Commentary & analysis of week's chip news, Aug. 27-31
Greetings from Down-East Maine, where we definitely are in a mini-recession. While we might not be hurting as badly as Silicon Valley, a gang of small manufacturers turning out everything from shoes to forest products have been closing their doors up here this summer. And our biggest business, tourism, was off in June.
I live in the poorest county in the Pine Tree state, and a nearby factory in Eastport is shutting down this fall--100 jobs gone in a town of 3,000! Now I hear a telemarketing company in Calais, a town of 3,000 that's 12 miles away, is closing its doors. That knocks out another 100 jobs. When that kind of disaster hits in a remote area like this, people usually end up with just one choice: Move south to find another job. To make matters worse, we've just been informed that Maine has the highest overall taxes in the nation.
But, the air is still pure, it's 49 degrees here this morning, the sea breeze is our window air conditioner. And it rained last night, so the fields are green again. Right now I can see Canada and the Bay of Fundy outside my home office window--a couple of fishing boats bobbing up and down at anchor off our beach, and two sloops sailing by from nearby Saint Andrews, New Brunswick. Thank god for telecommuting.
Why did 3 key Motorola
execs leave the company?
Motorola managers must now be wondering whether the next upturn in chips and cell phones will come in time to help their troubled company. At least three vice presidents have left the company in recent days as second guessers now question whether Motorola president Bob Growney over-reacted last week by responding to reports the firm was thinking about selling or spinning-off its troubled semiconductor unit.
Whatever the case, few companies can afford this kind of executive drain. Daniel Artusi, vice president heading up the Austin-based networking and computing systems group, has left Moto to join Silicon Laboratories as COO to run operations.
Thomas Conn Jr., who worked 20 years at Motorola Semiconductor, has resigned his post as vice president and head of the imaging systems division, which designs system-on-chip products. He will become CEO of Cynergy System Design, which makes embedded systems simulation solutions for hardware and software co-verification.
And Fairchild Semiconductor, headquartered up here in god's country, has named Hans Wildenberg, a 16-year Motorola veteran, to EVP in charge of sales and marketing and to its executive committee. Wildenberg had been VP and head of global sales at Motorola Semi.
(See Aug. 28 story.)
Competition just doesn't
get much hotter than this!
Intel has always been the toughest kind of competitor--I remember the 1980s when the big chip maker "requested" customers to buy its commodity DRAMs along with its short-supply microprocessors. But I don't think I've appreciated just how nasty and bloody its current battle is with Advanced Micro Devices over the PC processor market.
Despite its overwhelming lead in this huge chip market, Intel is going all out to beat AMD in the race for faster microprocessors--even though it still isn't clear what "killer apps" will need this kind of power in the immediate future. Heck, I just purchased a 500-megahertz PC and it's the phone lines that are killing me. Although the two chip makers account for nearly all the global MPU market now, their frantic price cutting is burning much of both company's current profits. Indeed, the stakes are high here.
This week Intel used its San Jose developer forum to promote its new line of processors. It not only rolled out its latest high-speed processor--the 2 gigahertz Pentium 4--but it hyped a product that is more than one year away. The chip giant demonstrated its Pentium 4 running at 3.5 gigahertz. The 32-bit chip is based on Intel's new 0.13-micron process, but reportedly won't be out until the second half of 2002. Before that hot chip arrives, the company expects to offer a 2.2 gigahertz chip by the end of the year.
But Intel didn't even stop there. EVP Paul Otellini says that "a 4-gigahertz processor is on the horizon."
Now the ball is in AMD's court and Intel's new chip is expected to pressure the No. 2 MPU maker to execute its plan to ship its 1.5-gigahertz Athlon processors.
One close observer expects AMD to begin shipping the 1.5-gigahertz chip "in the next 45 days." He says he has talked to PC makers that have tested both processors, "from what I can understand, if you would conduct a blind taste test, customers won't be able to tell the difference" between machines built around the 2-gigahertz P4 and the 1.5-gigahertz Athlon. Now if they just don't ship a $10 bill with every MPU.
(See Aug. 28 story.)
New round of consolidation
begins in DRAM business . . .
It seems just like yesterday when the Japanese semiconductor giants pushed Intel out of the world's largest chip market Yeah, I'm getting old.. Intel figured there wouldn't be enough profit in a commodity DRAM market. It took the Japanese a few years, but now they're acknowledging the same problem. But it is taking an awful year for DRAM to convince them.
