Hello from Down-East Maine, where heavy rain finally arrived this past week, ending our year-long drought. Leaves have started turning--we already have the flaming reds along with many shades of orange. The magnificent scenery and magnificent, early fall days help to get us through this critical period. Up here, it's still hard to accept the fact that our way of life was changed perhaps permanently by the events of September 11th. But Americans are united and strong--adversity always brings these attributes out.
Tomorrow night is the annual fall bean supper of the Robbinston Historical Society, held in the 180-year-old, Ridge Methodist Church now owned by the society.You're all invited. For six dollars, you can fill up on homemade beans, macaroni and cheese, salads, and all kinds of desert I recommend my wife's apple pie. As you can see, this is not your Atkins-friendly meal--this is stick-to-your ribs, old-fashion eating, Maine & Nebraska Cornhusker style.
fabless ASIC business . . .
Intel is trying to get a bigger piece of the semiconductor pie than it already has--it's getting into the fabless ASIC business. The chip giant has set up a new business--called Intel Microelectronics Service--to provide fabless ASIC, global logistics, and other allied services.
To provide these comprehensive services, Intel has forged some powerhouse alliances with EDA software companies, intellectual-property (IP) providers, silicon foundries, and IC-packaging and test providers. Among its partners are Synopsys, Chartered, TSMC, UMC, Amkor, ASE, and ChipPAC.
The whole idea of the radical turnkey service is to help IC makers and OEMs bring their application specific integrated circuits (ASICs), application specific standard parts (ASSPs), or system-on-a-chip designs to market faster. The new service will concentrate on communication chips.
"This is a recognition of the outsourcing model in the industry," says Craig Peterson, co-general manager of the new service. Third-party companies are fabricating and packaging a fast-growing number of chip products.
Intel won't use its own fabs, but will work with foundries such as Chartered, TSMC, and UMC. The chip company will manage every aspect of the IC design and production cycle, as well as distribution for its customers, Peterson says. In other words, it will offer "one-stop shopping" of IC design, production, chip-packaging, and test at a cost that is equal or better than that charged by traditional ASIC houses such as IBM and LSI Logic, he says.
Intel already is developing ASIC designs for several customers, including TranSwitch, Peterson says.
So why else is Intel doing it besides expanding its base? The company may be looking to bolster its communication-chip business. The word is that it is still losing money and struggling in this market. Stay tuned.
(See Sept. 24 story.)
. . . But fabless ASIC
model still unproven
The idea of a fabless ASIC model has been kicked around for years, but there is little or no evidence that it will work in the marketplace.
While the fabless ASIC business is still in its infancy, it does has some advantages over the traditional ASIC model, according to Dataquest. "The key advantage of the fabless ASIC company business model is that it transfers a number of the fixed costs that are associated with a traditional ASIC company to third parties," says Dataquest. "The approach avoids the capital burden of owning costly fabs and packaging plants, developing process technology, creating intellectual property and tools," the researcher says.
So far, the embryonic fabless ASIC business has been dominated by smaller companies, such as Dialog Semiconductor and eSilicon. But "LSI Logic is moving towards a fabless ASIC model as well," says Dataquest analyst Martin Reynolds, "so it too is threatened" by Intel's new service.
He figures Intel has a good chance of succeeding. "Intel's fabless ASIC model seems like a good idea," the Dataquest analyst says. "You can get Intel to help develop your ASIC and they manage the entire project."
Reynolds believes that Intel's ASIC service could grow from "tens of millions of dollars in sales" for the company in the first year "to perhaps $100 million in revenues in the next couple of years." But the chip giant's new service is still unproven, he adds, so it could also face an uphill battle in "gaining the confidence of potential customers."
(See Sept. 24 story.)
ASICs won't suffer as much,
but won't rebound as fast
The application-specific integrated circuit (ASIC) market is expected to do better than the overall chip business this year, but it will not bounce back as fast.
At least that's the word this week from iSuppli. Global ASIC revenues will fall 22% in 2001, compared to a drop of 28% for the entire chip business, says Jordan Selburn, researcher at the El Segundo company.
