More important, more complex: MEMS metrology
03/01/2008
EXECUTIVE OVERVIEW
Just as MEMS developers are waking up to the fact that metrology tools can add value to development and fabrication of MEMS devices, tool suppliers are facing the challenges-and opportunities-of serving an increasingly complex market.
Finding commonalities among the various challenges that come with inspecting, measuring, and characterizing the myriad MEMS structures in production is not always a straightforward task. Many manufacturers have moved away from the idea that metrology and inspection are “necessary evils” that add little value to the process and have come to the realization that such measurement tools play a vital, enabling role for the successful development and fabrication of their devices.
Although some measurement synergies and overlaps exist among the various process schemes, the critical metrology needs of a fab running accelerometers or gyroscopes are quite different from those of a factory producing RF MEMS, micromirror arrays, or microfluidic systems. Then there are the questions of volume, CMOS compatibility, and IDM versus fabless/foundry business models. Are the devices in question being manufactured in quantities of thousands of wafers per month in a modern 200mm fab or produced in small lots on a customized flow? Is the process largely CMOS-compatible or a complex, specialized micromachining-heavy process with nonstandard or otherwise unusual measurement requirements? Are the devices in question being fabricated in a traditional integrated device manufacturing facility or as a result of collaboration between a fabless MEMS company and a foundry operation?
Tool suppliers tune in
MEMS manufacturers push their metrology/inspection equipment suppliers to provide enhanced capability solutions, both customized and off the shelf, for the entire range of micromechanical devices and their respective measurement and process control applications. As a result, the vendor companies have become increasingly tuned in to what their customers want and what’s going on in the fabs.
“As the market evolves and bigger players are involved, the need for accuracy, reliability, and failure analysis has become critical,” explains Noushin Dowlatshahi, MEMS product manager for Veeco Instruments. “One of the biggest barriers to more widespread adoption of MEMS in many application spaces is their perceived lack of robustness. Quantitative measurement of performance and device-failure modes is critical to continued growth in the industry.”
“With the introduction of MEMS processes in high-volume manufacturing, process control becomes more and more important,” says Volker Knorz, director of marketing for Vistec Semiconductor Systems. “Starting from development, prototyping, and first-pilot production, [the need] to control and stabilize the processes and to increase the output of good devices has come into focus now. Also, with MEMS devices for automotive applications, quality criteria and their verification are a must.”
 Figure 1. FRT combines optical metrology, AFM, point and field-of-view sensors, and topography and film thickness probes in one automated instrument. |
FRT president Thomas Fries says that since customers ask for a variety of measurements, a multi-sensor metrology tool approach (Fig. 1) “fits the MEMS community best. There is always a need for more than one feature to be measured. MEMS are 3D structures, are coated, have functions, etc. Many MEMS wafers are not according to SEMI standards, so special equipment for automated wafer handling is needed. We combine optical metrology with AFM, point sensors with field-of-view sensors, topography with film thickness, and we integrate all of these in one fully automated instrument.”
Veeco’s Dowlatshahi acknowledges that “different devices do have their own special challenges in terms of critical parameters to measure, failure modes to be characterized, and device motion. For instance, surface roughness, tilt, and curvature are important for micromirrors, but inkjet heads typically only require spacing, width, volume, and depth characterization of features. Other devices require multiple drive frequencies or unusually high currents or voltages.”
MEMS sensors employing bonded wafers also present some tricky challenges when it’s time for final-manufacturing inspection, according to Vistec’s Knorz. “The sensors get sealed by a ‘cover,’ which means that a silicon wafer is attached by using glue. The special metrology challenge is to check whether or not the sealing (gluing) has been done correctly-that is, without bubbles in between. A method [like Vistec’s infrared system] is needed that allows one to look through the silicon down to the gluing plane. Other challenges are linked with measuring the overall accuracy of elements on the wafers that have been glued together.”
The 200mm transition
Skyrocketing demand for such consumer-market MEMS devices as multiple-axis accelerometers, digital micromirrors, and silicon microphones has finally pushed some manufacturers to establish high-volume production plants. Several companies have moved or plan to move from processing on 150mm substrates to processing on 200mm wafers, seeking the economies of scale the larger silicon platters provide.
 Figure 2. Metrology plays a key role in STMicroelectronics’ Agrate, Italy, facility, one of the first 200mm MEMS fabs. |
STMicroelectronics has been ramping its new 200mm MEMS line in Agrate, Italy, for well over a year (Fig. 2). The fab, which produces MOS and MEMS in the same shell, is running at least 5,000 wafers per month, but “we have room for increasing capacity to more than 10,000 wafers,” says Orio Bellezza, the company’s group vice president of front-end technology and general manager of advanced R&D, nonvolatile memory, smart power, and MEMS. The choice to make the transition from 150mm wafers to 200mm wafers was “a natural decision for us, a natural evolution,” given the improved productivity, faster ramp times, and better utilization of installed assets that the larger wafer size affords, he believes.
