Taking MEMS sensors to market

NORTH BENNINGTON, VT. The burgeoning microelectromechanical systems (MEMS) field is full of good ideas, as researchers pursue innovations ranging from smart-weapon components to miniature medical devices. But turning good ideas into products that can be manufactured and tested in quantity can be difficult. To address the production and test challenges, the Bennington Microtechnology Center (BMC) was founded in 2004 to provide assembly, packaging, and test solutions for MEMS devices used in biotech, military, commercial, and industrial applications.

Henry Klim, executive director of the Bennington Microtechnology Center, and Dr. Harry Stephanou, director of the Automation & Robotics Research Institute at the University of Texas at Arlington and BMC founder and chairman, collaborate on MEMS design, production, and test.

Founded with congressional funding championed by Vermont's US Senator Patrick Leahy, BMC's first customer was the US Navy, for which BMC builds safe-and-arm devices, which monitor environmental inputs to control the deployment of weapons systems. Used on a torpedo, for example, a safe-and-arm device can detect acceleration due to launch from a vessel and disable its safe mode; it then arms itself on encountering deceleration when contacting a target.

Klim, who joined BMC in the summer of 2006, said that he is now working to move the firm into biotech and commercial markets as well as military ones. In conjunction with ARRI, the organization is developing devices and processes for inertial, optical, and RF applications. BMC now occupies 12,000 ft², including a 3000-ft² class 10,000 clean room.

BMC engineer Manoj Mittal works with BMC’s M3 (macro-meso-micro) multiscale robotic assembly and packaging system, which he helped develop when he was a graduate student working at ARRI.

Other clients may turn to BMC for one stage of an assembly and packaging process. One customer, for example, takes advantage of BMC's Kulicke & Soffa ball bonder to make I/O connections to an imaging chip.

Such projects develop revenue for BMC, providing employment opportunities for equipment operators and helping BMC meet its economic development goals. But the sweet spot for the client, Klim said, is to come in at the beginning and take advantage of the combined capabilities of ARRI and BMC in bringing a concept to fruition.

ARRI's primary domain is in concept creation and design analysis, Klim said, with BMC taking the lead in process development and production. But the two organizations cooperate throughout the entire process, from concept creation and design analysis, through process design and development, and on to production, quality assurance, delivery, and supply-chain management. Stephanou pointed out that although BMC serves as a client's primary point of contact, the interaction between BMC and ARRI is seamless, so clients don't have to worry about the organizations pointing fingers at each other.

Typically, Klim said, an inventor will come in with one device he has made in his laboratory, and now he wants to make a hundred. And often, added Stephanou, the original design makes the product more difficult to assemble and test than is necessary. In such cases, BMC and ARRI can make design recommendations with regard to manufacturability and testability, and they can also help address basic functionality decisions.

For example, a customer might propose to power a device using batteries, when energy harvesting might be a viable approach. By enlisting BMC and its ARRI partner, Klim said, a customer can get the benefits of experts who are working on a fivefold improvement in energy-harvesting efficiency.

Commenting further on BMC and ARRI interaction, Stephanou said, “We bridge the proverbial valley of death,” which often swallows up technologies developed at university labs before companies can convert them to marketable products. The key, he said, is to address not just products but rather the assembly and packaging techniques as well as the equipment necessary to manufacture them. Further, ARRI and BMC address product, process, and equipment development in ways compatible with their respective organizational strengths.

A BMC engineer works with a device being fabricated on BMC’s semi-automated robotic system.

Traditionally, Stephanou said, MEMS manufacturers tend to use legacy equipment—designed for microelectronics—that's expensive and not flexible enough to be effective for manufacturing the variety of MEMS devices that are emerging. In fact, the raison d'être for BMC, he said, is to serve the fragmented MEMS market with the ability to cost effectively and reliably handle small production runs. “Anybody can automate a process to make trillions,” he said, “but the challenge is to make small batches.“ The key, he said, is modular equipment that can operate at multiple scales—meters, centimeters, microns, and nanometers.

Said Klim, “We are not selling commodities. Our business is the development, assembly, and packaging of odd and unusual devices. The vast majority of our work is done developing processes.” And that work, he said, involves significant input from ARRI. “Our alignment with ARRI makes a lot of sense,” he added. “They are developing robots that in turn build series of successively smaller robots.”

