I wrote Al that Cathy was so accurate I didn't laugh, I cried. No one ever listens to the poor consumer. The time displays on VCRs have blinked for decades because they're too difficult for consumers to set and manufacturers won't listen. I've got two $400 printers that jam, a PC that locks up and has to be rebooted at least twice a day. And now my wife gave me a new digital camcorder for Christmas that has a 400-page manual I am still trying to comprehend. P.S. It's snowing again--hard.

Micron and Hynix will likely
compromise on price of fabs

Micron Technology and Hynix Semiconductor are still dickering hard over the price the big U.S. memory supplier would pay for seven of the South Korean firm's wafer fabs. At least that's the way it looks to me on Friday.

The big South Korean chip maker, which is nearly underwater with more than $6.5 billion in loans outstanding, still holds nearly 20% of the world's DRAM market with its seven DRAM plants.

The two firms were negotiating the proposed sale in California last week, but were unable to reach an agreement. Micron says there are no plans to resume the talks. For all we know, the negotiations might be dead. But I don't think so.

It looks to me like two poker players playing hardball on setting the final price. But let's face it--where else can the financially troubled Hynix and its beat-up creditors go? Infineon doesn't have any money.

What's happening now most likely is part of the South Korean firm's "brinkmanship" negotiating strategy. Observers believe Micron has left the door open for Hynix to return to the negotiating table if they want to discuss terms that are closer to the Micron offer.

Industry observers believe the negotiations broke off when the two sides couldn't narrow their differences over the price for the fabs. Micron reportedly offered between $2.5-to-$3 billion, but the South Korean firm's largest creditors wanted $2 billion more--or around $5 billion.

But as big a gap as that is, it isn't expected to be a deal breaker, and some observers believe the two firms will end up compromising on the price for the deal, which would turn Micron into the No. 1 DRAM supplier, ahead of Samsung Electronics. Analysts expect a deal that falls midway between the two sides' proposals--around $4 billion.

(See Jan. 28 story.)

So what is Hynix up to
by allocating DRAMs?

It's really hard to believe--particularly after a year of horror in the DRAM business--but Hynix Semiconductor has put all DRAMs on allocation through March. The reason, the No. 3 DRAM maker claims, is "increased demand and a supply shortfall" for both single-data-rate and double-data-rate memories.

At this point, the South Korean chip maker is the only DRAM vendor to impose an allocation system. It was also strange that Hynix put all types of DRAMs on allocation--the usual practice is to apportion specific devices in critically short supply, not the entire DRAM lines.

But Hynix says it acted after a number of things happened. Personal computer prices became more affordable as a result of the introduction of the Brookdale chip set that supports the Intel P4 processor with SDR and DDR, P4 pricing was reduced, and a growing consumer appetite developed for more memory to operate new operating systems such as Microsoft Windows XP. Also fueling memory demand, the DRAM vendor says, is "an increase of sales in consumer electronics such as DVD, set top boxes, and handheld devices."

Also helping to tighten the market, Hynix says, were falling DRAM inventory levels which were bled off as memory suppliers consolidated and cut back capacity.

(See Jan.29 story.)

No other DRAM vendor
is allocating DRAMs yet

We weren't been able to find any other DRAM supplier this week that has gone along with Hynix Semiconductor's surprising move to put DRAMs on allocation.

Vendors admit the supply of DDR and 256-megabit chips is tight, but they say they've been able to keep up with customer needs. Says Grant Johnson, marketing intelligence manager at Converge: "I haven't seen any other producers on allocation. Most are still delivering to the spot market, although perhaps not in as large quantities as a few months ago." In fact, he says, "we have a lot of Hynix products available now. I don't know why they should claim they have to go on allocation."

"There is an adequate supply of DRAMs in the market, although it is tight for certain types," says Sherry Garber, analyst at Semico Research. "I was surprised by the Hynix announcement. Maybe they are trying to change their industry image as a low-price vendor."

Micron Technology doesn't have any products on allocation, according to Mike Sadler, sales vice president, but he acknowledges that in some cases his company hasn't been able to agree to deliver the full level of quantities that the customer has wanted. On these orders, he says, the gap between what Micron can deliver and the customer wants has been in the 10-to-20% range.

