Our historical society in December also held a community Christmas tree lighting, a Christmas Tea for village seniors, and a party for the village's 45 younger school children at the society's Grace Chapel. The highlight was a visit from Santa, who arrived on the town's newest fire truck.

Bullish Semico convinced
chip recovery is underway

The ever-optimistic Semico Research now believes that October chip sales data shows evidence that a solid recovery may already be underway. Further, the Phoenix market researcher now expects that semiconductor sales will jump by 20% in 2002. Too bullish for me.

Most forecasts so far for next year call for only slight growth at best--or on the order of 1%-to-6%. That would be coming after this year's big plunge of between 32% and 35%.

But Semico says it's now detecting stronger-than-expected growth in the final months of 2001. It discovered them by analyzing the typical "saw-tooth pattern" of monthly semiconductor sales that frequently obscures trends. Some semiconductor product types--flash memories, standard cell ICs, micrologic, and analog devices--already are pointing to a recovery, the market researcher says. Analog chips often have been a leading indicator of market conditions, according to Semico.

(See Dec.12 story.)

Infineon, Toshiba
close to DRAM deal . . .

There seems to be real progress being made this week in the DRAM consolidation world. Germany's Infineon Technologies, world's fourth largest maker of DRAMs behind Samsung, Micron, and Hynix, has agreed on the broad outlines of a joint venture combining its DRAM operations with Toshiba's.

Infineon CEO Ulrich Schumacher confirmed to EBN at midweek that the two companies were discussing the DRAM merger, but he wasn't sure when a deal would be completed. The plan is being presented to both corporate boards, and if they go along with it, final details will then be worked out. Current talks don't include Toshiba's flash memory operations, Schumacher said. "We're having enough trouble concluding DRAM negotiations without trying to take on flash memory."

When Schumacher was asked why Infineon was pursuing Toshiba's DRAM operations, he responded: "Why shouldn't I, if I can get it for free?" There has been speculation that it would be more costly for Toshiba to shut down its DRAM operations than to spin them off.

Infineon is interested in a deal with Toshiba because it would bring Infineon a key customer base in Japan, Schumacher said. He also was interested in hiring the Japanese company's DRAM engineers. Earlier this week, Reuters reported the two companies were close to announcing a DRAM deal where Infineon would take as much as 80% of a joint venture with Toshiba.

(See Dec. 11 story.)

. . . as the German DRAM maker
ramps up two 300-mm fabs in '02

It sure looks like Infineon Technologies is deadly serious about grabbing a bigger slice of the DRAM pie. The German chip maker now expects to ramp up 300-mm wafer production in a second, joint-venture fab, ProMOS in Taiwan, only a few months after the December launch of the new-generation wafer in Dresden.

Running two 300-mm wafer fabs will put Infineon "far ahead" of its DRAM rivals, CEO Ulrich Schumacher tells EBN this week. He maintains Samsung Electronics, his nearest 300-mm DRAM competitor, won't launch its 300-mm DRAM production until the fourth quarter next year.

The two 300-mm fabs in production next year will more than double the number of DRAM die per wafer that Infineon is now getting with its existing 200mm fabs. But Schumacher doesn't think that will increase the global DRAM oversupply at a time when suppliers are hoping supply would finally balance demand. "The extra die from our 300-mm wafer fabs will only increase Infineon's global market share by 2 percentage points," he says. "That won't have a major impact on the global supply of DRAMs," he claims.

Wafer starts at the new Dresden fab will start at several thousand a month, then ramp up to 17,000 12-inch wafer by the end of 2002. In full production, it will start 26,000 wafers a month. Infineon is delighted with its first 300-mm fab, claiming that the $1.2 billion fab was "built on schedule, under budget, and without any major problems except the normal hiccups."

As far as the higher production costs that many worry about with 300-mm wafers, Infineon figures that when its Dresden fab reaches 4,000 wafers a week, production costs per wafer would drop below those of the older 200-mm wafers. Infineon maintains that as production increased, it would reach its targeted 30% costs savings.

(See Dec. 12 story.)

