The foundry business is coming back
faster than expected
"But demand is changing so fast that it's nearly impossible
to track what's going on; one foundry executive calls it "a
case of market whiplash"
By James B. Brinton
Forecasting market demand never is an easy task. But the job
was never harder than it is now in the pure-play foundry
business. This rapidly growing part of the chip market is now
emerging much faster than expected from the worst slump in its
short history. Demand for these services is changing so fast that
market watchers are now finding it nearly impossible to track
either the degree or the kinds of changes that are influencing this
sector. It's the sort of change that burns out managers and can't
even be quantified by market analysts.
As late as two months ago, nearly everyone was talking about a
foundry capacity glut. But that glut is now being wiped out by an
unexpected surge in demand and big changes in demand patterns.
This upswing is coming hard on the heels of a severe business
contraction that had wiped out the bulk of capital spending plans.
Foundry capacity now appears roughly equal to demand, but
forecasting either one is proving to be extremely difficult. To the
experts, the situation is both confused and contradictory, holding
out both the possibility of a capacity shortfall and the hope of
unprecedented growth.
The good news is that the plunge in
demand which forced some foundries
to run at below 50% utilization rates last
year is history. There's no doubt about
that. "Last year was bad," confirms one
bruised industry executive. "At one
point, we were at 48% utilization.
Thank God the market's back."
Foundry utilization rates last year
averaged less than 70% with overcapacity exceeding 40% at
some companies, according to industry analyst Jim Hines at
Dataquest. But the picture began to change in the fourth quarter,
he says. Now the figures seem to rise almost hourly.
Hines reckons that industry-wide overcapacity in the first quarter
dropped to an average of about 20%, and that by the middle of
the second quarter demand should be running far more than
foundry capacity. The turnaround has been almost instantaneous,
as these things go. One bemused foundry industry executive calls
it "a case of market whiplash."
Foundry customers are noticing it too. "Wafer demand clearly
has shot up in the last three months," notes Dennis Segers, senior
vice president of FPGA development at Xilinx Inc., the San
Jose-based programmable logic supplier. "Two factors appear to
be at work," he says. "A recovery in the overall semiconductor
market, and the success of pure-play foundries . . . in attracting
business from [the integrated device manufacturers]."
Whatever happened, "it happened fast!" declares Jim Ballingall,
vice president of worldwide marketing for the UMC Group
(USA) in Sunnyvale, Calif. "We got the expected orders for the
Christmas and Chinese New Year's seasons, and frankly,
expected a fall off in demand afterward. But instead," he says,
"demand stayed high."
Now UMC is booked solidly through mid-June, and the upward
trend is expected to continue, Ballingall predicts. As late as last
January, UMC was projecting that it would turn out about 2
million 8-inch wafers in 2000. Now, due to the turnaround, its
model predicts that between 2.3- and 2.5-million wafers will be
produced at the Taiwan company that year.
The picture is much the same at Chartered Semiconductor
Manufacturing Ltd., in Singapore. "We're running near capacity
now, and we expect to be at or near capacity for the foreseeable
future," says Kevin Meyer, vice president for worldwide
marketing.
"We saw the first signs of the turnaround [late in the third quarter
of 1998]," claims Meyer. "Because our customer base consists
mostly of telecom companies [and a 'leading-indicator' market],
our business probably picked up earlier than [it did for] most
others. And it picked up fast," he adds.
In fact, orders turned up so fast that Chartered is increasing its
production capacity. So are UMC and Taiwan Semiconductor
Manufacturing Co. Ltd. (TSMC), the other two of the "big three"
foundries.
"The business has changed dramatically since January first,"
confirms Magnus Ryde, president of TSMC USA in Sunnyvale,
Calif. And he's happy about it. "We went through 1998 at a
utilization rate of about 73%, and though we were profitable,
that's not a place you want to be."
Like the other big foundries, TSMC's business kept increasing in
January instead of going into the seasonal slump that usually
follows the holiday season. In fact, by February, orders were
rising so fast that they were running was well beyond TSMC's
internal forecasts, Ryde notes.