While the strong market has whittled down inventories, Micron feels that it needs to support customers and the company is continuing to sell to the spot market. While Hynix insists it stopped selling to the spot market a month ago, there are still Hynix parts available on the commodity exchanges, according to Nam Kim, memory specialist for iSuppli.

(See Jan.30 story.)

Are DRAM suppliers
'coordinating' output

The last time I looked, price fixing was still illegal in the U.S. But listen to this. The biggest DRAM makers have been putting their heads together in an effort to "coordinate" their production capacity. Maybe this kind of thing isn't price fixing exactly, but it sure sounds close to me.

These capacity talks are intermittent and unofficial, according to one memory chip executive in Taiwan, and haven't touched on the issue of pricing. The participants include Samsung Electronics, Micron Technology, and Infineon Technologies, the executive says. "There are getting to be fewer and fewer suppliers so we have to work more closely," he says. "There is a lot of contact with these guys," he adds.

What suppliers are doing is comparing notes about production quantities across different product lines, capacity plans, and the expected supply of Intel's Pentium 4 chip in an effort to balance supply with demand. "We are not talking pricing because that is illegal," the Taiwanese executive claims, "but we are talking about how to cooperate--how to ensure that the supply side does not get crazy."

None of the big DRAM makers would own up to the meetings. Says a Samsung spokeswoman: "we have never considered the coordination of production capacity" with other suppliers. And officials at Infineon and Micron couldn't be reached for comment.

Right now, something in DRAM pricing is working right, at least as far as vendors are concerned. DRAMs have shot up more than 250% from their November low, bucking the traditional first-quarter dip. But there's no agreement on how long this momentum will last. If end-user demand doesn't materialize, then DRAM prices, especially for those memories going into PCs, are expected to fall again.

Something a lot of people believe would help keep DRAM prices from falling is Micron's planned acquisition of the seven Hynix fabs, a move they predict would help keep DRAM capacity in check. "Nobody can afford the breakage of this deal," declares Charles Kau, EVP with Nanya Technology, a major supplier of double-data-rate memory.

(See Jan.30 story.)

So what is Infineon
talking to Hynix about?

You really need a scorecard to follow what's going on this week in the crazy DRAM business. On top of everything that's happening or not happening this week in mergers, alliances, pricing, and product allocation, now we hear reports from Seoul that South Korea's Hynix Semiconductor is in discussions with Germany's Infineon Technologies about a possible chip alliance.

Tokyo business daily Nihon Keizai Shimbun quotes Infineon CEO Ulrich Schumacher as saying his company was now in talks with Hynix about a joint venture and other potential alliances. The talks included a potential DRAM alliance, he says. What else could happen now in this nutty business?

(See Jan.31 story.)

AMD and UMC to build
300-mm joint-venture fab

Like we predicted two weeks ago, Jerry Sanders does have a plan to keep up with Intel and its huge, multi-billion fab expansion program. This week the Advanced Micro Devices CEO spilled the beans and it looks like a pretty good deal to me for AMD. The big PC processor maker signed an agreement with Taiwan's United Microelectronics to form a joint venture to build a 300-mm wafer foundry in Singapore.

For AMD, the alliance is aimed at overcoming a major hurdle for it in competing with Intel, which now has nearly a half dozen 300-mm factories in the works or on the drawing boards. AMD and UMC will form a joint venture called AU Pte in Singapore that will begin commercial production by mid-2005 in their new 300-mm fab using 65-nm (0.065-micron) process technology. Current plans call for the fab to turn out 10,000 wafers per week.

In a separate foundry agreement, UMC will start producing PC processors for AMD, augmenting the output from its Fab 30 in Dresden, Germany. AMD expects to get its first 0.13-micron processors from UMC's fabs in Taiwan by the end of this year.

"First and foremost, we wanted to make sure that we removed Intel's one remaining advantage, which was volume capacity beyond what people thought we might be able to get," explains Sanders. "The joint venture will make available to AMD for its half of the 300-mm fab more square inches of silicon than our Dresden facility," he maintains.

