THE WAY WE INNOVATE NOW
Who says American manufacturing is dead? We tour three U.S. plants that use high technology and sound management to succeed in a competitive global marketplace. Whether the game is kids' car seats, semiconductors, or fiber-optic cable, it's all about working smarter.
By PHILIP SIEKMAN

(FORTUNE Magazine) – BUILDING A BETTER CAR SEAT

DOREL

INDUSTRIES

Montreal

2004 sales $1.7 billion

Employees 5,000

Key products Bicycles, juvenile items, furniture

Making the world safe for baby onboard is a major line of business at Dorel Industries, a Canadian company that is the largest manufacturer of children's car seats in the U.S. Dorel's plant in Columbus, Ind., produces some six million a year (more than the number of babies born annually in the U.S.), along with diaper pails, baby bathtubs, and potties.

Three years ago the Columbus plant was a mess. Aged injection-molding presses churned out plastic parts that then gathered dust on a 300,000-square-foot mezzanine overlooking the factory floor, often taking weeks to make their way into finished products. Trucks transported the products to a warehouse 45 miles away, where they sat before being shipped to Wal-Mart, Kids "R" Us, and other retail clients. Not surprisingly, the factory "wasn't performing financially in a way the shareholders were expecting," says Bruce Cazenave, who was recruited from Black & Decker in 2002 to take over Dorel's juvenile-products group and fix the business.

Like many consumer-goods makers before him, Cazenave considered moving the entire production line to China. Besides wishing to cut costs, U.S. manufacturers often feel obliged to shift production offshore to retain Wal-Mart and other big-box customers that might otherwise cross the Pacific and buy directly from Chinese plants. But car seats are a special case because they are so often the target of product-liability suits: Big retailers and their insurers are loath to face aggressive tort lawyers by themselves every time a child is hurt or killed in a car crash. Dorel defends numerous such cases each year, as do all its competitors, according to Cazenave. Manufacturers often settle these cases, because they know that juries tend to sympathize with bereaved parents even if the car seat meets or exceeds the tough standards set by the National Highway and Traffic Safety Administration, and the injury occurred because the seat was not properly installed or the child wasn't properly restrained. Cazenave says high insurance and litigation costs help explain why the number of U.S. car-seat manufacturers has dropped from 15 to four since 1990.

Some of Dorel's rivals sell Chinese-made seats under their own brands, thus keeping the big retailers comfortable by standing in the line of legal fire. But Dorel became concerned about delivery times, given that its customers sometimes insist that orders ship in a week. "We felt we could react to opportunities much faster if we were right here," says Cazenave. "[When] an order comes in, we can ramp the factory up right away, as opposed to waiting for five weeks. We can also shut a line down faster if a seat is not selling."

Dorel decided to use Chinese suppliers for car-seat components, especially the labor-intensive sewn-fabric pads that cover seat shells. That enables the Columbus plant to focus on capital-intensive injection molding. U.S. workers also handle final assembly work, minimizing litigation risk by carefully checking each seat before it ships. (One missing warning label is all a tort lawyer needs to charge negligence.)

Dorel also considered building a new U.S. factory, possibly outside Indiana. That got the attention of Brooke Tuttle, head of Columbus's economic-development effort. Since 1985 he has persuaded 52 companies to locate in Columbus, including 20 from Japan. But in 2003 his big job was to stop this manufacturer from leaving town. The city and state offered Dorel an incentive package of loans, tax credits, training support, and grants, including one that covered half the cost of a railroad spur into the plant so that resin could be pumped directly from cars to storage bins and into the presses.

Given that support, Dorel chose to stay in Columbus. Since late 2003 the company has invested more than $26 million in the plant, including $10 million to install 40 brand-new injection-molding presses, $8.5 million for a new, 326,000-square-foot warehouse adjacent to the production lines, and $1.5 million for a state-of-the-art testing facility where mannequins of children are strapped into car seats and hurled into a barrier at some 30 miles an hour.

