(FORTUNE Magazine) – Unnoticed, like dust mites on a couch, are growing numbers of tiny mechanical gadgets with amazing capabilities. Rugged motion sensors smaller than a fingernail. Micromirrors, 1.2 million of the little devils crowded onto the area of a postage stamp, that pivot billions of times on hinges that never wear out. Electric motors with gears the size of a pollen grain that whir at an unprecedented 350,000 revolutions per minute. A museum of these devices would fit in a thimble.

They're taking over important jobs. The midget sensors trigger automotive airbags, and the micromirrors are used in projectors that beam an unusually bright, crisp image. Tiny motors, if their designers' hopes are realized, will prevent the accidental detonation of a city-destroying nuclear bomb. What these mechanisms have in common, aside from smallness, is that they are made using the same well-understood processes by which silicon and metals are built up like layers of a cake to form computer chips. They're produced with the same kind of photolithographic equipment, in clean rooms tended by the same breed of bunny-suited workers.

Except that this silicon moves. Computer chips, for all their data-handling prowess, are a visual bore. Computational miracles take place as electrons zoom through their millions of on-off switches. But even when viewed through a powerful microscope, there's no show for the eye to see. With the arrival of micromachines, more properly called micro-electromechanical systems, or MEMS, fanciers of mechanics and motion need gripe no longer about the nondrama of silicon. These chips, when magnified, resemble Erector Set creations.

Some 600 organizations worldwide are working on MEMS, according to consultant Roger Grace in San Francisco. About 150 are companies--half in the U.S.--pursuing commercial markets. Grace estimates the current global market for MEMS devices at $6 billion to $8 billion a year and believes the figure will reach $20 billion by 2002.

Impressive as these projections are, developing MEMS has taken longer than expected because of maddening difficulties arising in the production and operation of microscopic mechanisms. The earliest MEMS exploited the fact that when silicon is flexed its electrical resistance changes. In the 1960s companies like Honeywell developed hydraulic-pressure sensors for aircraft flight-control systems that were more accurate than their predecessors. By the 1980s automakers and their suppliers began making similar sensors to monitor engine intake-manifold pressure in fuel-injected cars. Other early MEMS applications included inkjet printer heads and catheter devices for measuring blood pressure in surgical patients.

MEMS took a big leap beyond mere flexing and into oscillating and spinning parts when a sophisticated production technique called surface micromachining was introduced in 1985. The method involves stacking layers of patterned and etched silicon structural material, which is acid-resistant, and alternating them with layers of a "sacrificial" silicon dioxide material that is dissolved away by hydrofluoric acid to create spaces between moving parts. What's left, wondrous to tell, are tiny mechanisms that require no assembly, including gears that spin around hubs.

One slice of the MEMS business, small now in terms of commercial sales but likely to grow especially rapidly, is microfluidics devices, or bio-MEMS, used by pharmaceuticals researchers. These laboratories on a chip are micromachined with fluid passages, valves, and chambers for carrying out chemical reactions on a tiny scale. More exciting to behold, however, are MEMS that are taking over essential mechanical chores outside the lab. Around the U.S., eclectic teams of scientists and engineers have developed marvels like these:

CHEAP LITTLE AIRBAG SENSORS

At a fab in Cambridge, Mass., Analog Devices is cranking out, at the rate of a million a month, tiny MEMS accelerometers that have slashed the cost of airbag controls. The company's accelerometer business, among the biggest of the recent MEMS innovations, has been growing 30% to 40% a year in dollars. Analog's MEMS design, the top seller among its peers, uses an intricate movable silicon structure called a proof mass, which has spindly fingers protruding from each side, making it look like a stylized fish skeleton. Interleaved with its fingers, but not touching them, are stationary fingers. Together, the sets of fingers form a differential capacitor that can electrically sense changes in the gap between them.

When the car in which the airbag is installed is jolted, the proof mass jiggles slightly on spring mounts that look like hairpins. The resulting change in the gaps between the stationary and moving fingers creates a signal varying in proportion to the g force of the jolt. Bumping a curb, say, produces a minor jiggle that the accelerometer disregards. But if the capacitor senses enough force, the accelerometer tells the airbag to blow.

Analog had an eye on the potentially huge automotive airbag market when it began working on MEMS accelerometers in the late 1980s. At that time, recalls Chairman Ray Stata, automakers were relying on mechanical devices that used a metal ball or cylinder in a tube. For reliable control, a car needed as many as five of the devices, each costing about $17. Stata says Analog is shipping its fingernail-sized accelerometers for less than $3 apiece.

Analog, a publicly held company whose sales hit $1.2 billion in 1998, expects airbag sensor volume to keep growing rapidly. The company also sees a big future for its accelerometers that sense gentler motion. One is used in a device called Back Talk, made by Bio Kinetics Corp. of San Antonio. Designed to reduce the risk of workplace back injuries, the pager-sized gadget clips on the user's belt and issues an audio or vibrating alarm when he or she makes an ergonomically unsound move, such as lifting from the waist instead of from the knees.

CHIPS THAT CREATE SHARPER IMAGES

Texas Instruments in Dallas has a fab exclusively devoted to producing its digital light processors (DLPs), which are MEMS chips packed with all those micromirrors. First launched in 1996, DLPs come in versions containing 0.5 million, 0.75 million, or 1.2 million tiny mirrors, depending on how much resolution the customer wants. TI sells the chips to 30 manufacturers, which have put them into more than 165,000 projection systems that sell for $4,500 to $120,000.

