AMD's High-Stakes Wager

The chipmaker capitalizes on advances in the server and desktop markets, betting $2.5 billion on a plant that could help it reach its goal of doubling market share in the

October 31, 2005

7 Min Read
Network Computing logo

The tech industry has a long history of David and Goliath stories. Yet few tech underdogs have started as far behind as Advanced Micro Devices Inc. did in its battle with Intel in the x86 chip market. And few have made such big gains so fast. In the past year, AMD beat Intel to the market with 64-bit x86 technology and with dual-core server chips. Those advances have paid off in substantial market gains.

In June, AMD chips were in only 20% of desktop PCs sold in the retail market. In September, AMD for the first time surpassed Intel, capturing 52% of that market, according to research firm Current Analysis.

Now AMD is making its biggest bet yet--spending $2.5 billion, or half of its 2004 revenue, to build a chip plant to substantially increase its manufacturing capacity. It's a gamble many believed not too long ago that the company wouldn't be able to afford. But AMD now believes its investment has set the stage for a major breakthrough in 2006 and beyond.

"We made progress at leaps and bounds in every quarter in 2005," says Hector Ruiz, chairman, president, and chief executive of AMD. "The perception of the company in the world has been elevated."

AMD's Fab 36 plant in Dresden, Germany, will let the chipmaker increase processor production.

Greater Capacity
AMD earlier this month officially opened Fab 36, a processor manufacturing plant in Dresden, Germany. The plant gives AMD the ability to produce a much greater volume of processors than it could before. Fab 36 demonstrates that AMD has the wherewithal to be a long-term player in the microprocessor market. Not only will the plant provide much-needed leading-edge capacity, AMD believes it will be able to use Fab 36 to demonstrate that it has implemented greater manufacturing flexibility than Intel by having the ability to adjust its manufacturing processes on at least a quarterly basis using what it calls an automated-precision-manufacturing model, a way of introducing innovation into the manufacturing process more quickly.As the plant ramps into full production, AMD will be able to build enough chips to meet its target of roughly doubling its market share in the overall x86 microprocessor market (servers, desktops, and notebooks) over the next two to three years, from about 16% today to 30% or more. That's assuming, of course, that orders continue to grow.

The plant is AMD's first to utilize 300-millimeter wafers, the largest used in the semiconductor industry. The move from the 200-millimeter wafers it uses in its existing Fab 30 in Dresden to the larger wafers gives AMD more than double the area to produce hundreds of individual processors. In addition, the company will shrink the transistor line widths of the processors that move into production next year to 65 nanometers from the current 90 nanometers, giving it the ability to cram even more processors onto each wafer.

AMD plans to phase out Fab 30 over the next year or so and rely entirely on Fab 36. The plant will carry AMD through the next three evolutionary steps of semiconductor manufacturing, from 65 nanometers to 45 to 32, giving the plant at least a six- to eight-year life span, Ruiz says. "It's been a gargantuan task that is on schedule and on budget, and has been well executed by our German team," he says.

The plant should help AMD battle with Intel, which has five 300-millimeter fabs that mainly create microprocessors. "Reputations are built one action at a time," Insight 64 analyst Nathan Brookwood says. "AMD has done reasonably well with their manufacturing for the last few years, and now they will be very competitive with Intel in terms of manufacturing costs once they move into full production."

Sonderman compares AMD's model to FedEx.

Tom Sonderman, director of automated precision manufacturing for AMD, says the next version of the manufacturing technology being implementing in Fab 36 is the "secret sauce" that will let AMD better compete with Intel, which has a capital budget this year of about $5.6 billion, compared with AMD's $1.5 billion."We have a unique model for manufacturing that we think can be as revolutionary as how FedEx changed the way people deliver packages," Sonderman says. "We aren't going to have as many fabs as Intel, and we aren't going to outspend Intel, so we need to leverage our assets more effectively."

Automated precision manufacturing uses concepts originally found in the chemical-process industry and includes AMD's own software to create control and scheduling capabilities designed for semiconductor manufacturing.

Continual Improvement

The typical method for introducing technologies into a manufacturing flow is to run them on smaller and separate R&D pilot fabrication lines. When the processes have proven themselves and consistently deliver high yields of usable devices, the technology is then transferred to a full-scale production facility.

With automated precision manufacturing, AMD is committing about 10% of its volume manufacturing capacity to research and development. Its unusual to run an R&D pilot line alongside mainstream production lines. The company also has moved from the traditional approach of making major technology changes within the manufacturing environment on a semiannual basis to making continual incremental improvements on the transistor level.

AMD believes it can make ongoing changes because it has developed five algorithmic analysis systems, which span equipment- performance optimization, integrated-production scheduling, and yield- management systems to implement new plans quickly. "The real difference in what we're doing is we've made a big bet that by connecting every piece of equipment in the fab to an automated decision-making environment, we can use that information in a much more productive way than has been possible before," Sonderman says.With automated precision manufacturing, AMD "does appear to gain certain advantages in flexibility and responsiveness that Intel doesn't have," Brookwood says. "Intel also has the challenge of trying to run four or five separate fabs with absolute consistency between them."

In addition to Fab 36, AMD is transferring its automated-precision-manufacturing process to semiconductor foundry company Chartered Semiconductor Manufacturing Ltd.'s 300-millimeter plant in Singapore in 2006, providing AMD with "flex" capacity it can tap into if market demand exceeds its own manufacturing capacity. AMD, which can make around 45 million processors this year, will be able to boost its production to about 60 million units in 2006 and 100 million units in 2008.

Market-Share Increases
There are indications that the increased capacity will be needed. According to research firm Mercury Research, AMD controlled 16.2% of the overall x86 microprocessor market in the second quarter of this year, compared with Intel's 82.3%. A year earlier, AMD's share was 15.1%. When counting only x86 processors used in servers, AMD's market share in the second quarter was 11.2%.

Much of AMD's recent market-share gains in the retail PC market came from the success Hewlett-Packard enjoyed with its AMD-based Media Center PC platforms, which accounted for 46% of all U.S. retail desktop sales in September, Current Analysis analyst Toni Duboise says. "Earlier in the year, AMD was having a lot of difficulty getting their products on the retailers' shelves," she says. "Basically, what we saw beginning in July, with the back-to-school season, was that if AMD's product makes it to the store shelves, they sell."

Progress came in "leaps and bounds in every quarter in 2005," AMD chairman Hector Ruiz says.

AMD will be able to build a successful PC business in commercial accounts, Ruiz says. The company decided to "go after the belly of the beast" and break into the enterprise by establishing a compelling value and performance proposition for its Opteron processor in servers. The strategy was accomplished when AMD was able to beat Intel to the market first with 64-bit-capable processors, and then with the first dual-core processor.Only in the past few weeks has Intel been able to get its first dual-core Xeon processors to market, giving AMD a yearlong advantage to build a growing base of servers in business deployments.

But AMD's gains in the past year or two can't be denied. That may be why Ruiz says he has only one major goal for 2006, and that's for the company to "stay the course" and capitalize on new opportunities expected to emerge from a year of advancement.

Stay informed! Sign up to get expert advice and insight delivered direct to your inbox
More Insights