But ASIC revenues will show a slower rebound than the total chip business, he says. There will be a small increase in ASIC demand before the summer of 2002, and a mild but steady increase in ASIC revenue the following year. ASICs will have a compound annual growth rate of just over 10% from 2003 through 2005, Selburn predicts.
"The current slowdown in ASIC strongholds, such as data processing, the Internet infrastructure, and other communications applications, will have a depressing effect on shipment of these devices for some quarters to come," he predicts.
"Historically, the electronics industry has designed its way out of a recession," Selburn says, "with ASICs as the core component. "ASICs will continue to be the key to next generation systems," he says, but "a number of factors are working to limit ASICs' role, with competing device, design, and fabrication models becoming ever more feasible."
(See Sept. 25 story.)
Why is Intel creating
new post of R&D czar?
Has Intel been unhappy with the direction and progress of its current R&D? If not, why did the company decide to name its first technology czar?
This week the microprocessor giant formed a new corporate-level group to direct its R&D, technology, and product efforts. Heading it up will be Patrick Gelsinger, former chief technology officer for the Intel Architecture Group.
He reports to CEO Craig Barrett and will ensure "consistency among Intel's emerging computing, networking and communications technologies. Convergence of digital computing and communications requires the coordination of many of Intel's technology and research efforts at a high level," Barrett explains.
Intel is not a company to add a new layer of management unless something needs to be fixed.
(See Sept. 25 story.)
AMD closes two fabs, cuts
15% of its work force . . .
So how much of Advanced Micro Devices' downsizing is due to the current chip recession and how important are its current market share losses? It's hard to tell from where I sit, but the current market appears to be largely responsible.
But no matter how you cut it, it is one giant restructuring of its manufacturing. AMD is closing its two oldest six-inch wafer fabs in Austin, eliminating 15% of its work force, or 2,300 jobs by the end of the second quarter of 2002.
The two 15-year-old fabs have been used mostly as foundries to produce programmable logic devices and communications chips based on 0.7-micron technologies for Lattice Semiconductor, which bought its PLD business in 1999, and Legerity, which was spun out of the company in 2000. Lower volumes of these devices no longer support these fabs, AMD says.
A realignment and restructuring of backend, chip-packaging operations in Penang, Malaysia, will result in 1,300 of the jobs cuts. The final assembly facility has been used to finish chips for Lattice and Legerity. AMD also is transferring final assembly of memory products to its facility in Suzhou, China.
This restructuring will enable AMD to concentrate on "our most promising opportunities--flash memory and PC processors," says CEO Jerry Sanders.
AMD will take a one-time charge somewhere between $80- and $110-million to cover these restructuring activities and other special charges in the current quarter. When fully implemented, the restructuring is expected to save the Sunnyvale, Calif., company about $125 million annually.
(See Sept. 25 story.)
. . . But what role did customer
defections play in AMD move?
Advanced Micro Devices confirmed this week that it has lost its single most important customer. Gateway will no longer use AMD processors in its PC lines.
AMD's Athlon processor will be phased out during the next two months as part of a plan to reduce the number of suppliers that Gateway is using in order to reduce manufacturing costs, according to the PC maker.
This means that Intel will be Gateway's sole source of microprocessors. It is not clear to me why Gateway would put itself in that kind of situation--relying on a single source for its most important PC part.
Gateway's decision to stop using AMD processors follows IBM's decision to stop selling its NetVista PC line in the U.S., which uses AMD's Duron processor.
(See Sept. 25 story.)
Atmel gives up this year,
lays off 26% of work force
It seems like many of the leading chip makers have finally given up on the chip downturn flattening out until later next year. So more drastic restructurings have begun. Like AMD has just announced, Atmel is closing two six-inch wafers fabs. It also will delay the startup of an 8-inch fab in the U.K., as part of its latest restructuring plan.
