AMD Makes its 65nm Move: Will It Be Enough?

AMD said it would be integrating strained silicon engineering processes -- stretching the atomic latticework of the semiconductor's silicon layer by attaching it to silicon germanium -- along with its silicon-on-insulator (SOI) technology, with the goal of producing the lowest power 65 nm processors in just 18 months' time. At least that was the goal, when AMD made that announcement in June 2003.

Today, the company that last year held the triple-crown of price, performance, and power-saving leader, announced that the 65nm fabrication transition is now underway. AMD's goal is to fully convert its Dresden, Germany, fabrication facility to 100% 65 nm by the middle of next year, with the first 65nm Athlon 64 X2 dual-core processors due in the first quarter.

It cannot afford any further delays, not if it expects to catch up with Intel's stated goal of starting its move to 45nm lithography next year.

Why is smaller necessarily better, if it isn't necessarily faster? Semiconductors that can be produced on smaller scales require less power to operate, as Intel's introduction of its Core 2 Duo, Core 2 Extreme, and now Core 2 Quad demonstrated most undeniably last July.

These Intel processors have already met the 65nm goal, and Intel has already weaned itself from the 90nm lithography on which its first dual-core Pentium D 8xx series was based.

Moreover, when CPUs are smaller, a foundry can produce more of them on a single 300mm wafer, increasing manufacturing yields without increasing the cost per unit.

One other potential benefit, however, has taken longer to realize than manufacturers anticipated: Microprocessors that use less power should theoretically leak less power, which is extraordinarily important in the race to make mobile computers both faster and more efficient. In practice, this hasn't always been the case.

As it turns out, foundries aren't exactly "EZ-Bake Ovens" - you can't just take a 90nm design and shrink it by 28%. Both AMD and Intel have had to develop design processes that embed insulators on their substrates, to compensate for the leakage that engineers surprisingly discovered could happen at lower and lower scales.

AMD –- working in association with IBM -- is known for making incremental design changes in its processor production methodology from month to month, some on an almost ad hoc basis. The company even has a marketing term for this: Continuous Transistor Improvement (CTI). Meanwhile, Intel makes careful, long, introspective analyses of its production processes, before making sweeping changes every three years or so. Last July, by all measures, Intel started producing a 65nm processor that performed better and used less power than AMD’s best 90nm products.

While AMD’s first 65nm Athlon 64 X2 processors next year -– available probably just in time for the consumer release of Windows Vista -– will sell for the same price as its 90nm line, they probably won’t outperform their 90nm counterparts right at first. But AMD’s goals are now firm: Even its Quad FX high-performance series, announced just last week, must be converted to 65 nm by the middle of next year.

These transitions will be but two steps on the road to surpassing Intel once again, which throughout its storied history AMD has always managed somehow to do. Last June, the company announced it will be building a new fabrication facility in New York, with the goal of that facility being able to produce 32 nm processors as soon as 2012.

But for Intel, assuming it makes its goals, the road between 45nm in 2007 and 32nm thereafter, may not be as long. AMD may be hoping, privately, that engineers on both sides encounter some more physical stumbling blocks, sending Intel back on one of its long introspectives, buying it time to make just a few more incremental refinements.