Dorel also hired a lean-manufacturing expert named Oscar Estrada and turned him loose on the factory floor. Estrada chose eight workers and put them through a concentrated four-week course, from which they graduated with the title "Dorel champion." To reduce the time lost when Dorel changes molds on its presses--something it does more than 200 times a month--Estrada assembled a kaizen, or continuous-improvement team, that included one of the champions. "Over three days they kept improving," he recalls. "Two hours, 1.5 hours, 1.25 hours, 45 minutes." By the time the team was ready to demonstrate the process to Cazenave, they had it down to 18 minutes. With the boss watching, they shaved off another four minutes.

As part of the lean-manufacturing effort, Dorel installed andon lights, which signal when a line needs help, and implemented the 5S program of a place for everything and everything in its place. (The term "5S" is derived from a Japanese phrase that's often translated "sort, store, shine, standardize, sustain.") The company also installed conveyors to lift seat shells directly from the presses to one of 16 assembly lines. On one line, 11 operators check the work of the person before them, do their part of the assembly, and pack the seat in a box--all in less than five minutes.

Work in process has diminished to the point that there is no need to store parts on the old mezzanine, which has been torn down. Receiving docks are now positioned on one side of the plant, shipping docks on the opposite side. The new onsite warehouse opened in spring 2004, eliminating inefficient road shipments.

All those changes are paying off. Total production time for a seat, from resin to packed box, has dropped below 15 minutes. In 2003, output on Dorel's lines averaged fewer than 60 car seats an hour. In 2004, output rose to 63. This year it's more than 70 and climbing. Thanks to design and engineering changes and improved quality control on the assembly lines, car-seat recalls declined from 100,000 seats in 2000 to fewer than 1,700 last year. There's more to do, of course--continuous improvement must be continuous. Estrada plans to train additional lean-manufacturing champions. He's also discovered that training is a two-way street. "Yes, the machines are phenomenal, and the warehouse looks great, but we also have more than 800 experts out there who keep teaching me," he says. "Just listen to them. Provide them the tools. And it's phenomenal."

BIG BLUE'S $2.5 BILLION SALES TOOL

IBM

Armonk, N.Y.

2004 sales $96 billion

Employees 319,000

Key products Computer services, chips, hardware, software

Bragging rights never come cheap in semiconductor manufacturing, but IBM has earned them at its flagship East Fishkill, N.Y., fab. It is, brags the computer maker, the world's most automated chip plant. Certainly it's a wonder to see. The production area, which could cover two football fields, is populated by large metal boxes. Overhead, green or red transparent plastic containers whiz past on a suspended monorail that snakes through the factory. Occasionally a container brakes to a stop and is lowered to the front of a machine, where it is quietly engorged. Some minutes later a machine door opens, the container slides out, waits briefly for a cable-hung device to lift it back to the monorail, then zips away.

You can't see what is happening inside the machines. But the containers are delivering racks of glass wafers, each 300 millimeters (about a foot) in diameter. Wafers this size are the largest in the semiconductor business, and this is the only IBM plant that uses them. The boxes in which they travel, known as front-opening unified pods, or FOUPs, each hold up to 25 wafers with a total value of as much as $2 million. The fab, as semiconductor plants are known, processes each wafer into as many as 1,000 chips, each incorporating millions of transistors and other devices as narrow as 90 nanometers--90-billionths of a meter. The wafers are never touched by human hands or even a molecule of ambient air. From here they are shipped to other factories, where they are packaged as individual chips and sent on to IBM equipment plants or to customers like Microsoft and Sony.

IBM spent nearly $2.5 billion literally lifting the roof and rebuilding an old plant to get East Fishkill into 300-millimeter production in early 2002. But the plant is also based on years of cooperative research across the semiconductor industry. Working through Sematech, an Austin-based manufacturing association, the major chipmakers jointly created the necessary technology to move from 200- to 300-millimeter wafers while progressively reducing the size of chip circuitry and components.

FOUPs and the machines here aren't what sets East Fishkill apart from the 20 other advanced fabs around the world processing 300-millimeter wafers. The dress code offers the most visible indication of what makes this plant unique. Because the production units are sealed, there is no need for bunny suits, the all-enveloping outfits that semiconductor technicians have traditionally donned before passing through air locks and into the production area. Other modern fabs take no chances and still encumber their workers. Workers here simply wear head nets, coveralls, and booties over their shoes. IBM's relaxed dress policy is "a real gutsy move," says Dan Hutcheson, president of VLSI Research, a Santa Clara, Calif., group that does economic analysis for the chip industry. The policy saves IBM money and is also a productivity boon, since getting in and out of the production area is, Hutcheson says, "a five-minute affair at most as opposed to 40 to 60 minutes."

