Manufacturing At Root of 'Microprocessor Flu' Technology Isn't Really
The Problem
May 17th, 1999
Although most of the San
Francisco Bay Area is
basking in spring's balmy
weather, a nasty case of the
microprocessor flu has
stricken more than one
Silicon Valley chip company.
In recent weeks, National
Semiconductor announced
plans to amputate its PC chip
business, while Integrated
Device Technology went
public with its own pleas for
third-party assistance.
While IDT's troubles will affect only a small portion
of the market, National's role has definitely shifted.
For many, National will retreat further into its
traditional role as a 'widget' supplier, one of the top
manufacturers of analog and mixed-signal products,
but lacking the public attention that PC
microprocessors still command.
Symptoms Of A Bigger Problem
Nevertheless, the problems experienced by both
National and IDT seem like symptoms of a more
widespread condition: that manufacturing, not
innovation, seems to be winning the day in the PC
microprocessor market. National acquired Cyrix in
1996, after co-founder Jerry Rogers had assigned a
separate design team to work on an integrated
processor, later named the MediaGX.
Under National, the team designing a stand-alone
processor would transfer its finished core over to
the designers of the integrated chip, which would add
peripherals such as core logic or some basic graphics.
Since National intends to sell off the design team,
which created its stand-alone microprocessors, the
MII, the future of National's microprocessor
product road map as well as its integrated products
could be subject to dramatic change.
Most Likely Buyer: ST Microelectronics
The most likely buyer for the MII team and/or
design seems like ST Microelectronics, which has
historically served as a foundry and second-source
option for several chip company, including Cyrix.
ST's expertise in consumer electronics also leaves
open the possibility that ST could outdo National in
its integrated strategy. IBM Microelectronics, once
a rumored suitor, just doesn't seem interested.
Ironically, National is left with an integrated
product road map, which is anchored by the
"MediaPC," supposedly all of the functions of a PC
integrated onto a single chip. Since National is now
concentrating upon the embedded set-top box market
and not the PC, the chip as well as the 'MediaPC'
name will probably require an overhaul.
When I interviewed Steve
Tobak, vice-president of
corporate marketing for
National, he said the set-top
box market was, from
National's perspective,
"ours to lose."
The field may be bigger than
he thinks. Executives at Via
Technologies, a maker of PC
core logic chip sets, have
already expressed interest
in bolting on a
microprocessor core to their core logic. And if I
heard National chief Brian Halla correctly on a
conference call, Intel's "Timba" integrated chip will
appear sometime next year. Meanwhile, there's a host
of embedded chips from Motorola, ARM, and MIPS
that aren't going to just give in.
Meanwhile, IDT's own WinChip microprocessor has
been characterized by some analysts as little more
than an enhanced 486 core, with significant cache
additions to improve its performance. The
first-generation WinChip was relegated to the
low-cost overseas market because of its low
performance; the WinChip 2 is now sold in the United
States, but only to third-tier OEMs and
distributors.
Low Rung Of The PC MPU Market
National and IDT represent two of the three
companies on the lowest rung of the PC
microprocessor market. In the first quarter of 1999,
National's unit market share was under 6 percent,
while IDT slipped to a meager 0.6 percent of the
total market, according to Mercury Research, in
Scottsdale, Ariz.
The reason for this low market share is part of the
vicious cycle of manufacturing. Since both companies
have had problems manufacturing chips at high enough
speeds to win customers, demand has weakened and
the manufacturing sites or 'fabs' are not run at full
capacity. As sales dried up, so did the additional
funds needed to shore up the company's market
position. As a corollary effect, both IDT and Cyrix
have found it difficult to fight their way out of a
sub-$600 PC into a higher-cost product.
About a week ago, I sat down with a venture
capitalist who used to track Intel for a prominent
securities house. After describing his company's
investments into so-called "infrastructure plays,"
"intellectual-property licensing company like
Rambus," he asked me about the types of companies I
cover. After I ran down a list of PC microprocessor,
multimedia, and storage companies, he looked at me
incredulously.
"Geez," he said. "You're
covering a dying industry,
Mark." Maybe so. I
certainly wouldn't be the
first to claim that the PC is
dead; that's been evidenced
by the number of companies
scrambling to be included in
the market for set-top
boxes and portable
computing devices.
