Funny stuff, yeah. But it illustrates some of the problems that make it difficult to invest knowledgeably today in the semiconductor industry. E-mail me your comments.

Infineon called 'little Hynix'
as well as 'Germany's Enron'

Infineon Technology came in for some tough criticism this week in an e-mail I received from reader Gero Breuer, a German financial analyst, who has been following the crazy global DRAM business.

To begin with, he compares the Siemens spinoff to Hynix in how it raises money. "Since Infineon is backed up by 'Deutschland AG', or Germany Inc. think Japan Inc., it was still able to raise money in January," Breuer comments. The company's shares were distributed to many retail investors by German banks, which sold it as a 'solid' high-tech company during the boom, he points out. "Since Germans like to blame the government and even the banks for stock losses--there are elections this summer--Infineon will always be able to raise money," he believes.

At the same time, Siemens sold a large part of the Infineon stock to its pension fund, Breuer says. "So Siemens used its pension fund as a trash bin." The reason this can happen, he adds, "is that the labor unions, who partly operate the pension fund, do not know anything about stocks." As a result, Breuer says, "I believe Infineon is Germany's Enron."

Micron, Hynix get closer
to deal, but still $1B apart

Micron Technology and Hynix Semiconductor are getting closer to a deal, according to reports from Seoul this week. But the two giant DRAM suppliers still remain $1 billion apart on the price that Micron would pay for the South Korean company's seven DRAM fabs.

Members of the Hynix restructuring committee are telling the South Korean press that Micron's latest offer is around $3.8 billion, while the committee, steeled by a firming DRAM market, is still asking for $4.8 billion. Micron's original offer reportedly was somewhere between $2.5 billion and $3 billion.

Micron officials confirm they are talking to Hynix again about the deal after negotiations broke off a week ago.

Hynix creditors, which hold $6.5 billion in the DRAM maker's debt, still think they can agree soon on a preliminary agreement and reportedly gave short shrift to overtures from Infineon Technologies late last week. Infineon CEO Ulrich Schumacher met with Hynix executives in Seoul on Feb. 1 to discuss possible sales and technology partnerships. But Infineon had nothing new to report from the meeting.

(See Feb. 4 story.)

Global chip sales may not
have bottomed as reported

We may not have seen the low point in industry sales last fall after all, despite what many suppliers are saying. The Semiconductor Industry Association released its December sales figures this week and monthly sales proved to be disappointing.

Worldwide chip sales in December fell 4% sequentially to $10.2 billion from $10.60 billion in November. The decline followed two months of rising sequential sales--up 2.5% in October and 1.6% in November.

That drop was unexpected. As late as November, the SIA was predicting that 2001 sales would end up around $141 billion, or a 31% decline from the previous year. Instead, global sales came in at a couple of billion dollars less, or $139 billion for the year.

But SIA officials were ready with an answer. "Except for 1999, monthly semiconductor sales in December for the past six years have registered a decline just as we are seeing this year," insists SIA president George Scalise.

I have to admit that when you look at sales on a quarterly basis, fourth-quarter chip sales were close to third-quarter revenues. They were still falling, although they did end three straight quarters of double digit declines. The SIA says fourth-quarter global chip sales were $30.54 billion vs. $30.57 billion in the third quarter--only $30 million less. But still going down.

But Scalise was ignoring it. "Key demand drivers, wireless handsets, and personal computers bottomed out in the third quarter and recorded double-digit increases in the fourth quarter," he says. The chips used in products like microprocessors, digital signal processors, and DRAMs, "also registered double-digit gains," he says.

Japan was still hurting global chip sales the most. Japanese sales in December fell 8.5% from November, compared to drops of 1.2% in the Americas, 2.2% in the Asia-Pacific, and Europe 5.3%. It was even worse in the fourth quarter, when Japanese sales fell 11.8% from the third quarter, while the rest of the world was up 3.7% sequentially.

(See Feb. 4 story.)

Did big IC makers err by closing
own fabs and going to foundries?

There's a lot of whispering going on these days about the problems that silicon foundries are having getting their 0.13-micron processes into production.

Several major foundries reportedly are having trouble reaching acceptable yields on their copper-based 130-nm (0.13-micron) processes. These foundries haven't admitted to problems publicly, but they do acknowledge they're still in the "debug" or "early-learning/yield-improvement" stages with their 0.13-micron technology.

