So, is Intel's Q3
good news--or what?

Unless you're from Mars or a Yankee fan, you already know that Intel reported its third-quarter results yesterday. They were better than most people expected, including even Intel managers who had warned Wall Street last month that third-quarter sales and profits would fall short of expectations. Wall Street then revised their estimates downward, but net proceeded to come in at the original earning-per-share forecast. Oh well, who ever said that forecasting was a science.

Despite flat microprocessor sales, weak European PC demand, and a series of manufacturing problems, Intel reported third-quarter sales of $8.7 billion, up 19% from a year ago but not the $9 billion-plus that analysts had been expecting earlier. Sales were up only 5% over the previous quarter. Intel is now projecting fourth quarter sales growing 4%-to-8% over the third quarter.

Net income, not including acquisition-related costs, amounted to 2.9 billion, up 52% over the third 1999 quarter but down 18% from the second quarter of this year.

Intel does seem to be more optimistic about the fourth quarter. "Looking ahead, we anticipate record revenue, with growth across most of our product lines," predicts CEO Craig R. Barrett.

(See Oct. 17 story.)

IBM goes to silicon-on-insulator
to supercharge new server line

The move to silicon-on-insulator (SOI) technology picked up more speed this week. IBM now says it is putting SOI to work along with copper interconnects to boost the performance of the central processors it is using in its new pSeries 680 servers. As a result, this new lineup of 64-bit servers will be rated 63% faster than the company's existing S80 systems, which are based on copper processors using standard silicon substrates.

The new servers, code-named Turbo, come with 24 SOI copper processors and will be delivered in volume to customers beginning Nov. 17. IBM claims its servers will beat the online-transaction performance of Sun and Microsystems and Hewlett-Packard systems with twice as many CPUs.

IBM is now running with an aggressive roadmap for servers. It expects to increase server performance in 2001 with the addition of IBM's new Power4 CPU, which integrates two processors, Level 2 cache, and other functions on a chip with 170 million transistors. Following launch of SOI copper-based Power4 processors, IBM plans to add low-k dielectric insulators to the CPU technology mix in 2002 for even higher levels of server performance.

Next year, the Power4 processor will built with SOI and seven-level copper interconnect technologies and will go into IBM servers. "We came out with copper three years ago and everyone tried to pooh-pooh that move, and now everyone is doing it replacing aluminum interconnects with copper," points out Joel Tendler, a program director for IBM's processor unit in Austin. "The same thing was the case with silicon-on-insulator," he says, referring to IBM's 1998 decision to put SOI into commercial production. "At the time it wasn't seen as important, but now it is," he says, adding "the same thing will happen with low-k."

IBM's primary objective here is to overtake the Unix market leader Sun Microsystems. It claims its new p680 system can outperform Sun's E1000 system by 38% in transaction processing at a slightly lower cost per transaction. Compared to HP's V2500, the p680 is 116% more powerful at two-thirds the cost.

(See Oct. 16 story.)

LSI Logic still sees 300-mm problems,
but expects to be in production by 2002

While Wilf Corrigan, CEO of LSI Logic, acknowledges the new 300-mm wafer still has its problems, he aims to move into production with the big disk by the end of 2002. He plans to go into 12-inch production with the company's next 0.13-micron process at its fab in Santa Clara, Calif.

Corrigan says the biggest problem now is the lack of production equipment. "There is a question of when real equipment will be available," he says. "There are a number of 12-inch fabs alleged to be in existence. They range from a patch of ground to a prototype line," Corrigan says. "We still have to see real production out of a 12-inch fab." But when 12-inch production finally arrives, he says, it will build up very quickly.

(See Oct. 16 story.)

Latest Pentium III
delayed until spring

Intel continues to struggle with its fastest microprocessors. This time it is its 1.13-gigahertz Pentium III.

It now turns out that the chip giant is reworking entirely this MPU, something that will delay the high-speed chip's introduction until the second quarter of 2001.

All of Intel's 0.18-micron Pentium III line is undergoing what's called a new core "stepping," or manufacturing revision, according to a product change notification notice sent out by Intel. This newer version will be used to fabricate the new 1.13-gigahertz chip. New steppings are usually used to fix errata in the chips, or allow those chips to run at higher speeds.

