AMD is leaning increasingly on IBM as it battles with Intel for next-generation microprocessor manufacturing leadership. And the payout to IBM is significant.
AMD fab
(Credit: AMD)First some background: On Tuesday, AMD announced that IBM had successfully produced a working test chip using next-generation Extreme Ultra-Violet (EUV) lithography for the critical first layer of metal connections across an entire chip. Previous projects utilizing EUV produced working chip components on only a very small portion of the chip.
Why EUV? The size of transistors and the metal lines that connect them is directly related to the wavelength of light that is used to project a chip design onto a wafer. EUV lithography uses a wavelength of 13.5 nanometers (nm), significantly shorter than today's 193nm lithography techniques, allowing the march toward smaller and smaller chip features to continue (though EUV has its own set of problems discussed below). EUV is currently targeted at the 22nm generation of chips, due in three to five years. Intel, a few years back, was targeting EUV for the 45nm generation of chips but abandoned it.
According to the Tuesday announcement, IBM and its partners patterned the first layer of metal interconnects (between the transistors), then, after other processes, the EUV device structures underwent electrical testing at AMD, with transistors showing characteristics consistent with those of test chips built using more standard techniques, the two companies said.
This development follows a series of joint disclosures over the past six years that highlight the crucial expertise that IBM provides to AMD. The two companies began cooperating on advanced chip manufacturing in 2002, when AMD was having trouble with silicon-on-insulator technology, or SOI. AMD got SOI to work with help from IBM and they have been renewing agreements periodically since then. First, in September 2004, to include development of technologies through 2008 for 32nm manufacturing and then again, in November 2005, the agreement was extended through 2011 for the 22nm process. In other areas, AMD is now cooperating with IBM on "high-k/metal gate" transistor technology for next-generation 32nm chips--a technology that Intel is employing in its current 45nm chips.
This kind of know-how is not cheap. AMD's 2007 10K form says the following about the agreement that extends to December 31, 2011: "We anticipate that, under this agreement, we will pay fees to IBM of approximately $400 million in connection with joint development projects between 2008 and 2011."
Even with the substantial payout of $400 million, AMD goes on to say in the 10K that a termination of the agreement "could significantly increase our research and development costs, and we could experience delays or other setbacks in the development of new process technologies, any of which would materially adversely affect us."
Here's the challenge for AMD. Intel's R&D budget dwarfs AMD's. Intel spends about $6 billion per year on R&D, AMD about one-sixth of this. On the manufacturing front, it's even more of a mismatch. For the 45nm generation of chips alone, Intel plans to eventually have four plants making 45nm processors, which Intel is currently manufacturing commercially. AMD doesn't even have 45nm out the door yet (commercially) and needs a chip manufacturing heavyweight like IBM to stay in the running.
As with any collaboration, part of the impetus is to reduce costs. AMD's work with contract manufacturer Chartered Semiconductor is another cost-saving measure. And AMD is by no means the first company to go outside for help and farm out development and manufacturing, said Don Scansen, a semiconductor technology analyst at Semiconductor Insights. But AMD may do more of this as it continues "to get hammered by analysts and the stock market," said Scansen. "IBM is shouldering most of the process development work out of the Common Platform partners--including AMD."
"IBM is a critical part of AMD's Asset Smart manufacturing (and) R&D strategy," an AMD spokesperson said. "By sharing the R&D cost for semiconductor process technology across the membership of the IBM Alliance, each of the parties, including AMD, get access to leading-edge manufacturing technology."
On its Web site, AMD lists transistors, chip connection, packing, and lithography as areas of collaboration. Much of the collaboration takes place at the Albany Nanotech Center. Not coincidentally, Albany is the same spot where AMD has plans to build a chip plant. The chipmaker said recently that if it decides to go forward with the $3.2 billion fabrication plant, construction would start next January, according to a recent report at Timesunion.com.
IBM has a state-of-the-art manufacturing line in East Fishkill, N.Y.--about 90 miles away from Albany--among other locations.
All this collaboration between IBM and AMD means of course that future manufacturing hurdles are high--for everybody. Including Intel, which has delayed procurement of an R&D EUV lithography tool from Nikon, according to a report in EE Times, casting doubt on the viability of EUV at Intel.