Today, TSMC's fabs are running at
utilization rates that average in the "high
90% range," Ryde says. In other words,
the foundry leader is close to capacity,
with only a little flexibility still remaining.
Capacity is becoming an issue. TSMC
originally planned to produce the
equivalent of about 1.6 million 8-inch
wafers in 1999. Now, says Ryde, "we
will surely exceed 1.8 million, and may top 1.9 million." The
situation changed very quickly. "We thought as late as
mid-January that we had the scheduled capacity necessary for
1999," he notes, "but since then we've enlarged our expansion
plans significantly."
It's not only the big three that are profiting - this tide is raising all
boats. Smaller foundries, such as Great Britain's Newport Wafer
Fab Ltd. and Israel's Tower Semiconductor Ltd., say they are
pulling in the new business and bringing idled capacity back on
line.
"Business is good," says Steve Della Rocchetta, Newport
executive vice president in its San Jose office. "In our case, it's
based on orders from several American and European IDMs.
Suddenly," he says, "we found ourselves doing whole product
lines for these companies. As a result, our utilization is ramping
upward at 6% to 10% per week. Projecting that [out], we expect
full utilization in several months."
The question that's now on the minds of industry managers is
just the opposite of what was worrying them last fall. With every
pure-play foundry becoming an optimist overnight, and with all of
them running at or heading toward full utilization, they now
wonder whether they are heading instead for a capacity shortfall.
Opinions are mixed, of course, and the statistics that usually
buttress such projections are moving targets, according to key
industry analysts such as Joanne Itow of Semico Research in
Phoenix. She is now convinced that the claims of fab executives
that they are running at high capacities are accurate.
"We assembled our last capacity/demand projection in
November," Itow says, "and since then things have moved so
fast and so far that I no longer trust it."
But even that 1998 report showed the possibility of an early
wafer shortfall. It all depended on how much of the IDMs'
production would move to the foundry sector. Itow's November
projection indicated a shortfall could hit foundries by January
2000 if the big integrated chip makers placed 10% of their
production with foundries this year.
That production shortfall, however, could arrive a lot sooner,
even though every member of the foundry community has begun
adding capacity. "I think there is reserve capacity in the industry,"
allows Itow, "but capital investment over the past year was
constrained. Even though some new plant is under construction
and some cleanroom space already exists, it's a question as to
how soon this capacity can become available."
There's the rub. Dataquest's Hines agrees that is hard to judge the
present ratio of demand to capacity. Like Itow, he finds it
difficult to make projections now because of the rapid pace at
which the market is changing. But his hunch is that the foundry
segment - with enough additional capacity brought on line - could
be heading for a period of approximate supply-demand balance.
Maybe. Taiwan's chip industry, most of which is pure-play
foundries, was expected to pour $4.5 billion into capital
investment this year. But that estimate came from a January
forecast and the real figure is probably much higher now,
observers say.
Foundry managers also are saying all the right things now about
adding capacity. Chartered's Meyer, for example, says that three
of his company's five existing fabs, which includes joint venture
fabs with Lucent Technologies Inc. and Hewlett-Packard Co., are
being "aggressively ramped up today."
The Singapore foundry also is able to add capacity much faster.
In recent years, it has been building plant and "facilitizing" only
about half of the potential capacity, eliminating the time it would
take to build clean-rooms from the ground up. Chartered's
"semi-equipped" plant is running at "about 600,000 eight-inch
wafers per year today," Meyer says. "During the coming three
years, we will be at least doubling that." That would give
Chartered a capacity of something like 1.2- to 1.3-million wafers
by about 2001.
TSMC also is adding a lot of capacity
fast. It has a fab under construction in
Singapore (with Royal Philips Electronics
NV as a partner) that's scheduled to come
on line at the end of 2000 with a capacity
of 40,000 eight-inch wafers per month.
