Semiconductor Business News summary for week of Dec. 1.................
techweb.cmp.com
A service of Semiconductor Business News, CMP Media Inc.
Story posted at 8 p.m. EDT/5 p.m. PDT, 11/26/97
ElectronicsAlert!
Weekly News Analysis for Senior Industry Executives
Monday, Dec. 1, 1997
Vol. 6, No. 25
c 1997, CMP Media Inc.
By Robert Henkel
The following is a summary and analysis of stories appearing in the Dec. 1
editions of Electronic Buyers' News and EE Times.
Taiwan ahead in 300-mm
The U.S. isn't getting off to a very fast start in the 300-mm race. The latest
score is U.S. 1, Japan 3, and Taiwan 6, according to Semiconductor
Business News. That's based on announced plans to build production fabs
and was tabulated by the Global 300-mm Initiative. Not every
announcement makes it. Some are pulled because I300I director George
Lee feels uneasy about a plan's "firmness." On his latest list is Texas
Instruments, Mitsubishi, Hitachi, Fujitsu; and the Taiwan gang-Powerchip,
TSMC, UMC, Macronix, Mosel-Vitelic, and World Semiconductor Mfg. If
these plans hold up, watch out!
Bells toll for Koreans
The falling won should cut part of the $2.5 billion loss that South Korean
DRAM makers are expected to incur this year. But over the longer term
Korea's economic crisis will only add to the woes of its chip industry and its
worsening competitive position, Andrew MacLellan and Jack Robertson
write in EBN.
An even bigger problem is that the Koreans "didn't invest very aggressively
over the past two years in quarter-micron technology," says Dan Hutcheson
of VLSI Research. Korea's Big Three IC makers will account for only 7%
of global capital equipment investments this year, down from 20% in '95,
says VLSI Research. But Taiwan will account for 15% and the U.S. 32%.
"From everything we've seen, the Korean chip industry probably isn't going
to recover for 7-to-8 years," he predicts.
16 months of growth ending?
The U.S. electronics industry is still perking right along. That's the word from
purchasers polled by EBN for its monthly Quest index. Quest indicated the
industry grew in November for the 16th consecutive month. The composite
index came in at 55.2 (a reading higher than 50 signals growth), Ismini
Scouras reports in EBN.
But the index dropped 2.6 points from October, a 30-month high, and off
0.2 points from a year ago. All the indices making up the composite index
fell from the previous month. But even though the November index was the
lowest point of the year, one month of data is no trend. Growth in
production and new orders, which had been driving growth for the past 16
months, slowed in November.
Copper R&D is smoking
R&D aimed at replacing aluminum interconnects with copper in the next
generation of ICs is moving faster and across a broader front than was
originally thought. Just two months after IBM disclosed it would use copper
in its upcoming 0.2-micron CMOS process, the giant already was quietly
using this technology to turn out samples of an unannounced 32-bit RISC
microprocessor--a souped-up PowerPC 750, Rob Lineback writes in
Semiconductor Business News.
Turns out IBM didn't have the lead observers believed it had. Motorola was
quick to enter the race by saying it had developed copper interconnects for
a new 0.2-micron process. Both chip makers will market products next year
with copper interconnects.
Grim outlook for Japanese
Japan's chip makers got troubles. They're not only getting walloped by the
dive in DRAM prices, but the outlook is made even grimmer by a sluggish
home market and rising overcapacity due to the growth of South Korean
and Taiwanese competitors. The Japanese are trying to cope with the red
ink from DRAMs by shifting production offshore and make alliances with
Taiwan foundries.
But even worse, the Japanese are falling behind in the capital investment
race, Jack Robertson writes in Semiconductor Business News. Its Taiwan
and U.S. competitors are spending more now. As Japan's IC makers fall
behind, they will lose global market share, predicts Robertson Stephens'
Susan Billat.
Wall Street still worries about TI
Financial analysts like Texas Instruments' aggressive push into DSP, but they
want to hear more about the rest of the company. The non-DSP side of TI
still accounts for 60% of its $7 billion IC business, but it doesn't talk much
about it, Rob Lineback writes in SBN.