What's unusual, say industry sources, is that new steppings are usually planned to prevent an interruption in the steady introduction of faster clock speeds. Intel in July announced "limited production volumes" of the 1.13-gigahertz Pentium III, but pulled the plug on the chip at the end of August after spotting an intermittent glitch that could result in data loss.

The chip apparently will still be manufactured with a 0.18-micron process. A second chip at this speed, the Coppermine-T, is expected three months later, using a finer 0.13-micron process.

(See Oct. 16 story.)

AMD's Athlon MPU speeds
away from Intel's Pentium

Believe it or not, the gap is widening in performance between Advanced Micro Device's Athlon and Intel's Pentium processors. This week AMD improved its speed lead by releasing a 1.2-gigahertz MPU.

The new chip follows AMD's recent introduction of the 1.1-gigahertz Athlon. All this new product action is coming from AMD as Intel's Pentium III chip development is stalled at 1 gigahertz and its hyped 1.4- and 1.5-gigahertz Pentium 4 won't be out until late November.

Compaq Computer, Gateway, and Hewlett-Packard are all planning to offer PCs based on the 1.2-GHz Athlon, AMD says, while Gateway began selling PCs using the new Athlon chip this week. Price is $612 each in quantities of 1,000, while the 1-gigahertz chip is priced at $460 each in lots of 1,000.

(See Oct. 17 story.)

This is $2.7 billion chip
deal, but who are they?

Looks like the big guys in local-area networking chips--suppliers such as Broadcom, Intel, and Lucent--are going to have a new competitor giving them fits. In a surprising deal announced this week, Marvell Technology Group in Sunnyvale is expanding its communications chip offerings by acquiring Galileo Technology for $2.7 billion with a B. I hate to admit it but I don't know a heckuva lot about these two chip makers, or the folks running them.

By acquiring the Israeli-based Galileo, Marvell will become one of the world's leading suppliers of switch-chip controllers for Layer 2 and 3 applications in Ethernet- and Fast Ethernet-based networks. A pioneer in the merchant switch-chip business, Galileo sells a line of both low- and high-speed switching ICs to such major OEMs as Cisco Systems and Hewlett-Packard.

"This combination brings together Marvell's industry-leading gigabit Ethernet Alaska transceiver technology with Galileo's Galnet family of advanced Layer 2/3/4/5 switched Ethernet processors," declares Marvell CEO Sehat Sutardia. "We believe the combined company will be the only supplier currently shipping both gigabit transceivers and switch ICs."

Avigdor Willenz, now Galileo CEO, will become executive vice president after the merger, heading up what will become Marvell's Galileo subsidiary. The deal is expected to be completed in the first quarter of next year.

A plunge in Marvell's share price after the deal was announced worried investers, but Galileo's CEO told Reuters that the acquisition would not fall through. "The deal will not close before January or February so we have a lot of time for the stock to recover and show true value," Willenz said.

Marvell investors apparently were disappointed by the sale to the Israeli company because they were hoping their company would be bought out by a larger competitor. The drop in the share price on Tuesday and Wednesday alone shaved off about $1 billion of the price of Marvell stock.

(See Oct. 17 story.)

Smug STMicro expects to grow
faster than chip market next year

STMicroelectronics is feeling a little smug about itself these days. I guess it has good reasons. This week the European chip giant reported a record $2 billion in third-quarter sales, up 60% over last year. Net profit results were even better, hitting a record $415 million, a 207% jump over the same quarter last year.

"The fourth quarter will be another period of record financial performance," predicts CEO Pasquale Pistorio. "Sequential revenue growth is expected to be in the high single digits."

"Looking ahead to 2001, we believe ST is positioned to grow faster than the overall semiconductor market," declares the optimistic CEO. This is based on current forecasts by analysts calling for a market growth next year of between 25% and 30%. ST reported strong demand across the board for its products. "Sales to each of our end markets--namely automotive, consumer, computer, telecom, and industrial--increased in the third quarter," Pistorio notes. Telecom and industrial posted sequential gains of 15.3% and 20.3%, respectively, he says.

(See Oct. 18 story.)