Atmel's new look will hit employment hard. About 26% of its workers in Europe and the U.S.--or about 2,500--will be laid off by the end of the year. Its goal is to reduce its manufacturing and operating costs by $150 million per quarter, or about $600 million per year.
"This is the most difficult business climate our industry has ever faced," declares CEO George Perlegos. "We do not expect a rapid improvement in market conditions, and accordingly we have decided to reduce our expense run rate to a level that should ensure positive operating profits by the first half of 2002."
(See Sept. 28 story.)
Micron gets flushed along
with rest of DRAM business
Even the best-run company can't do much when its primary market collapses. So when the global DRAM market plunged by an incredible 78% in July to just $600 million, Micron Technology fell right along with it.
So it shouldn't come as any surprise that the Boise company had an awful fiscal fourth quarter ended August 30th. It reported a net loss of $576 million on sales of $480.3 million, down 40% sequentially from previous quarter and 79% lower than a year ago. The main culprit was average selling prices, which dropped 55% sequentially and 85% from the same quarter a year ago. Wow.
But Micron, which may be getting used to the roller coaster ride in the memory business, is defiant. CEO Steve Appleton says that the DRAM giant would fight back and make investments of $1 billion in new technologies in its new fiscal year to be the lowest cost producer. This is down $800 million, however, from previous plans.
Micron, he maintains, has "one of the strongest balance sheets in the industry, an excellent complement of people resources, an industry leading process technology, and a resolve to emerge from these troubled times as the strongest semiconductor memory manufacturer in the world."
But the current market isn't looking any better. The back-to-school push in the PC market never materialized, Micron says. While its PC customers tell the DRAM maker that they expect a normal holiday season sales pick up, Micron is "taking it a day at a time."
(See Sept. 25 story.)
'Tensions' have developed
in microprocessor design
There are growing tensions today in deciding which way to go in microprocessor design and no one, not even Intel, has the answer yet.
Microprocessor designers have engaged in a decade-long debate over the relative merits of pushing frequency or instruction-level parallelism, known as the "speed demon versus Braniac" debate. But what is really needed now, says John Shen, head of Intel's microarchitecture lab, is both instructions-per-cycle and frequency. Getting this, he says, is a "really delicate balancing act." So the question really is, "what new and clever ways can we combine the two?"
To begin with, the trade-off between instructions-per-cycle and the increasing emphasis on microprocessor clock frequency needs a thorough re-examination, he says. Speaking to computer designers this week at an IEEE conference in Austin, Shen says that researchers are now at a crossroads, moving toward deeper pipelines with higher frequencies, a trend which could hurt instruction-level parallelism and overall processing efficiency.
Instruction-level parallelism requires a wider pipeline, which increases the complexity of each stage, rather than the thinner, multiple stages of a deep pipeline. "As you slice the pipeline thinner, you increase the latency and lower the instruction-level parallelism. There's a point at which you don't want to go."
Tensions have developed between each of the ways being developed to solve this problem, Shen notes. "There are tensions between hardware complexity and doing more in the compiler. It may take several years to develop new hardware. To recompile all of the existing software may take much longer."
Power consumption also may dictate the course that MPU design teams take, Shen says. "Forces are pulling us in various directions, and it is not clear what is the obvious path."
As processors evolve over the rest of this decade into huge chips incorporating several billion transistors operating at frequencies of 10-to-30 gigahertz, power consumption will increase exponentially, he says, and this will put greater emphasis on processor efficiency rather than brute frequency increases.
(See Sept. 25 story.)
TSMC blames terrorism
for missing growth targets
It's growing increasingly difficult to make sense out of Taiwan Semiconductor Manufacturing's comments on the state of its business. Just days after telling investors that its operating income would jump dramatically in the months ahead, TSMC now says it will probably miss this year's profit forecast as the terrorist attack on the U.S. dashes hopes for a fourth-quarter recovery.
The world's largest silicon foundry was pinning its hopes on the Christmas season to help it to achieve its $746 million net profit target. With only one-third of this forecast achieved through June, TSMC says it won't be able to meet the full-year forecast.
