Wow! Lehman's Niles sees flat
to down chip market next year

Everyone, obviously, has written off this year for the global chip business. Some analysts even call it the worst decline ever for the industry. So how can it get any worse? Easy. Up until now, nearly everyone was forecasting a mild to strong recovery next year. But maybe the chip business might keep falling. Now at least one expert seems to think so.

Dan Niles of Lehman Bros. shocked his audience this week at a San Jose manufacturing conference by predicting that the worldwide chip business in 2002 would run "flat to 10% down" from this year's dismal results. And he expects the IC market this year to decline 31% from 2000--"and it could be closer to 35%," he adds.

On top of that, the analyst figures it could take chip makers and chip-equipment companies three years to rebound and fully recover from the impact of the current chip recession. Globally, the picture is still "fairly bad" in the IC industry, he says. With inventories slowly getting under control, "the biggest problem is demand--not inventories--in the channels."

While the global IC industry in the first half of 2002 looks positively awful, Niles expects the business to rebound in the third and fourth quarters. "We are not looking for a 'V' shaped recovery," he says, "we're looking at a 'bathtub' shape recovery."

The analyst bases his chip forecast on PC shipments growing only 7% in 2002. He also expects the telecom sector to remain depressed. Even more troubling to Niles is that it could take years before some IC and chip-equipment companies to recover fully from the current downturn.

(See Oct. 10 story.)

Chang is accelerating
TSMC's 0.10-micron effort

He's doing it again. In the 30-odd years I've known Morris Chang, chairman of Taiwan Semiconductor Manufacturing, I've always considered him a risk taker. To make it big in this business, you have to take chances and stick your neck out. Now Morris is doing just that. Despite being caught up in the worst downturn in chip industry history, TSMC is speeding up the development of a new generation of chip processing even before the silicon foundry is finished ramping up the previous process generation.

Morris says the foundry giant is accelerating the development of its 0.10-micron efforts to handle system-on-a-chip (SoC) designs. The Hsinchu-based company--which is still ramping up its 0.13-micron designs--plans to offer its initial 0.10-micron silicon foundry services in the second quarter of 2002.

The aggressive company will accelerate the development of its 0.10-micron process "modules" at a new and record pace, Morris claims. Process modules include the intellectual-property (IP) cores, embedded memories, and processes technologies. "Our target for 0.10-micron is to make the modules available within about nine months of each other," he says. That compares with two years that it took TSMC to develop all the modules for its 0.18-micron technology and 13 months for its new, leading-edge 0.13-micron process.

(See Oct. 10 story.)

Does Intel have leg up
on 2006 IC packaging?

Intel did some long-range hustling over the past weekend, leaking a story to everyone from the New York Times to the Bugletown Gazette. The chip giant claims to be doing something that no one else, including a gang of companies from General Electric to NEC, has been able to do.

Intel's claims it has already developed chips based on its new packaging technology--called "bumpless build-Up layer" or BBUL--that could make a 20-gigahertz microprocessor possible in five years. The proprietary bumpless-bonding manufacturing process may be able to turn out 1-billion transistor MPU that's only 1-mm thick--or about the thickness of a dime, the firm says. But Intel doesn't plan to put such a chip on the market for at least five years.

GE and others have been trying to develop bumpless-bonding in chip packaging for military applications for years, but most everyone has regarded this technology as too costly for commercial markets. "It is an exciting technology, but nobody has been able to manufacture it," says Marcos Karnezos, CTO at ChipPac, the Santa Clara chip-packaging specialist. It has been difficult, if not impossible, to turn out these chips relatively inexpensively.

"It's going to be very difficult to develop a 20-gigahertz processor with today's packaging technology, but it's also not impossible," Karnezos says. "What Intel is talking about makes a lot of sense," he says. "The name of the game in chip-packaging is to make the paths shorter and shorter and with lower noise."

What makes it expensive is that the process builds up layers on the substrate on the die, using deposition or sputtering, photoresist, photolithography, and etching. But if anyone can, Intel should be able to do it. After all, just like IBM back in the good old days, it has a bunch of people working in R&D. The MPU giant has 900 employees alone working on IC packaging. While it doesn't need the technology anytime soon, Intel figures that "existing packaging technologies will run out of gas over time," says Koushik Banerjee, technical advisor for Intel's assembly technology group. "We can scale existing packaging technology" for the foreseeable future, he adds. This BBUL technology could pave the way for a new class of devices called "system-on-a-package" products. In contrast to current IC packages, which consist of three basic layers-a core layer of vias sandwiched between two copper interconnect layers--BBUL has two layers, the device and a copper interconnect. The silicon is embedded in the package core and only the lower interconnect layer needs to be built up, Intel explains.

