Greetings from Down-East Maine, where the economic signals are pointing in all directions. On the positive side, a small, but growing band of chip companies are beginning to see daylight four of them are discussed below.
But on the negative side, inventories don't look to be in as good a shape as some experts had thought. And layoffs continued this week with IBM Microelectronics eliminating 1,000 jobs--or nearly 5% of its total work force. Even The Wall Street Journal gave the industry "a kiss of death" this week by writing a big story suggesting that the chip business is turning up. I still feel the same way I did in January--the chip business won't be rolling until the third quarter of next year.
Meanwhile, it is life as usual up here in the Maine woods. Our stores are crowded and seem to be doing better than expected. An amazing number of homes--some of them very modest--are turning into blazing displays of lights in the evenings.
And, thanks to the excellent fall we're still enjoying, my team of guys are still going full speed on the rehabbing of the 180-year-old home down the road we purchased last summer. The new el was insulated and cedar shake siding, a labor intensive job, is nearly completed. The kitchen got a new ceiling this week. The dooryard, which looked like "Ma and Pa Kettle's place" when we acquired the property this past summer, has been completely cleaned up--with 40 dump-truck loads of debris and brush hauled away. Indeed, a new lawn planted just two weeks ago is now turning bright green as it comes up.
Chip inventories are
higher than expected
The stock market may be getting too far ahead of the chip market. It now turns out that semiconductor inventories are not going away as fast as some people had expected and that could delay the recovery.
A new report from iSuppli concludes that excess inventories in the supply chain--between chip makers and the systems sales channels--are actually higher than a previous forecast of $5.9 billion, or 17-days worth, according to the El Segundo firm's market intelligence services.
The weak demand has pushed the timeframe for supply-chain corrections out to the first quarter of next year, iSuppli now says.
"Approximately 27% of the excess inventory in the supply chain was worked off during Q3," says Greg Sheppard, iSuppli vice president. "But this was less than we had previously projected." The slowdown was due primarily to "much weaker system demand in the consumer and enterprise sectors," which was worsened by the terrorist acts of Sept. 11th. This happened, he says, even though "the electronics supply chain actually lost about one week of shipments in September, as logistics shut down in the U.S. and between its trading partners after the attacks."
"Masking a far more serious excess inventory position," Sheppard points out, are financial write-downs by several manufacturers of commodity chip products--such as discrete transistors, analog circuits, standard logic, and memories. iSuppli now believes that "many of these products remain in the supply chain," he says, "so 'effective' excess inventory could be as high as $7.4 billion at the moment." That isn't good.
Nearly half the inventory in the supply chain continues to be held by contract electronics manufacturers, according to iSuppli. These assembly houses are holding these inventories for clients, "even though some major contract manufacturers were able to make customers move several billion in inventories back onto their balance sheets," Sheppard notes.
(See Nov. 27 story.)
DRAMs go south again, Hitachi
slices Singapore output by half . . .
After the brief run up in DRAM prices earlier this month, the DRAM market has turned very weak again. This unexpected softness is now causing a fresh round of problems for DRAM suppliers.
For starters, slice another 10,000 wafers a month out of global DRAM production. Hitachi caved in this week and cut production of 64-megabit and 256-megabit DRAMS by half at Hitachi Nippon Steel Semiconductor, its venture with Nippon Steel and EDB Investments--and its only fab outside Japan.
The Singapore fab had been running at 60%, or about 2,000 wafers per month. Hitachi also will cut 430 permanent employees at the plant, reducing the staff in Singapore to 550.
In October, the Singapore facility lost 300 temporary workers as part of the strategic plan by Hitachi to exit the commodity DRAM market and move into more profitable memory and logic products. The fab will now concentrate on producing AND-type flash memory for Hitachi's range of media cards and SDRAM.
Hitachi's latest DRAM cutback surprised observers, who say it shows just how bad this year's DRAM market is. "This move means Hitachi no longer has enough money to sustain foreign DRAM production," declares Soo Kyoum Kim, analyst at International Data in Seoul. "They've had more trouble than they expected.
(See Nov. 29 story.)
. . . And Fujitsu closes Oregon fab
because flash recovery is delayed
Flash memory may still look good enough for Hitachi to commit its old DRAM fab in Singapore, but the market looks so bad to Fujitsu that it is closing down its Oregon fab.
The Gresham plant, Fujitsu's first overseas wafer fab, was built in 1988 to crank out DRAMs. But the decline of that market caused the Japanese producer to convert the plant over to flash memory. The plant has been operating "well below capacity," due to the current downturn and the market "isn't recovering as fast as the company had expected," a U.S. Fujitsu official says. The plant will be shut down in January, throwing 670 employees out of work.