A less obvious difference in this fab is the extent to which computers run the place. In older fabs, technicians handle such tasks as moving around racks of wafers and transcribing ID codes. Being human, they make mistakes. But as processing speeds increase and circuitry shrinks to vanishing scale, each wafer becomes more valuable, and there is no room for human error. As a result, computers have become increasingly important in all fabs. But East Fishkill is entirely automated, with data flowing through wireless networks and 600 miles of cabling into more than 420 servers with more computing power, says IBM, than NASA uses to launch a space shuttle.

The most important piece of hardware in the plant is a small glass vial containing an RFID transponder. One of the devices is embedded in each of the plant's 5,000 FOUPs, as well as in similar containers used to ship wafers or transport reticles--images of integrated circuits that are projected onto wafers. Each transponder emits a signal that can be read from a few inches away by 60 receivers along the monorail and in each of the plant's 1,500 machines. When a FOUP arrives at a processing machine, the computer system tells the machine how to treat the wafers it bears. Whistling while it works, the machine sends progress reports and also alerts the computer when it's ready for another load. "A lot of people ask, 'How do you justify it? How do you [calculate] a return on investment for RFID?'" says Perry Hartswick, a senior development manager who led the team that automated the plant. "It's like asking, 'How did you ROI the wires in the walls?' Without RFID, I couldn't have done any of this stuff."

East Fishkill's servers run a dozen major software programs and dozens of smaller applications, all networked through a central program called SiView. The system looks at orders and schedules production runs. It keeps tabs on what's where when. It monitors machines and shuttles FOUPs out of the way when an order needs to be expedited. It adjusts schedules to allow for planned maintenance and provides portholes for customers to look into the system to check on their orders. It even allows equipment suppliers to remotely debug their machines. And it feeds vast reams of production data into enterprise-wide management and financial-reporting programs. All this work goes on continuously and in real time.

The plant's ability to manage itself sometimes awes even its creators. Just weeks after the fab started up, a major snowstorm hit the area. Plant managers sent the first shift home early and told the second shift not to come in. Then they shut down any possibly dangerous equipment, left the rest running, locked the doors, and went home themselves. Once there, Hartswick logged onto the Internet and started monitoring a flow of data and graphic displays that told him exactly what was going on back at the fab. "It was the most amazing thing," he recalls. "Nobody in the fab. The FOUPs getting dispatched, moving around, going to the tools. I didn't end up going to sleep at all that night. It was great watching the thing just hum. It ran all night until it ran out of work. We came in that morning, and everything was just hanging out."

Hartswick says the plant provides the service that IBM's customers deserve. IBM says the plant also provides an example of what it can teach other manufacturers about automation. Last year the company created a unit called Sensor and Actuator Solutions. Its mission is to aggregate IBM hardware, software, and consulting expertise that clients can draw on to automate any industrial enterprise, be it refining petroleum, formulating drugs, or making windows. Any skeptic who thinks his business is too complex and arcane for automation is invited to check out East Fishkill.

PERMANENT REVOLUTION

CORNING

Corning, N.Y.

2004 sales $3.9 billion

Employees 25,000

Key products Optical fiber and cable, LCD glass, technical glass

The fiber-optic business was great while it lasted. After figuring out how to make optical-glass fiber for telecommunications, Corning erected a wall of patents, opened a plant in 1979 in Wilmington, N.C., and eventually built an extremely profitable franchise. By the late 1990s the plant covered nearly a million square feet. And even though competitors added capacity as Corning's patents expired, Corning bought out partners in fiber joint ventures abroad, opened a second U.S. plant in Concord, N.C., and planned another for Oklahoma. Just about then, the telecom bubble popped.

In the second half of 2001, demand for optical fiber shrank and prices fell. By 2002 global fiber production had dropped to 64 million kilometers from a high of 120 million kilometers the previous year. The price for the most common type of fiber dropped from $32 a kilometer in mid-2001 to $22 a year later and is now around $11, according to KMI Research in Nashua, N.H. By the end of 2001, most of Corning's fiber lines were idle. It scrapped the Oklahoma project, closed the foreign plants, and mothballed Concord. Fiber sales slid from $2.9 billion in 2000 to $755 million in 2004.