Now it seems the DRAM business is following the same course the automotive market took more than 60 years ago to consolidate into just a handful of supplier survivors. Toshiba apparently has contacted several large memory chip makers, including South Korea's Samsung, the world's largest, in an attempt to get rid of its troubled DRAM business.
While Toshiba ain't talking, a Samsung official acknowledges that Japan's largest semiconductor company contacted the DRAM giant to explore the possibility of selling its DRAM business to the Korean company. "As far as I know," a spokesman tells Reuters, "Toshiba is contracting several ranking DRAM makers around the globe."
Tom Quinn, Samsung Semiconductor marketing chief, says that what's going on in the DRAM marketplace "indicates the seriousness of the current market downturn." No matter what develops in the Toshiba matter, he expects the global DRAM market to go through wide scale consolidation in the near future.
There was buzz too regarding another big DRAM vendor, Infineon Technologies, that reportedly was also talking to Toshiba about its DRAM business. The German chip maker confirmed it was holding negotiations with Toshiba "to cooperate in DRAMs," but it wouldn't comment on any joint venture or merger.
(See Aug. 28 story.)
. . . as DRAM market
collapses this year . . .
The current incredible-shrinking DRAM business is due to weakening PC sales, a lack of new software that needs more main memory, and by careening prices due to overcapacity. This crisis has caused many observers to predict a major shakeout among suppliers. Some of them now say that only two-or-three major DRAM vendors are needed globally.
Meanwhile, DRAM suppliers--especially the Japanese--are drowning in red ink as average selling prices for memories continue to fall like a rock. Worldwide DRAM revenues are now expected to plunge 56% this year to $14 billion, down from $31.5 billion in 2000, according to Dataquest. DRAMs hit their high point in 1995 when global sales shot to more than $40 billion.
The latest consolidation in the DRAM industry began a year ago when NEC and Hitachi combined their two struggling DRAM operations into a new joint venture called Elpida Memory. The South Korean government previously had orchestrated a merger of LG Semicon into Hyundai Electronics, which was renamed Hynix Semiconductor. But that combine is now caught up in a severe financial crunch and needs money desperately to meet more than $2 billion in short term debt that's maturing in the next 12 months.
. . . and Winbond is hit by reports
Toshiba will pull out of DRAMs
Taiwan's Winbond Electronics, and perhaps even the island's entire DRAM industry, may be in a pickle by Toshiba's reported decision to pull out of the DRAM market.
Because of the Japanese giant's changing direction, the Taiwanese DRAM vendor is pulling out its technical team from Toshiba's lab and has put its plans on hold to build a $3.5 billion, 300-mm wafer fab in the Hsinchu Science-based Industrial Park next year.
The move by Winbond, which licenses its process technologies from Toshiba, comes just days after Toshiba reportedly opened talks to sell its DRAM unit to Samsung and Infineon Technologies. Winbond says it is altering its deal with Toshiba after the Japanese company started negotiating a "new collaboration model and process technology strategy."
Some analysts took a negative view of the whole situation, raising concerns that Winbond would lose its ability to license next-generation technologies. "The most significant effect would be once Toshiba sells its DRAM business, Winbond will be in a tough situation since it has no ability to develop new technologies on its own and it's very expensive to license from somebody else," says Barro Liou, industry analyst with Prudential Securities in Taipei.
That could apply to the entire Taiwanese DRAM industry, as makers depend heavily on foreign partners for technology transfers. "None of the Taiwan players is sophisticated enough to have their own design capabilities," Liou points out.
Winbond should also brace for a sales decline, since half of its DRAM output is shipped to Toshiba, according to analysts. "Winbond has to look for new customers for its DRAM products," which account for 50% of its sales, says Alfred Yin, analyst at BNP Paribas Peregrine Securities in Hong Kong.
(See Aug. 29 story.)
It's tough out there now for
Japan's electronics giants . . .
It isn't that hard to figure out why Toshiba was shopping its DRAM business this week. The Japanese giant is in big trouble. It unveiled a sweeping reorganization plan this week that will mean 10% of its work force--or 18,800--will go over the next three years. And that's all in Japan.
That's going to hit the bottom line hard. Toshiba now estimates than it will have a net loss of nearly $1 billion for the year ended next March 31. That's nearly $2 billion away from the previous year's $800 million net profit. Sales for fiscal 2002 will fall 3% to $48 billion, Toshiba estimates.
(See Aug. 26 story.)
. . . as four of them slash
54,000 jobs, lose $ billions
Toshiba isn't the only Japanese electronics giant that's facing real trouble. In fact, it is the fourth to reorganize operations and chop its work force in recent weeks. NEC is slicing 4,000 workers, Fujitsu is reducing its staff by 16,400, and late this week Hitachi followed suit.