AMD's Dresden fab plays big
role in keeping up with Intel . . .

Advanced Micro Devices' huge wafer fab in Dresden, Germany has become a major factor in how the processor maker keeps up with its giant competitor Intel in process technology. In fact, AMD execs contend that it's proof the chip maker no longer has to depend on Intel for process technology advances.

AMD even claims that, as a result, its pioneering ramp-up of copper processing and silicon-on-insulator (SOI) wafers at Fab 30 will benefit its archrival by enabling Intel to get mature and lower cost production equipment as it moves to the new technologies. Every wafer turned out by the Dresden fab since it started production in late 1999 sport copper interconnects. Now it is starting to process SOI wafers and will use SOI to make all of its Claw Hammer and Sledge Hammer 64-bit processors.

While Intel is still spearheading 300-mm wafer fab development, AMD's much smaller processor die size doesn't force the company to rush into 300-mm wafer production as rapidly as Intel, maintains Hans-Raimond Deppe, AMD vice president. "The very large Pentium 4 and Itanium die size means Intel has to go to 300-mm wafers to cut costs by getting more die on a wafer." AMD, on the other, he says, can scale down its die size using next-generation 0.13-micron processing and later 0.09-micron processing "to stay ahead of Intel in die size."

AMD "took great risks" in bringing up Fab 30 in late 1999 exclusively with copper, Deppe says. "The tools were not mature, and we had to work very closely with equipment makers to iron out all the bugs." He figures Intel is going through the same learning curve as it ramps up copper processing. "But they get a mature tool set, thanks to all the work AMD and other copper pioneers did," Deppe adds. And . . . the tools are less costly than we had to pay."

AMD's move next year to 100% SOI wafers will give it a jump on Intel, which has yet to put SOI on its manufacturing roadmap. SOI is not the manufacturing challenge that copper posed, since AMD buys its SOI wafer blanks from Soitec. "Still you have to design the processor circuitry differently for SOI," he adds.

(See Dec. 11 story.)

. . . as AMD eyes locating
300-mm effort at Dresden

There's a good chance that Advanced Micro Devices will locate its 300-mm efforts at its Dresden complex. A pilot line could be running in 2003, followed by a production fab one year later.

"It's a natural choice to start AMD's 300mm work here," says VP Hans-Raimond Deppe. But "many other factors" could influence the site selection for the firm's entry into the next-generation of wafers, he says. The many local governments holding out large tax incentives and assistance could determine the ultimate location for AMD's first 300mm wafer fab, he concedes. Also influencing site selection will the partner that AMD joins with in the 300-mm fab.

Dresden is the only fab producing AMD's high-end Athlon family of processors, and next year it will be the only fab making its K8 generation of 64-bit Hammer processors. The200-mm fab will be running at full capacity of 5,000 8-inch wafer starts a week by the end of December.

(See Dec. 12 story.)

Intel tries to resuscitate
'artificial intelligence'

Back in the late '70s and early '80s, one of the most exciting and controversial high-tech stories around was artificial intelligence. AI even made the cover of major business publications such as BusinessWeek. Alas, it became more the stuff of science fiction than developing into a new computing tool.

But it's hard to keep an exciting idea like AI down. Intel among others is now poking around in the ashes and has started exploring promising new ideas in biometrics and AI--a move that some observers say could result in a new class of equipment some day that could be compared with HAL, the computer that totally ran the space ship in the sci-fi movie classic "2001: A Space Odyssey." These guys must have got that analogy from one of my 20-year-old stories.

This week Intel released an updated line of 500 image-based software tools and code aimed at developers working on "computer-vision" equipment that would use its future microprocessors to power advanced inspection systems and sophisticated airport security gear. These systems could spot and identify terrorists, for example, using software running on standard PCs, the chip giant claims. They would do this by viewing the world in stereoscopic "depth" and "flat" images and in three dimensions--a technique that would support face recognition and object tracking.

