(Business 2.0) – When the bottom dropped out of the technology market two years ago, Wim Roelandts of Xilinx was one of the few tech CEOs who kept their cool. Even as chip giants Intel and AMD eliminated more than 9,000 jobs in 2001 and 2002, Xilinx held on to every one of its 2,600 employees. It's not that the San Jose chipmaker wasn't hurting: Three-quarters of its revenue came from the telecom industry, which explains why its top line shrank from $1.7 billion to $1 billion between March 2001 and March 2002. The sales collapse led to a stomach-turning $114 million loss.

But the 58-year-old, Belgian-born Roelandts was willing to bet he could replace his flagging telecom revenue by cracking a new set of clients--consumer electronics companies. And he'd need the goodwill of every engineer and marketer he could keep. So rather than pass out pink slips, he asked workers to take unpaid vacations and salary cuts. He even lopped $113,000 off his own $675,000 paycheck. Last year, sure enough, Xilinx put all that brainpower and productivity back to use.

By the end of its 2003 fiscal year in March, Xilinx had completed a stunning rebound: $126 million in profit on $1.2 billion in sales. And its stock, which had fallen to as little as $14 a share in 2002, climbed as high as $31 this summer. Investing in workers' bonhomie may sound good to the rank and file during a downturn, but that's not the only reason Xilinx is counting cash again. Its comeback was a product of smart management, technological innovation, and, especially, getting on the right side of Moore's Law. Xilinx sells a radically different kind of computer chip--called a programmable logic device, or PLD--which JP Morgan analyst Chris Danely says will make the company "one of the leaders of the next upturn" in semiconductors. Xilinx's top consumer-products executive, Robert Bielby, takes that a step further: "Programmable logic is doing to conventional chips what digital cameras have done to 35-millimeter film."

While the circuitry of computer chips like Intel's Pentium cannot be changed once they emerge from the plant, PLDs are "softwired." That is, they can be reprogrammed to carry out new tasks--even after they're already in a customer's hands. Such chips have obvious advantages: It's a lot easier to upgrade electronics hardware by reconfiguring chips over the Internet than by hiring technicians to rewire motherboards. But they have drawbacks too: They're more expensive than comparable hardwired chips, especially in large volumes. And their flexibility comes at the expense of space--to perform the same tasks, they require 6 to 10 times the number of transistors used in conventional chips.

When Xilinx got its start in the mid-1980s, its chips were limited to chores in semiconductor design and testing. PLDs were pricey, but not as expensive as signing off on a conventional silicon design only to discover later, when the first chips emerged from the foundry, that they had some unforeseen flaw. With a PLD, engineers could test different designs on a prototype before sending out a final version.

Eventually, the chips took on roles in products such as high-end routers and cellular base stations. Cell-phone companies loved upgrading their systems by simply beaming up new software to equipment in cell towers. Telecom, of course, would soon prove to be the wrong industry to rely on. By the end of September 2001, with the economy reeling from 9/11 and telecom choking on overcapacity, Xilinx's quarterly sales had sunk by half, to $225 million.

But the downturn hit just as conditions were right for Xilinx's expansion. Thanks to Moore's Law, PLD manufacturing costs, like those of other chips, have been in a virtual free fall. In 1998, Xilinx's most powerful chip cost $2,000. By 2001 the price for the same chip hit $100. Today it's approaching $20. Suddenly, it has begun to make sense to use the chips in mass-produced consumer gizmos.

Problem was, Roelandts first had to steer Xilinx through the crash--when no one was buying chips of any kind--and bring costs in line with radically lower revenue. At the same time, he knew he had to keep his top engineers if he was going to be able to meet demand once the recovery arrived.

So Roelandts, who spent 30 years in the people-first culture of Hewlett-Packard (most recently as head of its computer business), got creative. He asked for tiered salary cuts: 20 percent for execs, 10 for directors, 7 for senior managers, and 3 for middle managers. The cuts yielded $18 million in savings between May and October of 2001. He asked managers to shave 50 percent off travel costs and 24 percent off energy expenses. Workers were encouraged to take unpaid sabbaticals, for which they were paid a $10,000 bonus; they kept their benefits, and Roelandts guaranteed them jobs when they returned. These moves helped cut Xilinx's gross expenses by 17 percent--enough to tide it over until sales picked up in 2002.

Meanwhile, Xilinx's main competitor, Altera, was laying off 7 percent of its workforce. The companies had previously run neck and neck in the PLD market. But when buying rebounded, Altera ran into software problems while Xilinx introduced new chips and began reeling in market share. According to Dataquest, Xilinx's piece of the PLD pie now stands at 49 percent; Altera's has fallen to 31 percent.

As Roelandts expected, much of the new business has come from consumer electronics. Sony is using Xilinx chips to bring the latest high-resolution displays to its Vaio laptops and Clie organizers. Nokia and Samsung are bringing cell phones to market faster by using Xilinx chips to mix and match features like built-in cameras. Consumer-goods companies now account for nearly a third of Xilinx's total sales, while the once-dominant telecom industry contributes just half. "I believe most digital consumer products will one day contain programmable chips," Roelandts says.

That's not to say that Xilinx chips will challenge the computing power of the mighty microprocessor anytime soon. Xilinx and Intel enjoy the same Moore's Law advances--such as 300-millimeter wafers and 90-nanometer lithography--that pack more chips onto a wafer and more transistors into a chip. Intel uses the advances mainly to improve processing speed, a goal at which it has succeeded almost too well. "Intel can build faster chips," Roelandts concedes. "But who needs them?"

Xilinx, on the other hand, uses the advances to make its chips smarter and more adaptable. Replay TV avoided a recall of its Xilinx-equipped video recorders by reprogramming the chips over customers' phone lines while they slept. Replay TV's vice president for engineering, Philippe Pignon, says the chips could also allow the company to download into its boxes features such as better algorithms for processing high-fidelity music or inserting ads.

You can't do that with a Pentium. And Roelandts can probably tell his execs they needn't worry about more pay cuts anytime soon. --ERICK SCHONFELD