I got up early this morning--about 6 am--and ran into one of our more beautiful "Sunrises over Campobello." The sun was just rising over the hills of New Brunswick and the Canadian offshore islands and it made beautiful patterns in the low clouds streaking along the horizon and over the Passamaquoddy Bay. Walked out to pick up the Bangor Daily News and the Ravens and Gulls were making a major racket--most likely it was over a fox or herd of deer that had invaded their domain. Just had a little excitement--you city folks don't have this kind of problem. A nearby thunderstorm just knocked out my computer, my telephone, my fax--so I just went outside and took a walk. Now the power is back on and I can finish my column and get it out of here before the power goes off again. Actually, we have a backup generator that we can turn on if the power doesn't come back. So my column should always get out--no matter what the weather is or what Eastern Maine Electric is doing. Have a great weekend.

Can vendors wait 3-to-4 years
for volume market to develop?

That noise you heard coming out of San Jose this week was just the marketing hype for a new generation of network processors at--you guessed it--the network processors conference. But you have to wonder about the timing here, given the sad state of the communications business and the fact that the current generation of network processors is expected to dominate this market for the next couple of years.

But everyone apparently has their eyes on 2004 and 2005, a period when the market may be ready for these 10-gigabit processors. Analysts at Cahners In-Stat figure the network-processor market will jump from $135 million in 2002 to more than $7 billion by 2005 as they replace costlier ASICs in systems. Not too shabby--if it happens.

More than a half dozen chip makers launched the new network processors, including Agere, AMCC, EZChip, IBM, and Mindspeed. At the same time, IDT, NetLogic, and Silicon Access, introduced packet-processor devices based on content-addressable memory (CAM) technology.

Another reason why these parts will be slow in coming is that the design cycles for targeted systems are long and painful. These devices are not expected to be incorporated in next-generation networking equipment until 2004 or later.

Such high-performing MPUs have been promoted in the past. "There have been many, many promises made by microprocessor vendors--big and small alike--that have offered OC48 and OC192 line-speed performances with a simple MPU, but they cannot get there from here," points out VP Rick Kepple of Integrated Device Technology.

(See Oct. 22 story.)

Network processor vendors
deliver less than they promise

Slowness of OEMs to design in the network processors is not because of any downturn, but is due instead to chip suppliers not living up to their promises. That was the message coming this week from one of those OEMs, Cisco Systems. This failure, analysts contend, has pushed out the mass adoption of network processors, forcing OEMs to continue relying on costly ASICs.

While Cisco uses a few network processors in its equipment now, the vast majority of its systems are based on its own, internally-developed network-processor. Called Toaster, the chip is an ASIC that handles the data- and forward-packet processing functions. "ASIC designers have done a good job in developing parallelism in our packet processors," says Cisco technologist Steve Fu.

The San Jose networking equipment giant would like to procure more merchant network processors, especially for its low-end to mid-range equipment, because they hold out a "huge potential" for cost reduction. But, there are problems with today's network processors. "Instead of focusing on bandwidth speeds like OC-192 and OC-768, network-processor vendors should focus more on features," Fu says. They need to bundle more services on these devices, he adds, such as Ten Gigabit Ethernet and MPLS.

Some chip vendors claim their processors chips can achieve OC-48 transmission rates, but it ends up that they can only support slower OC-3 and OC-12 speeds in the real world, he claims. "We need more headroom," Fu says. "Network processors need to be more programmable, the software-run cycles are too long, and there is also a lack of mature software tools." I guess he told them.

(See Oct. 24 story.)

Equipment glut to stymie faster
network processors for 3 years

Even if network processor vendors could deliver what they promise, the glut of capacity in long-haul backbones could push out the need for higher-speed devices for another three years.

Over the past few years, network-processor vendors have rolled out parts that run at speeds up to 2.5-gigabits-per-second (OC-48). Many of these devices are integrated in switches, routers and other equipment, which are being deployed in backbones and metropolitan-area networks.

But new generations of network processors--10-gigabits-per-second devices (OC-192) and 40-gigabits-per-second (OC-768)--are designed for backbones or long-haul networks. And these systems are "overbuilt" right now, declares Armando Garcia, IBM Micro VP. "Only 25% of the fiber is lit in the worldwide market," he notes.

"We don't see OC-192-enabled network processors taking off until 2004," predicts Garcia. While he sees a good business for network processors in slower-speed switches and routers, it's not enough to cause the mass adoption of these devices in the marketplace, he says. To get things moving, Garcia says, the network processor vendors will have "to step up and move towards standard interfaces and applications."

(See Oct. 24 story.)