(See Oct. 8 story.)

Moto's Q3 chip results
are no great shakes . . .

I don't know what deadline Motorola corporate gave its semiconductor unit to get back on track, but it doesn't look like the chip giant made much progress in the third quarter in its all-out attempts to improve performance.

Moto certainly didn't make many people happy this week, predicting its third straight quarterly loss and cutting a lot more jobs. But it did report some progress with its wireless handset business, which returned to the black in the third quarter and gained market share.

Overall, third-quarter revenues overall struggled to $7.4 billion, flat with the $7.5 billion in the second quarter, while net losses jumped from $153 million to $1.4 billion.

But the semiconductor outlook has grown darker. Instead of a recovery for the global industry later this year, Moto now doesn't see one coming until the first half of 2002. It now expects global chip sales to fall by 25%-to-30% this year, much weaker than its previous forecast of a 15%-to-20% drop. Next year isn't looking as good as it did a few weeks ago. Moto expects global chip sales to rise by 5%-to-10% in 2002, instead of the 15%-to-20% increase it had previously forecast.

Moto's semiconductor business continued to struggle in the third quarter, showing an operating loss of $355 million on sales that fell 14% sequentially from the second quarter. It did manage to slightly narrow its losses, racking up an operating loss of $381 million in the second quarter.

Motorola Semi revenues were $1.08 billion, a sequential decline of 14% from the second quarter and a fall of 48% from the $2.07 billion in the year-ago quarter. Orders were up slightly on a sequential basis to $1.1 billion, but fell 49% from those booked in the year-ago quarter.

(See Oct. 9 story.)

. . . As 4th quarter looks like
flat sales, another big loss

The bad news at Motorola wasn't confined to the just-ended quarter. The electronics giant is warning investors that it now expects to post a loss for the 4th-straight quarter on sales that will be only flat to 3% higher than the just-ended third quarter. Wall Street had expected a profit for the quarter.

As their problems deepened, Moto executives decided to cut another 7,000 jobs this year. That will make a total of 39,000 positions eliminated this year, or more than one-quarter of the company's total global workforce.

But these major moves to get the company back in line with current business conditions won't be enough, if business keeps getting worse. Moto says its new guidance assumes "no further deterioration in global political or economic conditions." I wouldn't take that bet now.

Motorola again lowered its forecast for cellular phone handset shipments, estimating that 380-to-400 million handsets will be shipped worldwide this year, down between 5% and 10% from last year. On the plus side, Moto says worldwide inventories are back to normal levels, with most regions having five-to-seven weeks of sales in backlog. Next year, Moto forecasts a global market of between 420-to-450 million handsets.

(See Oct. 10 story.)

Will Alcatel give Moto trouble
in one of its top IC markets?

One of the rare chip markets that Motorola clearly dominates is automotive electronics. But there may even be trouble brewing in that rare Moto bright spot.

Alcatel Microelectronics is gearing up to drive deeper into automotive electronics, which is still growing despite the industry's severe downturn. The chip subsidiary of Alcatel, the Paris-based telecom giant, is getting ready to launch its first standard automotive ICs--chips for in-vehicle networking and electrical motor control.

The fourth largest European chip maker, which has already built up a large presence in European vehicles using custom and semicustom ASIC designs, is mounting a major drive to break into the North American auto electronics market. It aims to use a wide variety of semiconductor technologies in its campaign, including its new 0.35-micron high-voltage CMOS process, ARM-based RISC cores, embedded flash memory, and EEPROM.

The Brussels chip supplier wants half of its automotive chip sales to come from North America by 2005, up from almost zero today. It already has a big chunk of the European automotive chip market, which should grow 53% this year. Its auto and peripheral chip business did about $91 million last year.

Alcatel Microelectronics already has begun doing business with Delphi Automotive Systems and one other major "Tier 1" supplier of automotive electronics in the U.S. "We are in discussions with Lear and we will be approaching TRW as well as all of the top suppliers for cars," declares Tony Denayer, who runs the company's peripherals and automotive business in Brussels.