It also kills a plan for the Fujitsu-Advanced Micro Devices joint venture to take over the fab. Fujitsu was exploring the possibility of turning the Gresham fab into a flash production site for the venture, but the plan was dropped. "There were discussions for AMD to take over a piece of the fab," confirms the official, but "AMD decided not to do it."
Fujitsu is bearish about a recovery in flash memories. It now seems likely to be delayed by up to a year longer than the previously expected upturn of late next year.
(See Nov. 29 story.)
ASM Lithography flushes
SVG's 193-nm scanner
ASM Lithography, which coughed up $1.6 billion in stock in May to acquire Silicon Valley Group for its scanner technology and Intel's lithography business, has killed the major SVG platform that Intel had originally intended to buy for its production fabs.
After Silicon Valley Group delayed shipping the new tool several times, Intel decided to scrap its production plans to use the 193-nm, argon-fluoride (ArF) lithography tool, pulling the plug on what was thought to be a $100 million tool order.
Now ASML has decided to drop the Micrascan V 193-nm lithography system and concentrate all of its 193-nm wavelength technology and talent on a single platform--the Dutch company's dual-stage Twinscan AT:1100 system, which was introduced only last summer. It now expects to ship this product early next year.
Industry observers figure Intel is now considering ASML's 193-nm tool as a replacement for the ill-fated Micrascan V in next-generation production lines along with exposure tools from Canon and Nikon. Intel has repeatedly denied that Micrascan V delays have delayed its move to 0.13-micron processes, but some observers disagree, saying the lack of 193-nm tools has increased Intel's costs for new microprocessors.
ASML will transfer Twinscan manufacturing and some R&D programs from the Netherlands to Wilton, Conn., where SVG Lithography was based before its purchase by the Dutch company.
(See Nov. 27 story.)
Is chip industry running
out of process technology?
The semiconductor industry may run out of process technology faster than expected. Transistor-gate lengths are shrinking so fast now in microprocessors that technology options may be exhausted for additional shrinks beyond 2007.
That pessimistic outlook comes from experts who have just completed the latest edition of the International Technology Roadmap for Semiconductors. The roadmap shows transistor-gate lengths in microprocessors shrinking to just 25 nanometers (0.025 micron) by 2007, six years earlier than the 1999 edition of the roadmap projected.
This acceleration comes partly from chip makers going to post-lithographic process techniques to reduce feature sizes beyond the minimum printed geometries in photoresist. The smaller gate lengths speed up the microprocessors, which enables vendors to raise their prices.
The new roadmap sets industry technology targets and milestones for the next 15 years, updating the last official ITRS issued in 1999. The latest edition accelerates the shrinking of feature sizes in processors, memories, and ASICs from the 1999 roadmap.
By 2016, the new roadmap projects the minimum physical gate length of transistors to be just 9 nm (0.009 micron). This is just about where the industry figures it can extend MOS transistors.
The new technology roadmap also speeds up DRAM technology by a one year, compared with the 1999 schedule. The 1999 roadmap called for the 100-nm (0.10-micron) DRAM generation to move into production by 2005, but now industry experts predict that DRAMs will reach 90-nm, half-pitch feature sizes by 2004. Two years ago, the 70-nm DRAM was expected to go into production in 2008, but now the 65-nm DRAM is forecast to hit wafer fabs in 2007, according to the new forecast.
"Lithography half-pitch and transistor-gate length scaling trends continue to accelerate," says Paolo Gargini, an Intel fellow who chairs the International Roadmap Committee. The prospects of 9-nm gate lengths in microprocessors by 2016, he adds, is now causing the industry to "consider technologies beyond planar or even post-CMOS devices."
(See Nov. 28 story.)
3rd-quarter chip-gear orders
just plain awful--down 75%
No one on this planet was looking forward to seeing the third-quarter shipments and orders for global semiconductor equipment. But it was even lousier than I expected.
Shipments dropped 20.2% sequentially in the third quarter to $5.63 billion from $7.07 billion in Q2, according to data released this week by the SEMI trade group. And compared with the year-ago quarter, these sales were 55.6% lower. Not good.
But wait until you get to orders. Bookings for the quarter totaled up to $3.92 billion, a huge decline of 75% from orders in the third quarter last year and 12% below the previous quarter. But SEMI research head Elizabeth Schumann wasn't surprised. "The significant decline in capital equipment shipments and third quarter results for the worldwide industry have been anticipated," she declares.