Plans for another new plant, Fab 6 in
Taiwan, have been accelerated by four
months and should now be coming on line
by the fourth quarter, reaching full capacity of 40,000 wafers
monthly by mid-2000. The capacity of TSMC's Fab 5 in Taiwan
is being roughly doubled to 28,500 wafers a month and is
expected to be running at that capacity before the end of this
year. Its Fab 4 in Taiwan, which now has cleanroom space
available, is already producing about 32,000 eight-inch wafers a
month but is not currently being ramped higher, according to the
company.
TSMC's Wafertek plant in Oregon is being ramped up now as
well and should be turning out 20,000 wafers a month by the end
of 1999.
UMC is also adding new plant. It took advantage of the lull in
wafer demand to acquire fabs at low prices last year, which
ended up improving its competitive position and capacity without
expanding industry capacity. When Taiwan's Holtek
Microelectronics Corp. went fabless, UMC bought its fab. It also
acquired Japan's Nippon Foundry Inc. from Nippon Steel.
In late April, UMC bought Oak Technology Inc.'s 9.3% stake in
United Integrated Circuit Corp. (UICC) for $51.2 million. Last
October, ESS Technology Inc. sold its shares to UMC in the joint
venture fab, which was originally set up between the Taiwan
foundry and seven U.S. chip companies.
The Taiwan foundry also is building new fabs. Under
construction in Taiwan is United Silicon Inc.'s 8-inch wafer fab
and its Fab 5. "We built the shell last year," notes Ballingall, "and
we're moving in the production equipment now." If all goes well,
Fab 5 could be producing 40,000 eight-inch wafers a month by
November, he says.
The smaller foundries also are beefing up their capacity. Even
though it was having financial problems last year, Newport was
able to finish its Fab 3, an 8-inch, 0.35-micron facility in
Newport, Wales.
With all this new foundry capacity being added now, the orders
are flooding into the semiconductor equipment manufacturers.
"March was our best sales month in the past 12," exudes Jack
Ghiselli, president of GWAssociates Inc., a Sunnyvale, Calif.,
company which supplies much of the factory automation
software to the foundry industry.
But the danger now is that if their order books suddenly fill up,
equipment vendors may not be able to supply production gear as
quickly as the foundries want it. There are sectors of the
capital-equipment industry that are unable to turn on a dime,
reminds one executive, and lithography is one of them.
A big reason is that those vendors are going to have to crank up
their own production lines. "Over the past couple of years, for
example, Nikon was forced to cut everything to the bone," notes
one chip executive. "They cut annual capacity from 2,000 units
to 800."
Another major lithography house has cut capacity by two-thirds
over the past 24 months, adds this executive. "Their ability to
produce is way down. So is their manpower, and their supply
chain isn't in good shape either. You can't just turn these
companies back on by flipping a switch."
Another industry observer agrees that while fab shells may
already be built, there may be less production gear available to fill
them than some chip executives are counting on. "Lithography
could be a bottleneck," she says. "Lead times on lenses are 12 to
24 months, so don't expect overnight delivery."
With all these imponderables, it makes it tough for anyone to
decide whether the additional capacity - if it materializes - will be
enough to meet the growing demand expected over the next year
or two. According managers closest to the problem, such as
TSMC's Ryde, the answer is that "nobody knows yet." It's just
too early to tell.
In Ryde's opinion, the foundry business is in its earliest stage of
the recovery. And the metrics are changing so quickly now that
all the industry and analysts can do now is monitor the situation
closely until the trends become clearer. To do that, they will have
to get a better handle on what's driving the recovery. It appears
they are still gathering data, rather than forming a consensus.
Confusing the picture even more is that product, customer, and
technology mixes vary from foundry to foundry, shaping each
company's view a little differently.
Demand patterns are still unclear. One of the few certainties is
that demand seems to be broad-based. Wafer orders are coming
from across the board - from the fabless houses to the IDMs and
the DRAM makers.
As far as markets are concerned, UMC's Ballingall figures that the
wafer-demand drivers include the PC chip market, which is
proving to be stronger than anticipated. "Everyone expected
demand from the graphics companies to contract," he says.