Hitachi, which called its massive cutbacks "emergency management measures," is cutting 10,200 jobs in Japan and 4,500 overseas. Some 2,000 of those jobs in Japan will come in semiconductor operations. The company says it is closing down a number of chip manufacturing plants in Japan and overseas, including its DRAM facility in Singapore called Hitachi Nippon Steel Semiconductor Singapore.
It appears that the Japanese electronics giant has all but come to a dead stop in this fiscal year's semiconductor capital spending. It is slashing this budget by 57% to $500 million from its current budget of $1.17 billion in the year ending March 31, 2002.
Hitachi now expects to post a net loss of $1.17 billion in fiscal 2001 instead of the net profit of $750 million projected just four months ago. That's a $2 billion swing! Sales this year are now projected to fall 10.3% to $65.4 billion from $72.9 billion last year. Those are tough numbers!
(See Aug. 31 story.)
Wilf figures IC bottom reached,
but U.S. will be off 40% in 2001
The good news, as far as Wilf Corrigan is concerned, is that the IC downturn appears to have hit bottom. But the bad news, adds the veteran CEO of LSI Logic, is that the IC market this year will be a disaster.
"The panic could be over," Corrigan says, because "a lot of our customers have told us that their business has bottomed." But he figures this year's global semiconductor business will fall "something north of minus 30% over 2000." And he notes, the market will be even worse in the U.S. The U.S. chip market could decline by about 40% over last year, he estimates.
Next year, Wilf is looking for an uptick. "I think it's fair to say the worldwide IC industry in 2002 could grow by 20%. But that will not be a cause to celebrate," he says. That "will just get us back to where we were the year before."
Wilf already is seeing initial recovery signs in some chip markets--especially in the storage and consumer segments. But he notes that communication chips remain soft.
(See Aug. 27 story.)
Sarnoff comes up with way
to shrink the unshrinkable
Sarnoff, the old Bell Labs in Princeton, N.J., has started licensing new technology it developed that will reduce by 30% the size of circuitry on a chip that hasn't been shrunk in years. And that would mean big cost savings.
"We've found a way to shrink . . . the I/O input/output and ESD electrostatic discharge protection areas that have been stuck at the same size for years," says Koen Verhaege, who is design director at Sarnoff Europe. "We reduce it at least 30%, while improving device performance."
Shrinking the I/O and ESD portions of a chip hasn't been keeping pace with the logic and memory of deep-submicron devices, notes Sarnoff. The labs' long years of experience in ESD and I/O design resulted in a brand new approach that increases electrostatic discharge protection by 30-to-60% while achieving at the same time the 30% cut in the silicon area needed for driver and ESD transistors, says Verhaege.
South Korea's Hynix Semi and Japan's Toshiba already have license the new technology called TakeCharge! Savings of $100 or more per processed wafer are possible now because chip; makers can place more devices on a wafer, the R&D outfit says.
The new technology enables designers to put $100 worth of additional devices on a wafer. And the savings is "often several times that figure, with no additional masks or process line changes," Verhaege claims.
(See Aug. 28 story.)
Nanoelectronic circuits may be
in production within a decade
Nanotechnology is this week's hot technology development. Research on atomic-level electronic devices got headlines this week when IBM scientists reported the first successful creation of a logic circuit within a single molecule. A few days earlier, researchers at Osaka University disclosed they had bulk-produced carpets of micron-long chains of crystalline globules and silicon dioxide stems that they predict may vie with nanotubes to produce next-generation semiconductor circuits.
IBM researchers predict their circuits could go into production in from two to 15 years, while the Japanese scientists predict a 10-year development time frame for their technology.
The creation of a logic circuit within a single-molecule carbon nanotube may lay the foundation for that technology to succeed silicon in electronic circuits within 10 to 15 years. Transistors had been created in carbon nanotubes, but functioning logic circuits has not.
"The ability to build a whole circuit on a single molecule has always been the holy grail of computing," says Phaedon Avouris, lead scientist on the IBM project. "This gives us hope that by the time silicon reaches its limits in another decade, we will be ready to bring nanotubes out."
The IBM team currently is focusing on nanotubes, which it says offer the potential to change computing, making; it possible to build switches just 5 nanometers across--about 100 times smaller than today's silicon switches. Chip makers could use these nanoscale circuits to produce higher-density electronic devices that consume less power and dissipate less heat.