The terrorist attack will "prolong the bottom of the global IC industry, and any significant pickup won't come before the first quarter of next year," TSMC chairman Morris Chang said recently. But I can't quite figure out how the September 11th assault on New York and Washington, D.C., could be held responsible for TSMC missing its revenue and profit targets.
(See Sept. 25 story.)
FlexICs' exciting new technology;
but startup faces major challenges
A tiny Milpitas, Calif., startup is laying claim to being the world's first and only "semiconductor-on-plastics" foundry.
The two-year-old company, which announced a pilot production facility in June, counts several financial backers, including giant Intel. Despite the "trailing-edge" fab gear and processes that are used to make these products, FlexICs technology is new and "revolutionary," claims CEO Magnus Ryde. The startup has developed a ultra-low temperature semiconductor process that can fabricate ICs on conventional plastic at temperatures of less than 100 degrees C.
FlexICs hopes to develop plastics-based, lightweight displays, which would be integrated with liquid-crystal display (LCD) driver ICs, memories, or other chips on the substrate, Ryde says. He says such products could be used in cellular phones, instruments, hand-held devices, lighting systems, PDAs, and even wall TVs. "Our first market will be flat-panel displays," he adds.
FlexICs faces major challenges in bringing its new technology to market, say analysts, but the company intends to shipping in October samples of the world's first semiconductors fabricated on six-inch plastic substrates using 2-micron technology.
The technology is exciting, but one wonders what chance a tiny, new company has in surviving in the market today against the established industry giants. Not much, I'd say. "Initially, people were highly skeptical about our technology," Ryde admits. But "over the past three or four months, the market has changed."
"It might be a promising technology, but they still have an uphill battle," says Dataquest analyst Martin Reynolds. Other technologies hold more promise, he says, including organic light emitting diodes (OLEDs), which are thin-film, light-emitting devices. They are now constructed on glass, but could also be fabricated on plastic or other flexible substrate films. But FlexICs claims that ICs on plastic substrates are superior to OLEDs as far as electron mobility is concerned.
(See Sept. 26 story.)
Chip gear biz may have bottomed,
but PC chips worst than expected
The semiconductor market was weak before the recent terrorist attacks sent consumer spending into a tailspin, but the signs are still there that the capital equipment market may have reached bottom.
The semiconductor capital equipment market may not be hit as hard as the chip sector by the September 11th terrorist attack, primarily because those companies have already battened down the hatches to weather this economic storm.
"The equipment companies are already running their businesses at maintenance levels," notes Eric Chen, JP Morgan's chip analyst, after they were forced earlier this year to cut costs and prepare for a year of slow sales. Now that the economic typhoon is hitting the rest of the high-tech market, he says, the gear vendors are ready for it. "For companies that sell into the capital spending markets, the financial impact of the recent tragedy may be smaller," he says.
Impact of the Sept. 11 attacks may not be felt immediately in the chip sector, he says. Sales figures for the third quarter will not be too far off target, he predicts, and the impact on the supply chain turned out not to be very significant.
Chen expects chip sales next year to remain flat, with a sluggish first half and some modest recovery in the second half. "The impact on chip revenue and earnings from the September 11th tragedy is likely to be worse for people who are serving the consumer directly," he believes.
One example is the PC business. Eric Ross, analyst for Thomas Weisel Partners, has just lowered his PC market forecast partly because of a due to a very weak back-to-school season. Instead of an earlier forecast of 33 million units, He now expects 30 million PCs to ship this quarter, down 10% from his original forecast. For the entire year, Ross is dropping his estimate for all of 2001 from 136 million units to 124 million units.
Motherboard vendors say back-to-school PC sales were disappointing, he says, and Intel's price cuts for its Pentium 4 processor didn't seem to have much impact on PC sales. "Back-to-school demand appears to have been slower than our earlier expectations."
(See Sept. 24 story.)
Applied's Etec plays catch up
with e-beam photomask tool . . .