Weakest results regionally compared to the second quarter came in South Korea (-37.7%), Europe (-28.9%), and Taiwan (-24.6%). Only the rest-of-world (ROW) segment showed sequential growth--rising 9.1% from the second quarter. This came mostly from Chinese spending. But the ROW segment was off sharply from the year-ago results, falling 54.7%.
(See Nov. 27 story.)
More bad news from Zilog;
it's now filing for Chapter 11
We figured last January that something had gone wrong between Texas Pacific Group and Zilog's CEO Curtis Crawford. Zilog had signed a merger deal three years ago with Texas Pacific, an investment group, which had then lured Crawford away from Lucent Technologies to take over as CEO.
We were right. The wheels were coming off the chip maker and management was fleeing the scene. Since then, Zilog has been working desperately to get back on track. It has tried to refocus its business on core products, rationalize manufacturing, lay off workers, and reduce operating costs.
Well, things seem to keep getting worse. Zilog is now preparing to file for Chapter 11 bankruptcy protection after striking a preliminary agreement with bondholders holding more than 60% of senior debt to restructure the debt on $280 million in senior secured notes. This latest move comes just a month after the company acknowledged it had not made an interest payment due in September on the senior notes.
The recapitalization of Zilog's debt was called a "prepackaged Chapter 11 filing." Under the plan, bondholders will exchange their notes for equity, plus a $30 million non-recourse note. "We have made significant progress in returning Zilog to full financial health," claims CEO Jim Thorburn. "This . . . gives us maximum financial flexibility to reinvest in the business," he says.
Thorburn figures the troubled company is now getting back on track. Its bookings in the third quarter were up 26% from the previous quarter and new orders exceeded product shipments for the first time this year, he says.
(See Nov. 28 story.)
Did Intel drop the ball
with one Pentium 4 part?
What's Intel up to these days with its Socket 478 Pentium 4? Shortages of the microprocessor has observers wondering whether the chip giant dropped the ball in production planning or if it is trying to get customers to use more expensive, high-end chips.
For the past two months, PC component makers in Taiwan have griped about shortages of the processor, which is mounted in a 478-pin, flip-chip pin-grid array with a heat spreader. Some observers say this allowed Intel to sell its inventory of the Socket 423 Pentium 4, which is housed in a 423-pin plastic PGA with a heat sink and rear case fan. Others speculate that Intel is having yield problems with its flagship product.
Shortages of Pentium 4s have also been reported in South Korea and in China, which represents a large Pentium 4 market. But it's beginning to look like the supply pinch is easing and should abate in January, or just as Intel begins to market its double-date-rate SDRAM chip set for the Pentium 4.
Intel may have missed an opportunity to boost sales of Pentium 4 systems with SDRAM, since it is the only supplier of a single-data-rate SDRAM chip set for the 478-pin processor. Some Taiwanese motherboard makers figure that the shortage cost Intel market share in October to competitor Advanced Micro Devices.
"I like to believe the conspiracy theories, but I think Intel is innocent on this one. They just didn't plan well," says Scott Thirwell, marketing manager for DFI, a Taiwanese motherboard maker.
Then there's the theory that says Intel simply wants its customers to buy the 1.7-gigahertz Pentium 4, which has less than an 8% premium over its 1.5-gigahertz counterpart. "This is part of a concerted strategy on Intel's part--it's not a yield problem. If people are still confused about that, they shouldn't be," says Tony Chen, analyst for Bank of America Securities. "The closer Intel can push people to the sweet spot of the market--2 gigahertz--the better it is for them."
(See Nov. 28 story.)
Surprise! Intel now likes
silicon-on-insulator . . .
For years now, Intel publicly dismissed silicon-on-insulator (SOI) technology out of hand. It took this stance even though competitors like IBM Microelectronics and Advanced Micro Devices were rushing to take advantage of this new technology for next-generation processor designs. But was Intel conning the industry again?
Could be. Intel shocked everyone this week by endorsing SOI technology for use in building a new transistor structure that also marks its first use of high-k dielectrics and epitaxial wafers. The new TeraHertz transistor structure was called the company's key building block for future, low-power microprocessors.
Intel execs deny the company is running behind IBM and AMD in SOI because it is skipping traditional SOI wafers and will move to a next-generation technology called "thin SOI" or fully-depleted SOI.