The new fab will be housed in an expansion of UMC's existing 300-mm fab facility in Singapore. That facility also houses a year-old joint venture between UMC and Germany's Infineon Technologies called UMCi Pte. But recent reports indicate tool installation for this fab might be delayed due to the industry's current overcapacity situation.

The new joint venture represents AMD's first move into 300-mm technology. Sanders believes only one or two chip makers outside the foundry business will be able to afford full ownership of a 300-mm wafer fab. One of those is Intel and the other might be DRAM maker Micron Technology, he says.

By the end of 2003, AMD's Dresden fab will be producing 50 million processor per year, Sanders says, which will be enough capacity for it to reach its target of 30% market share in PC processors. That's up from just over 20% today, he says. "We think we can do better than 30% market share over time by working with UMC," a bullish Sanders predicts.

(See Jan.31 story.)

TI's chip sales increase 3%
sequentially in 4th quarter . . .

Texas Instruments did better in the fourth quarter than Wall Street expected, with chip revenue rising 3% sequentially over the third quarter. Chip sales hit $1.5 billion, down 34% from the year-ago quarter. For the year, TI semiconductor sales fell 34% to $6.78 billion.

The Dallas chip maker had cut its operating losses a bit from the third quarter, losing $204 million vs. $219 million. But a year ago, it had an operating profit of $679 million. TI overall posted a net loss of $116 million, flat with the $117 million net loss in the third quarter.

Semiconductor orders weren't too swift. They totaled $1.3 billion and were flat with the third quarter, but only half of year-ago orders of $2.45 billion.

TI expects its revenues will be flat sequentially flat in the first quarter of 2002. Last year was a "tough year," comments CEO Tom Engibous, but "it ended on a much more positive note than it started."

"We maintained a strong balance sheet, and our actions cut about $600 million out of TI's costs on an annualized basis," he says. "As the market recovers, TI's revenue growth should fall through to the bottom line at a high rate."

. . . but it has enough capacity
to cut capital spending by 56%

Texas Instruments is slicing its capital spending this year by 56% from what it spent in 2002. The total will run $800 million, down from last year's $1.8 billion.

Its factory utilization rate is now just over 50%, giving the company plenty of room for production capacity in 2002 without significant investments, points out CFO Bill Aylesworth. Capacity utilization in the third quarter was about 45%.

Most of this year's capital spending will go for "technology enhancements," he notes. Last year TI finished installing its first 300-mm frontend and the conversion of three analog fabs to 200-mm wafers. TI expects to be moving its 300-mm DMOS6 fab in Dallas into initial production during the first half of 2002, Aylesworth says.

R&D will not be cut this year, however. R&D spending will run about $1.5 billion in 2002, the same as last year.

Fourth-quarter analog chip revenues were even with the third quarter, while DSP sales were 10% higher that the previous quarter. About 40% of TI's semiconductor sales were in analog circuits, while 25% were from DSP products.

(See Jan.28 story.)

Once again, Chartered
delays 300-mm fab . . .

Chartered Semiconductor Manufacturing appears to be falling farther behind in the high-stakes race between the big silicon foundries to go into volume production with the next generation of fabs. Once again, the Singapore company has delayed its initial 300-mm wafer fab project by at least two quarters until the end of next year.

Chartered's Fab 7 was supposed to be up and running in Singapore by mid-2003, but the big fab will not move into production until the "end of 2003," the company now says. A year ago, Chartered said it was going to upgrade Fab 7 from a 200-mm wafer plant to a 300-mm facility and would be up and running by mid-2002. The company planned to invest $3 billion in the giant fab once it was fully equipped. But last July, Chartered said it was postponing the start up of the 300-mm fab by one year--or to mid-2003, because of the severe downturn in the chip foundry business.

But Chartered CEO Barry Waite played down the decision to delay the big fab again. The timing of the 300-mm fab is more of a "capacity statement" on when new volume is needed rather than a technology issue, he says. All the company's technology modules for 300-mm processes have finished development, he notes.