But that fact wasn't evident
two to three years ago when
National and IDT entered the PC microprocessor
market by acquiring Cyrix and Centaur Technology,
respectively. By chaining themselves to the
manufacturing treadmill, they let Intel -- "a
manufacturing superpower" -- to run them out of the
race.
"Believing that Intel would not and could not compete
in the low-end [PC] space was a mistake," Halla told
Wall Street analysts in a conference call. And
Intel's fabs are quite literally squeezing the life out
of its competition. National, IDT, and Advanced
Micro Devices all have one key fab -- or did.
(National also said they would sell a majority stake in
its South Portland, Maine production fab. Rise
Technology, the other low-end PC microprocessor
maker, still has not disclosed where its chips will be
manufactured.). Any mistake and production suffers.
MPU Manufacturing Costs: Seemingly Straightforward
On the surface, calculating the manufacturing cost of
an MPU is relatively straightforward. There is a
fixed manufacturing cost for processing a wafer, and
each wafer can produce a certain number of MPUs.
The number of chips per wafer depends on the chip's
size. Because of errors during wafer processing, only
a certain percentage of the dice will function, which
determines the "defect density" or "yield." In
reality, companies also factor amortized R&D and
other fixed costs into the manufacturing cost of a
chip.
Intel's Celeron is a good example of the premiums PC
microprocessors can still command. With 128
kilobytes of on-chip Level 2 cache, the Celeron
performs about as well as Intel's midrange Pentium
II. The price of a Celeron ranges from $187 for a
466-MHz chip to about $81 for the 333-MHz part,
in lots of 1,000. But even the lower price is far
higher than the cost to produce the processor. The
manufacturing cost of the Celeron is extremely low,
even when the MPU is being produced on Intel's
older 0.25-micron process, according to Mercury
Research. The cost is about $30 for each raw die,
and about $45 when test and packaging costs are
added.
Because the wafer cost remains about the same for
each MPU supplier -- approximately $3,250,
according to Dean McCarron, an analyst at Mercury
Research -- a chip's die size and the functional
yields play critical roles. Yields, in fact, determine
much of the price bands that separate one speed
grade from the next, McCarron said.
Given the fact that the Celeron's 154-sq.-mm die size
is public information, the only mystery is the number
of functional dice each wafer yields. Intel's yields
generally hover around 70 percent, while those of
AMD have slipped to about 40 percent, according to
industry sources and analysts.
This means that Intel's manufacturing cost per
Celeron is about $41, or $56 after packaging and
test -- still well below the sale price of even the
oldest Celerons.
As Intel moves to the more
advanced 0.18-micron
manufacturing process, the
Celeron's die size should
drop to about 80 sq. mm, or
roughly the same die size as
AMD's older K6-2 MPU,
McCarron said. The cost per
unpackaged Celeron die
could then fall to as low as
$13, even less than that of
the K6-2. "At that point, the
packaging cost is more than
what's needed to
manufacture the chip," McCarron said.
AMD has had an awful time converting from
0.35-micron processing equipment to 0.25-micron
lines, and yet it is Intel's toughest challenger in the
stand-alone PC microprocessors.
AMD Faces 0.18 Micron Conversion
Still, AMD again faces a conversion to 0.18-micron
equipment and its chairman and CEO, W.J. "Jerry"
Sanders III, has promised that 0.18-micron parts
will begin to be produced in its Fab 25 in Austin by
the third quarter. We'll see; AMD's been long on
promises and short on results.
According to Intel's customers, the company's
confidential road maps show that a 500-MHz Celeron
for low-cost PCs will be introduced in September.
Performance PCs will also take advantage of the
smaller die sizes afforded by the 0.18-micron
process. Intel's 600-MHz Coppermine MPU, also
scheduled for September introduction, will come with
256 KBs of on-chip cache. And a 667-MHz Pentium
III Xeon for the workstation market will be the
first Xeon available with integrated cache, also at
256 KBs.
Why Not Manufacture In Asia Instead Of
Maine?
Halla's argument in favor of exiting the PC market
and essentially selling its South Portland fab was
that third-party Asian foundries could provide the
manufacturing technology that National was reluctant
to invest in.
Let's hope so. If Intel designs an integrated
processor similar to National's, then price becomes
the only feature that distinguishes both products. At
that point, Intel can apply the same manufacturing
stranglehold it used to help drive National out of the
stand-alone PC microprocessor business. National
must take advantage of the nine months or so before
Intel enters the integrated processor market. If not,
National's second bout with the microprocessor flu
could be its last. |