Execs say no signs yet
chip biz is cooling off

Despite the growing number of "Chicken Littles" like me, the mood was definitely upbeat at the Semicon Southwest trade show in Austin this week. Strong global demand for chip-production systems is showing no signs of cooling off, according to industry executives, who say that concerns about an early downturn in the chip business were overblown by industry analysts and the press.

"We're not seeing any slowdown in PCs, and we are not seeing any slowdown in chip demand," declares C. Richard Deininger, AMD's director of strategic equipment. "And we're not seeing any slowdown in the cellular phone area for flash memories, microprocessors, and other communications chips." To Deininger, it is "unbelievable that people in the rest of the industry would be saying in the press that they are seeing a slowdown or we've reached the peak."

"Every customer I've spoken to is running flat out, and they are trying to squeeze as much capacity as they can out of their existing infrastructure," maintains Don Mitchell, CEO of FSI International, a Minnesota-based supplier of fab tools. But he is concerned over the industry's ability to keep up the current strong growth in new orders. "While you can't keep up at this pace, I still see this cycle as having a lot of legs on it."

(See Oct. 18 story.)

Strength of 300-mm orders
surprises chip gear makers

One development that is surprising some chip production equipment people right now is just how strong the 300-mm tool business is coming on. Analysts earlier were expecting about 10% of wafer fab system orders would be for the new generation substrate in 2000, but this business already is running as high as 15% at such companies as FSI International.

"There is no question that 300-mm is really providing momentum," FSI CEO Don Mitchell said at a panel discussion this week at Semicon Southwest. Fourteen chip companies already are building 300-mm fabs around the world right now, he says, and 300-mm tool orders will be half of the industry's business by 2005.

But it won't be easy to make the switch. While AMD already has said that it plans its first 300-mm logic factory to be on stream between the end of 2003 and 2004, "there is still a long way to go between having a plan to go somewhere and having a commitment to break ground," says C. Richard Deininger, AMD's director of strategic equipment. Even so, he adds, "based on the demand we are seeing right now, we have the need for that kind of factory."

"Lithography tools still have a way to go--no question about it," Deininger notes. "Some of the deposition tools have a way to go too. But there are working tools. Etch is in its second or third generation for 12-inch diameter substrates." He sees no show stoppers, but that "doesn't mean everything is perfect." He doesn't think that anyone can turn on full production today, "but I don't think we are far from it."

(See Oct. 18 story.)

Fourth OEM introduces PC
based on Transmeta's Crusoe

No wonder Intel is changing its mind about silicon-on-insulator, it needs a lower-power microprocessor fast to compete with its tiny new competitor--Transmeta. The Santa Clara-based startup, which is soon to go public, revealed another big OEM customer this week.

NEC, Japan's largest PC maker, has introduced a new notebook computer built around a 600-megahertz, x86-based MPU from Transmeta. This makes the fourth major design win for Transmeta. In the past month, Sony, Fujitsu, and NEC have each rolled out a notebook PC based on Transmeta's Crusoe line. And another Japanese OEM--Hitachi--reportedly is developing a line of products based on the Crusoe.

Transmeta is not going after the high-performance desktop PC market, but is concentrating on the notebook PC and Internet appliance. And it is getting Intel's attention there now. While Transmeta is grabbing a lot headlines these days, there are reports circulating in the industry that its chips are not running as fast as the company originally expected.

One OEM customer dismissed these rumors, however. "NEC is very impressed with the battery life and performance capabilities provided by Crusoe," says Osamu Iseki, a senior manager at NEC. NEC's new notebook PC, called the LaVie MX, is a three-pound unit claiming a battery life up to 11 hours.

(See Oct. 17 story.)

Intel now calls relationship
with Rambus 'a mistake'

Wow! According to the Financial Times of London, Intel CEO Craig Barrett admitted this week that "we made a big bet on Rambus and it did not work out. In retrospect," he said, "it was a mistake to be dependent on a third party for a technology that gates your performance."

Intel is now hoping to partner with a company that is a technology innovator, Barrett said, rather than "seeking to collect a toll from other companies."

It now looks like Intel gave up on Rambus as long as six months ago. A spokesman for the chip maker believes Barrett's comments were accurate and representative of a viewpoint that the Intel CEO has held for six months.

