November 3, 2005 – Eighteen months and $2 billion later, Intel is ready to show off its success in remodeling an aging 0.18-micron/200mm fab into the company’s newest leading-edge 300mm/65nm production site. Fab 12 in Chandler, AZ, joins Intel’s four other 300mm operations — Fab 11X in New Mexico, D1D and D1C in Oregon, and Fab 24 in Ireland — and will produce a mix of dual-core and single-core processors targeting mobile, desktop, and server applications.
David Aires, Fab 12 manager, declined to reveal specifics of Fab 12’s current or future capacity, but indicated that the site will gradually ramp to maximum capacity in early 2007, maintaining a 30 tool/month installation schedule through 2006. Nearly all of Fab 12’s tools are new purchases, supplemented with a handful moved from Intel’s D1D operation in Oregon, which are moving up to more advanced 45nm technologies. Officially, only the D1D site is currently shipping 65nm products for revenues, but Aires said Fab 12 would ramp up to equal D1D’s output by year’s end, and by January would become Intel’s highest-volume production source for 65nm chips. Intel also spent millions to send roughly 800 employees, all of whom volunteered, to other facilities for retraining on the 65nm processes, pointed out.
How much Intel ultimately saved in terms of time and money by converting Fab 12 vs. building a new 300mm plant is unclear. Intel wouldn’t share its internal cost comparisons, instead pointing to plans for its $3 billion Fab 32, a 300mm/45nm sister fab in Chandler, AZ, projected to come online in 2H07. However, George Burns, president of Santa Cruz, CA-based Strategic Marketing Associates, surmised that Intel likely did save some time and money with the Fab 12 conversion, when compared with similar 300mm/65nm fabs — Texas Instrument’s fab in Richardson, TX ($3 billion/18 months), AMD’s recently opened facility in Dresden, Germany ($2.5 billion/~21 months), and Intel’s $2 billion Fab 24-2 in Ireland. Some of the Fab 12 cost savings would have been offset by architectural complications with converting an older site (e.g., raising the ceilings), Burns noted. Working within an existing site also likely saved Intel some time in terms of local approval — getting permits for a new fab for construction, water, and other infrastructure needs is a more complicated and lengthy processes with a new fab, he explained.
Although Intel has proved that a large-scale conversion can be done — as long as there’s enough available floor space — it has no plans to do another one any time soon. For an upgrade to make business sense today, with capacity relatively tight across all its fabs, the company would have to roll a fab off a technology and consolidate that capacity into other factories, cutting back overall volume, said Aires.
For now, Intel is keeping busy converting its D1D development facility to early 45nm work, along with a slew of upgrades at other facilities in the US. In addition to Fab 32, the company is spending $105 million to add capacity to a test facility in New Mexico; spending $345 million for upgrades to its Fab 23 and Fab 12 200mm fabs in Hudson, MA, and Colorado Springs, CO; and investing $650 million to boost capacity at Fab 11X in New Mexico. — James Montgomery, News Editor