"Instead, demand is solid and their prices have firmed up." And
demand for consumer-product ICs and networking
semiconductors is also strong at UMC, Ballingall adds. One hot
category now is in low-power parts for portable
communications.
What's paying off at Tower, it seems, is the Israeli company's
strategy of stressing embedded-memory-intensive technology, as
well as its CMOS image sensor technology. Business in both
areas is showing "significant increases in demand,"
Paraskevopoulos notes.
Another area of uncertainty is the impact of IDMs as they move
to foundries for more of their manufacturing needs. No one
seems equipped - or willing - to estimate just how much foundry
capacity they are soaking up at present. The consensus now is
that the impact on the foundry business will be a major one if the
big IDMs make good on their highly publicized intentions -
Motorola's plan, for example, to move half of its IC production
from its own lines to outside foundries in four years.
Memory is turning into a strong growth market for foundries.
Dataquest's Hines says major foundries such as TSMC and UMC
already are doing "a fair amount of DRAM work." Foundries like
the DRAM business because of its cookie cutter nature, large
volumes, and because DRAMs usually require their
highest-resolution production equipment. At this point, Hines
figures, DRAMs may be eating up 8% to 10% of total foundry
capacity.
The possibility that demand will exceed
foundry capacity sooner than many
people expected isn't worrying some
chip executives. Mike Hackworth,
chairman of Cirrus Logic Inc., believes
there is sufficient latent capacity in the
industry to permit the foundry industry
to raise its capacity by two to
two-and-a-half times between now and
mid-2000. Some of this would come from new foundries being
built, but mostly reflects the addition of production gear to
existing fab shells. In fact, Hackworth is so confident of this
latent capacity that his Fremont, Calif., company is now
renegotiating its joint-venture fab agreements with IBM Corp. and
Lucent in an attempt to move closer to a pure fabless model.
Cirrus wants to become more of a customer than a partner in
these two fabs.
Xilinx Corp. managers also don't seem worried, but for different
reasons. "While wafer production might become a problem for
some, it won't be for Xilinx," says Segers. "Xilinx has long-term
relationships and substantial equity investments with our foundry
partners to lock in capacity beyond our immediate needs."
Then there are folks like Goodloe Suttler, marketing vice
president for Analog Devices Inc., who doesn't see any shortage
of foundry capacity materializing yet. His Norwood, Mass.,
company does most of its foundry business with TSMC. "I'm not
that sure that fabs are running in the 90% range yet," he notes.
"And I don't see the price or the lead-time pressure that usually
goes with that. There's a little, but not an amount consistent with
90% utilization rates."
Those chip companies with the most at stake in the event of a
capacity shortfall, of course, would be the fabless firms - no
group is more dependent on foundries. But foundries managers
are all quick to describe the fabless firms as their market base and
that their interests would be given priority if overall demand ever
exceeded capacity.
Perhaps that's why the fabless companies don't seem concerned
now with any possible capacity shortage.
"Fabless firms aren't necessarily worried, but they want to be
sure that foundry capacity is a recognized issue," comments
David Angel, chairman of the Fabless Semiconductor
Association. "We know that capacity crunches can take place,
and hopefully, we have some time to work the system to make
sure that we avoid such a situation."
"Communication is very important at times like these," adds Jodi
Shelton, the FSA's executive director, based in Dallas. "Chip
companies can't always predict their production needs any more
accurately than the foundries can predict demand;" she says. This
was certainly borne out by the events of the past three months.
To get around this problem, the fabless firms are following
conservative strategies. "Some chip companies over- or
under-order; some foundries over- or under-expand," Shelton
says. "Hopefully, if we can be more open and accurate about
capacity, demand, and timing, we can avoid a shortfall."
That may prove to be a difficult strategy to carry out. Even the
FSA acknowledges it.
"Nobody really knows what the probabilities are now for a
wafer-capacity shortfall," comments FSA chairman Angel. Not
only is the basic situation fluid, but "many companies have been
reevaluating their business models and market strategies over the
past two years," he points out. "You don't know where some
companies are going to be coming from in the future and that
multiplies the uncertainties."
|