Klim said ARRI's approach melds well with BMC's concept of modular automation: “We have large robots with a wide range of motion but fairly course resolution as well as smaller ones that have restricted movement but very fine resolution. Our goal is not to have a robot complete one task and then launch another, but rather to dedicate an island of modular automation to each task.” That approach, with manual intervention handling the transfer between modules of automation, serves BMC well in its role of a prototype foundry making hundreds or thousands of devices, he said.

Figure 1. BMC’s multiscale robotic assembly and packaging system includes a semiautomated M3 (macro-meso-micro) packaging platform that the company uses to develop modular automated processes.

Such robots help BMC engineers automate many assembly processes, ranging from those involving individual packages to those fabricated using wafer-scale techniques. An example of the latter is BMC's work on a dual-optical-microphone device. The original concept called for assembly of the individual glass and silicon devices, which entailed significant handling problems. As an alternative, Klim said he investigated wafer-scale processing in which the devices' nine layers of silicon structures are built up on a 4-in. glass substrate; singulation occurs after fabrication is complete.

That process presented its own challenges. One was maintaining sufficient flatness and proper alignment as multiple layers are built up on the glass substrate. Another was finding a way to dice a wafer consisting of nine layers of dissimilar materials. “If you want to stack up nine layers and are off by 12 degrees from bottom to top, you're not going to have many good parts after you dice them,” Klim said.

The dicing problem was relatively straightforward. “I knew that people are doing a lot of wafer stacking, so I realized, 'somebody has already solved this problem.' I did some research and learned what special saws and surfactant materials we could use to reduce chipping.”

As for flatness measurements, when an engineer proposed a dial gauge, Klim insisted he find one with a USB port so the process could be automated. Engineers are good at solving problems, Klim said, but he added that he wants to make sure each problem is solved just once. “If a dial gauge is the answer, then I want it mounted on one of our automated positioning devices, and I want the results sent to a computer. If another engineer comes in, I don't want him to have to relearn how to use the dial gauge. I want every problem solved once. It's a lot harder, initially, to do this up front, but it's necessary. Our processes need to be personnel-independent.”

At each stage of the process, BMC ensures quality by implementing what Klim calls “in-process test gates,” each of which must be passed before beginning the next process stage. “We have to qualify the process somehow, and the best way to do that is to have in-process gates where we make measurements. You don't go through that gate unless you have passed some testing criteria; then you go on to the next process step. The concept is simple, but the implementation of these gates can be painful. You have to put a lot of thought into it.”

One example of a test gate is the flatness measurement in the multilayer, wafer-scale optical microphone process. Another is the wafer probing used to functionally test MEMS dies before packaging them.

Much of the test that takes place at BMC is destructive, involving bond-pulling and shear tests. “We want to pull wires to make sure they are bonding correctly, and we have a shear machine that we use to twist components we've glued together to see how strong that bond is.”

One test aspect that hasn't been emphasized at BMC is final functional test. “The first question I asked when I arrived here is, 'you assemble parts and send them to customers and don't know if they work?'” said Klim. BMC's parts do work, because they have successfully passed through the in-process test gates. Klim explained, “The theory is, if you put in enough gates you'll have 100% yield.” Further, he said, functional test is not something customers have demanded. Despite that, Klim is looking to an increased emphasis on final functional test, for both BMC's and its customers' benefit. “There is significant value in acquiring physical test data from functioning devices,” he said.

To get that data, BMC is investing in a PXI-based test system to be supplied by Metrikos (see “Test focus shifts to MEMS.”). At Sensors Expo in June, BMC and Metrikos demonstrated a preliminary version of the system performing a closed-loop optical-alignment application. At the request of BMC's engineering staff, the system will employ LabView, although Klim, a self-described lab rat, said he would prefer a text-based language like LabWindows CVI, since his experience has indicated that the latter may be more suitable for a production environment because its execution is more deterministic.

Going forward, test will be an increasingly important component to BMC's offerings as the organization builds devices such as accelerometers and RF switches for next-generation cell phones. Roger Grace, a consultant who advises BMC, said his research has shown that assembly and test represents anywhere from 50 to 80% of the cost of a device.

“Silicon is cheap,” he said, adding, “What BMC is doing here is attacking the highest contribution to sensor cost.” Concluded Stephanou, “Testing has involved little more than voodoo in many MEMS applications. Our goal is to replace the voodoo with science.”