Intel's big shift went public earlier this week when execs acknowledged the chip maker is looking "very seriously" at using DDR (double-data-rate) memory with its desktop PC chipsets. "We are adopting DDR technology for servers, and exploring DDR on desktop" vice president Paul Otellini said. "We're looking very, very seriously at it," he added.

SDRAM is considered to be a lower-cost alternative to the Direct Rambus memory that Intel pushed so hard for next-generation, higher-speed memory. DDR DRAM is seen as an evolutionary means to achieve performance parity with Direct Rambus, while keeping costs down. Memory vendors such as Micron Technology Inc favor DDR. For its part, Rambus was laying low this week.

(See Oct. 18 story.)

CMP may end up getting replaced by
'electropolishing' in copper processes

A key problem in finding a low-k dielectric insulator for copper interconnects has been to find one that can hold up under the stress and strain of chemical mechanical polishing (CMP). Many insulating materials with lower dielectric-constant ratings can't meet the mechanical-strength requirements for CMP, which is used to flatten layers in interconnects and remove excess amounts of copper from vias and trenches.

But help may be on the way. ACM Research, a Silicon Valley startup, has developed new current-controlled electroplating and electropolishing technologies that it claims "will dramatically ease" processing steps for next-generation copper and low-k dielectrics on ICs. "Our electropolishing system is 'stress free polishing,'" claims ACM's CEO, David Wang.

These processes and the new tools to go with them use electrical current to control locally the deposition of copper, plating steps, removal of metal, and polishing sequences. The novel electropolishing process and tool is said to be capable of removing materials at the atomic layer down to 80 angstroms.

Switching from CMP to these electropolishing techniques for planarization in dual damascene copper processes opens up a wide range of choices for low-k dielectric materials that couldn't cut it before because they couldn't withstand the stress of CMP.

Now, softer dielectric films such as spin-on, porous Xerogel materials have become viable candidates for integration in damascene processes with copper. These porous films can be "tunable" and have low-k dielectric ratings less than 1.8 vs. above 3 for many of the current low-k insulators being used in prototype copper processes.

Other companies also are pursuing alternatives to CMP. If the new technology catches on, it could threaten a huge wafer tool segment that is expected to top $1.2 billion in sales by 2003, according to VLSI Research. "There may not be a need for copper CMP," Wang adds.

He believes his company's patented concepts also will extend the use of copper electroplating to the 0.035-micron technology node, which is still expected to be in use more than a decade from now. The cost of his alternative to CMP will run less than $2 per wafer vs. $4-to-5 with CMP tools, Wang claims.

(See Oct. 18 story.)

TI sees chip growth
slowing in 4th quarter

Texas Instruments did good in its latest quarter. For one thing it outdid the Street, hitting profits of 38 cents a share vs. the consensus figure of 33 cents.

Semiconductor sales grew 29% to $2.7 billion in the third quarter; overall revenues hit $3.5 billion, up 25%. For the third quarter, TI's net income was up 30% to $591 million vs. $453 million in the period last year.

But sales growth will slow down in the fourth quarter, predicts the Dallas company. TI now expects chip revenues to grow sequentially by "a few percentage points" in the fourth quarter. In the third quarter, chip sales grew 8% sequentially due to strong demand for digital signal processors and analog devices.

(See Oct. 18 story.)

Chartered on track to hit
$1 billion in sales in 2000

Chartered Semiconductor's third-quarter sales keeps the No. 3 silicon foundry on track to hit the $1 billion sales level this year and marks its seventh consecutive quarter of solid growth.

Demand in the communications and consumer markets pushed up third-quarter sales 67% over last year to $306 million. Net profits came in at $72 million vs. a loss in the third quarter last year.

For the nine months, Chartered earned $167 million on sales of $815 million. This compares to a loss of $55 million on sales of $478 million last year.

Capacity utilization was 103% in the quarter as the Singapore-based company shipped 246,000 eight-inch-equivalent wafers, up 39% over last year. Average selling price rose 22% to $1,242 per wafer in the quarter.

(See Oct. 18 story.)

If you have any comments or questions, don't hesitate to E-mail us at bhenkel@aol.com. Have a great weekend!

(Click here for last week's Semiconductor Alert!.)