Is industry cutting back too far?
The latest capital spending cuts may lead to capacity shortages
By Jack Robertson
The massive size of the cuts made in this year's semiconductor capital spending during
the first half had stunned veteran managers and analysts. As much as $20 billion in
wafer-fab investments were deferred or eliminated. But since July, the picture has gotten
even worse. In fact, the budget slashing has accelerated to a point where people are
now beginning to worry about severe shortages in chip-making capacity developing
around the turn of the century.
"I've never seen such a dearth of new fab starts," commented analyst George Burns of
Strategic Marketing Associates in Santa Cruz, Calif. "The value of new fab starts this
year will be less than $15 billion -- half of the $30 billion in new starts two years ago."
Wafer fab systems suppliers are digging in for another drought of fab spending -- one
that many industry observers expect to continue for another 12 to 18 months as chip
makers try to find the right balance between supply and demand. Since July, the chip
industry has canceled, delayed, or closed down 19 wafer fabs, based on a tally by
Robert N. Castellano, president of The Information Network.
"The cancellations and postponements were huge [in early 1998], but things settled
down for a while before this next big wave of cuts," said the analyst based in New
Tripoli, Pa. "Back in January and February, the uncertainty revolved around a number
of factors, including whether fabs should be 200- or 300-mm, but those issues are gone
now and the overriding concern is whether there is any need for new capacity."
For many semiconductor companies, it became easier to delay or cancel future capital
projects than to close down existing facilities. But many analysts and some industry
managers believe that this trend is setting the stage for a major shortfall of
next-generation wafer-processing capacity when new lines are needed.
"The fab postponements and cancellations are now at about $38 billion -- meaning
plants that would have gone into production between 1999-2001," estimated
Jean-Philippe Dauvin, vice president and chief economist at ST Microelectronics in
Paris. "This is a major problem because it will certainly create undercapacity," he
warned.
The majority of fab closings that were announced in the past couple of months, mostly
by DRAM suppliers, employed older technologies, noted Dauvin, who is also the
president of the World Semiconductor Trade Statistics (WSTS) organization. "These
[fabs] were not leading- edge, and while I don't want to say these actions were
cosmetic, all the closings since August represented less than 5% of the market," he
noted. "But we need another 20% more [of production cutbacks]," he said. "Now,
finally, we are seeing serious moves to do that," Dauvin said, referring to the steps being
taken by Japanese and South Korean memory chip makers to scale back their current
production levels.
However, in a dramatic move made late in September, Philips Semiconductors and
Taiwan Semiconductor Manufacturing Co. (TSMC) took a contrarian approach to the
capacity crisis by saying they would start building a $1.2 billion joint-venture fab next
year that would open late in 2000 and be in full production in 2003. The fab will be
located in Singapore -- backyard of TSMC rival, Chartered Semiconductor
Manufacturing Pte. Ltd. In contrast, Chartered and two of its U.S. partners have
delayed the opening of two joint-venture fabs there until the chip business turns around.
"Timing is critical with major investments like this," commented Arthur van der Poel,
CEO of Philips Semiconductors in Eindhoven, The Netherlands. "Our projections show
that by the time this facility comes on-line late in 2000, the market for logic chips will be
strong."
Capital spending forecasts have fallen sharply every month this year. Worldwide capital
spending on equipment and plant construction is expected to plunge between 25% and
30% in 1998 compared to last year. VLSI Research Inc. ended up slashing its forecast
from -8% to -28% after chip companies began chopping their capital spending plans
and closing down fabs. Now it looks as if global capital spending will fall to $30.6 billion
this year compared to the $44.7 billion the industry spent in 1997, predicted analyst
Risto Puhakka, director of chip-making markets at the San Jose market researcher.
The drop is so big, in fact, that he estimated that investments in chip manufacturing will
not get back to the 1997 level until around 2001, the VLSI Research analyst predicted.
Chip makers will continue to make only selective investments in existing plants until they
grow more confident in the future.
"If you do a one-generation CD [critical-dimension] shrink, the investment cost is about
$300 million to $500 million," Puhakka estimated. "That investment essentially doubles
your capacity in terms of die shipments and the cost is much lower than building a new
fab, which last year averaged $1.4 billion."
Not only can fabs print more die on a 200-mm wafer with device shrinks, but they also
are able to get much higher yields from new technology much faster than they could in
the past, noted Bill Bottoms, CEO of test equipment supplier Credence Systems Corp.
in Fremont, Calif. "In the past, they would have a yield loss with a design shrink and
they would have to work for 6-to-18 months on the learning curve to get acceptable
yields," he said. "But chemical mechanical polishing [CMP] has made planarization
possible for each lithography step, and that has immediately increased the yields from
shrinks."
Ironically, the drive to shrink die sizes in order to cut costs also is pumping a lot more
chips into the marketplace. And this higher production is driving the need to reduce
capacity, according to market observers. And that trend is leading to a new round of
cutbacks in spending plans for 1999.
"Chip companies are now overcorrecting," said analyst Bill McClean, president of IC
Insights Inc. in Scottsdale, Ariz. "Everyone has taken on a death march mentality [in
terms of capital spending]," he said. "Hitachi, for example, has announced 'zero
spending' on semiconductor capacity in the second half of the fiscal year [ending March
31, 1999]."
And like other observers, McClean believes that these cutbacks will probably come
back to bite the industry in 2000. "We saw the same situation in 1992," he recalled,
referring to the shortage of new wafer-processing capacity that preceeded the
1994-1995 boom.
Underscoring the bleak outlook for fabs was United Microelectronics Corp.'s decision
in mid-September to freeze its plans to spend $14.5 billion on a half dozen new chip
plants in Taiwan over the next 10 years. UMC is continuing the construction of its
newest Fab 5, but the silicon foundry is unsure at this point whether it will equip the new
plant with production gear once the shell is completed next year or wait for the market
to improve.
Instead of building so many new fabs, UMC is now looking at acquiring existing plants
outside of Taiwan. "We are looking at two fabs in the U.S., one in Europe, and one in
Japan," stated Alex Hinnawi, assistant to the UMC chairman. "UMC won't buy all four
-- that would be too much to absorb all at once. However," he said, "we think it is a
good opportunity to expand quickly by taking over an existing leading-edge fab to give
us the capacity we need."
Another major cutback in expansion was just made by Motorola Inc. Its Semiconductor
Products Sector is postponing for a second time the construction of a $3 billion megafab
complex at West Creek, Va., near Richmond. The facility was scheduled to begin
production in the middle of 2000.
Motorola insisted that it is ready to restart the project once it sees clear signs of a solid
recovery in the chip markets. "We still intend to build a site that's on the same magnitude
as announced last December, and it will be the third major hub for [Motorola]
manufacturing and R&D in North America," said Sean Hunkler, director of the West
Creek site. "The only thing that's changed is the timing."
Timing is what worries industry observers most about the construction delays. Most of
the canceled projects were expected to begin production in a year and a half to two
years, noted Burns of Strategic Marketing Associates. "That's precisely [when] many
forecasters call for a strong upturn in chip sales," he said. "If the upswing is as big as
predicted, we should be breaking ground right now on new fabs to have the new
capacity ready at that time."
"But with the present extremely low level of new fab starts," Burns was reasonably sure
that "we could end up with shortages." That in turn could launch the next
boom-and-bust cycle. Shortages in 2000, he believed, "could touch off a big new round
of fab starts leading to another chip oversupply." -- Additional reporting by J. Robert
Lineback
semibiznews.com |