(See Oct. 10 story.)

SVG's 193-nm tool
delayed once again . . .

I'm beginning to wonder just how ASM Lithography's Silicon Valley Group unit can keep from blowing it with its latest generation of lithography tools. It now looks like SVG has delayed the delivery of its 193-nm system by another two or three months.

Acquired by the Dutch company for $1.6 billion earlier this year, SVG at one time had been the prime 193-nm tool supplier for Intel. SVG had received a $100 million order for its Micrascan 193 line late in 1999 from a "leading microprocessor manufacturer" that sources identified as Intel.

SVG originally was scheduled to deliver the Micrascan 193 late last year, but it has yet to ship the tool to Intel. Product delays forced SVG to push back delivery of the systems several times. In July, after ASML confirmed the SVG system had been delayed again by another three to four months, shipments were set for October. Now, the tools will not be shipped until the end of this year at the earliest.

"We will ship limited quantities of the scanner by the end of the year," insists ASML CEO Doug Dunn. He acknowledges the scanner is late, but denies industry reports that his company has scrapped it after several delays and technical snags. "I confess that it's late," he says, but insists the SVG 193-nm tool project "is still ongoing."

(See Oct. 8 story.)

. . . But one analyst calls
SVG tool 'dead in the water'

Intel originally had hoped to use the new SVG 193-nm lithography tool in its 0.13-micron production process. But after the production delays, the chip giant decided to extend the life of its existing 248-nm tools from SVG and Japan's Nikon and use them instead. Intel officials insist that the tool delays did not impact the company's 0.13-micron product ramp.

Intel now plans to use 193-nm tools to handle critical layers in its upcoming 0.10-micron process, but is still evaluating tools from ASML, SVG, Canon, and Nikon, according to analyst Dan Hutcheson, president of VLSI Research. "The 193-nm production tool orders have not been cut by Intel yet," he says, predicting that "Intel will wait until the last minute before it places an order."

The veteran market researcher is pessimistic over whether the new SVG tool will ever go into wafer production at Intel. "ASML will ship it, because Intel is such an important customer, but I don't think ASML will develop the Micrascan 193 any further," he predicts. "It looks to me that the Micrascan 193 is dead in the water."

(See Oct. 8 story.)

Hynix now says it may
sell some of its fabs

Oh, the trials and tribulations of Hynix Semiconductor, South Korea's debt-ridden DRAM supplier. While some observers wonder just how much longer the company can keep going with its huge debt load, Hynix now acknowledges that it is "considering . . . the possible sale of any redundant fab to a variety of potential public or private buyers," as part of its financial self-rescue effort. But no decisions have been reached as yet, it notes.

Hynix also says it is also "considering possible strategic alliances with other semiconductor manufacturers." But it is not giving any details, other than to deny published reports it is negotiating to sell one of its 8-inch fabs to a Chinese consortium of government-owned and private chip makers.

Hynix has tried for years to sell an empty fab shell in Wales, but its latest comments indicate that it might sell an existing production fab. This fab is likely to be an older 8-inch fab, sources say, with the equipment removed and installed in a new facility outside of South Korea by any potential buyer.

Hynix, which has denied it is trying to sell its sole U.S. fab in Eugene, Ore., has closed that facility while it installs upgraded production gear that will process 0.13-micron wafers.

(See Oct. 8 story.)

Sagging foundry biz may be
showing modest rebound

Maybe Morris Chang got a signal to speed up his 0.10-process development from a firming foundry market. For the second month in a row, foundry revenues grew sequentially in September at Taiwan Semiconductor Manufacturing and United Microelectronics. This is evidence to some observers of a modest rebound going on in the trouble foundry business, a market which has been struggling through its worst downturn ever this year.

TSMC's September revenues were up 3.1% to $270 million, its third straight month of sequential growth. UMC, the world's second largest pure-play foundry after TSMC, reported sales grew 7.2% in September to #121 million, its second month of sequential growth.

But both foundries are still a long way from last year's revenue levels. TSMC's September revenues were off 43% from last year, while UMC's revenues for the month were down 60% from last year. For the first nine months, TSMC revenues were off 17.5% from the same period last year, while UMC reported its revenues down 31%.

(See Oct. 8 story.)