Capital spending plans for 2002 have been trimmed 18% by Chartered from $490 million to $400 million. Most of that will go to install advanced processes in its 200-mm Fab 6 and for R&D systems to develop 0.10-micron processes, Chartered says. Waite told analysts that The company now plans to begin the installation of Fab 7 equipment in early 2003 so that the facility will be ready for initial production at the end of next year.

(See Jan.28 story.)

. . . as 4th quarter sales fall
76% and losses hit $127M

Are Chartered Semiconductor Manufacturing's miseries behind it now? You'd think so by listening to CEO Barry Waite this week.

There were positive signs for the company, he claims, in the fourth quarter. "We were pleased to see growing signs of stabilization in our business and in the broader semiconductor market, after three quarters of sharp fall-off." Waite says that it also is "quite encouraging" that the silicon foundry "was able to meet its fourth quarter revenues and earnings guidance."

First quarter sales are projected to be flat to up in the "mid-single digits," he predicts, although he still expects a loss of $136-to-$139 million--not exactly a turnaround.

Fourth quarter sales amounted to $76.1 million, down a gigantic 76% from the fourth quarter in 2000 and a 4% drop from the previous quarter. From a $77 million net profit in the year-ago period, the foundry lost $127 million in the final 2001 quarter.

Results for the full year were just as bad. The Singapore company reported sales of $463 million in 2001, a 58% drop from last year. And it posted a loss of $384 million in 2001, compared to a profit of $245 million last year--a swing of $630 million.

(See Jan.28 story.)

Toshiba again slashes
sales and net forecast

Toshiba's miseries aren't behind it, that's for sure. Things started off badly last year and the Tokyo giant had projected a $1.5 billion loss for its fiscal year ending March 31, 2002. That was a shocker, but listen to this. Now the company figures its net loss will hit $1.96 billion--33% larger.

It isn't only net that's still going down hill fast. Toshiba says its sales also are going to be lower than originally predicted, falling to $40.6 billion, down 10% from the $45 billion it racked up in the prior year. The company had estimated earlier that sales would amount to $41.4 billion. The global slump in communications and semiconductors takes the blame, as well as the high costs related to the company's restructuring.

Here's how sales look by business segment for the third quarter just ended: information and communications unit had sales of $1.42 billion, down 10%; social infrastructure systems had sales of $1.37 billion, down 2%; electronic devices and components mostly semiconductors had a 31% sales drop to $1.96 billion; digital media showed a 2% sales decline to $2.8 billion; and power systems sales dropped to $823 million.

(See Jan.29 story.)

Would you believe SiGe is now being
replaced by CMOS in some new tasks?

For years now, I've been hearing how silicon-germanium (SiGe) was going to take over the world. SiGe technologies promise to deliver the next wave of high-speed, communications chips, many experts were saying. But a funny thing happened on the way to the forum.

Greg Winner, senior engineering vice president at Advanced Micro Circuits (AMCC), now says that some unexpected power issues with SIGe could force IC suppliers to fall back on traditional CMOS. Some of the new SiGe communications devices consume too much power, he says, especially in next-generation 40-gigabits-per-second (OC-768) networking applications.

"SiGe has some power issues in that we can't get the technology down to 1.8-volts in OC-768 applications," Winner notes. These issues are forcing AMCC and other chip makers to develop the more critical parts on traditional CMOS technologies, he points out.

AMCC already is working with foundry giant United Microelectronics to develop devices for OC-768 applications based on next-generation, 0.10-micron, CMOS-based processes, he notes. UMC already is making several other CMOS-based devices on a foundry basis for AMCC.

One of the biggest problems with today's modules is that the separate SiGe-enabled transmit and receive devices consume too much power, he said. Even the latest SiGe-based processes will not lower the power consumptions sufficiently, he says.

But AMCC is not throwing in the towel on SiGe. It will continue to use SiGe for the amplifier and other portions of the module, Winner says. But to reduce power consumption, the company is looking at traditional CMOS processes to build the high-speed send and receive devices.

(See Jan.31 story.)

