Manufacturing MEMS

MEMS processing: not quite CMOS
Optical MEMS
The micromachined future

For several years now, the technical and trade press have been filled with fascinating pictures of tiny motors, mirrors, and gears. This article was supposed to be about how these miniature mechanisms are poised to revolutionize everything from cell biology to telecommunications. Only they aren’t.

Microelectromechanical systems (MEMS) have great potential in many applications. They allow electronic systems to interact with their environment directly, measuring and manipulating much smaller volumes than previously possible. A micromachined accelerometer, used in automotive airbag sensors, is simpler and more accurate than its macroscale predecessor. Optical MEMS can reduce the number of electrical components in telecommunications networks while improving data transmission. Biochips could transform medical diagnostics and drug delivery.

All of these designs are exciting on paper, but many manufacturing and implementation issues lie between paper and commercial product. As Karen Markus, VP of technology strategy and JDS Uniphase’s Fiberoptics Products Group, explained at IEEE’s Optical MEMS 2000 conference, it’s possible to make almost anything once. Making hundreds or thousands of devices for a commercial application is another matter. This article is about the obstacles that must be overcome for MEMS devices to achieve their promise.

MEMS processing: not quite CMOS

MEMS processing at least superficially resembles silicon integrated circuit manufacturing. Start with a silicon wafer, add or remove material using more or less familiar etch and deposition processes, add electrical leads, and package.

In accelerometers, possibly the first and most successful commercial application of MEMS, interlocking fingers serve as parallel plates of a capacitor. Acceleration moves the fingers relative to each other, changing the spacing between plates. From a manufacturing standpoint, this is one of the simplest of all MEMS applications. Acceleration acts on the fingers without the need for an intervening medium like air or fluid. The sensor still works if it is sealed safely inside a hermetic package.

Accelerometers also rely on the most integrated circuit-like MEMS process: surface micromachining. This process deposits polysilicon onto a sacrificial oxide layer, patterns it by conventional lithography and etch methods, then releases the structure by selective etch of the sacrificial oxide. It’s possible to build complex multilayer structures this way, releasing them all at once at the end of the process.

Even though it evolved directly from IC processing, surface micromachining still faces different challenges and constraints. A single particle can cause an electrical circuit to fail, but will not necessarily affect a mechanical system. At the same time, good mechanical performance typically requires thicker layers than commonly used in ICs. Small variations in thickness and residual stress will dramatically effect mechanical behavior.

Pressure sensors are the next step up the complexity scale. These devices typically place a thin membrane over a small volume. Changes in pressure on the membrane cause it to deflect, and the deflection is measured. Other sensor designs can measure temperature, concentration of a particular species, and so forth.

While accelerometers can function well in hermetic "can" packages, other kinds of sensors need to be exposed to the environment in order to measure it. The need for contact with the environment conflicts with the need to protect sensitive structures from contamination and damage.

Further up the complexity chain, we come to actuators, motors, and other devices with moving parts. Rather than merely measuring their environment, these MEMS are designed to change it in some way. Optical MEMS are some of the most commercially interesting members of this category. As Tellium’s Evan Goldstein and colleagues explained in the March, 2001 issue of Optics and Photonics News, optical MEMS offer solutions to several pressing optical telecommunications problems.

Back to top

Optical MEMS

For example, wavelength-division multiplexing (WDM) allows many data signals to squeeze through a single piece of fiber. WDM works by assigning each signal its own spectral channel. As data traffic grows, each channel becomes narrower while the total optical power increases.

Unfortunately, optoelectronic amplifiers introduce noise that is strongly coupled to channel count, power levels, and other system parameters. Nonlinearities in the fiber itself introduce additional variables. The resulting end-to-end transmission characteristics are difficult to predict or control.

How can MEMS help? As Goldstein and coworkers explained, a micromechanical silicon nitride film can function as a tunable dielectric mirror, filtering out amplifier nonuniformities. Other optical MEMS applications include polarizers and switches. As optical devices, these structures share several characteristics.

First, alignment of components relative to the optical path is critical. Feature profiles and sidewall angles must be carefully controlled. Moving parts, like switchable mirrors, must move in the same way every time. Second, surface characteristics like roughness can have a big impact on reflectivity and other optical parameters. Finally, these devices are typically free-standing structures with a free space optical path. The package must protect the device from mechanical damage while still leaving optical windows.

Many optical MEMS structures are beyond the reach of surface micromachining. Alternative processes like anisotropic wet etch and reactive ion etching (RIE) must also meet stringent requirements for residual stress and optical properties. Large mirror arrays used for switching can easily span larger areas than typical integrated circuits. Since a single defective mirror will ruin the array, film properties must be maintained over the entire area.

Back to top

The micromachined future

Despite all these obstacles, MEMS are slowly entering the commercial landscape. Bosch, Cronos (now part of JDS Uniphase), and other companies are beginning to develop standard processes and standard design building blocks. A few companies are even offering MEMS foundry services, though it is not yet clear if the processes are mature enough to support an arms-length foundry-customer relationship.

MEMS technology has a long way to go before it will fulfill the breathless promises of all those tiny mechanisms. It is now taking the first steps, comparable to the first transistors and perhaps the first primitive integrated circuits. The evolution of MEMS will almost certainly take longer than boosters of the technology would like, but not as long as critics claim. Like the transistor itself, MEMS devices are too useful to fail.