IBM efforts, so far, have been aimed at proving out the concepts, and not geared toward the creation of devices in production volume. IBM plans to study the physics of nanotubes for no more than two more years, then it wants to move into development of methods to mass-produce such circuits.
The IBM researchers are certain that carbon nanotube technology offers the best alternative today to silicon. "There are physical limits to how small you can go with silicon, and we expect to reach those limits in about 10 to 15 years," Avouris notes. "Right now, this technology is by far the most promising of all the different molecular alternatives that have been proposed."
(See Aug. 28 story.)
Intel product plans could kill
standalone USB chip market
I guess some folks never learn. They forget about the 11th commandment of chip design. That sooner, rather than later, system chips will expand their functions to gobble up any nice little businesses that were making a market in supplying peripheral chips to go along with the central chips. I may be missing something, but I have always wondered why developers of graphics chips and the like figured they had a long-term business.
Of course, the biggest spoil-sport here is Intel. Back a few years ago, for example, the chip giant integrated the graphics functions into chip sets, a dramatic move that played a major role in the eventual shakeout and death of the standalone, graphics IC suppliers.
Now the world's largest chip maker is at it again. This week it revealed plans to integrate more functions on PC chip sets--to drive down the costs of personal computer platforms, of course.
Intel expects to integrate Universal Serial Bus (USB) 2.0 technology into its PC chip sets by the fourth quarter, according to VP Louis Burns at the Intel Developer Forum. The move obviously poses a big threat to suppliers of standalone USB chips, which includes a long list of vendors such as Cypress and NEC.
Burns made a big point of saying that the next challenge for its OEM customers is to leverage USB 2.0 and related I/O bus technologies rather than just focusing on the higher speeds of new CPUs. "We think the CPU side is done," he says. "The thing that needs work is I/O," he told developers.
There are several ways to boost the I/O in a system, he says, including the IEEE 1394 serial link, Infiniband-based communications, and Serial ATA connections for mass storage. Intel also is pushing its 3GIO bus, This "Arapahoe" chip-to-chip bus I/O technology designed to replace the exiting PCI architecture, which was just endorsed by Compaq, Dell, and Microsoft. And Intel is moving along here. "The final 3GIO spec will be out by mid-2002," Burns says.
(See Aug. 30 story.)
Moto claims industry's
first EUV photomasks
It looks like Motorola Semiconductor may have taken a big step toward moving chip making to extreme ultraviolet (EUV) lithography and making it possible to produce integrated circuits with features sizes less than 100 nanometers (0.10 micron).
Its researchers have fabricated what Moto claims are the industry's first EUV photomasks. The EUV mask sets were built with a "very complex IC design layout," the company says. The development project is being used to gain experience with EUV photomasks, which will start being used with a prototype EUV litho tool early next year. . "Our objective is to develop proficiency in the fabrication of these masks. Ultimately, we expect to transfer the processes to mask suppliers," says Joe Mogab, a Moto lab director.
Advanced Micro Devices and others are working with U.S. Dept. of Energy's labs to accelerate the development of EUV as a next-generation lithography technology. The Extreme Ultraviolet LLC consortium has developed a working alpha tool using EUV.
Industry executives have been urging the consortium to speed up its timetable for manufacturing tools from its current target of 2005 to 2003 or 2004. Moto expects EUV lithography systems to be ready for manufacturing by 2005.
Moto researchers demonstrated a mask patterning process that's suitable for current industry-standard reticles--substrates 6-inches square and 0.25-inch thick. The masks were produced with an extension of today's wafer-processing tools.
(See Aug. 29 story.)
Hell-bent to take over market,
Samsung shows 1-gigabit flash
Samsung Semiconductors, long considered a "one-note Johnny" with its huge DRAM business, continues to beat its marketing drums harder for its flash memory business. Now it says it is shipping engineering samples of the world's first 1-gigabit NAND flash memory device, a key product in its grand strategy to take over the lead in the global flash market.
The potential low cost of its 0.12-micron NAND flash device, Samsung says, should give it a big advantage over emerging nonvolatile memory technologies such as polymer ferroelectric RAMs (PFRAMs), Ovonics unified memories (OUMs), and magnetic RAMs (MRAMs).
NAND flash memories, which have fizzled this year by dropping an estimated 28%, are due to resume their meteoric growth next year, analysts figure. Semico Research predicts that revenues for NAND-based flash memories will grow 138% to $2.74 billion in 2002. By 2004, NAND flash memory will shot up to $8.25 billion, according to the Phoenix market researcher.