Many of the metrology challenges are not necessarily wafer-size dependent and were dealt with in the 150mm line or via small-lot experiments in the 200mm line, says Paolo Ferrari, ST’s director of MEMS process development. “To minimize the risk of transfer from 150mm to 200mm processes, we took some actions. The first one was the formation of a dedicated team,” with people from R&D, MEMS specialists who developed the 150mm process, and individuals from semiconductor manufacturing “who had a lot of experience involved in high volume and process transfer from 150mm to 200mm. The competencies coming from these different groups were fundamental in the success of transition to 200mm mass production.”
They also conducted a “deep feasibility study of the start-up of the advanced process flow in the CMOS fab and did work in the labs,” says Ferrari. “We were able to anticipate possible problems that could occur in the manufacturing of 200mm MEMS,” such as the increased stress on thick, poly-silicon structural layers resulting from the use of the bigger silicon slices and the need for improved handling of wafers with through-hole structures. The 150nm to 200mm transition did result in changes in inspection tactics at ST’s MEMS fab, some of them expected, some not. “For what we call standard inspection, we adopted the standard 200mm sampling. We went from measuring five points per wafer to nine points, but that is the standard scale-up for 150mm to 200mm,” says Ferrari.
But other inspections “are performed on 100% of the wafers and 100% of the die,” such as a newly instituted step during the company’s proprietary Thelma process, which incorporates a hefty epitaxial layer structure on a 350-micron-thick bonded wafer. “When the sensor wafer and the capping wafer are bonded together, this inspection is carried out with a scanning acoustic microscope and is done to detect defective die,” explains Ferrari. “For this inspection, we internally developed all of the software and the automation protocol. This screening is done to help in the quality and long-term reliability of the device.”
Ferrari would like to see “higher throughput, higher quality of the image, and software that should be capable of analyzing the images that are taken” in future optical inspection/metrology tools. “From what we are running now, we would like to have at least double the productivity; today, we do not reach 10 wafers per hour. With the small devices that are coming, the throughput will be even lower, and the quality of image will not be so good. Two- and three-dimensional inspection will become, for MEMS, more and more important to guarantee high quality.”
Role of specialized foundries
Not every device requires volume production levels, let alone a 200mm fab. Much of the MEMS world still operates in the traditional “one-process, one-product” environment. As Veeco’s Dowlatshahi says, “MEMS is not really one industry but a collection of industries.” Although leading pure-play semiconductor foundries such as TSMC and niche contract chipmakers like X-Fab have shown a recent interest in attracting MEMS business, the microelectromechanical foundry space remains largely the domain of specialist manufacturers. One of the more successful of these is Tronics, which has built a profitable enterprise working with customers on developing and fabricating many demanding custom components, most of which are built on thick silicon-on-insulator (SOI) wafers (Fig. 3).
 Figure 3. Custom devices such as inertial sensors present difficult metrology challenges for Tronics and other MEMS foundries. |
Jean-Marc Gilet, Tronics’ process and supply chain manager, explains that the company uses “mostly standard and traditional semiconductor metrology tools.” He identifies five “recurrent challenges or specific needs” in the company’s manufacturing metrology practices. “One recurring challenge lies in the handling of double-sided patterned wafers on the metrology equipment...which were originally built for single-sided patterned wafers and are not always adapted. Next, the metrology of high-aspect ratio topography and patterns remains an unsolved challenge so far, [although] we overcome this by destructive controls on test wafers and electrical tests on the products.
“Then with deep-etched structures in silicon, we are mostly measuring lateral dimensions in the range of 2 to 3 microns,” says Gilet. “Even though optical metrology equipment may still be fine for those dimensions, we are reaching the limits of optical tools to measure and control the tolerances.” Another challenge is “measuring and controlling the mobility of moving structures. For example, we can measure vibrating modes to validate the behavioral model of certain mechanical devices. And infrared inspection is also specific to key MEMS operations, such as microstructure release controls and through-wafer controls before and after wafer bonding.”
An additional concern comes to the fore when designers “forget that the processes have tolerances or are not always fully aware of those tolerance values,” says Gilet. “As a result, they think their chip will be a perfect replication of their design, and so they push the geometrical tolerances below the metrology tool capabilities. We spend a lot of time with the designers at the beginning of each project to find the right design-process trade-offs.”
 Figure 4. Tronics’ failure mode effects analysis approach plays a critical role in the MEMS foundry’s yield enhancement efforts. |
Tronics’ closed-loop, failure mode effects analysis (FMEA) methodology (Fig. 4) remains a cornerstone of the company’s yield-enhancement efforts, as Gilet explains. “We work in close interaction with the customer, and we conduct electrical tests and functional device characterization and testing [at least part of it] in order to identify failure modes, then track the causes of the failure modes at the material and process levels. This approach allows us to set up statistical process control procedures based on metrology to control the right things-and not more than necessary-in the right way with the right tools.”