October 19, 2010 – Days after landing a $187M promise from the Israeli government to help upgrade its Fab 28 in Kiryat Gat to 22nm process technologies, Intel has turned its sights back to the US — with a $6B-$8B pledge to invest in a new development fab in Oregon and upgrades to four other fabs to make them 22nm-process-technology capable. The projects will support creation of 6000-8000 construction jobs, and about 800-1000 permanent high-tech jobs, the company says.
"Today’s announcement reflects the next tranche of the continued advancement of Moore’s Law and a further commitment to invest in the future of Intel and America," said Intel president/CEO Paul Otellini, in a statement.
"Contrary to conventional wisdom, we can retain a vibrant manufacturing economy here in the United States by focusing on the industries of the future," added Brian Krzanich, SVP/GM of Intel’s manufacturing and supply chain business unit.
Among Intel’s plans is a new development fab ("D1X") in Hillsboro, Oregon, scheduled for R&D startup in 2013. Also getting upgrades will be fabs D1C and D1D in Oregon, and fabs 12 and 32 in Arizona. The company currently slates its first 22nm microprocessors ("Ivy Bridge") to enter production in late 2011. Note that Intel is on pace for $5.2B in capital expenditures in 2010.
Gartner analyst Bob Johnson offers some further insight into Intel’s 22nm plans:
— Initial 22nm work will be split between Fab D1D, currently home to INTC’s 32nm/"1268" basic MPU process, and D1C, where Intel has a slightly different 32nm/"1269" process, adding things like SoC capabilities and other chipsets. (Note that GlobalFoundries proudly does its process flavors all in one place, not split between facilities.) Once the processes are locked and ready for production, they’ll be shipped down to Fabs 12 and 32 in Arizona.
— Note that Fab 12 in AZ is currently a 65nm site after upgrading a few years ago from 200mm/0.18μm; moving 22nm in there will probably require a significant retooling, e.g. new steppers, equipment to handle high-k/metal gates (HKMG), etc. "They’ll pretty much replace everything," Johnson said — though fellow Gartner analyst Dean Freeman points out that INTC will probably upgrade with as much existing equipment as they can vs. buying everything new. (And what to do with Fab 12’s 65nm toolsets? Don’t be surprised to see some or all of those tools "finding their way on a ship or plane across the Pacific" to Intel’s proposed 65nm fab in China, Johnson speculates.)
— Despite Intel’s commitment from Israel to upgrade its Fab 28 in Kiryat Gat to 22nm, Intel made it clear that its US sites will get first dibs on the technology, Johnson said. With development work in Oregon it’s a lot simpler to have work starting in its Arizona fabs. And Johnson noted that for the publicity about Intel’s Kiryat Gat fab funding, there was nothing close to the definitive timeline offered for its US plans.
— The new fab, D1X, will likely see some level of groundbreaking quickly, perhaps by end of this year, Johnson notes. But with R&D startup planned for 2013, that puts its work well beyond 22nm, and probably beyond the ~16nm node as well, into the ~11nm and lower realm, Johnson notes. And that also puts it squarely in the perceived wheelhouse of next-generation lithography, whose current frontrunner is EUV (Intel has said it thinks it can push current litho with tricks to the ~16nm node). INTC execs told Johnson that the new fab’s construction "would be appropriate for any given tool," which would include EUV and its unique requirements of size, weight, etc. (To house its elephantine EUV alpha system, Albany CNSE contracted with a construction company to design a special crane & transport/rigging setup.)
— Aside from next-gen litho, another new semiconductor technology is on the horizon: 450mm wafers. While not offering much detail, an INTC exec told Johnson that the D1X fab also would be built with the size of things like 450mm wafer processing in mind, though perhaps requiring future expansion for it. (Reading in between the lines: D1X will probably start primarily with 300mm, but eventually when — or if — INTC does 450mm, it’ll be done there too.)
— Why name the new development fab "D1X," and not maintain sequential letter-naming convention (D1C, D1D…) Because, Johnson says, INTC thought that "D1E" might invoke an unfortunate necrotic association!