From this perspective, the IBM-AMD EUV statement could be seen as promising for AMD. "AMD's ability to get a full chip done with an EUV tool is pretty significant," Scansen said. Some have even been speculating that EUV would not be ready for 22nm, he said. "(But) this announcement might suggest that development is quickening."
AMD and IBM said Tuesday that the next step in proving viability of the EUV lithography for production will be to apply it not only to metal interconnects but to all critical layers to show that an entire working microprocessor can be made utilizing EUV lithography.
Intel has other ideas on how to get to 32nm and beyond using more conventional lithography and "clever design tricks," according to Scansen. And Intel is already moving forward aggressively with 32nm. The company will switch to immersion lithography at 32nm on a "couple of critical layers," according to a statement made in December of last year by Brian Krzanich, vice president and general manager of manufacturing and operations at Intel. The more traditional dry lithography will still be used on less critical layers.
Update: Suntech Power Holdings, an aggressive, rising star in the solar world, said in a conference call yesterday that it hopes to build plants in the U.S. to help it break into the market here.
"We are currently in discussion with the governors of three different states who have been recruiting us to build factories," said Roger Efird, president of Suntech America, the company's U.S. subsidiary, according to a report on Greentech Media. Efird was speaking on a conference call for the Solar Energy Industries Association. More details will likely emerge on November 15 when Suntech reports earnings.
The factories could employ up to 1,000 people in the states in the next few years, Efird reportedly said.
Update: In a phone interview with News.com, chief strategy officer Steve Chan said that they company is in the decision making stage. "The question is how to get to the point where the costs make sense," he said. Suntech might not move to the stage where construction begins for one to two years.
Earlier this week, green drywall maker Serious Materials raised $50 million to build factories in the U.S.
U.S. factories will also let Suntech compete in what many believe will be the largest solar market in the future. The company right now sells most of its products to Europe. Five years ago, Suntech was an asterisk. Now, it is the third largest producer of solar cells in the world and one of the fastest growing. It benefits from cheap Chinese labor, but also depth in research and development (the science behind its panels came out of the University of New South Wales), strong ties to equipment makers, and volume discounts on silicon. Executives in China make less too: there probably aren't 10 people in the company who make more than $200,000, said Chan earlier this year.
European and Japanese manufacturers have trouble competing on price, and Chinese manufacturers have trouble keeping up with their quality.
U.S. factories, though, will challenge the company's low cost position, as Chan noted. Suntech, though is testing out different strategies to get around this. One idea: Suntech is developing its own manufacturing equipment that could reduce costs.
The clean-technology revolution will likely make a lot of people in Silicon Valley rich, but it's also going to help bring back some of the factory jobs that have disappeared.
Why? Weight, for one thing, explains Kevin Surace, CEO of Serious Materials, which recently landed $50 million in funding to build factories for its eco-friendly drywall.
Although labor is cheaper in China, shipping costs are going up, primarily because of fossil fuels.
"You could spend $2 to $3 a panel just to ship it (drywall), and that's just to get it to the dock. You'd then have to spend another $3 to $4 to ship it by rail," he said. "You can't do that if you plan to sell it for $10 to $20 a board."
As a result, Serious Materials will open its first factory, which will be capable of churning out 400,000 square feet of drywall a year, in the United States. It will likely also build its next two factories in the States as well, regionally spaced out to serve different markets.
Shipping materials from China also "blows the whole point about zero carbon dioxide," he added. "You're on the wrong side of the energy curve."
State governments are also offering substantial incentives--free rent in industrial parks, tax holidays, loans, grants--to woo these companies. "States do not want to be left out of the next industrial revolution," Surace said.
Some of the most aggressive states include New York, California, and New Mexico.
The heartening part of all of this is that Surace isn't alone. Bruce Jamerson, CEO of Mascoma, which wants to make cellulosic ethanol, has said the same thing. Mascoma is building plants in Michigan, New York, and Tennessee because that's where the wood chips and vegetable matter are. Several analysts have said shipping is one of the big barriers for Chinese solar-panel makers.
Granted, it's not like these companies are staying in the States because the CEO woke up one day to a Bob Seger song playing on the radio and started getting misty-eyed over the disappearance of the industrial heartland. They are being encouraged to stay stateside in part because of subsidies.