Sematech pushes harder on
rising mask-making problems

It's getting a lot tougher to develop new generations of photomask production gear for the ever-finer IC geometries. For one thing, the limited size of the photomask production equipment market has made it difficult to build a strong business that's capable of investing enough money to keep up with the semiconductor industry roadmap.

Because the problem is continuing to worsen, International Sematech has decided to spend more than half of its lithography R&D budget to solve the growing problems in mask making and reticle materials.

"The problems now go beyond just masks--it's all about optimized image delivery," says Kurt Kimmel, mask strategy manager for the Austin-based semiconductor consortium. "There has to be a strong partnership between the mask makers and lithographers," he says. "That market must be supported if lithography and semiconductor growth are to continue at the same speed that they always have," he adds.

To help maintain the pace, Sematech has adopted a mask strategy that is even more aggressive than the chip industry's roadmap for 70-nm and 50-nm process technology nodes, which are scheduled to start building products between 2005 and 2007. For 70-nm technology, Sematech plans to use 157-nm optical lithography technology. And for 50-nm, it will use a next-generation lithography (NGL) technology, such as extreme ultraviolet (EUV) systems or, perhaps, electron-beam projection lithography (EPL), Kimmel explains.

(See Jan.29 story.)

AMD buying Alchemy to get back
in embedded processor business

Advanced Micro Devices is moving back into the embedded processor market. It dropped out of the competition back in the mid-90s when it was having trouble selling its RISC-based Am29000 line.

AMD will do this by acquiring Alchemy Semiconductor and pit that company's MIPS-based processor architecture against Intel's Xscale processors. The acquisition also will also put AMD in competition with other MIPS licensees such as Integrated Device Technology, NEC, PMC-Sierra, and Toshiba.

Three-year-old Alchemy recently began shipping standard embedded products based on its Au1000 processor core, which dissipates 1 watt at 500 megahertz or 0.5 watts at 400 megahertz. That performance-to-power ratio puts Alchemy's products in the mid-to-high performance range that Intel is going after with its Xscale processor.

While Alchemy gained market success last year--achieving more than 15 design wins--the 75-person startup was having difficulty trying to raise funds for its second round of financing. While it attracted venture capital firms willing to invest, finding a lead investor proved to be a long and complicated process.

(See Jan.29 story.)

China foundry moving quickly
into state-of-the-art products

China foundry startup Grace Semiconductor Manufacturing , which reportedly is getting its chip-making technology from Japan's Oki Electric Industry, has now closed a deal with Avanti to obtain the use of its 0.25- to 0.13-micron chip libraries.

Avanti will provide its Libra-Visa (timing-specific) and Libra-Passport (multi-foundry access) digital design libraries, both being compatible with Avanti's Milkyway-based line of EDA tools for system-on-a-chip (SoC) designs.

GSMC is now building its first fab, an 8-inch, 0.25-micron and-below plant in Shanghai. The fab should be completed by the end of this year.

(See Jan.28 story.)

Will TSMC beat Intel in making
volume 0.10-micron products?

Looks like Taiwan Semiconductor Manufacturing has a shot at becoming the world's first chip maker to deliver 0.10-micron technology--ahead of even Intel and IBM Microelectronics.

This week TSMC revealed it has produced its first 0.10-micron chips and is accelerating development of this technology so that it can move into initial production by the end of the third or beginning of the fourth quarter. In contrast, Intel won't be out with its 90-nm (0.09-micron) technology until early in 2003, according to analysts.

But analysts don't believe the Taiwanese foundry will move into volume production until the following year, or early- to mid-2003, but they acknowledge the firm still appears to be far along with its new process. The 0.10-micron technology is targeted at system-on-a-chip (SoC) designs and will include copper-interconnects and low-k dielectrics.

TSMC plans to use 193-nm lithography tools from its long-time scanner and stepper vendor--ASML. It will use the Dutch company's Twinscan 1100 AT line of 193-nm tools for the critical layers in its next-generation designs. There are a lot of critical layers in 0.10-micron technology, TSMC acknowledges.

(See Jan.30 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!.)