For AMD, more money means more problems

Demand for its current chips has AMD scrambling to get capacity online just as Intel gets ready to win the race to four cores.

Tom Krazit Former Staff writer, CNET News
Tom Krazit writes about the ever-expanding world of Google, as the most prominent company on the Internet defends its search juggernaut while expanding into nearly anything it thinks possible. He has previously written about Apple, the traditional PC industry, and chip companies. E-mail Tom.
Tom Krazit
5 min read
Success has its pitfalls, especially when you've awoken a sleeping giant.

For the past two years Advanced Micro Devices has made Intel, one of the world's most prominent companies, look bad. Better products and better timing have brought AMD significant market share and prominent new customers like Dell. But Intel is on the comeback with new processor designs better suited for the power-efficient multicore era, and it will beat AMD to the quad-core punch by using a design strategy that makes purists scoff but accountants happy.

The next six months will be tricky for AMD CEO Hector Ruiz. For starters, his company is taking a hit in the stock market after disclosing that its gross margin fell five points from the second quarter to the third, which ended last week. Part of that was due to the price war in the PC processor market, but AMD also is facing a challenge of overall demand rising before new manufacturing technologies are completely ready.

"It's a good problem to have," said Dean McCarron, an analyst with Mercury Research, since it means people want your products. "But AMD needs more factories." The company's design strategy for the quad-core era requires that it successfully navigate the transition to building smaller transistors at the same time it gets a new factory up and running.

AMD's current pickle is the result of its success, which makes it a little easier to swallow for company executives. Demand is high, but the company's dual-core processors still use its 90-nanometer manufacturing technology. Intel's chips, on the other hand, are built using the smaller transistors provided by its 65-nanometer manufacturing technology. Not only is AMD using larger transistors, but its dual-core Opteron and Athlon 64 processors contain two processing cores integrated onto a single piece of silicon, or a die. This design has given AMD great performance during the past few years, but resulted in processors that were almost twice the size of its single-core chips.

Individual chips are cut from round silicon wafers. Manufacturers obsess over reducing the number of defects on those wafers, but there's always going to be some number of chips on a wafer that simply don't work. The problem is that when each individual chip is relatively large, there's an increased risk that a portion of that chip might contain a defect. Since it costs the same to make a wafer whether a chipmaker gets six chips or 60 chips from that batch, maximizing yields--or the number of good die per wafer--is essential to this business.

AMD's 90-nanometer dual-core Opteron and Athlon 64 processors have a die size of 199 square millimeters. By chip design standards, that's considered a little large, McCarron said. When AMD starts making dual-core Opterons on its 65-nanometer manufacturing technology, that die size is expected to go down to something a little more comfortable that will allow AMD to produce more chips per wafer. An AMD representative declined to comment on the die size for its first 65-nanometer products.

On a conference call following AMD's earnings results last week, Chief Financial Officer Bob Rivet noted that the company would see a cost benefit from its move to 65-nanometer processors in the fourth quarter, since the cost of building the wafer can be spread over more chips. He also pointed out that AMD still hasn't made the full transition to 300-millimeter-wide wafers from 200-millimeter wafers. Obviously, the larger the wafer, the more chips that can be cut from that wafer, and--not counting the one-time expense of purchasing 300-millimeter equipment--the extra costs of the larger wafer are negligible.

"Every day will make the wafer costs go down because we will have better utilization and the die costs will go down due to the conversion to 65-nanometers," Rivet said. The company will also be able to produce more chips from the same wafers in order to satisfy demand, and performance will also likely be improved.

Intel, however, made the move to 65 nanometers in the fourth quarter of 2005, and to 300-millimeter wafers some time ago. Its Core Duo chip was its first processor built using the smaller transistors, and it announced earlier this month that it is now shipping more 65-nanometer processors than 90-nanometer chips.

This has given Intel the flexibility to leapfrog AMD to the quad-core generation of processors. Intel plans to build quad-core chips by taking two separate dual-core processors and putting them together in a single package, which it calls a multichip module (MCM).

The MCM allows Intel to get its designs out into the market faster than AMD. Intel's first quad-core processors are expected to become available next month, but AMD is waiting until the middle of next year to unveil its quad-core server processor.

It also allows Intel to maximize its yields by building smaller chips. For example, if one of the dual-core processors in the MCM gets knocked out by a defect, the whole product doesn't have to be tossed. Intel will still need to build dual-core chips for the mainstream and lower ends of the market for several years. It can build dual-core chips for those markets, and simply package two dual-core chips when it wants to ratchet up the performance.

AMD's chips are based on a different design than Intel's, and so the company believes it benefits most from an integrated core design, where all four cores on its quad-core chips will live on a single piece of silicon. This means that data can be exchanged between cores at the chip's clock speed, since the data doesn't have to leave the die. Intel's design means that if a core on one processor wants to exchange information with a core on the other processor, they have to do so at rates slower than the chip's clock speed, since the signals have to travel through the package.

Some chip enthusiasts--who occasionally resemble architectural critics--aren't too impressed with the MCM approach, since it fails to address Intel's reliance on external communications links to exchange information between processors. This was one of the factors that led to AMD's performance advantage up until the introduction of Intel's Core microarchitecture processors, which outdo AMD's chips on several benchmarks.

Intel thinks its quad-core processors will be extremely competitive on performance and power consumption, in part because the company will boost performance between now and when AMD's quad-core chips are ready, said Bill Kircos, a company spokesman. Still, Intel will use a mix of monolithic and MCM quad-core designs in the future, depending on the need for performance, low-cost chips, volumes, and speed, he said.

AMD's monolithic design also means its die sizes will likely increase when it moves from dual-core 65-nanometer processors to quad-core 65-nanometer processors, raising the cost and margins issue once again. Earlier this year, AMD showed it recognized the need to catch up to Intel's manufacturing pace with the announcement that it plans to introduce 45-nanometer processors just 18 months after rolling out its 65-nanometer chips, a transition that usually takes at least two years. But if it can continue to deliver performance that makes server customers drool with its quad-core chips, the cost questions will be overshadowed by the revenue padding its bottom line.