"We have always said that SOI was not useful for our current 0.13-micron generation chips," maintains Intel technology analyst Rob Willoner. "We believe that traditional SOI wafers cost many times more than bulk silicon--sometime five to six times the cost," he estimates. "Thin SOI" wafers may make it possible for Intel to lower the cost of this technology for mainstream chip applications, according to the company.
Willoner claims Intel never ruled out using SOI wafers despite dismissing the technology in numerous interviews.
"Thin SOI" has several advantages over competitive technologies, according to Intel. It prevents leakages through the chip substrate, lower junction capacitance, and reduce overall voltages in chips, the company claims.
(See Nov. 25 story.)
. . . And will use SOI
in new transistor design
Intel's new TeraHertz transistor is expected to become the company's key building block for the development of 10-to-20-gigahertz processors by the second half of this decade. And the company may even use some of this technology in its next-generation, 0.09-micron (90-nm) chips due out in 2003.
The big advantage will be overall chip performance, of course, but the new transistor architecture will also solve a fast-growing problem plaguing Intel and other processor suppliers--heat dissipation. "Power dissipation is getting out of hand," acknowledges Rob Willoner, Intel technology analyst. "The power issue is the number one challenge for processor design in this decade," he says.
Intel claims the TeraHertz transistor will keep power dissipation levels somewhat constant in future processor designs. To do this, Intel had to do what it called re-inventing the wheel in semiconductor technology. The TeraHertz transistor appears to be a souped-up conventional transistor with three new elements: chip-scaling technologies, high-k dielectrics, and a so-called "depleted substrate transistor."
To build this design, Intel will need SOI as an insulator and epitaxial wafers to "raise" or "thicken" the source and drain of the transistor. It will replace the gate-based silicon dioxide layer of a transistor with high-k materials.
To solve the power problems, Intel is having to re-engineer the transistor. Current transistors have many problems--they're subject to excessive gate and transistor leakages, off-state leakages, and soft-error rates, according to Willoner. While Intel claims it has the technologies in place to come up with the TeraHertz transistor, it still faces major hurdles ahead. "We need several years before all of the bugs are worked out," figures Willoner.
(See Nov. 25 story.)
Why am I not surprised
18-inch wafers will be late?
After writing or editing at least 493 stories over the past several years on the chip industry's expensive and excruciating slow shift to 12-inch wafers, it surprises me not at all that the move to the next size is going to take a lot longer than many have predicted.
A couple of years ago, a senior executive from International Sematech predicted chip makers would move from 12-inch wafers to 18-inch pizza pans around 2008. Now, it looks this shift will not take place until 2013, or five more years, according to the new 2001 International Technology Roadmap for Semiconductors.
Analysts say the 450-mm transition has been pushed out for good reason. The current switch from 200- (8-inch) to 300-mm wafers has been slower and more painful than expected, they say. In the late-1990s, several chip makers were scrambling to build 300-mm fabs to reduce manufacturing costs. And chip-equipment vendors followed suit by pouring billions of dollars into the development of new 300-mm tools.
But the current IC downturn, coupled with aggressive die shrinks, forced most chip makers to delay their 300-mm fab projects, leaving equipment vendors in the lurch. A good many of the production equipment makers had developed 300-mm tools back in the late-1990s, but even today only a handful of chip makers are in pilot production at their 300-mm fabs now. Can you imagine the cost of a 18-inch wafer fab?
(See Nov. 30 story.)
Fairchild still sees earnings
improving now through 2002
The chip market picture isn't changing for Fairchild Semiconductor--and these days that's good news. The Maine chip maker is still expecting its fourth-quarter revenues to be flat to down 5% sequentially from the $325.4 million reported in the third quarter, but it believes a recovery is already underway.
"Our turns bookings were strong in October and have slowed in November as the seasonal demand winds down," says CFO Joe Martin. "Quarter-to-date book-to-bill ratio continues to be slightly above 1:1."
Fairchild's 26-week backlog has continued to grow throughout the current quarter. "While we expect our book-to-bill ratio may drop below 1:1 during December and January as a result of normal seasonal reductions in demand . . . we are already building backlog for first quarter 2002," Martin says. "We have seen a slight uptick in pricing in October, mostly due to a stronger mix of power, interface, and 'TinyLogic' product sales."
As a result, he says, Fairchild is "on track to meet our guidance of improving gross margins by 200 basis points this quarter." The company believes it will "turn the corner" on earnings performance in the fourth quarter, Martain says, and expects "improved results sequentially in this quarter and throughout 2002."
(See Nov. 28 story.)