Output at the massive Wilmington plant sagged for a couple of years. But now that Corning's other plants are closed (except for a smallish one in China), it is making as much fiber as it did during the boom. Meanwhile Corning continues to act as if fiber is a growth business. The company hasn't stopped inventing new equipment to make better optical fiber at lower cost. It's as though Corning is addicted to betting on fiber, can't be dragged from the table, and ought to sign up for a 12-step program.

Optical-fiber manufacturing requires precision at the submicron level. The primary raw material is vaporized silicon tetrachloride burned as an open flame with methane and, for the core portion of the fiber, a smidgen of germanium tetrachloride. In the soot from that flame is silicon dioxide, the purest form of glass, which is collected in an enclosed lathe on a rotating ceramic rod. The glass builds up to a cylinder that looks like the frost your mother melted off the inside of her refrigerator. Removed from the lathe and sintered down to solid, transparent glass, this material will become the signal-carrying core of the optical fiber.

The process is repeated on different equipment to clad the core with an outer layer of glass. The cylinder, known as a preform, is then placed at the top of a multistory tower and heated until a dripping gob can be captured and pulled downward, thinning into fiber in much the way that mozzarella strings out when you pick up a slice of pizza. On the ground floor the fiber is wound on spools and sent on for testing before shipment to another facility, where multiple fibers are gathered into cables.

The original preforms were about the size of a rolling pin. Improvement efforts focused on building larger preforms from which more fiber could be produced. Today they are a couple of yards long, and each contains enough material to draw many hundreds of kilometers of fiber. But as the cylinders grew, the process got tougher to control. A technician once manned every lathe. Now computers have taken over, and a single technician covers a roomful of lathes. "It simply wasn't possible to get precision control having people turning knobs, so we took the knobs away," says plant manager Tom Nettleman.

The next challenge was to boost drawing speed. The faster the fiber is pulled, the thinner it gets, if nothing else is changed. But the finished fiber must be 215 microns in diameter (like button thread) with a tolerance of less than one micron. To maintain consistency, Corning uses a computer system that measures the hot fiber several hundred times a second, continually adjusting speed and tension. Corning won't say how fast the fiber is moving, but one can conclude that the rate would probably attract a trooper on most U.S. highways.

Pervasive automation has enabled Corning to reduce production time and defects. The company has always tested every meter of fiber to ensure that its tensile strength is at least 100,000 pounds per square inch. But Corning once checked optical characteristics by sampling the fiber made from each preform. Now testing has been automated, and every spool is checked. Even so, the time elapsed from drawing to warehouse is down from weeks to just hours.

Thanks to the upgrades, output of kilometers of fiber per Wilmington employee has doubled since 2001. Returns, the best measure of quality, have declined from more than 6,000 kilometers per million kilometers of fiber shipped in 1986 to 50 per million today. And while Corning's fiber business has not yet returned to profitability, its cash cost (before depreciation and other noncash charges) is once more lower than the selling price.

There might be light at the end of the fiber. Prices seem to be leveling out. Demand is creeping up, especially in the U.S. Eric Musser, the Corning vice president who runs the optical-fiber business, says the industry can still produce as much fiber as it sold worldwide in 2001. But how much of that capacity is out of date? Though Corning continues to upgrade product and reduce costs, Musser doubts that's true of his rivals. "Some of their capacity is going to age," he says. "How useful and cost effective it will be in the future is an open question." At Wilmington, it's not.

FEEDBACK fortunemail_letters@fortunemail.com. Stories from the Industrial Management & Technology section can be found at fortune.com/imt. Executives in manufacturing and research and others eligible to receive FORTUNE's Industrial Edition can subscribe by calling 888-394-5472.

In 2003, output on Dorel's lines averaged fewer than 60 car seats an hour. In 2004, output rose to 63. This year it's more than 70 and climbing.

It's as though Corning is addicted to betting on fiber, can't be dragged away from the table, and ought to sign up for a 12-step program.