Computer vision, which looks like an offshoot from the ill-fated AI business of the '80s, is one of several technologies trying to dominate the emerging biometrics market. Biometrics makes use of several technologies--such as face location, fingerprint identification, retina scans, voice recognition, and others--to identify a person in a particular application. So far, however, there has been more talk than action.

Intel hopes to change this with the computer-vision software and other advanced technologies that its engineers are working on in Russian and China, for example. The technology is not limited to airport security. Computer vision today is being used for quality control and industrial inspection, such as spotting bad potato chips before they go into the bag. But future computers, Intel believes, should be able to do such things as read sign language, which, in turn, could be automatically fed into a system. Well, at least such technology makes for good news stories.

(See Dec. 11 story.)

Bluetooth 'industry'
now sees 2003 start . . .

I've been wading in so much hype for so long about Bluetooth that I tend to over react to this everyman's com link. But after all, the low-powered wireless system was supposed to have hit the street more than two years ago. The Street hyped this short-distance technology because it was going to connect cell phones and notebooks over a 1-megabit-per-second wireless link. More than 100 chip makers and OEMs started developing products.

Despite continuing interoperability difficulties, high cost, and other problems, a slew of new products were unveiled at this week's Bluetooth developers conference in San Francisco. Despite all the hoopla, industry experts and analysts don't think the "mass adoption" of Bluetooth will happen until 2003. Chip makers are shipping or expected to ship Bluetooth parts in 2002, but volumes will be lackluster at best.

"Bluetooth will happen, but not in 2002," admits Dwight Decker, CEO of Conexant Systems, a supplier of Bluetooth chips. "Bluetooth is now a 2003 product," he says. If memory serves, some chip execs were predicting a year ago that Bluetooth was a 2001 high-volume product.

The delay has resulted in a lot more competition. Bluetooth has taken a backseat to rival wireless protocols such as 802.11a, 802.11b, and 802.11g. "Bluetooth is still important, but it is not as important as 802.11," Decker adds.

Market researchers keep expecting the moon. Cahners' In-Stat Group expects 13 million Bluetooth chips to ship this year, and by 2005 sales of these chips will hit an incredible 780 million units.

(See Dec. 10 story.)

. . . as TI promises
$5 'solution' in '02 . . .

Texas Instruments is one of the big guys gunning for mass-market consumer products with a new Bluetooth processor. The Dallas company showed its new ROM-based baseband processor that will provide full data rate Bluetooth links at a "solution" price as low as $5 each in 2-million unit orders.

TI calls it a 4th-generation baseband device that will deliver 723 kilobits per second. The $5 solution includes a new baseband controller as well as a radio-frequency transceiver. That's quite a promise.

The new baseband processor is fabricated with a 0.18-micron (drawn) CMOS process technology.

TI doesn't see any delays to Bluetooth adoption and says a study by Strategy Analytics says that U.S. cell phone users were willing to pay a 27% premium for Bluetooth connectivity in their handsets. I wonder. Anyhow, engineering samples will go to customers in December, with volume production planned for mid-2002.

(See Dec. 11 story.)

. . . and Motorola shows
set that stops interference

One nagging headache for Bluetooth designers has been trying to avoid any conflicts caused by 802.11b wireless LANs and Bluetooth units operating too close together. This is certainly a problem that will increase rapidly as these wireless technologies are deployed.

Motorola Semiconductor says it has fixed this problem in its new Bluetooth chip set unveiled this week that features a gallium-arsenide power amplifier and power management chips. The chip set is optimized on the link between mobile handsets and wireless headsets.

To get around any interference, Moto has come up with an adaptive scheme that allows the chip set to intelligently hop around channels to avoid interference. And it has added joint detection and maximum likelihood sequence estimation (MLSE) techniques to the RF front-end to increase sensitivity, which in turn increases resistance to noise. The better sensitivity also gives the transceiver a range of 120 meters, four times better than the Bluetooth spec.

Motorola has working silicon and plans to sample the chip sets in the first quarter with production planned for the second quarter. The chip set will be priced at $5.90 each in million-unit quantities.

(See Dec. 11 story.)

