"You had some particular causal events that drove us to the situation of a year and a year and a half ago," Paul Otellini, general manager of the Intel Architecture Group, said in a recent interview. Struggles with Rambus problems and chip shortages were just some of the issues on Intel's plate at that time.
"We went back to basics on basic engineering disciplines, and we're now doing the stuff that for all but two years of our 31-year history we did very well."
Analysts agree, but caution that problems may still lurk. A 900MHz Xeon chip for multiprocessor servers, for instance, was pulled off the market in July because of a bug. The processor re-emerged last week.
A larger hurdle for the chip giant may lay in soft demand for processors and plummeting prices. Intel is scheduled to give its midquarter sales update to Wall Street analysts late Thursday.
"They still are having some problems, but they sound more like design problems rather than process (manufacturing) issues," said Kevin Krewell, an analyst with MDR/Instat. "So far it looks OK. Part of it, though, is the recession. When demand is slack, you don't push your yields. It is not the time to detect flaws in the system."
Avoiding manufacturing problems will be crucial in both curbing competition from Advanced Micro Devices and fulfilling a fairly ambitious product strategy. Some of the company's upcoming projects:
Pentium 4 notebooks are expected by the first half of next year.
A new chip called "Banias," designed to dramatically reduce power consumption, is slated for 2003.
A new version of the XScale processor will appear in handheld computers early next year.
In servers, demo units of McKinley, the touted successor to the current Itanium chip, will come out at the end of this year.
As for desktops, the first desktop chips based on the 130-nanometer (0.13-micron) manufacturing process--which allows smaller features to be printed on chips--come out in the fourth quarter. These chips will feature copper, rather than aluminum, wires. Copper chips are already inside notebooks, but selling them into the desktop market will require Intel to produce these chips in far larger volumes.
"From an execution standpoint, they have pretty much resolved their problems. There's been a focus on it and some heads have rolled," said Joe Osha, an analyst at Merrill Lynch. "But the big question is pricing...Prices are just going to keep going down."
For the most part, Intel's lingering illness derived from two non-recurring issues, according to Otellini: Rambus memory (RDRAM) and the aging of the Pentium III design.
"There were two root causes. One was the difficulty in coming out with RDRAM systems," he said. "The other was with extending the life of the Pentium III. The Pentium III was at the end of its life...From 900MHz to the end of that one was very difficult."
The Rambus problem has been ameliorated through a variety of factors. In 1999, Rambus was slated to be the memory of choice for Intel desktop chips, including an upcoming chip for budget PCs called Timna.
The honeymoon didn't last long. Bugs in the 820 chipset, which paired the Pentium III with Rambus-based RDRAM, delayed the release of a number of desktops in late 1999. Around this time, AMD began to gain market share with its Athlon chips.
To get around the problem, Intel released a chipset that worked with standard memory and canceled projects such as Timna, which had a built-in Rambus controller.
Although the Pentium 4 still needs to be paired with RDRAM, chipsets that will let manufacturers couple the Pentium 4 with cheaper types of memory, such as SDRAM and DDR DRAM, are coming to market. These products will be hugely popular with PC makers, according to, among others, Dean McCarron, principal analyst of Mercury Research.
Meanwhile, RDRAM has continued to drop in price.
The problems with extending the life of the Pentium III were conquered by time and the arrival of the Pentium 4. The severe chip shortage of late 1999 began to ease up toward the middle of 2000, ironically just before the downturn in the economy.
"They ran into problems with Coppermine (the code name for the last generation of Pentium IIIs) because of sudden and rapid demand," Krewell said.
Even though sales began to slow, Intel still had to battle with AMD over performance bragging rights, a key element in chip marketing. The new chip allowed Intel to move past the 1GHz level and, eventually, firmly ahead of Athlon. The Pentium 4 now runs at 2GHz, while the Athlon tops out at 1.4GHz. In terms of performance on common computing tasks, the 2GHz Pentium 4 still wins out, but not by as wide a margin as the megahertz numbers might suggest, according to benchmark testers.
The company also reorganized its internal divisions and tightened up quality control. In the old organization, chipset engineers and processor engineers worked in separate divisions, Otellini said. Chipsets and processors, however, are interdependent: The chipset exists to fetch and deliver data to and from the microprocessor. Coordination between the product groups began to decline.
In 2000, the groups were realigned by product categories. Engineers designing chipsets and processors for notebooks were put in the same group, for instance, while server and desktop chip development were reorganized in a similar fashion.
"The execution got better because the chipset guys talked to the processor guys," he said. "You've seen us execute like clockwork on the Pentium 4."
Unexpectedly, reorganization led the way toward a form of technological cross-pollination. Energy-efficiency technology from the notebook group, for instance, is ending up in servers.
"There are architectural techniques that were invented out of need that are yielding circuit technology that can be used across the line," Otellini said. "That is a second-order effect that we are now just getting to touch."
So far, the company appears to have managed the transition from the 180-nanometer to the 130-nanometer manufacturing process. The new process, which involves shrinking the architectural elements of the chip, will lead to smaller, faster and ultimately cheaper chips. Next year, the Pentium 4 desktop is slated to hit 3.5GHz.
"The (130-nanometer) process is ramping like a hose," said Frank Spindler, vice president of Intel's mobile products group.
Most analysts contacted have generally agreed that the company's manufacturing issues have been put to rest. Still, demand is going to remain an open question.
"There will still be some potholes in the road, but it won't be anything on the magnitude of the past two years...The bigger issue is, when does the demand turn?" said Ashok Kumar, an analyst with U.S. Bancorp Piper Jaffray. "Economic cycles overturn manufacturing cycles."