Intel chip will be bigger, more expensive to make

Intel's upcoming Willamette processor eventually will reach speeds beyond 2 GHz, but the size of the chip is raising questions about how much it will cost to make.

Willamette, Intel's next-generation desktop processor expected to emerge later this year, will debut at 170 square millimeters, according to estimates from analysts. At that size, it will be 60 percent larger than today's Pentium III. While the larger size will allow Intel to pack more transistors onto the chip and add more features, it also means fewer chips can be produced out of a single wafer, which raises Intel's costs.

That has some worried that the transition to Willamette, Intel's first all-new desktop architecture since the Pentium Pro in 1995, could affect Intel's profit margins in the future.

"Intel's microprocessor line should be extremely profitable in (the second half of this year), but as the company transitions to Willamette in 2001, margins will again begin to fall," U.S. Bancorp Piper Jaffray analyst Ashok Kumar said in a report today.

Intel would not comment on the size or manufacturing cost of the chip.

Others think that the large die size won't hurt Intel that much, as Willamette will initially be a rather high-priced item. Credit Suisse First Boston analyst Charlie Glavin said that by the time Willamette is in the cost-sensitive segment of the market, Intel will have shrunk the chip as it has with previous processors.

"Don't be myopic about the first die size and product," Glavin cautioned investors. "You throw out the first one. It's like making pancakes."

It's not unusual for Intel to debut a processor at a large size, waiting for the next generation of thinner wiring to allow the chip to shrink to a less costly size. Intel is expected to move sometime next year to a 0.13-micron process, from today's 0.18 micron manufacturing line. The numbers refer to the size of components on the chip.

"They will aggressively shrink that die," said MicroDesign Resources analyst Kieth Diefendorff. "It's going to spend most of its life at 0.13 micron."

Kumar said the large size will prevent a major launch of Willamette processors when the chip debuts later this year. In April, Paul Otellini, general manager of the Intel Architecture Group and the chief of processor programming, said that Willamette systems would be available in volume toward the end of the year.

"Systems from major manufacturers based on Willamette and Timna, a low-end version of Celeron with an integrated graphics chip, will be in the market in high volumes for the peak selling season for 2000," he said at the company's spring analyst meeting.

Otellini said the chip will debut at 1.4 GHz. The first Willamette chips are expected to emerge in late September or October, according to sources.

By contrast, Kumar predicted quantities of Willamette would be limited this year, partly because of the size of the chip. A shortage of factory capacity has also created a processor shortage, although Kumar noted that the company is investing heavily to bring more facilities on line by next year.

"The more wafers Intel devotes to Willamette, the fewer total processors it can produce," he said. "This problem will prevent Intel from selling significant quantities of Willamette this year, as the company is already capacity constrained and expects to be so until the end of the year."

Willamette is crucial to Intel's plans to stay ahead of rival Advanced Micro Devices. The two companies have been battling for the fastest clock speed since AMD debuted its Athlon last June. At 1 GHz, the current Pentium III design is nearing its peak. Willamette is slated to debut at a speed of at least 1.4 GHz.

Chip competition

As for performance, Intel's chips have done better on benchmark tests of late, largely because AMD's Athlon has lacked extra on-chip memory, known as Level 2 cache. That will change on Monday, as AMD debuts a revamped version of Athlon, code-named Thunderbird, that contains 256 KB of on-chip secondary cache.

While it adds some performance enhancements, one of Willamette's chief benefits is that it has been designed to easily climb to faster clock speeds, most likely to 2 GHz and faster.

"Every transistor in Willamette has been designed from scratch to meet three goals: Clock speed, clock speed and more clock speed," Kumar said.

In addition to the new processor core, Intel is planning a faster system bus, which will enable the processor to exchange data faster with memory. Intel is hoping that will drive demand for Rambus-based memory and has said its companion chips will only support the expensive, high-speed memory.

"Dual-Channel (Rambus-based memory), we think, is a real nice match-up," said Intel spokesman George Alfs.

However, Kumar predicts that Intel will probably relent if the industry clamors for another memory option, such as double data rate (DDR) memory.

"If there is severe (manufacturer) resistance, they will pull a DDR chipset out of their back pocket," Kumar said.

Intel has already said it will support DDR for machines running Foster, a version of Willamette for servers. Alfs would not say whether Intel is developing a DDR chipset but reiterated that the initial chipset that will accompany Willamette at launch supports only Rambus-based memory.

Once again, the question remains whether all of the performance Intel can deliver is actually needed.

"No one in the volume PC space today needs a 1-GHz processor," Diefendorff said. "It's just useless."

On the other hand, Diefendorff said boosting performance beyond what most applications use is not necessarily a bad thing, since software designers tend to design for the types of computers most people are using.

"Having that much excess performance capacity would certainly enable some applications that aren't being done today such as artificial intelligence and speech recognition," he said.

News.com's Michael Kanellos contributed to this report.