Microchip looking better;
sequential sales, orders up

There was a real overload this week in grim financial news--particularly from the Far East. And it's been going that way with this month's flood of third quarter reports. But a small but steadily growing number of positive signs are beginning to emerge that indicate the bottom of the slump is at hand.

Microchip Technology reported a 2% sequential increase in net sales to $141.7 million for the quarter ended Sept. 30th, although they were down 27.2% from the year-ago period. Net income totaled $23.1 million, down 50% from a year ago.

But the trend was up. "Orders were up 15% sequentially with the book-to-bill ratio being very close to parity," says CEO Steve Sanghi. For the December quarter, Microchip expects revenues and net to run about flat with its September quarter.

"Our second quarter results were at the upper end of the original guidance we gave in our first quarter earnings release in July," Sanghi notes. "The performance of our proprietary products was especially encouraging," he says, "with sequential revenue growth in both our microcontroller and analog product lines running 3% and 18%, respectively."

(See Oct. 24 story.)

Fairchild's book-to-bill up,
as order bookings rise 23%

Fairchild Semiconductor not only outperformed the industry in sales performance vs. a year ago, but it also recorded an increase of 23% in order bookings in the third quarter vs. the previous three months.

The Maine chip maker showed a 12.6% sequential drop in revenues to $325.4 million in the third quarter and posted a net loss of $19.1 million, including restructuring charges.

Fairchild revenues were down 32% from the third quarter last year, while industry sales were down more than 45% for the same period, says CEO Kirk Pond. New products helped. "We introduced more than 145 new products this quarter," he notes, "raising our total through nine months to more than 385, well ahead of our 2000 run rate."

The increase in product books pushed Fairchild's book-to-bill above 1:1, Pond says. "Our turns business, which is comprised of orders for shipment within the same quarter, was unusually strong," he points out. "We turned 28% of our third quarter bookings into revenue during this quarter."

Pond is optimistic about growth prospects. "My experience through several semiconductor cycles has been that a combination of high turns, an improving book-to-bill ratio, low backlog visibility and aggressive pricing has always marked the beginning of the end of the downturn," he comments.

(See Oct. 23 story.)

Chip gear biz backs into
slightly better book-to-bill

New orders for IC production equipment received by North American suppliers fell again in September, but shipments dropped even more, meaning that the book-to-bill ratio for tool suppliers actually had a small increase. We'll take it anyway we can get it.

The September book-to-bill was 0.65, or that $65 worth of new orders were received by suppliers for every $100 of products shipped, according to the SEMI trade group. In August, the ratio stood at 0.63, not good but a lot higher than the book-to-bill low point in the current downturn last April, when it bottomed out at 0.44.

Year-to-year, September's tool bookings were 78% lower than the $2.89 billion in orders in September, 2000. Shipments dropped 60% below the $2.48 billion recorded in the year-ago month.

Bookings for semiconductor equipment dropped 11% in September to $644 million vs. August's revised $724 million, said SEMI. Worldwide shipments of systems by North American suppliers dropped 13% to $993 million vs. $1.15 billion in August.

Stanley Myers, CEO of SEMI, usually has some optimistic words on a turnaround to go with the monthly book-to-bill report. But not this time around. "Industry analysts have been fairly unified in pushing out the timing of recovery for the semiconductor and capital equipment industries," he notes. "It is still too early to determine the long-term fiscal effects of ongoing geo-political events," he adds.

(See Oct. 23 story.)

Agere loss results
knock my sox off

I never thought I'd see the day when a chip maker would lose $3.35 billion--in a year, much less three months. But Lucent spinoff Agere Systems accomplished this dubious achievement in the quarter ended Sept. 30th.

Yeah, I know, much of the loss was from taking a writeoff of goodwill and intangibles related to its acquisition of optical components maker Ortel. But the Allentown, Penn., company still had $500 million in operating losses and $153 million in losses from restructuring and layoffs.

Revenues in the quarter fell 35% sequentially to just $600 million. IC revenues dropped 28% sequentially to $469 million and optoelectronics sales fell 53% to $131 million. Grim stuff. Revenues in the quarter ending Dec. 31st will continue falling, but at a slower rate-about 10% from the Sept. 30th period.

ICs showed a pro-form loss of $173 million in the Sept. 30th quarter--a 50% bigger loss than the previous quarter showed.

But CEO John Dickson sounds optimistic, even after reporting such bad financial results. "We have lowered costs and expenses, as well as reduced our capital expenditures by approximately one-third from the June quarter," he says. "As a result, we exited fiscal 2001ended Sept. 30th with more than $3 billion in cash." I guess that's good.

(See Oct. 23 story.)