But other factors--like shipping costs, the low prices of their products, and the proximity to local markets--could conspire to get the manufacturing arm of the country moving again.
While AMD isn't giving away any information on its future fab plans, a major chip foundry is gearing up for a big new customer.
TSMC, the largest chip foundry in the world, is apparently planning to take on a new customer that wants to use the high-k dielectrics and metal gates introduced earlier this year by both Intel and IBM. Sumner Lemon of IDG News Service sat through a TSMC earnings call in which company executives mentioned a mysterious new customer would be coming on board in the second half of 2008.
There are not a lot of folks planning to use that technology at 45-nanometers in 2008. Intel is, but it's got its own manufacturing facilities. IBM is, but it also grows iits own in New York. Sun and TI are, but they've got plans to build their chips at TI's facility in Texas. Then comes AMD, which is planning to use metal gates in the future and has been thinking about relying more on foundries.
Novel technologies like metal gates--which are actually the first changes to the basic materials in a transistor since the 1960s--generally enter production on processors that are gunning for every last bit of performance. If performance isn't as much of an issue, it makes more sense to wait until the kinks are ironed out. Companies like Intel, IBM, and AMD are generally the ones pushing the envelope on both performance and new transistor technology, so it's pretty likely that TSMC's customer is someone like that.
AMD's losing a lot of money these days, and rumors have been flying that the company wants to outsource more production to foundries. TSMC already makes graphics chips for AMD. Both companies use the letter "M" in their names. Seems like a done deal.
But yesterday, AMD said it hadn't decided whether to use the metal gates on the second iteration of its 45-nanometer chips, or the first generation of its 32-nanometer chips. AMD's first 45-nanometer chips are scheduled to come out in the middle of 2008. If in fact that decision hasn't been made, then TSMC is probably looking at another customer in 2008, since AMD's 32-nanometer designs won't be ready until 2009 at the earliest.
All of this speculation comes because AMD hasn't said exactly what it wants to do to obtain future chip-making capacity. It has danced around something called "asset-light," but hasn't really explained what that will mean in the future.
IBM's chip federation has grown again.
Switerland's STMicroelectronics will collaborate with IBM to develop manufacturing processes for the 32-nanometer and 22-nanometer generations of chips. 32-nanometer chips should start coming out in 2009 or 2010. (the number refers to the size of the average feature on the chip. A nanometer is a billionth of a meter.)
A whole cavalcade of companies--Advanced Micro Devices, Chartered, Infineon, Samsung, Sony, Toshiba, Freescale-- already have existing semiconductor alliances with IBM and all of these alliances overlap and leverage each other in various ways. Under the ST-IBM alliance, researchers from each company will be transferred to the labs of the other.
The cooperation, ideally, will help spread the costs and risks in the chip industry. Designing a new process can cost hundreds of millions and a new chip can cost several millions to design.
Sometimes these alliances work, and sometimes they don't. ST, Philips and Freescale once cooperated on research and manufacturing in a plant in France; it has since been dissolved. AMD once had a far-ranging alliance with Taiwan's UMC that was dissolved before the first chips came out the door.
On the other hand, the Cell processor came out of the cooperation between Sony, IBM and Toshiba. IBM also helped AMD implement strained silicon in its chips.
Dell continues to make wholesale changes to its executive roster.
On Wednesday, the company announced that it has hired Michael Cannon, the chief executive officer of manufacturing powerhouse Solectron, to lead Dell's worldwide manufacturing operations. Cannon will oversee all of Dell's global manufacturing sites, which were previously run by local leaders, a company representative said.
Cannon is the latest hire in an executive shakeup that has seen CEO Kevin Rollins and CFO Jim Schneider--among others--leave Dell after a tumultous year. Last week, online marketing chief John Hamlin became the latest longtime Dell executive to seek other pastures.
Even before his return to the CEO spot, Michael Dell has been busy recruiting new executives--such as Steven Schuckenbrock from EDS, Donald Carty from American Airlines, and Sona Chawla of Wells Fargo--from outside the technology industry in hopes of injecting new life into the organization. As head of Solectron, Cannon oversaw a huge technology manufacturing organization, but Solectron also designed and built products for other industries such as automotive and medical.
- prev
- 1
- next