Analysts say they're frustrated over the dearth of details from TI on chips
falling outside its strategic focus on DSP solutions. TI has "gone to the
extreme in touting DSPs," says Brown Bros. Harriman's William J. Milton,
Jr. "There is a perception that TI is not telling the full story," he adds. "They
may be hurting themselves." Analysts also don't understand TI's joint-venture
model. It has never "been very specific" about how it works, Milton adds.
Fairchild buys Raytheon Semiconductor
This past week Kirk Pond did what he said he was going to do. The
Fairchild Semi CEO had told EE Times in August he was looking for an
analog IC business that fit his market focus. He also wanted to grow the
Maine chip maker faster.
On Wednesday, he came up with $120M in cash to buy Raytheon's
semiconductor business and complete his portfolio of multimarket products,
Ismini Scouras reports in EBN. Raytheon turns out analog and mixed-signal
ICs and runs an R&D lab in San Diego and a 4-in. fab in Mountain View.
Doing business at $76M annually, Raytheon could add $100M to Fairchild
revenues next year. Fairchild is now going at an $800M run rate
Japanese lithography advanced
Japan seems to be making surprising progress in advanced lithography.
Hitachi says it has tripled the writing speed for a direct-write, electron-beam
lithography system for volume 0.18-micron production, Yoshiko Hara and
David Lammers report in EE Times.
Throughput is 10 eight-inch wafers an hour when the E-beam system writes
all layers. E-beam lithography is gaining favor among Japanese chip makers
for producing the critical layers in new chips. Advantest is now developing a
twin-head, E-beam system for commercial chip making. NEC may use
E-beam litho for several mask layers of 1-gigabit DRAMs before going to
X-ray for critical layers of 4-gigabit DRAMs.
Out of the firing pan...
With the DRAM business drowning in a tidal wave of red ink, what's a
DRAM maker to do? In Taiwan, the wafer-foundry capital of the world, the
answer seems simple. DRAM makers are running not walking to the foundry
biz where margins can still be obscene. Latest to join the rush is Acer, which
next year will break ground on a $1.3B fab to turn out commodity logic ICs,
Mark LaPedus writes in EBN.
It hopes to buy 6-in.-equipment from its joint venture with TI, which recently
shut down its 6-inch lines. While Acer believes there's still money to be
made in fabbing low-end logic parts in a 6-inch foundry, one analyst
disagrees. Acer is not only late, says ING Barings' Donald Floyd, but faces
6-inch overcapacity. Other Taiwan chip makers are expected to follow Acer
and Nan Ya Technology, another DRAM maker to offer foundry services.
Litho vendors aren't happy
Art Zafiropoulo, Ultratech Stepper CEO, isn't a happy camper. He wants to
be part of Intel & Co.'s new $250 million Extreme Ultraviolet R&D
program, but doesn't like the terms and conditions. He supplied the litho
platform for the original EUV demo two years ago, but says he's
"disappointed" by his talks with EUV Corp. in joining the follow-on
program. Intel wants full data rights to any lithography platform and this,
Zafiropoulo says, "isn't very agreeable." Other litho vendors, including
Nikon, Canon, and ASML, also have balked at the EUV project's terms,
Jack Robertson writes in Semiconductor Business News.
No surprises in IC roadmap
There were no real surprises to the '97 update of the chip industry's
technology roadmap, Rob Lineback writes in Semiconductor Business
News. For the most part the latest roadmap, put together by the SIA, largely
resembles the draft that emerged last summer that shows the U.S. chip
industry blowing past previous set targets a couple of years ahead of
schedule.
But the new forecast for the first time maps diverging courses for DRAMs
and MPUs over the next 14 years. Both are designated drivers, with
memory focused mainly on the shrinking of features on a die and MPUs
pushing performance with smaller transistor gate lengths and additional layers
of interconnect. Until now, DRAMs were regarded as the driver for the
entire chip industry.
