eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
SAN FRANCISCO—Who knew that semiconductor manufacturing fabs could turn on a dime?
Intel Corp., long known for its grinding manufacturing strategy, revealed that it has facilities dedicated to quickly making changes in its manufacturing output to meet sudden changes in supply and demand.
The three “flex fabs”—Fab 11 in Rio Rancho, N.M., and Fabs 10 and 14 in Leixlip, Ireland—are designed to switch gears and produce an emergency supply of components when necessary. While most of Intels fabs are in a state of slow transition between two process technologies, Intel executives said the flex fabs can maintain up to four different manufacturing lines.
The flex fabs arent designed to turn out microprocessors. In most cases, according to company spokesman Chuck Mulloy, the company needs to quickly respond to demand for supporting components, such as chip sets.
“One of the advantages I think that Intel has is in our technology pipeline,” said Steven Grant, vice president of the technology and manufacturing group at Intel. “Weve gone and invested heavily in that area.”
Fab 11 in New Mexico is an older fab that still produces chips on 0.18-micron and 90-nm lines. Intels fabs in Leixlip, Ireland, produce 0.13-micron and 0.18-micron parts. These fabs still use 200-mm wafers, an older technology. Mulloy said Intel doesnt want to “risk” dedicating 300-mm wafer technology to the flex-fab model.
Neither Grant nor Mulloy indicated how fast the lines could be turned on or off to meet demand. However, the fabs are designed to run on the same tool set, part of the “Copy Exactly” strategy to facilitate production.
Roughly 75 percent of Intels annual R&D budget—expected to total approximately $5.2 billion in 2005—is spent on its manufacturing technologies. About 70 percent or so of the companys R&D budget during the fourth quarter 2004 was dedicated to putting the finishing touches on its 65-nm lines, the next-generation process node that will begin yielding wafers by the end of the year.
“All indications are that well be good for 65 nm,” Grant said.
To save money, Intel reuses as much as it can inside a fab, saving 30 percent to 50 percent of the cost through retrofitting the fab with a more advanced line. After that, the fab is either left unchanged or closed. Intel still maintains a 150-mm fab in Jerusalem, dedicated to both flash and logic, that manufactures wafers using 0.35-, 0.50-, 0.75- and even 1.0-micron production equipment, Mulloy said.
In total, Intel owns 16 fabs at nine sites around the world, including Israel, Ireland, Colorado, Oregon and New Mexico. Twelve additional sites in Southeast Asia and China serve as test and assembly facilities.
Each fab can cost billions of dollars; in a keynote address at the Intel Developer Forum here on Tuesday, Intel CEO Craig Barrett recalled how an early 1970s fab was built in Livermore, Calif., for roughly $15 million. Today, Intel cant build a cafeteria inside the fab for that amount, he said.