But these reports of widespread yield problems at 0.13-micron are raising questions about the ability of these third-party foundries to get their next-generation processes up and running quickly at a time when a growing number of IC suppliers are opting to shut down their own fabs and use contract manufacturing instead.

Taiwan Semiconductor Manufacturing is said to be having problems getting its 0.13-micron technology up and running after rolling out the process last year. And other chip makers are complaining of lower-than-expected yields with TSMC's 0.13-micron process.

TSMC is still having some "reliability" issues with its 0.13-micron process based on low-k dielectrics, confirms Shang-yi Chiang, R&D vice president for the foundry giant. "We hope to solve that within the next three months," he tells SBN.

There also are reports that TSMC's chief competitor, United Microelectronics, also is having problems with 0.13-micron technology. UMC officials acknowledge the foundry is still in an early stage of tweaking this technology. "We are still in the early stage, but we are in pretty good shape for this 'pre-production' or 'early production' right now," maintains Fu Tai Liou, UMC's chief marketing officer.

The same question was raised about Chartered Semiconductor Manufacturing. While the world's third largest silicon foundry announced last summer that it had fabricated its first 0.13-micron copper IC for a customer, it still has not officially sold a 0.13-micron wafer for revenue nine months later.

But, CEO Barry Waite maintains that Chartered is "literally days away from making our first revenue shipments in 0.13-micron. I'm expecting 'full qual' in the late third-quarter time period."

(See Feb. 4 story.)

Foundries in best position
for turnaround, says iSuppli

Even though several major wafer fabs have been closed in recent months, there's still more than enough global capacity to meet anticipated customer needs for nearly all chip products this year. That's not exactly news and is certainly what's worrying equipment makers these days.

But there is a bright spot. Just as in previous downturns, there is that one exception--there's never sufficient capacity for the latest, state-of-the-art chips. And that means there is not enough capacity now to build the latest products requiring process geometries of 0.18-microns and less. But new facilities to build these parts are expected to be on line by June, according to iSuppli's Market Intelligence Services.

And the producers that may be in the best position this year are the silicon foundries, says Len Jelinek, iSuppli analyst. Since Integrated Device Manufacturers (IDMs) were unable to invest in keeping their internal fabs up to current competitive levels, they will be forced to outsource their new product manufacturing requirements to the foundries to remain competitive, the analyst predicts.

As a result, iSuppli sees foundry usage "rising rapidly" this year and foundry factory utilization passing 70% by the fourth quarter. One of the biggest foundry bulls around right now, iSuppli expects foundry revenues to grow fivefold by 2006 at a compound annual growth rate of 38%. "That is extremely good news for the six new foundry companies expected to come on stream outside of Taiwan in the Far East by the end of Q2," Jelinek notes.

While equipment makers have taken their usual dive during this chip market downturn, current shortages in 300 mm and fine-geometry fabrication facilities are beginning to turn chip gear book-to-bill ratios around, he says. Jelinek expects to see the semiconductor equipment book-to-bill ratio reaching one in the second quarter, and maintaining this ratio throughout the second half of 2002. Leading the pack out of the doldrums, he says, are those equipment suppliers that developed strong relationships in China since the investment there is expected to be "the strongest of any market in the world over the next few years."

(See Feb. 4 story.)

Optical interconnects won't be
used for at least another decade

For some time now, researchers have predicted that optical interconnects would be needed for new high-performance silicon chips that otherwise would be input-output limited. But optical interconnects were dismissed as a possible solution by researchers who declared the interconnect problem has been overstated.

Metal wires may prove to be scalable and able to provide acceptable performance far longer than some people have predicted, declared panelists at the International Solid-State Circuits Conference this week

By optimizing the thickness and spacing of on-chip interconnect, and carefully guarding the quality of the shielded cable and printed circuit board for chip-to-chip and system-level interconnect, "10 to 15 gigahertz is already quite implementable" in microprocessors, says Alena Deutsch, technical staff member at IBM's T.JU. Watson Research Center.

Optical links are not faster, since both electrons and photons travel at near the speed of light, and optical links draw more power, points out Christer Svensson, a professor at Sweden's Linkoping University. "Why on earth would anyone want to use an optical interconnect?" He fails to see any merit to optical interconnects, except for long distances.

Also downplaying the commercial potential of optical interconnects for at least the next decade is Sam Naffziger, engineer at Hewlett-Packard's microprocessor development center. "Putting optical wave guides on-chip presents too big of a technical hurdle and is not going to pay off."

(See Feb. 5 story.)

