide a starting point from where priorities can be analyzed and rationalized according to budget and schedule goals.
Several fundamental characteristics of MEMS devices influence the overall structure of the high-volume test and calibration equipment. MEMS de-vices require electronic interfaces. Performance is highly dependent on the manufacturing process. Automated material handling solutions must be customized too.
With respect to the electronics contained in a MEMS device, there's a clear trend toward a common structure. The purpose of the electronics is to apply a calibration correction function, perform analog signal conditioning, and transmit the output signal. Therefore, the structure of each MEMS device can then be separated into three parts.
The first, the MEMS element, is the raw transducer or actuator. The next is the signal conditioning circuit, which may include amplifiers and filters. These circuits must be adjusted, or trimmed, on a per-device basis. The third part is memory, which is used to store the calibration constants determined during the trimming process.
Early versions of signal-conditioning circuits were laser trimmed. That gives only simple correction capability. Next came custom fixed-function circuits with memory registers. The trend has continued to include fixed-state machines with multiple register sets. Now we are seeing fully programmable microcontrollers. Eventually there will be full microprocessors. All of this is geared toward providing better performance. It's this performance advantage at a low price that determines which MEMS manufacturers will be successful. When all of these parameters are taken into consideration, the determination of requirements can commence.
Given the intense competition in any of the high-volume MEMS applications, manufacturers are faced with compressed development schedules. This means that production capability must be flexible to accommodate changing specifications and production rates.
Therefore, test requirements must be clearly defined for the life of the MEMS product and some consideration must be made for future requirements. Typical functional requirements categories for sensors include sensitivity correction, offset correction, and temperature-coefficient compensation. The electronic requirements categories for typical electronically trimmed sensors include the number of device pins, power supply, dc source and measure, ac source and measure, and digital I/O communication.
By providing the essential data for each of these requirements categories, the specification of the high-volume test system can be extracted. If specified correctly, there should be enough capacity and capability to support the production effort throughout the entire production life of the MEMS device.
One of the challenges to MEMS manufacturers is determining how much testing is required. In order to minimize cost, it's desirable to minimize the amount of testing as long as yield, performance, and quality levels can be maintained.
There are certain parameters that can be guaranteed by design and others that can be determined by sample-based testing. These generally apply to situations where statistical analysis reveals an acceptable level for the standard deviation in the production process.
If the degree of repeatability is too low, then each device must be tested explicitly on an individual basis. It's the nature of many MEMS devices to fall into this category. There's usually a high degree of variability that's an intrinsic characteristic of the MEMS element and its sensitivity to the mechanics of the production process.
One role of the production test equipment is to provide a mechanism to collect data for analysis. This is to determine if there are test optimizations that can be applied.
It's important to understand variability of the MEMS product on a continual basis. So, the capability must fully characterize devices and be able to change tests and measurements on an ongoing basis.
Although many solutions exist for testing standard semiconductor devices, which are strictly electronic, MEMS devices have both an electronic instrumentation element as well as a mechanical, optical, or chemical instrumentation element. This introduces a whole new perspective on physical instrumentation. Much of the existing physical instrumentation is geared for laboratory use and may not be easily scaled for high-volume use. These might exist as individual instruments, but may not be easily integrated into production machinery.
For example, as mentioned earlier, accelerometers and pressure sensors are well established high-volume MEMS devices. They can each be discussed to highlight how their differences affect the implementation of high-volume production test equipment.
Automated material handling systems for MEMS device production are a significant and costly element of any production system due to their high level of customization. The key consideration is how many devices must be processed at one time.
For pressure devices, the tendency is to manipulate as large a batch as possible, partly because of the dynamics of controlling the pressure stimulus. It also is due to the need to have robust pneumatic connections. Furthermore, it's quite possible to connect many pressure devices in parallel.
This is a sharp contrast to accelerometer devices. Because the effective area of the accelerometer stimulus is limited, only a small number of devices can typically be processed at one time.
The device I/O pins include all electrical contacts to the device. Both the pressure devices and the accelerometer devices share similar requirements. Each has power, ground, and, typically, a digital communication function and an output pin.
One important difference relates to the output pin. The accelerometer is a device whose operational signal bandwidth is high in comparison to that of a pressure device. How sophisticated the pin electronics and signal processing electronics must be is dictated by these differences.
Most calibration compensation circuits have provisions to compensate for temperature coefficients. The degree to which the temperature effects behave in a linear pattern determines how many temperature set points are required in the calibration process.
Temperature behavior is primarily determined by the intrinsic method of sensing—namely, resistive or capacitive. Normally, piezoresistive devices require more temperature compensation. Basically, because the typical pressure sensor is piezoresistive and the typical accelerometer is capacitive, that pressure sensor requires more complicated temperature compensation techniques.
The role of the device fixture is to provide electrical and mechanical connections to the device. The major concern for pressure fixtures is pressure leaks, while the major concern for accelerometer fixtures is mechanical resonance.
Another important yet subtle aspect of the fixture design is that the fixture itself becomes part of the signal measurement path. As a result, a wider range of technical disciplines is required to develop a robust production solution.
We have focused our attention on the testing of the currently popular MEMS devices, specifically pressure sensors and accelerometers. Major challenges still exist in the development of high-performance and cost-effective testing of these mature devices. As the commercialization of more-complex nonphysical input-stimulus devices becomes a reality, much-more-complex testing system and reliability analysis will be required.
The level of complexity of this problem will be reduced on one hand by the wealth of knowledge gained to date regarding the testing of physical sensors. Complexity will be dramatically increased by the difficulties in creating and calibrating systems associated with RF, chemical, and photonic input. Success for MEMS device manufacturers will depend on the ability to specify and procure the specialized equipment that meets their detailed requirements.
MEMS testing and reliability will be addressed at Commercialization of Microsystems 2000, to be held in Santa Fe, N.M., Sept. 5-9. Information on the conference is available at www.asm.unm.edu/mot/coms/COMS2000.htm.