Samsung predicts mass production of its new 1-gigabit flash will begin early next year for such applications as Windows NT-based computing devices, third-generation (3G) cellular phones, personal digital assistants, digital music systems, such as MP3 players, and digital cameras.
(See Aug. 30 story.)
TSMC changes mind and is
ready to build fab in China . . .
Looks like foundry giant Taiwan Semiconductor Manufacturing sees the Chinese writing on the wall.
While chairman Morris Chang said just weeks ago that TSMC had no plans to invest in China over the next 3-to-5 years, he acknowledged this week the foundry would be ready to launch its first operations in China if the Taiwanese government gave permission. "The time is coming very soon" for TSMC to start investments in China, said Chang, widely regarded as the chief architect of Taiwan's IC industry.
TSMC increasingly is feeling the competitive heat from multinationals such as NEC and Motorola, which are forming joint venture or building fabs in the surging Chinese market. There already are 14-or-so fabs either under construction or reading to break ground in China.
"China offers not only tax incentives, but high-tech personnel, land, water, electricity, and above all, a huge market," Chang said. "Even if we don't use these resources, our competitors will, undermining TSMC's leading edge."
"China has great potential to become a market as big as the U.S.," declares Rick Hsu, analyst with Nomura Securities in Taipei. "If TSMC and UMC don't get started soon, they will have a big price to pay years down the road."
(See Aug. 27 story.)
. . . and Cirrus Logic gets
China bug, signs with CSMC
Fabless chip vendor Cirrus Logic also sees China in its future. This week it signed a five-year foundry deal with China's Central Semiconductor Manufacturing to fab CMOS-based mixed signal consumer ICs initially for the Austin company.
No word on why Cirrus pulled out of its existing contracts, except the move would give it an edge in China's fast-growing market. "CSMC strengthens our consumer entertainment strategy by adding a significant local manufacturing resource for us in the China market," says Jerry Gray, VP for Cirrus Logic.
Four-year-old CSMC is based in Hong Kong, and its 6-inch and 5-inch fabs are located in Wuxi, China. The two firms expect China to become the second largest market in the world by 2010.
CSMC, which claims to have pioneered the open foundry model in China, says that it has racked up a fab utilization rate of more than 80% for the first half of this year, more than double that of the Taiwanese foundry giants.
(See Aug. 30 story.)
New DDR-like DRAM
pushed for JEDEC standard
Intel, which certainly has had its troubles with new generations of memory Rambus comes to mind, has been working hard with a secretive group over the past two years to come up with a new DRAM standard.
Intel and leading memory chip makers are now circulating to PC makers the specifications for a double-data-rate-like DRAM that the group hopes will succeed Rambus and DDR-2 memories. This Advanced DRAM Technology group will propose the spec to the Jedec standards body by as early as September.
OEMs have asked that ADT's technology be made a standard, Tabrizi said. "OEMs want to see ADT in Jedec," says Farhad Tabrizi, marketing VP for Hynix and an ADT group member. "The key to success is making it an open standard, and we believe that means having it become a Jedec standard."
ADT hopes to merge its new spec with Jedec's DDR-2, Tabrizi says. "It's similar to DDR-2 with some minor modifications to allow more headroom," he says. "We've incorporated some new design techniques so that we can go to much higher frequency than DDR-2." The goal is to drive the new DRAM technology into mainstream PCs starting in 2003 or 2004, Tabrizi says.
"I definitely don't want two memory standards," comments Fred Leung, marketing VP for Acer Laboratories, one of three chip-set vendors advising the ADT group. "Otherwise the memory controller will get too complex, or I would have to offer multiple versions of the chip."
Jedec has yet to hear ADT's proposal, but leading DRAM makers such as Elpida, Hynix, Infineon, Micron, and Samsung, who all participate in Jedec, also are working with Intel in the ADT group to develop what they hope will become a new standard. Unlike Jedec, the ADT group restricted the number of participants and kept quiet about its work. This was done to streamline the standards-making process and avoid patent conflicts, says Tabrizi.
However, memory chip vendors already are planning to sample DDR-2 devices by year end, and the new ADT parts aren't expected to arrive until 2003 at the earliest. And if Jedec declines to replace its current DDR-2 with ADT or creates a separate standard for ADT, it will likely cause more fragmentation that could further confuse the issue.
(See Aug. 29 story.)
We welcome your feedback, comments, criticisms, or questions. E-mail us at firstname.lastname@example.org. Have a great weekend!
(Click here for last week's Semiconductor Alert!.)