Are foundries going for title of
biggest boom & bust IC business?

It's beginning to look like the silicon foundries are challenging the DRAM companies and production equipment industry for the title of the industry's biggest boom and bust business.

Chartered Semiconductor Manufacturing, which was running flat out along with the rest of the big foundries a year ago, has watched output fall to just 22% of its capacity in the third quarter. Making the picture even worse is that the Singapore company expects little or no improvement for the rest of the year.

That is really bad news for profits. The world's third largest pure-play chip foundry has to run its fabs at least 70% of capacity before it reaches the break even point.

The best news the company could offer this week was that management believes it is approaching the bottom of the current slump. They expect revenues in the fourth quarter to be flat-to-down 5%.

(See Oct. 23 story.)

Intel is trying to increase its
'two-generation' lead in flash

I'm always surprised when some people wonder why a chip maker keeps building fabs during a period of overcapacity. Intel, for example, didn't cut back its huge capital expansion program this year.

The reason, of course, is quite simple. The new fabs can turn out denser products that outperform current generation parts, and cost less money as well. The older fabs just can't compete.

Intel, which has been going all out to add the latest 0.13-micron capacity for its microprocessors, is now moving another product line to the 0.13-micron process.

This week the chip giant introduced the world's first flash-memory devices based on 0.13-micron technology--a move that will ultimately enable it to develop chips storing more than 512-megabits. Its first flash devices are nearly 50% smaller and consume less power than the 0.18-micron chips Intel has been making. said Scott McCormack, product manager at the Santa Clara-based company.

Intel, which is the world's leading supplier of flash memories, now claims it is "at least two generations ahead" of its rivals in terms of process technology in this market. "Our goal is to ship 0.13-micron flash-memory devices before our competitors ships their 0.18-micron chips," brags Scott McCormack, flash product manager. "We want to break away from the competition," declares Scott Dunagan, another flash product manager.

By 2003, Intel figures that 50% of its overall flash-memory shipments will be based on 0.13-micron technology. By then, it hopes to ship the world's first NOR-based, 512-megabit device. It will be in production on the first 0.13-micron flash chips by the spring of 2002.

(See Oct. 23 story.)

Cisco doesn't pick Rambus
for next-generation products

Can we get an inkling of where the next-generation DRAM business is going by checking out the latest plans of Cisco Systems? Could be. The networking equipment giant now says that it will use new memory technology from Fujitsu, Infineon, and Toshiba, but not from Rambus.

It plans to use fast cycle random access memory (FCRAM) backed by Fujitsu and Toshiba and reduced latency DRAM (RLDRAM) from Infineon. "There are no plans to use Rambus high-speed RDRAM technology--for now," says Steve Fu, technical leader in Cisco's chief technology office.

There are "some technical reasons" why Cisco won't use Rambus, he says, but adds the company could use Rambus in the future.

Cisco--one of the world's largest consumers of chips--has little or no choice but to move to new memory technologies. Sisco and other networking-equipment OEMs have been using SDRAMs, SRAMs and other standard products in their systems. But now with the move to higher-bandwidth systems, they need faster memory architectures to boost throughput. Rambus is promoting a version of its Direct RDRAM for networking applications, while double-data-rate (DDR) SDRAMs are also in the running. Analysts figure Rambus still has a chance in networking gear. Vitesse Semiconductor's network-processors support RDRAMs and Intel's next-generation, network-processor line will have a RDRAM interface.

(See Oct. 26 story.)

Japanese DRAM makers may file
dumping charges against Koreans

Would you believe four Japanese DRAM makers of all people have started an "investigation" that could lead to dumping charges being filed against two South Korean giants--Samsung Electronics and Hynix Semiconductor. That would be a big deal in Japan.

The "investigation" is being called a "trail balloon" that's aimed at protecting the Japanese semiconductor industry's future ability to compete, according to one observer. Japan's share of the DRAM market has fallen to less than 20% in the past five years, and current low prices are causing losses and jeopardizing these companies' ability to invest in logic and LSI platform technologies, he says.

The four companies--Hitachi, NEC, Mitsubishi, and Toshiba--are only investigating possible legislation and hadn't yet approached the Japanese government.

The Japanese companies have two major grievances against their Korean competitors, according to an NEC source. They believe Samsung and Hynix have been guilty of flooding the memory market, especially during the past quarter. "It is not just Korean companies selling product below production prices--Micron and Infineon are also doing it, as are all the Taiwanese," a Hynix official claims.

NEC also has become incensed that Hynix--staggering under heavy debt--has not been allowed to go bankrupt. "The government support of Hynix has been raising quite a few concerns," the NEC source adds. A Hynix collapse could remove about 18%-to-20% of the world's DRAM capacity.

