News

Intel Revamps Factories for Next-Gen Chips

• By The Associated Press
• May 30, 2006

An army of robots suspended from the vast ceiling glide from one refrigerator-sized machine to the next. Their cargo: thousands of 12-inch silicon platters that form the raw material for Intel's most sophisticated computer microprocessor to date.

Inside this chip fabrication plant on the outskirts of Phoenix, engineers clad in what look like space suits are six months into a dramatic overhaul that could determine Intel's future as it faces its stiffest competition in more than a decade.

Intel closed the factory, officially known as Fab 12, for 18 months and spent $2 billion to retool it with more than 800 machines that follow a new manufacturing recipe cooked up more than four years ago and is already in place at a plant in Oregon. By year's end, the process will be up and running in a total of four fabs.

"Nobody ramps a technology at the rate we do," says Intel Vice President Tom Franz. "I'd be willing to stand up and say that in front of anybody, including our competitors."

The overhaul is part of Intel's and rest of the semiconductor industry's relentless quest to shrink the size of its circuitry so more transistors fit onto the same size chips. For decades, the industry has doubled the number of transistors on a chip every two years or so, a pace that has become known as Moore's Law, after Intel co-founder Gordon Moore predicted it in a 1965 article.

Because it allows a new generation of smaller, faster products at roughly the same cost as earlier ones, Moore's Law has provided a growth engine that separates the electronics industries from virtually every other business.

But no other company spends as much money as Intel adhering to the law's rigorous demands, and as a result the payoff from more efficient factories is higher. Intel, which has spent $25.3 billion on new equipment over the past five years and is the world's largest chip maker, also gets important competitive advantages from its uncontested role as manufacturing champion.

"If you're the person that's setting the pace and setting the course, everybody else is chasing you and it's a lot easier to stay in the lead," says analyst Rob Enderle of the Enderle Group.

Thanks to Moore's Law, Intel's Core Duo microprocessor, being manufactured in Chandler, is small enough to fit on the nail of an adult pinky finger. If it was made using the process considered state-of-the-art in the early 1990s, its 151 million transistors would take up as much space as compact disc jewel case.

Under the recipe being rolled out in Chandler, a chip's average circuitry measures 65 nanometers, small enough that 100 transistors would fit into a single human blood cell.

Within the next few months, the majority of Intel's processors will be made using the new process. That puts the Santa Clara, Calif.-based company about 18 months ahead of its chief competitor, Advanced Micro Devices Inc., and up to five years ahead of other chip makers, says VLSI Research analyst Dan Hutcheson.

Although some of the gear arrived just weeks ago and is nothing like the tools used in the past, the equipment is already intimately familiar to the thousands of engineers who work at Fab 12. That's because about 400 "seed" employees have already spent more than a year working in what amounts to a carbon copy of the plant in Oregon.

Now, under a process Intel executives call "Copy Exactly," the seeds are back in Chandler, where their job is to duplicate even the subtlest manufacturing variables found in Portland, from the color of a worker's gloves to the type of fluorescent lights used.

One of those seeds is Erica Anderson, a five-year Intel employee who's responsible for the performance and upkeep of two machines that wash silicon platters _ also known as wafers _ in a chemical bath to remove impurities. In January 2004, she left Chandler for a 16-month stint at a development facility in Oregon, so she'd know her part of the new process cold by the time Fab 12 reopened in October.

"There's peace of mind in knowing that your equipment is set up exactly and that the process worked up in Portland," says Anderson, 27. "You feel pretty confident that your process is going to work down here."

All the hard work is paying off. The "yield," or percentage of chips on a 12-inch wafer that function properly, rose more quickly during Fab 12's transition than the rollout of any new process in Intel's 38-year history, Franz said.

There's little margin for error. Over the past few years, Intel's edge in manufacturing has been blunted. A series of new chip designs over the past few years has allowed AMD's market share to rise more than 3 percentage points, to 18.2 percent, versus the 80.2 percent held by Intel, according to Mercury Research.

The most notable new feature was the ability for the smaller competitor's chips to handle larger blocks of memory needed by many corporations and scientific customers while remaining compatible with software designed for earlier systems.

While no one disputes the important advantage Intel gets from outspending its competitors on factory gear, AMD's gains are an important reminder that manufacturing prowess alone is no guarantee of success.

"It doesn't matter if you have a lot of manufacturing capacity if nobody wants to buy what you're selling," says Dan Niles, an investment manager with Neuberger Berman Technology Management.

Intel is countering AMD with a host of new processors based on a new chip design due in the second half of the year. One of them, for desktop PCs, will deliver 40 percent better performance while reducing power consumption by the same margin.

The company is also hard at work on an even more compact 45-nanometer recipe slated to be rolled out in 2007 that will ensure Intel maintains its manufacturing lead.

Intel Chairman Craig Barrett, whose "Copy Exactly" technique is credited by industry watchers as a key reason for the company's unmatched manufacturing muscle, says the challenge to get things right grows with each new transition.

"It's a little bit like the baseball player with a batting machine dialing up the speed of the pitches," he says. "Each generation we dial up the speed by 10 mph, but in spite of that, we're able to hit the ball more often."