Foundries may be heading for a
two-tier system
IDMs, fabless customers vie for same capacity
By Crista Souza
EBN
(05/03/02 16:02 p.m. EST)
As pure-play foundries get
more chummy than ever with
IDMs, will there be room on
the playground for fabless
customers when wafer
supplies become constrained?
The answer may become clear
by the second half of this
year when the two leading
foundries-Taiwan
Semiconductor Manufacturing
Co. Ltd. and United
Microelectronics Corp.-warn
they could run out of capacity
in the hottest process
technologies, 0.18-micron
and below.
Utilization rates at these
geometries rose to between
80% and 90% in the most
recent quarter, according to
TSMC.
For now, at least, fabless
companies aren't that
concerned. In the past, they
said, foundries remained
loyal to their core fabless
customers when the time
came to allocate wafers.
But even then, those with the
most clout with the foundry
stood a better chance of
having their needs met.
Now IDMs are engaging foundries in long-term
process R&D efforts in exchange for future
capacity at those advanced nodes. With the
stakes higher, some worry that foundries may
begin to shift their loyalties or simply be
stretched too thin.
“Fabless companies always have to fight hard for
capacity,” said Jodi Shelton, executive director of
the Fabless Semiconductor Association. “I think
a lot of people are worried about the leading
edge, especially the smaller companies who tend
to get the short end of the stick.”
IDM proliferation
Though Shelton said fewer fabless customers
were shut out during the 1999-2000 boom than
in previous capacity shortages, the increasing
IDM presence at foundries has the Dallas-based
FSA keeping close tabs on the situation. Foundry
use by IDMs has grown to the point that the
organization now includes IDMs in its annual
wafer demand survey.
“In the last cycle, a lot of IDMs were
outsourcing, but this time they're forming more
permanent relationships with foundries,” Shelton
said. “I think it will be tempting for the
foundries” to allocate more capacity to IDMs.
But they may not be ready to put full trust in
IDMs, which tend to be fair- weather customers.
“Foundries have been burned before,” said Gina
Gloski, vice president of worldwide
manufacturing operations at eSilicon Inc., a
fabless ASIC supplier based in Santa Clara, Calif.
“They realize their lifeblood is the fabless
customer, and they don't want to shoot
themselves in the foot.”
Nevertheless, IDMs are expected to surpass
fabless in foundry wafer demand within two
years, said analyst Joanne Itow of Semico
Research Corp., Phoenix. By 2004, Semico
projects IDMs will consume 48% of foundry
wafers, while fabless consumption will drop to
43%.
Itow isn't expecting to see a mad scramble for
capacity this year, but she does worry that the
growing influence of top foundries like TSMC
creates the perception there are no alternative
sources.
“The supply side needs to be more than just
TSMC and UMC,” she said. “There are several
other companies that will have 0.18-micron
capacity available. The problem is, everyone
gets drawn into the TSMC fold, and there's the
feeling that it's the only place they can get
leading-edge technology and service.”
TSMC said it is committed to an internal capacity
balance of 65% fabless, 34% IDMs, and 1%
system houses-a ratio the company has
maintained for more than five years, according
to a spokesman for TSMC North America Inc. in
San Jose.
While nudging customers toward newer
processes by closing down older lines, the
company is also trying to avert a severe
shortage at the leading edge by ramping
300mm-wafer processing lines this year and
upgrading its 8in.-wafer fabs to smaller
geometries to accommodate more chip designs
on a wafer.
Capex up
In March, the Hsinchu, Taiwan-based company
raised its capital spending budget by nearly $1
billion and plans to invest $2.65 billion this year
on new wafer facilities and process installations.
Fab 12, which produces 300mm wafers, is in
volume production and is being ramped further,
while Fab 14, also a 300mm fab, is currently
being equipped. Meanwhile, TSMC's 8in.-wafer
fab is continually being migrated to
accommodate process geometries smaller than
0.18-micron, the spokesman said.
UMC executives were unavailable for comment,
but the Hsinchu company recently announced it
will double its capital spending budget this year,
to $1.6 billion, to boost leading-edge capacity.
But until enough new capacity comes on line,
some fabless companies may have to figure out
how to design next-generation products using
mainstream or legacy technology, FSA's Shelton
said.
That's a scenario AMI Semiconductor Inc. is
hoping for. As a supplier of foundry services at
0.35-micron and older generations, the
Pocatello, Idaho, company is gradually adding
capacity to meet customer forecasts and expects
some customer fallout from the major foundries.
“We've been telling our fabless customers that
the world is changing, and they need to provide
better forecasts so we can be prepared for the
demand,” said Al Morrison, vice president of
foundry operations at AMIS.
The company also buys partially proc- essed
0.18-micron wafers from TSMC. “There's already
a capacity crunch there,” Morrison said, though
he's taken his own advice and given TSMC a
three-year demand forecast. “With our strong
relationship with TSMC, we're fairly secured of
them being able to meet our demand.”
Fabless companies said they have begun working
more closely with foundries to establish demand
forecast credibility. It's also wise to ensure
designs can be ported to multiple fabs in the
same foundry in the event a capacity crunch
causes dislocation, said Hugo Chan, president
and chief executive of fabless SRAM supplier
NanoAmp Solutions Inc., San Jose. “You can't
just assume they'll do it for you,” he said.
siliconstrategies.com |