Now Micron is talking to
Toshiba about Virginia fab

Someone told me recently that more than 45 wafer fabs worldwide are now on the block. That's a heckuva lot of capital equipment! Now you can add Toshiba's Dominion Semiconductor fab in Manassas, Va., to the list.

Sources tell EBN that Micron Technology is negotiating to take the big fab over. These negotiations are surprising since Micron is already talking with Hynix Semiconductor to take over the South Korean DRAM maker's fab. Micon might be hedging its bets in conducting all these negotiations.

The Virginia fab would not be included in any DRAM joint venture that Toshiba works out with Infineon Technologies, according to the sources. Infineon already has a big Virginia DRAM fab, White Oak Semiconductor outside Richmond. Dominion would be a redundant fab if the Infineon-Toshiba deal goes through.

It isn't known whether Micron is talking to Toshiba about an outright acquisition of the Virginia fab or running it as a joint venture, similar to Micron's arrangement with Hewlett-Packard and Canon in Singapore.

Toshiba's Dominion Semiconductor fab has two lines--one making DRAMs and a second turning out NAND flash memory chips as part of a joint venture with SanDisk. Toshiba reportedly would transfer flash production back to its Japanese fabs, freeing up the second stage of the Manassas fab for Micron to make DRAMs.

Dominion was originally launched as a joint venture between Toshiba and IBM, but Big Blue pulled out when it cut back its DRAM operations.

(See Dec. 14 story.)

Lattice jumps into FPGAs
fast by buying Agere unit

How does a chip maker get into the field programmable gate array business fast? By buying a going business, of course. So Lattice Semiconductor, which said in November that it wanted to enter this market, decided this week to acquire the FPGA business of Agere Systems for $250 million in cash.

As far as Lattice Semiconductor president Steven Laub is concerned, it was a marriage made in heaven. "Agere's products are proven solutions that will complement our internal FPGA development effort," he says. "Agere also offers a deeply experienced and talented team capable of extending the technology and building the product portfolio." Only about 100 Agere employees will join Lattice, however.

The deal apparently will help Agere, formerly Lucent Microelectronics, to get its troubled house in order. It currently is refocusing its chip business, cutting costs, and downsizing its operations. It also is slashing another 950 more jobs on top of the 6,000 positions eliminated earlier this year. "While FPGAs have been a strong business for us, we are redirecting our resources to areas where we can focus our skills on system-level integration for advanced communications applications," explains EVP Sohail Khan. However, Agere's FPGAs sound to me like the kind of products that would fit into Agere's new focus.

The transaction includes Agere's general-purpose ORCA FPGA product portfolio, its field programmable system chip (FPSC) series and all related software design tools. Also going to Lattice will be intellectual property cores and patents that pertain to Agere's FPGA business. Its FPSC advanced system-oriented products combine generic FPGA logic with embedded cores dedicated to advanced communication protocols and high-speed input/output (I/O) functions.

Lattice is counting on Agere's FPSC series to penetrate key FPGA applications in communication systems. "This FPSC portfolio will immediately provide Lattice an important means of differentiation in the FPGA market," Laub says.

(See Dec. 10 story.)

Intel, Sun designing multicore
processors for high-volume markets

Engineers are finding that it's becoming increasingly difficult to wring more performance out of single microprocessors. One way to get more out of processors, experts say, is a multicore architecture that would simplify the increasingly complex design and validation work of processors. This would both lower chip costs and slash time-to-market.

Now for the first time, mainstream computer companies are talking about using this approach for their high-volume markets. Intel and Sun Microsystems are each laying plans to deliver multiprocessing computers on a chip.

Both companies plan to put two or more processors on a die with simultaneous multithreading (SMT), a design approach that lets a processor handle two or more threads of an application simultaneously. Intel is working on its version of SMT to handle the memory access problems that crop up in multiprocessors, while Sun is going farther, designing new Sparc processor cores especially for use as multiprocessing chips with four or more cores on a die.