It was nearly two years ago when Applied Materials decided to make a run on the e-beam photomask systems business. It did it the easy way, swapping its stock for that of Etec Systems, world's largest supplier of e-beam photomask gear.
But things didn't go quite as planned. For the past year, Etec has been losing market share to some of its rivals, notably Toshiba. One reason why: The Hayward, Calif., company has had major delays in introducing new products.
These product delays caused the company to miss a critical market window, Etec execs acknowledged. "We missed the 130-nm development window," says Howard Neff, president of the Applied subsidiary.
But now Etec hopes to make up for lost ground. It is rolling out a new electron-beam-based, reticle-writing tool for 0.10-to-0.07-micron IC processes and is entering the emerging plasma-etch photomask equipment market.
The photomask-writer tool, developed with International Sematech, is the world's fastest e-beam system in terms of overall throughput, Neff claims. This eXara tool can produce a photomask in less than seven hours, something it takes competitive tools up to 20-to-40 hours.
But industry observers say the tool--called eXara--is a year late in getting to market--a delay that has opened the door to such e-beam rivals as JEOL, Hitachi, Leica, and Toshiba.
(See Sept. 27 story.)
. . . As photomask-gear market
keeps growing at 20% annually
Despite the current IC recession, the worldwide photomask-equipment business is expected to grow 21% from $578 million this year to $698 million in 2002, says Etec Systems. And by 2005, this market will hit $1.1 billion, predicts the Applied Material subsidiary.
The mask-writer tool segment of this market, including e-beam and pattern-generation tools, is expected to jump 23% from $292 million in 2001 to $360 million in 2002, the company says. By 2005, this market alone will reach $489 million, Etec projects.
(See Sept. 27 story.)
First 576-megabit RDRAM is
price competitive with SDRAM
The DRAM business may be in the toilet, but that doesn't stop the flow of new products. Samsung Electronics sounded the trumpets this week, claiming the industry's first 576-megabit Rambus DRAM.
The South Korean company, which doesn't plan to begin mass production for at least six months, says its new memory is built with 0.12-micron design rules and operates at 1.066-gigahertz, nearly eight times faster than the PC133 synchronous DRAMs now on the market.
It's not talking prices yet, but Samsung claims that the 0.12-micron design rules have "greatly removed" the cost penalties normally associated with RDRAM architecture. Helping matters is the larger die size required for the Rambus design was shrunk to less than 1% more than competing SDRAM-based memories.
The high-performance RDRAM will be targeted at high-end PCs, workstations, and servers. And it is expected to find its way into applications for digital video and high-performance graphics systems.
(See Sept. 27 story.)
Americans don't like debt-equity
swap in Hynix restructuring plan
Guess what? Yet another meeting of creditors will be held next week to wrestle with the proposed rescue package of Hynix Semiconductor. A lot of speculation is going on now in Seoul on whether new proposals will be put on the table that would make the debt restructuring plan more palatable to U.S. interests.
The debt-ridden DRAM giant confirmed a week ago that a $2.3 billion debt-to-equity conversion and the rollover of $1.9 billion in loans had been approved by creditors. The plan also calls for Hynix to sell $770 million in additional shares in a new stock offering.
The debt-to-equity swap is strongly opposed by the U.S. Treasury Department and U.S. Trade Representative, as well as Hynix's U.S. rival, Micron Technology. The Idaho DRAM maker is calling for U.S. agencies to file a complaint against Korea with the World Trade Organization, if the debt-to-equity swap proceeds.
The Americans believe that the swap violates Korea's pledge not to bail out domestic chip makers using local banks, many of which controlled by the Korean government. The banks themselves were bailed out in 1998 using public funds as part of an International Monetary Fund rescue package.
The Hynix debt breaks down as follows: Korean banks owed $4.7 billion, foreign investors owed $1.3 billion, investment trust firms carrying $915 million, and non-banking financial firms with $480 million.
(See Sept. 27 story.)
We welcome your feedback, comments, criticisms, or questions. E-mail us at email@example.com. And remember: God bless America!
(Click here for last week's Semiconductor Alert!.)