Flextronics hits big time
CEO Michael Marks thinks acquisitions are a bad way to grow a business,
but it hasn't hurt Flextronics, a contract electronics manufacturer that made
seven acquisitions in the past three years, Ismini Scouras writes in EBN.
The CEM will do nearly $1B in sales this year. If it hadn't gone on its
acquisition binge, Flextronics could not have become a major player in the
industry, he says. The firm has made only 2-1/2 mistakes in its big
acquisition drive, Marks says. It paid too much for nChip, it bought a plant
in Wales ("the wrong location"), and made a "languid start in building a
global sales force," he says. But that hasn't stopped Flextronics from
growing 100% annually since '91.
Ambitious Internet plan
Some 250 telecom carriers and regulators from 175 countries will meet next
Sunday in Vegas for the first official technical meeting on a highly ambitious
and costly plan to deploy an alternative "super-Internet" in the next 3-to-6
years. Promoters claim the global fiber-optic network could erase the
boundaries between Internet and traditional telecom, allow true connectivity
from anywhere in the world, and shift the telecom profit model from voice
service to data and video.
But critics question where the money will come from, not just for building the
network but for maintaining it, Larry Lange writes in EE Times. Jersey
telecom startup CTR Group has issued a set of specs and an invitation to bid
on the proposed network under an initiative called Project Oxygen.
Estimated cost is a cool $14 billion.
Good news from Commerce
The U.S. computer industry will grow 11% to $111B next year, rising to
$17B in '02, the Feds predicted this past week. PC unit shipments will rise
at the annual rate of 15-to-20% to hit 60 million PCs by '02, predicts the
Commerce Dept. PC revenue will grow to $100B by 2000, Jack Robertson
writes in EBN.
U.S. firms make 67% of the world's desktop PCs, 60% of the mainframes,
60% of the peripherals, 63% of the network devices, 87% of disk drives.
and more than 88% of the workstations. In chips, sales of U.S. vendors are
predicted to grow 9% to $89.7B next year.
Supercomputers aren't dying
What ever happened to the supercomputer industry? Business tailed off after
the cold war and leaders like Cray Research and Convex were acquired.
Surprise! The industry is regaining some of its luster and many companies
are now trying hard to get a piece of what was a fairly closed market, Terry
Costlow writes in EE Times.
Both Compaq and Sun had big exhibits at SC '97. While the market is far
smaller than servers, it's still a $3 billion business. More jobs are needing the
performance that once required a supercomputer. Improved parallel
processing make it possible to scale up servers with 2-to-8 processors and
add CPUs to meet supercomputing needs. It's this scalability that makes it
possible to compete now in this market.
Europeans discover CEM
European OEMs were slow to embrace contract manufacturing, but when
they did all hell broke loose. North American contract electronics
manufacturers (CEMs) are taking advantage of this momentum and are
furiously adding new plants, Darrell Dunn writes in EBN.
"It's growing like crazy," declares James Savage, BT Alex Brown analyst.
Western Europe shortly will jump to the 2nd-largest CEM region in the
world, passing Japan and trailing only North America, says Technology
Forecasters. Europe's CEM sales will grow from $10.5 billion in '96 to
nearly $30 billion by '01. Central and Eastern Europe also seems ready to
jump on the CEM bandwagon. The bulk of CEM work in Europe originally
was done for North American OEMs, but now large local firms like
Ericsson, Nokia, and Siemens are becoming big players.
Boundry scan problems
"The semiconductor industry is doing a terrible job in conforming to the
boundary scan description language standard," charges Hewlett-Packard's
Ken Parker, acknowledged expert in boundary scan. As a result, HP is
making some back-end verification tools available on the Web for front-end
IC design, Stan Runyon reports in EE Times.
While some test experts say HP's tools are no panacea for these thorny
problems, most agree the move spotlighted an issue of rising importance as
chip and board level designs move to high density and greater complexity.
Reason for the problem, Parker adds, is a lack of tools that could identify
potential problems before a design is frozen.
To contact Bob Henkel, click on byline at top of this page, e-Mail to
bhenkel@cmp.com or bhenkel@aol.com, fax to (207) 454-8047, or
phone (207) 454-8121. |
