To All,
Overcapacity won't last, fabless companies predict
By Craig Matsumoto
The need for fab capacity has cooled considerably after the
semiconductor industry's 1996 recession, but the glut isn't expected to last
long. EE Times takes a look at the worldwide capacity situation with a
report on the latest findings of the Fabless Semiconductor Association,
below, and sidebars from Eastern Europe and the bellwether Asian
region, on page 22.
---
As balance is restored to semiconductor supply and demand, fabless
companies expect modest, steady growth, according to the latest survey
from the Fabless Semiconductor Association (FSA; Dallas)
Speakers at FSA's annual luncheon predicted that foundries will account
for a growing percentage of the industry's output and that the fabless
model has proved it can withstand the next cycle of capacity shortages.
As the price of cutting-edge fabrication plants climbs above $1 billion,
startups are compelled to contract out their manufacturing to a host of
willing foundries, most of them in Asia. Many analysts have embraced the
"fabless model" as the future of the semiconductor industry.
FSA's primary metric is CMOS wafer demand, which it expects to climb
39 percent this year, to more than 2.3 million 6-inch wafers. Its survey
predicts 45 percent growth in 1998 and 36 percent in 1999.
Fabless revenues grew 10 percent in 1996, FSA's survey found,
compared with 61 percent in 1995 and 57 percent in 1994. Analysts
expect a return to better times, projecting revenue growth of 25 percent
this year and 27 percent in 1998.
Many companies suffered patiently last year as customers used up swelled
chip inventories. "We still think it's on the overcapacity side, but it's
moving to a more balanced environment," said Charles Boucher, an
analyst with investment bank UBS Securities (San Francisco)
Spending on capital equipment is down 14 percent, Boucher said, a trend
that should lead to a tightening of capacity beginning in 1998. That could
be the start of another capacity shortage, but probably not a severe one.
Boucher pointed out that 14 percent is small compared with the 30
percent corrections of the mid-'80s.
As the fabless model takes hold, said Dataquest analyst Clark Fuhs,
foundry business will boom. Pure-play foundries-manufacturers hired to
make other companies' chip designs-represent 5 percent of
semiconductor revenues, Fuhs said. He expects that to grow to 6.3
percent by 2000 and perhaps as high as 25 percent in 2010.
With that change comes an increased influence of fabless companies.
Historically, DRAM or microprocessor designs have driven manufacturing
innovations in the semiconductor world. But it appears system-level
integration, largely the province of the fabless sector, will be the next
source of technology drivers.
"We believe that the foundry model will drive technology from here going
forward," Fuhs said. "This is a major change."
Probably the biggest surprise in the 1996 figures was the acceptance of
0.35-micron process technology. "Migration to 0.35 is moving pretty
quickly, faster than some of us thought," said Robert Pepper, founding
chairman of the FSA and chief executive of Level One Communications.
Capacity for 0.35-micron manufacturing is "fully spoken for and will be
fully spoken for well into 1998," Boucher said. |