ASML adds Dongbu
to its customer base

ASM Lithography, which has a large installed base of chip production equipment at Samsung Electronics, has cracked another account in South Korea.

Foundry Dongbu Electronics has ordered ASML's i-line systems and step and scan scanners, a move that could threaten Japan's Nikon, which has been Dongbu's incumbent supplier of scanners.

The scanners will manufacture standard logic, analog, and embedded flash products at the foundry's EumSung facility, an 8-inch, 0.25-to-0.13-micron plant that has just started production. The company has two processes, one licensed from Japan's Toshiba and a so-called "TSMC-compatible" process. Analysts believe Dougbu will use its installed base of Nikon scanners in the Toshiba process and the new ASML equipment for "TSMC-compatible" products.

(See Oct. 9 story.)

Brad Mattson resigns suddenly
as CEO of Mattson Systems

Here's a real surprise. Brad Mattson, chip equipment veteran, is resigning as CEO of Mattson Technology and plans to retire. The 46-year-old not only founded the Fremont, Calif., company back in 1988, but also was a cofounder of deposition tool supplier Novellus Systems.

But Mattson isn't disappearing entirely. He will continue as vice chairman and work with Mattson on strategic issues. He says the move will make it possible for him "to spend my efforts to focus on strategic issues and pursue outside personal goals that I've delayed for so long."

His sudden departure comes just nine months after his company tried to expand via a difficult three-way merger with Germany's Steag Electronics Systems and CFM Technologies in Exton, Penn. This expanded Mattson's tool line into rapid-thermal processing and wet-wafer cleaning, and on paper, the company became the 14th largest supplier of semiconductor equipment in the world.

But this year's severe industry downturn created huge losses at Mattson while it was trying to digest the mergers.

David Dutton, president of Mattson's Plasma Division, was named acting CEO until a permanent head is selected. Dutton has worked at Mattson Technology for seven years, and was COO prior to the three-way merger. He also was acting president of the company when Brad Mattson was on his recent sabbatical.

(See Oct. 9 story.)

Intel's Barrett warns standards
will stifle chip industry growth

Here is a classic semiconductor industry hobby horse. This time around, it comes from Intel's CEO, Craig Barrett, who is warning the industry that any move towards setting standards in chip process technologies and manufacturing will "stifle" semiconductor industry growth and innovation.

Instead, Barrett insists that IC and chip-equipment makers must continue to focus on "competition and innovation" to drive the future growth of the semiconductor industry. "The semiconductor industry has thrived because of innovation," he says. "Competition and innovation give you the best technology in the market," he adds.

While Barrett doesn't give any examples of companies trying to forge industry-wide standards, analysts suspect the Intel CEO is referring to Japanese chip consortiums. Others believe he was taking a swipe at silicon foundries, which are now trying to drive process-technology standards. Taiwan Semiconductor Manufacturing, for example, is working with other companies to set new process standards for the 0.10-micron process node.

"The concept of joint development is here to stay in the semiconductor industry," said Barrett, during the ISSM keynote address. There is a mistaken mindset, Barrett says, "that the industry should coalesce around industry standards. This concept is the worst thing that could happen in our industry," he says, because you will stifle competition." By forging standards, the IC industry will sink to its "lowest common denominator," he maintains.

(See Oct. 8 story.)

Is AMD's effort to change way
PCs are rated a marketing ploy?

Here's another chip industry argument that I doubt will ever do much more than confuse the poor customer. Advanced Micro Devices is launching another effort to change the way that PC processors are measured. Instead of rating them in clock speeds by megahertz--something that has been going on for nearly two decades--AMD wants to come up with a better way to measure performance in standard PC applications.

AMD figures this would help its MPUs to better compete with Intel processors. It maintains that the megahertz-only benchmark puts its Athlon CPUs at a disadvantage in the marketplace next to Intel's Pentium 4. A chip maker can usually get a higher price for a faster processor.

"For most of the PC's first 20 years, megahertz was a reliable indicator of PC processor performance because the major players used the same architecture for product design, and clock speed was a good proxy for performance. This is no longer true," claims AMD CEO Jerry Sanders.

What AMD wants to do is set new standards for measuring PC processor power--what it calls TPI for true performance initiative. The chip maker is now trying to rustle up outside support for its new benchmarks, but it undoubtedly will run into opposition in the industry--certainly from Intel.