The systems include home theaters that project TV programs onto large screens, portable computer-driven projectors for conference-room presentations, and more powerful projectors for concert halls and stadiums. A prototype of a DLP cinema projector, built by Digital Projection Ltd. of Atlanta, was recently demonstrated, which raises the prospect of commercial theaters' switching from expensive film prints to movies delivered in electronic form.

In TI's system, light focused by a lens is shone through a spinning transparent wheel to break it down into a stream of rapidly alternating red, green, and blue primary-color segments that are beamed at the mirror chip. Responding to electronic signals from a laptop PC, a CD-ROM, or a videotape that are sent through a processing board, the mirrors on the chip play the same role as stadium spectators who display color cards on command to create images.

Tilted by electrostatic attraction as many as 10,000 times per second, the mirrors form pictures by flipping into or out of each color's light path at the correct instant. They behave like reflecting pixels to form a much brighter and crisper image, with truer colors, than is possible with conventional equipment. The image the DLP creates is projected through a second lens onto a viewing screen.

Who could have thought up such an audacious thing? It was physicist Larry Hornbeck, working at TI's Central Research Laboratories, who invented the DLP in 1987. Last year it won him an Emmy award for outstanding engineering development from the Academy of Television Arts and Sciences.

BETTER FIBER-OPTIC SWITCHES

MEMS populated with mirrors may be destined for an even more important role. One of more than 60 MEMS startups in the U.S., Optical Micro-Machines in San Diego has figured out a better answer to the fiendishly delicate problem of switching fiber-optic signals. The 14-employee company, backed by prominent venture capital firms that include Sevin Rosen Venture Partners, predicts that the MEMS devices it is developing will help telephone companies keep up with the explosion in Internet traffic.

The so-called "bulk optics" mechanical switches currently used by the telecom industry have a lot of shortcomings. Using what looks like a speedometer needle, they move a fiber strand along an arc, coming to rest where it aligns with one of several other strands. Switches that can shunt signals from four incoming strands simultaneously among four outgoing ones--called four-by-four matrix switches--cost about $20,000 each. They are easy to ruin if dropped.

Working with prospective telecom customers, OMM has developed smaller, more rugged MEMS switches that will cost half as much initially, and less later on, to do the same job. Customers will test prototypes later this year. The first model to go into volume production will probably be a four-by-four matrix switch with 16 tiny pop-up mirrors that can reroute signals bounced off them at a 90-degree angle. They pop up in different patterns to shunt four incoming signals as desired.

A TINY SAFETY LOCK FOR BOMBS

Should an airplane crash with a nuclear bomb on board, the weapon absolutely must not explode in a glowing mushroom cloud. Sandia National Laboratories in Albuquerque, which conducts weapons-related research and various cooperative programs with industry, has come up with a cheaper, space-saving MEMS mechanical safeguard.

With its first-rate fab and an annual MEMS budget of $100 million, Sandia is one of the dream playgrounds for micromachine researchers. All who see it are captivated by a videotape of the lab's tiny prototype MEMS "stronglink" built last year, a safety lockout device designed to prevent a nuclear weapon from going off accidentally.

The stronglink currently being used is an elaborate mechanism reminiscent of the Ultra encoding machine used by the Germans in World War II, with 450 clockwork parts machined from stainless steel at a cost of about $25,000. "This is what stands between you and a blinding white flash," says Paul McWhorter, Sandia's deputy director for microsystems science, handing a visitor one of the baseball-sized gadgets. "The question is, Can we achieve this same function, with higher degrees of safety and reliability, and build the entire system on a chip for a dollar? That's our motivation for working in MEMS." Sandia is sharing with industry the experience gained in developing a cheaper stronglink, which it hopes will refine the manufacturing know-how. Says McWhorter: "The designers tell us they are never going to put these into a weapon until they see them in cars."

The brainchild of Sandia technician Steve Rodgers, the MEMS stronglink can best be described as Rube Goldberg meets Gordon Moore. Peering through a microscope, you can see the device's prime movers, seven unorthodox electric motors called comb drives. Consisting of paired comblike silicon structures with their teeth intermeshed, the drives perform reciprocating motion when a fluctuating voltage applied to them creates an electrostatic field. A connecting rod joins each comb drive to an eccentric gear, which converts the reciprocating motion to rotary motion. Then a train of reduction gears slows the rotary motion while multiplying its torque, or power, to a useful level. Finally, a rack-and-pinion gear converts this to linear motion.

The stronglink is designed to cooperate only when someone enters a 24-number sequence correctly, leaving just a one-in-16 million chance of defeating it by entering random numbers. Each time a correct number is entered, motors switch on and steer a pin through the correct turns in a slotted maze. If a wrong number is entered, the pin enters a dead-end path in the maze, and the device locks up permanently. But when all 24 digits are entered correctly, the pin negotiates the entire maze, and the rack and pinion--which has a pair of gears at its tip--meshes with another gear train. Now the second gear train can drive a second rack and pinion, which raises a folded mirror. And the mirror deflects a beam of light, directing it to an optical sensor that arms the bomb. It all adds up to a tiny lock no thief can pick.

"We're on the verge of the second silicon revolution," reflects Sandia's McWhorter. "The first one has made transistors smaller and smaller for 30 years. The second one is creating a whole new generation of integrated circuits that not only think but can sense and act as well." When silicon can move, the possibilities can only be guessed at.