Intel is looking directly at many challenges ahead in the next 18 months,
according to industry analysts.
The company faces a shift in product lines, a change to
its manufacturing models and other assorted problems that it will need to address before the end of 2005.
Intel’s roadmap seems aggressive enough. This year alone, it has reported its ramping up
production of new desktop and mobile microprocessors in Prescott and Dothan, respectively. In addition,
the company is increasing production on a 90-nanometer (nm) process using 300mm wafers in 2004.
Intel has been enjoying some of its best sales
in several years. But the driving factor in the last three months was a stronger-than-expected demand for
its Flash memory chips
the company’s main microprocessor business were lower than expected.
CEO Craig Barrett announced that Intel would use the rest of the year to expand the company’s presence in chipsets,
Flash and other communications products. Barrett is even willing to splurge as much as $4.8 billion on R&D and
infuse approximately $4 billion into its capital spending coffers to make the necessary investments.
But analysts point to several of Intel’s setbacks. The company has seen leakage problems associated with
the continued development of its 90nm technology. Intel has also reported product delays of its mobile chipset
code-named Alviso. This follows the holdup of its “Dothan” Pentium M processor. Intel was also forced to recall
some of its Grantsdale desktop chipsets last month.
A new report filed today by In-Stat/MDR researcher Jim McGregor points to more than a few challenges ahead
for Intel’s product portfolio and manufacturing strategy.
“Intel will face a new capacity model starting in the [second half of 2005] when it begins shipping dual-core
processor solutions into desktop, mobile and server platforms,” McGregor said. “The switch to dual-core solutions
is being driven by increasing static leakage problems as the process technology transitions to 90-nanometer and beyond.”
With four 90nm fabs in production and two 65nm fabs ramping toward the end of 2005, McGregor said Intel also
has more capacity than it can use for just IA-32 and IA-64 processors. As a result, Intel said it will begin
shifting other products, such as chipsets, to the 90nm process.
“The new process increases the number of potential units four times over the older 0.13-micron process using
200mm wafers,” McGregor said. “Intel’s fab expansion strategy is also changing to favor the retooling of older
fabs over building new megafabs.”
Gordon Haff, Illuminata senior analyst and advisor, also put out a brief this week suggesting that Intel will
need to be more flexible in its new economy.
“It seems unlikely that Intel will ever attain the same level of direct influence on this new ecosystem that
it developed over time with its older, more computing-centric, one,” Haff said. “Intel would certainly like a strong
hand in setting the agenda for ubiquitous mobile computing. However, it’s easy to overstate the influence that Intel
ever had in setting the agenda for how and where its processors were used.”
Overcoming the Obstacles
Intel is not alone when it comes to performance problems. The static leakage problem is inherent in the current
semiconductor manufacturing process. But McGregor said with billions upon billions invested in perfecting a mature
process, it is unlikely to change anytime soon.
“The best solution for now is to look for ways to reduce the power consumption of the semiconductors, such
as moving to multi-cores in microprocessors,” McGregor said. “By sharing the processing load between multiple
processor cores, each can run at a lower frequency but achieve a higher overall performance level.”
The shift to dual core is a strategy that both Intel and its chief rival AMD are pursuing.
The choice brings an end to the MHz/GHz marketing campaign that has dominated the minds of the PC consumers for
the past 20 years. It also raises questions about the future direction of the standard microprocessor.
“Certainly Intel and others are facing technological challenges brought on by shrinking process technology
and changes in design approach, such as moving to multi-core,” Haff told internetnews.com. “Intel’s broader
strategic challenge, however, is the inevitable erosion of its ‘I’ll buy the fastest chip you can make’ desktop
franchise in favor of a more fractured and probably lower-margin, distributed-client market such as handsets and
the like.”
John Enck, Gartner vice president and research director, also agrees that the trend is to move away from increasing
frequency and cache as a means of improving server performance, adding that it is an ecosystem problem in addition
to a processor problem.
“As it stands today, one of the biggest problems data centers face is deploying adequate power and cooling for
the current and planned generation of processor,” Enck said. “If AMD, IBM and Intel continue to push performance
by increasing frequency and cache, the power and thermal envelopes continue to expand and expand.”
Enck said he sees the move toward threading and multi-core processors as a way for manufacturers to deliver
overall improved performance without pushing the power and thermal envelopes beyond the capabilities of data centers.
“At the same time, we also see vendors investing in cooling technology (e.g. bottom-to-top air flows in racks
and even the return to both open and closed-loop water cooling) and in DC power solutions,” he said. “The final
factor to consider here is the death march toward server density. If the industry were to turn away from its never-ending
quest for greater server density and servers became less dense — thereby offering more space for cooling and taking
less per-square-foot power — the problem would also be minimized.”
Enck noted that this scenario isn’t very likely to happen, as the buying community has become all too conditioned
to buying small server packages.
As for the product delays and recalls, McGregor pointed out that no technical solution works the same in the lab
as it does on paper.
“Shifting to any new technology has inherent risks in terms of development time,” McGregor told
internetnews.com. “The delays that many are facing with the double transition to 90nm and 300mm wafers are
to be expected to some extent, but this is just a small speed bump in the hardware evolution. Although not as highly
publicized, delays in introducing new hardware technology and products are common, but not as drastic as some of
the delays the market has experienced by some software vendors.”
In related product news, Intel introduced four new processors designed for mini-notebooks, sub-notebooks and
tablet PCs that weigh around three pounds. The processors range in speed from 1.4GHz to 900MHz with as much as 2MB
of cache available.
Three low-voltage Pentium M “Dothan” processors — the 738, 733 and 723 — work in conjunction
with the Intel 855 chipset family and its PRO/Wireless network connection family. The new Celeron M ultra
low voltage processor is compatible with the Intel 855 chipset family, as well as the Intel 852GM chipset.
All four chips
are based on 90nm process technology.