Surprise--TI is shifting
production to foundries

Of all the chip makers I've covered over the years, the last one I would have picked to take an "assets light" approach to manufacturing is Texas Instruments. But the Dallas chip giant will increase its reliance on foundry capacity during the next capacity crunch, says CFO Bill Aylesworth.

"This time around, we plan to use our own capacity for what we think are the stable demand levels and use foundries to address peak demand, providing a buffer," he points out. "It's a tactical change. But we're going to continue our own process development and the bulk of our capacity will remain internal."

TI currently outsources about 5% of its production to foundry partners, the bulk of which going to Anam Semiconductor, a South Korean supplier. Aylesworth says there is no target for future foundry usage, but says it's unlikely to grow to more than 20% of total production.

It is significant but not surprising that a company like TI would increase its use of foundries, maintains Mark Edelstone, Morgan Stanley Dean Witter analyst. "There's no question that the severity of this downturn was a wake-up call for everybody, and now everybody is trying to build in some degree of flexibility," Edelstone says.

But TI is still working hard to move its latest processes into production at its own fabs. It plans to move into full production with its 0.13-micron process in the second quarter. And it is ramping its 90-nanometer, 0.09-micron process at its 12-inch Dallas fab. Qualification of the 90-nm line is expected in the third quarter next year with production beginning by the end of 2003.

(See Feb 5 story.)

Here's Dr. Feelgood:
'Chips up 19% this year'

Believe it or not, Semico Research is not backing off its earlier bullish forecast for the chip business this year. Even though the general economy seems to be edging up again, I still have a hard time believing the Phoenix market researcher's numbers.

It is looking at 19% growth for this year over 2001 at a time when just about everyone else is expecting sales to be flat to up by only 6%.

The big drivers here, according to Semico, are the PC and cellular-phone markets, which it predicts will return to double-digit growth in 2002. "The market will innovate its way out of the doldrums this year," says Jim Feldham, Semico president.

Global shipments of handsets fell by 7-to-10% last year, Semico says, dropping from 420 million units in 2000, to between 380-to-390 million units in 2001, the researcher estimates. But this year, Feldham says, the market will shoot up between 20-to-25% and reach 480 million units, he says.

And he projects a 9-to-10% growth in the PC market for 2002, after falling in 2001. Desktop PCs will grow between 4-and-5% in 2002, while sales of notebook computers will jump 13% this year, he says.

There are several other hot markets, he expects for chips. They include global positioning systems, MP3, home entertainment, and home networking. But some communications markets will continue to be soft for the rest of this year, he adds.

(See Feb. 6 story.)

Low-power technology Intel
needs to build 10 Ghz MPUs

Intel has talked a lot in recent months about future processors that will run at blazing speeds such as 10-to-20 gigahertz. But it will never get there unless it can slash power consumption on these chips dramatically. So this week a gang of its researchers were at the ISSCC in San Francisco describing the new low power technology they are developing.

Three new, low-power building blocks that Intel says will make possible high-speed processors such as a 5-gigahertz integer execution core and an integrated 6.5-gigahertz arithmetic-logic unit and scheduler.

The company is aiming these technologies for next-generation processor designs. "We would expect to see products at the 90-nm (0.09-micron) node, and most certainly, at 65-nm (0.065-micron)," says Justin Rattner, head of Intel's Microprocessor Research Labs. The company plans to be producing 90-nm chips by 2003 and 65-nm designs in 2005. And by 2010, Intel expects to be shipping processors that run at speeds from 10-to-20-gigahertz.

The low-power technology Intel described this week includes what it calls a TeraHertz transistor, which marks its initial use of high-k dielectrics, epitaxial wafers, and silicon-on-insulator technology. Researchers also discussed what they call "forward and reverse body bias" technology in new low-power chip designs. Forward bias will make possible a 23% reduction in active power in chip designs, Intel claims, while reverse bias can cut stand-by leakage by three-and-a-half times over current technologies.

Intel has designed a CMOS-based, router chip that features an integrated body-bias. The 10-by-10-mm device was built with 0.15-micron technology and runs at 1-gigahertz and has 6.6-million transistors. This technology does not require any new manufacturing steps, according to the company, and it doesn't take that much real estate.

(See Feb. 4 story.)

Intel talks up McKinley,
but still mum on Yamhill

It was McKinley "show and tell" this week at the International Solid-State Circuits Conference. Intel has been talking about McKinley, follow-on chip to Intel's 64-bit Itanium, for a year but this week provided more details of the 1-gigahertz, 221-million transistor chip.

