The Camino Chipset Is Still Bugging Intel

SAN FRANCISCO -- There are bugs and then there are billion-dollar bugs.

Intel (INTC) has suffered through plenty of bugs in its time, but the one plaguing its Rambus (RMBS)-based Camino chipset is a corker.

See Also
Intel Bedfellows Are Unhappy With the High Costs of Rambus Technology

Rambus' memory-chip designs seemed destined to become the gingko biloba of semiconductors -- a leap forward in computer memory and data processing speed. In 1996, Intel vowed to back Rambus chips, making it tough for Intel customers to choose an alternative, and set to work designing the Camino chipsets that would connect PC chips to Rambus memory. Rambus, which went public at 12 in May 1997, rose as high as 117 1/2 last July.

In recent months, however, the tide has turned. Concerns of delays in the Camino gained momentum despite Intel's assurances. Then, on Sept. 27 -- Camino's planned launch date and two days after Intel told the Los Angeles Times that all was on schedule -- Intel acknowledged the Camino bug. Rambus closed down 2 3/16 at 70 9/16 Friday. Intel closed down 2 15/32 to 70 7/8.

A Bug's Life
March 1990	University of Illinois professor Mike Farmwald and Stanford University professor Mark Horowitz found Rambus to design high-speed memory chips.
March 1992	Fujitsu, Toshiba and NEC license Rambus designs.
July 1994	Nintendo announces it will use Rambus DRAM in its gaming console.
October 1995	Hitachi, LG Semicon, NEC, Oki Electric, Samsung and Toshiba announce they will begin volume production of Rambus DRAMs for multimedia computers in 1997.
December 1996	Intel signs agreement with Rambus says the memory design will become the industry standard by 1999.
February 1997	Alternatives appear: Samsung to develop double data rate DRAM. Industry forms alliance for SyncLink.
May 1997	Rambus IPO offered at 12, opens at 23 3/4 and closes at 30 1/4 in the biggest IPO of the year.
July 1998	Toshiba says Rambus will have 50% of memory market by 2001.
January	Rambus stock hits all-time high of 109 15/16.
February	Intel announces three-month delay of Rambus-based chipset until third quarter.
August	Intel announces it will produce PC133 chipsets as an alternative to Rambus in the first quarter but Rambus' Sept. 27 launch is on track.
September	Bug in Intel's chipset delays Rambus launch indefinitely

Slowly, the toll of the Intel-Rambus fiasco is becoming clear. Executives in the computer-memory industry told TheStreet.com they now believe the Camino problem is so intractable that Rambus technology may never work, at least not in any way that would make Rambus chips marketable or profitable. The problem goes beyond dollars and cents; many in the semiconductor industry are critical of how Intel has effectively forced Rambus technology on the marketplace. The cost of a Rambus failure to the computer industry, they say, could reach billions of dollars in potentially wasted inventory and development and lost revenues.

TSC interviewed executives and representatives at 12 companies that make memory chips, modules, testing equipment and personal computers, as well as several other sources with years of experience working with Intel and the memory industry. Many asked for anonymity because of nondisclosure agreements with Rambus. Most assert that Intel and Rambus are in denial about the scope of the problem and the potential cost.

While most companies declined to estimate how much they have spent on Rambus, one executive at a maker of DRAM chips says his company has spent about $500 million on development so far. The latest delay has cost his company $45 million in products that now have to be scrapped, plus another $180 million in estimated lost sales from standard memory parts that would otherwise have been made and sold using an alternative to Rambus.

Intel and Rambus acknowledge delays, but deny there is a long-term problem. Paul Otellini, Intel's director of its architecture business group, assured investors in an Oct. 12 conference call that the company is addressing the problem. "We are continuing to do validation activities" on the Camino, Otellini said. "We anticipate being able to ship that when we finish the validation in the fourth quarter."

Rambus CFO Gary Harmon says Rambus DRAM chips manufactured so far are still usable, if not in PCs then in the upcoming Sony (SNE) Playstation 2. "No one is suggesting the DRAMs have to be tossed," he says. "There doesn't seem to be any issue with the Rambus DRAM we can see." Harmon acknowledged that the PC market represents 60% of the DRAM market, while video games account for just 10 percent.

Industry sources say the quick fixes for Rambus that Intel is likely looking at will either result in a slower system or limit the amount of memory in a PC because they will reduce the number of memory slots. And while Rambus chips might be usable for these chipsets, many say the motherboards made might still have to be scrapped.

Logistically, it would be impossible to produce new DRAMs this quarter even if Intel can launch its chipset now, executives at four memory chipmakers say. Outside of Intel, all Rambus DRAM production has stopped, they say. "A DRAM takes four months to make," says Jim Sogas, director of the DRAM division at Hitachi (HIT). "At the very, very minimum it would be 90 days."

Even as Intel sorts through Rambus, the company is readying an alternative called PC 133 SDRAM. And Hitachi and Micron (MU) are preparing to demonstrate computers with another standard, called double data rate, at the Comdex trade show in November.

For Rambus, which makes a royalty of about 1.5% on the sale of every memory chip using its design, the wide-scale abandonment of its chips could be devastating. Harmon dismissed such a bleak scenario. "We don't see it as even a remote possibility," he says. "That will only happen if PCs don't need performance. [Alternatives] don't really come close to Rambus performance." Some industry forecasts project Rambus could take 50% of the memory market by 2001.

Intel acknowledges that the market will ultimately decide Rambus' fate. "We still see [Rambus DRAM] as the best-performing [technology], but it will be up to the market to decide," says Intel spokesman Daniel Francisco.

And if the market decides Rambus doesn't make sense? "There are options on the landscape," is all Francisco would say.

>To order reprints of this article, click here: Reprints