Original URL: https://www.theregister.com/2009/03/02/amd_no_foundry/

AMD: 'At heart, we're a design company'

Look, Ma! No foundry!

By Timothy Prickett Morgan

Posted in On-Prem, 2nd March 2009 23:49 GMT

Calling AMD a chipmaker is no longer accurate. The proper term is chip designer and seller.

Today was Advanced Micro Devices first day of business without its chip foundry. In the short term, little will change between AMD, the chip designer, and the Foundry Company, the chip fabrication outfit the company has spun out. But over the long haul, plenty of things can change as AMD is now free to react to market conditions and the Foundry Co. will be looking for new clients to keep its production lines ramping.

Nigel Dessau, chief marketing officer at AMD, says the relationship between AMD and the Foundry Co. is by necessity a symbiotic one, with AMD having a 34.2 per cent stake in the foundry and an equal number of voting shares as the foundry's buyer, Advanced Technology Investment Company.

ATIC is the investing arm of the government of Abu Dhabi, one of the United Arab Emirates that is flush with oil and cash - just the kind of partner you want in a chip fab. It takes billions of dollars to invest in chip manufacturing technologies and move processes forward. Moreover, Middle Eastern governments know that sooner or later, the oil runs out and they are going to need to participate in (and perhaps control) other industries. Buying AMD's fab operations makes at least as much sense as anything else Abu Dhabi might do to bootstrap a tech industry.

"AMD is at its heart a design company, not a foundry," says Dessau, who only joined AMD in March 2008 after a stint at Sun Microsystems running its storage marketing and two decades at IBM working mostly on mainframe marketing.

"It's a great business, but not for AMD. This may be a 300 million unit semiconductor market this year," Dessau concedes, talking about the global market for chips of all kinds, "but it can be a 400 million, 450 million, or even 500 million unit market five years out from now. There is no doubt in anybody's mind that the appetite for semiconductors is going to continue."

Which is why the backers of the Foundry Company, which will launch formally later this week - hopefully with a better name - would kick $825m into AMD's coffers. AMD needed the cash, to be sure. But who really needs a foundry more than a chip maker whose products are tightly tied to the next one or two generations of process technologies in their own foundries? Foundry Co. has a guaranteed customer for quite a few years.

AMD got its start back in 1969, when a bunch of ex-Fairchild Semiconductor executives formed the company and tapped Jerry Sanders to be its president (and eventually its CEO). And in 1975, the company cloned Intel's 8080 microprocessor. (Gordon Moore and Bob Noyce left Fairchild only ten months earlier than the AMD crowd to form Intel, and they have been locked in mortal combat since that time).

Back in the early days of the microprocessor (which was invented by Intel when it created the 4004 chip in 1971), being in the microprocessor business tended to mean being in the foundry business. This was not necessarily so, but it was the way Intel did it, and it is the way IBM did it with mainframes and even earlier electro-mechanical data processing equipment.

But over time as each new generation of chip making technology became more expensive, some microprocessor companies (like Sun Microsystems) decided to be fabless because of the high cost of making chips. Texas Instruments was always happy to have someone like Sun to test out its most advanced tech on large, low-volume chips. (TI has since itself decided to go fabless). Still other microprocessor makers partnered to either keep their fabs full of extra work to make them pay their way better (think IBM) or to actually get chips made by others (IBM made the last few generations of PA-RISC chips for Hewlett-Packard and still makes Unisys' mainframe engines).

'Real men don't need fabs'

That is by no means meant to be an exhaustive list. But there are a lot of different ways chips can get made these days, and it is increasingly true that no chip maker can stand alone. Note with humor that Intel and TSMC have inked an agreement today that will see the Taiwanese chip maker create variants of Intel's Atom x86 processor that will in turn be sold by Intel into its OEM channel.

Basically, TSMC is a partial Foundry Co for the Atom processor. So even big, rich, and proud chip makers like Intel, which was bragging three weeks ago that it was shelling out $7bn to upgrade its U.S.-based fabs has to nonetheless deal with the harsh economic realities of chip making. Everybody is going to need friends, and this is going to be a cooperative effort.

"I know this violates Jerry Sander's rule, that real men have fabs," says Dessau. "But markets change. Real men work in ecosystems, and it is always better to spend someone else's money than your own."

One part of AMD, the ATI graphics chip business it acquired in July 2006 for $5.4bn, never had its own foundry. In fact, it uses TSMC to make most of its chips as well as another Taiwanese foundry, the slightly older UMC. And, soon, Foundry Co. will be getting a little taste of that ATI graphics chip action too, as a sweetener for the ATIC-backed deal. You can see now why Intel was keen on working a deal with TSMC today, of course.

AMD is, of course, bound pretty tightly to Foundry Co. for its current 45 nanometer processes, used to create the current "Shanghai" quad-core Opterons, among other chips, as well as the future 32 nanometer chip making processes coming down the pike. Dessau says that AMD and Foundry Co. are exploring their options for 22 nanometer processes. "At this point, we are not sure we will go anywhere else," says Dessau. "But in five years time, who knows? We have designed our model so we have options."

With the slip in the "Barcelona" quad-core processor launch in late 2007 into early 2008 and the fact that AMD is trailing Intel in chip making processes for several years, there has been plenty of talk about AMD losing some of the mojo it had when the Opteron chips busted onto the scene and scared the living hell out of Intel in April 2003. "We have three chips in a row of not just meeting, but beating our dates. AMD has its mojo back, and we have a great set of products out and on the way," says Dessau.

Focusing on chip design, sales, and marketing is more important than ever, and it is going to require some freedom. "The single most important factor in the semiconductor business today is that the speed of the processor no longer determines the experience of the end user," Dessau explains. "And AMD can't be an agile company if we have huge debts and have to run a manufacturing core."

In a separate but related announcement, AMD has appointed Bruce Claflin, formerly chief executive officer at 3Com and a long-time exec at IBM and Digital Equipment, as AMD's chairman of the board. He replaces Hector Ruiz, who has been AMD's chairman since being hand-picked by Sanders as his successor back in 2000. Ruiz has retired from the AMD board and is now chairman of the board at the Foundry Co.

AMD also said that Waleed Al Mokarrab, chief operating officer at Mubadala Development Company (one of the backers of the Foundry Co from Abu Dhabi that is kicking in funds alongside ATIC, but which has no shares in the company), has been appointed to AMD's board of directors. ®