Cirrus gives Chartered
its mixed-signal business
Here's something that you'll be seeing a lot more of. Cirrus Logic has picked Singapore's Chartered Semiconductor Manufacturing as its foundry for mixed-signal ICs under a new multi-year manufacturing agreement. Under the new pact, Chartered will supply Cirrus with a range of processes, from 0.35-micron to 0.10-micron technologies for products aimed at consumer applications such as video, set-top boxes, and Internet gateway devices.
"This relationship gives Cirrus Logic a leading mixed-signal process platform," says CEO David French. And it helps to fill up Chartered's fabs, which have been running at less than 25% capacity in the current industry downturn. "This agreement goes beyond capacity-it's about providing complete solutions that address specific needs of our customers," maintains Barry Waite, CEO at Chartered.
(See Nov. 27 story.)
New orders brighten
TSMC's 4th quarter
Some good news out of foundry country for a change. Taiwan Semiconductor Manufacturing is raising its forecast for the fourth quarter because of a surge in orders. TSMC had been expecting a 15% sequential increase back in September, but this week it hiked before-tax income to more than 20% higher. This would give it a fourth quarter gross profit of about $271 million.
The increased forecast comes from greater use of TSMC's more sophisticated process technologies by such customers as Nvidia, which supplies core circuitry for Microsoft's new Xbox game console, according to a TSMC spokesman.
A few OEMs like Nvidia and Via Technologies have pushed PC-related chips to 49% of overall TSMC sales. The Taiwanese company also estimated that nearly 50% of its capacity will be in use in the fourth quarter, up from 41% in the third quarter. The first few months of next year should be about the same as this quarter, TSMC believes.
(See Nov. 26 story.)
Another vote
for rising revenue
Here's another brightening outlook. ESS Technology has bumped its revenue estimate up 3-to-8% for the fourth quarter. It is now looking for a range of $74- to $76-million vs. its previous forecast of $68-to-$74 million, compared to third quarter results of $72.4 million. It is more optimistic due to the strong demand for chips used in DVD video players.
"We believe our fourth quarter DVD shipments will set another record," declares CEO Robert Blair. He also expects profit margins and earnings to exceed earlier estimates in the fourth quarter. Gross margins should top the company's previous prediction of 37%-to-38% as higher margin DVD business continues to go up. ESS also expects net per share to run between 22 and 24 cents in the fourth quarterup from its previous guidance of 17-to-22 cents a share. ESS posted 28 cents a share in the third quarter.
(See Nov. 26 story.)
Conexant Systems joins
good news bandwagon
Business also is looking up for a chip maker that has had almost nothing but bad news to report for many a moon. Conexant Systems has upped revenue estimates for its quarter ending Dec. 28th to a 5%-to-7% increase from the $201 million it reported in the prior quarter.
Reason for the upgrade are stronger sales from the Newport Beach company's wireless communications division. Wireless product revenues are now predicted to grow sequentially by 25% in the quarter, which follows a 24% sequential rise in the quarter ended Sept. 30th. "This is the result of strong demand across our wireless product portfolio, including power amplifiers, complete GSM cellular system solutions, and radio frequency subsystems," says CEO Dwight Decker.
(See Nov. 29 story.)
Intel builds set-top
market for its flash
Leading flash memory vendors saw this business crash and burn this year because they were selling most of their product into the sagging cell phone market. So major job this year was to develop new flash markets.
One flash supplier that seems to be successful in doing just that is Intel, which has had its flash memory technology adopted by four major set-top box OEMs. Already, set-top boxes are Intel's second-largest flash revenue generator.
The four new Intel customers--Scientific-Atlanta, Motorola Broadband Communications, Thomson Multimedia, and Hughes Network Systems--are using a multi-level cell flash technology called StrataFlash.
This business gives Intel a lead position as a flash supplier in the set-top box segment, according to Curt Nichols, vice president of Intel's flash group. "Two years ago we didn't sell to any of these people," he notes. "In the past 24 months, StrataFlash has opened the door and they are all using it."
"It's the flash memory that allows the set-top box to be reprogrammed in the field, which extends the service life of the box," he notes. "As a non-volatile memory, flash retains the program features, whether code or data settings, even if power is interrupted."
But it will be a long time, if ever, before set-top boxes challenge cell phones as a flash business. Cell phones, with estimated sales of about 400 million units, will continue to be the dominant market for flash memory, says Jim Cantore, analyst at iSuppli. Set-top boxes, by comparison, will only reach about 40 million units this year, he adds.
(See Nov. 27 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!.)