McKinley has a 400-megahertz, 128-bit-wide bus and eight pipeline stages, giving it a total bandwidth of 6.4-gigabytes-per-second. "McKinley will deliver one-and-a-half to two times the performance than Itanium," says Intel fellow John Crawford. It has a Level 2 cache bandwidth of 64-gigabytes-per-second, roughly four times the performance of RISC designs, Intel claims.

Analysts seem to be impressed with McKinley. "It's the Itanium done right," declares analyst Nathan Brookwood of Insight64. Itanium has been a big disappointment to Intel. Developed at a reported cost of $1 billion, it was two years late to the market.

McKinley currently is sampling McKinley for what the company calls "pre-production pilot systems," and is scheduled to ship in mid-2002.

Intel didn't talk about another 32/64-bit microprocessor technology code-named Yamhill it reportedly is also developing. It could go on the market if Intel's other 64-bit chip lnes fail to gain market acceptance.

While Itanium and McKinley are full-fledged 64-bit chips, Yamhill is a software and hardware technology that will add "64-bit extensions to Intel's x86-based 32-bit architecture, which is similar to what AMD is doing," says analyst Nathan Brookwood of Insight64. He was referring to AMD's 64-bit processor line called Hammer due out early next year.

(See Feb. 4 story.)

Dongbu knows when
to start new fab--now!

At least one semiconductor company apparently has learned from the old story about Manhattan real estate. You know the story--I tell it every chance I get. What it boils down to is the experienced builder starts building skyscrapers in New York at the darkest part of a downturn, or when vacant space is flooding the market. Get the connection? Let me spell it out. The time to build new fabs is when business looks absolutely awful. But it's those long lead times that will kill you when you're adding new capacity.

Anyhow, South Korea's Dongbu Electronics, which is just now ramping its first 8-inch fab in Eumsung, already is starting to raise a new round of funding so that it can start building its first 300-mm fab next year. The plant probably won't go into production until 2004.

So far, Dongbu has raised $1 billion, including $410 million from a group of 10 financial institutions led by the Korean Development Bank. It already is processing 0.25-to-0.18-micron wafers and plans to develop a 0.13-micron process by the third quarter.

The Korean foundry startup figures its timing on the 300-mm fab is dead on the money. The overall foundry business is picking up steam and "the bottom is behind us," declares Peter Hillen, EVP at its U.S. headquarters.

(See Feb. 6 story.)

China IC imports
soar 25% last year

Do you wonder just how much of a chip industry China has? And how they did last year compared to the rest of the world? It's hard to figure out with the kind of numbers the government releases, but you can safely say they are becoming "a contender."

China did relatively well last year, importing $16.6 billion worth of ICs, up 25% from 2000. And it exported $2.5 billion worth of ICs, a 10% drop from the year earlier, according to the Chinese government. China itself produced 6.4 billion ICs, a 9% rise over 2000, while it imported 26.18 billion units.

This growing chip demand reflects China's huge contract manufacturing base. And in contrast to the rest of the world, computer exports grew. It exported $13.1 billion worth of computers, up 19% from 2000, and $6.8 billion in computer peripheral equipment, which was up 8%.

(See Feb. 6 story.)

Will Dupont be first to make
0.18-micron reticles in China?

DuPont Photomasks is getting ready to push the envelope in the production of lithography reticles in China. The U.S. firm is planning a major expansion of its mask-making facility in Shanghai and it expects to be the first merchant supplier of reticles to install a production line for 180-nm (0.18-micron) masks in China. DPI hopes to begin things in the current quarter.

But first it has to begin the export licensing process that's required by the U.S. government. And that could be a problem. U.S. export controls over advanced lithography technology in China have been a bone of contention for years. Under older U.S. export guidelines, 0.25-micron technology was the limit, but in the past couple of years many chip-making startups, joint ventures, and major IC makers have started planning to install 0.18-micron and below processes for their China fabs.

DuPont is supporting its expansion on the Mainland by forecasts that show the Chinese semiconductor market running at a 28% compound annual growth rate--from $15 billion in 2001 to $40 billion in 2005. It adds that China's admission to the World Trade Organization is expected to increase chip production significantly in that nation.

(See Feb. 7 story.)

We want your feedback, comments, criticisms, or questions. E-mail us at bhenkel@aol.com.

(Click here for last week's Semiconductor Alert!.)