The investigation comes at a crisis point for the global DRAM industry. Spot prices of some memory parts have fallen to less than half of their manufacturing costs, leaving every DRAM player awash in red ink. Despite several bailouts of Hynix, analysts predict the company will run out cash in two quarters.

(See Oct. 24 story.)

Hynix slashes more staff,
requires unpaid time off

The clock is ticking for Hynix Semiconductor. The debt-ridden DRAM maker apparently is doing everything it can now to cut costs. This week it reached an agreement with its labor union to require workers to take 30 days of unpaid leave on a scattered basis over the next five months.

The South Korean company also is reducing its overseas workforce by another 20%. So far this year, Hynix says it has slashed its total staff by one-third to 14,000. Some 5,000 of the 7,000 released employees went along with the spinoffs of Hynix non-semiconductor units.

And the memory maker is consolidating its operations by reducing its program offices and teams by one-third. It currently has 281 teams.

(See Oct. 25 story.)

Intel's capital spending may be
higher than expected next year

Intel--the largest purchaser of chip production equipment, by far--may be cutting its capital expenditures next year by only 10-to-20% from the record $7.5 billion it is spending this year.

This lower spending level of between $6-to-$6.8 billion in 2002 would be higher than earlier estimates by nervous analysts. A report quoted Intel CTO Patrick Gelsinger as saying Intel's 2002 capital spending plans could be down 10-to-20%, according to a newsletter from Prudential Securities.

"This could be positive as many analysts were expecting Intel's capital expenditures to be down significantly between $4.5-to-$5.5 billion after an extraordinarily high capital spending level in 2001 of $7.5 billion," Prudential says. Capital spending by the entire global chip industry, on the other hand, was projected by the firm to drop "in excess of 20%" next year. The initial 2002 estimates from chip makers "are likely to appear abysmal," Prudential says.

Intel's 2002 spending will likely be high because it is ramping up three 300-mm plants, including Fab 11x in Rio Rancho, N.M., D1D in Hillsboro, Ore., and the D2 development facility in Santa Clara.

(See Oct. 22 story.)

TSMC fab to be able to turn out 4,500
12-inch wafers a month by end of year

Taiwan Semiconductor Manufacturing continues to make good progress in ramping its 300-mm production facility. The silicon foundry now claims its new Fab 12 has become the industry's first, full-scale 300-mm wafer plant capable of processing 0.13-micron ICs with acceptable volume-production yields.

Fab 12 has produced its initial 0.13-micron SRAM devices using an all-copper metal interconnect process with yield levels equal to the company's 8-inch fabs, TSMC says. It has already taped out one customer's 0.13-micron SRAM product for production in Fab 12 with several other devices now scheduled to be produced in the 12-inch wafer fab.

TSMC said Fab 12 has been designed to produce 25,000 twelve-inch diameter wafers per month. The Installed capacity by the end of 2001 for Fab 12 is expected to reach 4,500 wafers per month. It was designed to produce 25,000 12-inch diameter wafers per month--an awesome amount of chips.

(See Oct. 22 story.)

Samsung doing okay
except for semiconductors

Samsung's semiconductor operation was the only one of its divisions to be unprofitable in the third quarter. Semiconductor posted an operating loss of $294 million. Revenues totaled $1.24 billion, 26% below second quarter sales due to falling DRAM prices.

The Korean electronics giant's debt-to-equity ratio stood at 46.3%, only slightly less than the 46% at the end of the second quarter. Its combined outstanding debt for all operations--including overseas subsidiaries--totaled $3.94 billion, down 5.5% from the end of last year.

(See Oct. 22 story.)

Brooks, PRI Automation
merge to survive and grow

Both Brooks Automation and PRI Automation have been struggling with the severe downturn in semiconductor capital spending as well as with the growing delays in 300-mm wafer plants. But now the two fab automation suppliers say they've figured out a better way to survive in today's market.

Brooks Automation is acquiring PRI Automation for stock valued at $380 million. They aim to create a larger supplier of fab software, materials handling systems, and other products for increased productivity in wafer fabs. The acquisition will create a new company, called--you guessed it again--Brooks-PRI Automation, headquartered in Chelmsford, Mass., with annual sales of around $700 million.

By combining the two companies, Brooks expects to save more than $20 million annually in "operational synergies" when the deal is completed in the first quarter of 2002. "Our customers will benefit from doing business with a global company that can provide the hardware, software, and professional services required to manage every wafer move in the fab," says Brooks CEO Robert Therrien, who will become CEO of the combined companies.

(See Oct. 24 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!.)