Intel expects to deliver its multicore processors with the 70-nanometer (0.07-micron) process generation. That chip would follow-up Intel's IA-32 Xeon processor, a single-core device coming next year that will use SMT to handle two application threads. Intel believes it can squeeze 10%-to-30% more performance out of single-core Xeons using dual-threaded SMT.

Sun's initial multicore device will be a dual-core UltraSparc 3 with SMT that will target its medium-performance servers and workstations, as well as middle-tier servers used for applications processing. It also plans to design new Sparc cores geared for multiprocessor chips featuring more than two CPU cores. They would be primarily for Web servers.

The trend to multicore designs has been firmly established, especially among network processor makers and academics. Researchers at Stanford University, MIT, and UC Berkeley already are trying to put 16 or more simple computing elements on one die to greatly improve processor performance.

One of the secrets to multiprocessor success is creating "an assembly line on chip" where one die handles as much work as possible, points out Stanford professor Bill Dally. Indeed, Intel's IXP 1200 network processor has six cores on a die, and its next-generation device is said to have as many as 16 cores. And startups such as EZChip Technologies and Internet Machines will use as many as 64 cores on a die.

(See Dec. 10 story.)

Applied Materials
signals lousy 2002

Those forecasters who see the semiconductor production equipment business getting much better next year are smoking something, as far as I'm concerned. This week the industry leader indicated this market wasn't going anywhere anytime soon by doing something it hates to do--layoff experienced staff.

Applied Materials is reducing its global work force by 10%, or about 1,700 positions. It is cutting about 450 positions in Silicon Valley and 600 in Austin. Applied also will offer other employees the option of taking a furlough program, an unpaid leave of absence, during which they will continue to receive some employee benefits.

Comments CEO James Morgan: "Unfortunately, the continuing downturn requires us to make some tough decisions to align our operations with current levels of demand for semiconductor equipment." The company already has implemented several cost-saving programs, including executive and employee salary reductions, restricted hiring and travel, mandatory shutdown days, a voluntary separation plan, and a reduction in the regular and temporary work force.

(See Dec. 12 story.)

TSMC now ramping
low-power 0.13-micron process

Foundries are continuing to introduce the latest state-of-the-art processes that are making the smaller fabless companies more competitive with the larger, integrated device manufacturers with their own fabs.

Taiwan Semiconductor Manufacturing says that it's the first silicon foundry to offer a 0.13-micron process option for low-power ICs used in wireless applications. The new process offers multiple, threshold voltages that make possible "fine-grained tradeoffs" in standby power leakage and performance for specific low-power applications, such as personal digital assistants and cellular phone handsets.

Demand for advanced processes with optimized power management has been strong, even during the current downturn, says VP Genda Hu. "These applications require a finely tuned balance of the speed/leakage tradeoff," he notes, pointing out that the new 0.13-micron low-power option "offers designers the flexibility to integrate more design blocks with lower power consumption."

TSMC says that it will ship at least 9,000 wafers processed with 0.13-micron technologies by the end of this year. High-volume 0.13-micron production will ramp throughput 2002, it says, with the new low-power option expected to be in volume production by early next year.

(See Dec. 11 story.)

Transmeta delays 0.13-micron MPU
due to production problems at TSMC

Even farming out chip making can sometimes result in production problems for a fabless semiconductor company. Transmeta is delaying volume shipments of its new 0.13-micron microprocessor lines by about four months, citing production problems with its foundry partner, Taiwan Semiconductor Manufacturing.

Transmeta execs didn't exactly blame TSMC for the glitches, but does say that the foundry's transition from 0.18- to 0.13-micron technology has proven to be a major headache.

"Although our unit shipments ... increased during November, we fell substantially short of our production goals," says CEO Murray Goldman. "Unfortunately, we concluded that it was necessary to make metal mask revisions in order to achieve our volume manufacturing goals using the new 0.13 micron manufacturing process." The microprocessor startup will continue to produce the MPUs in small volume until its new mask set is in production, but mass production of the new MPUs was delayed from November to February. TSMC reportedly is making the reticles for Transmeta at its own mask shop.

(See Dec. 12 story.)

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

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