Any effort to set new processor benchmarks historically have been extremely difficult, if not impossible, because any measurement potentially favors one CPU architecture over another. AMD's Athlon processor architecture "demonstrates that clock speed is only half of the performance equation," Sanders says. And he notes that the latest Athlon XP processor, formally introduced this week, contains new, patented features that extend its performance beyond the traditional megahertz benchmark.

The new Athlon XP processors use a new architecture called QuantiSpeed that delivers up to a 25% performance advantage over a comparable competitor in such real-world applications as digital media, office productivity software, and three-dimensional (3-D) gaming, according to AMD. Let the marketing begin!

(See Oct. 9 story.)

Fairchild keeps expanding,
through good times and bad

It wasn't too many years ago that Motorola Semi was always bragging about having the broadest semiconductor product line in the world. But that didn't turn out to be too cool, so Moto decided to specialize and spinoff its old products. Nowadays, it almost looks like Fairchild Semiconductor wouldn't mind having bragging rights to offering the broadest semiconductor product line.

They keep adding products, right through industry boom and bust. This week, the Maine my home state company began offering foundry-manufacturing services for electromechanical systems (MEMS), based on SUMMiT IV process technology it has just licensed from Sandia National Laboratories.

Fairchild plans to ramp MEMS foundry production into high volume at its 6-inch wafer fab facility in South Portland, because it figures these products will turn into one big business. MEMS devices such as sensors and actuators will represent a $7 billion worldwide market by 2004, Fairchild predicts.

"Utilizing the SUMMiT IV process technology will enable Fairchild to develop RF and optical MEMS devices," said W.T. Greer, senior vice president and general manager of the company's Interface and Logic Group.

The Sandia process will enable Fairchild to turn out higher performing MEMS products such as RF and optical MEMS devices because of tighter control over film stress and device topography, the company says.

(See Oct. 8 story.)

U.S. EDA revenues
flat in 2nd quarter

The EDA revenues didn't go anywhere in the second quarter. But at least they didn't fall off a cliff.

Second quarter EDA revenues for products and services rose in the U.S. by less than 1% to $973 million, slowing from a sequential gain of nearly 14% in the first quarter, according to the EDA Consortium's Market Statistics Survey. The four largest publicly-held EDA companies accounted for nearly 80% of all second quarter revenues.

Revenue from computer-aided engineering (CAE) tools, the largest products category with half the total EDA revenue, fell by 5%. This category was up 8% in the first quarter.

IC layout, which accounted for 21% of EDA revenues last year, rose 43% in the second quarter to $258 million, about the same rate of growth in the first quarter. But PCB-layout-tools fell by 8% in the quarter, after gaining 12% in the first quarter. Also falling were design-services consulting, which fell 20%.

Merrill Lynch estimates that the EDA business will increase 6% in the third quarter to $2.85 billion and then flatten out in the fourth quarter.

(See Oct. 11 story.)

This foundry lives next door
to NFL's Minnesota Vikings

PolarFab is not only based in Bloomington, Minnesota, of all places go Vikings!, but claims to be the only U.S.-based pure-play silicon foundry. Now the company has fired up its new 8-inch wafer fab and started production.

The foundry will not compete in the plain-vanilla, CMOS foundry business, but aims to be a niche player competing in analog and mixed-signal foundry work against Taiwan Semiconductor Manufacturing and X-Fab Semiconductor Foundries. Its 0.35-micron process supports analog, BiCMOS, CMOS, and mixed-signal ICs.

"We specialize in RF, power-management, Bluetooth, and related applications," says sales head Jerry Thimsen. Because "we're a niche player, we're not getting hurt as badly in the downturn as the Asian foundries," he claims. The privately-held company says its fab-utilization rate is now running between 65-to-70%. The big Asian foundries are running at 20-to-40%.

PolarFab, which initially was part of Control Data Corp., has been in the foundry business for more than 30 years. Several years ago, the business was spun-off as part of VTC Inc., an IC supplier for the disk-drive industry. Later, VTC sold the bipolar fab to the foundry. PolarFab currently has two fabs--6- and 8-inch plants. The 8-inch plant was built last summer.

(See Oct. 11 story.)

We welcome your feedback, comments, criticisms, or questions. E-mail us at bhenkel@aol.com. And remember: God bless America!