"We are looking out onto a broad landscape of new technologies," he says." The first one is wireless, which I entitle 'radio-free Intel.' The second is wireless sensor networks where you take the wireless technology and build self-configuring, dynamic networks using on-chip MEMS sensors. And the third is silicon photonics, where we imagine the opto-processors of the future."

Despite its exotic competition,
silicon keeps humming along

If there's anything I've learned in recent years about the semiconductor industry, it's not to write off silicon too soon. This week IBM surprised a lot of people by revealing that the world's fastest IC doesn't use one of the new high-speed compound semiconductors, but was made instead with silicon-germanium technology.

IBM disclosed its latest SiGe design, a ring oscillator, that operates at speeds over 110 gigahertz, which means it can process a signal at 4.3 trillionths of a second. That delay time is 0.3-picosecond faster than the 4.6 picosecond speed of the previous record holder--an indium phosphide ring oscillator circuit.

This new circuit proves that silicon-based technologies will continue to outperform compound semiconductors such as InP and gallium arsenide, maintains Bernard Meyerson, vice president at IBM's communications R&D center. "People have badly underestimated the extendibility of silicon," he says.

Even Meyerson was surprised with the performance of IBM's SiGe 8HP-based ring oscillator. "I was astounded that it outran the fastest circuit ever built in any technology--silicon, InP, or GaAs--it doesn't matter."

"That 0.3-picosecond advantage is substantial," he says, but even more important is that "when the circuit is operating at lower power, it still has a speed advantage and a factor of 300-to-500% lower power" consumption, Meyerson says.

The first chips fabricated with SiGe 8HP technology, which is IBM's fourth generation SiGe technology, will be introduced later this year. IBM already is working with early access customers such as Sierra Monolithics. IBM first talked of its SiGe 8HP process in June, and Meyerson says this week's disclosure is the "real-world" indication of the actual benefits of a process technology vs. transistor measurements that can be controversial.

The ring oscillator is the best way to demonstrate performance, he says. "An oscillator delay is an actual delay measurement, period," indicating that many high-speed transistor measurements have been more theoretical than real. IBM has been critical of Conexant System's recent claims that it developed the world's fastest SiGe process using a loosely defined "unilateral maximum available gain" technique to measure maximum frequencies, which he says can inflate performance.

(See Feb. 25 story.)

Good news from Orient:
DRAM prices rising fast

The two South Korean companies that dominate the global DRAM business are sounding almost happy this days. Why? Rising prices, that's why.

Samsung Electronics and Hynix Semiconductor claim they've succeeded in raising the supply price of 128-megabit memory chips by 19-to-25%. That means that 128-megabit chips that previously had sold for around $4 are now going for almost $5.

After the two DRAM makers hiked their contract prices, spot prices for DRAMs also rose sharply. The price of 128-megabit synchronous DRAMs rose by 7% in mid-February on the Asian spot market.

As a result, Samsung says that its semiconductor operation turned profitable in January. Hynix expects to go into the black in February.

(See Feb. 27 story.)

Micron confirms DRAM
prices are still going up . . .

That little upturn in DRAM prices that began late last year isn't dieing out like some observers had expected. DRAM demand from PC makers is exceeding expectations in the first quarter, according to Micron Technology.

"Going into last December, our customers were setting us up to think about a 10% down first quarter compared to the fourth quarter," says VP Kipp Bedard. "They have now revised that to think more in the down 5% to flattish" range, he reports at Robertson Stephens' San Francisco technology conference.

Most of that new strength is coming from the consumer PC and the notebook computer segments, but "recently a couple of customers have mentioned a little bit of renewed corporate spending--a little bit of IT spending," Bedard says.

Another positive note is a surge in the amount of memory-per-system, according to Dave Parker, Micron's investor relations manager. The average amount of DRAM per box at the end of 2001 was 170 megabytes, he says. Now, "based on what our customers tell us, we're looking at a pretty clear path to 256 megabytes per box and possibly higher" by the end of this year. That's a 50% increase in DRAM per PC!

The introduction of Windows XP is the primary driver. "Most of our OEMs are telling us they're not really shipping any XP-configured boxes with anything less than 256 megabytes of memory," Parker says.

Despite rising DRAM prices, Parker expects them to continue to go up. "Elasticity is still in place," he believes, pointing out that for just $36, a PC owner can move from 128- to 256-megabyes of DRAM.

The OEM contract price for a 128 megabit, synchronous DRAM chip is now in the $4-to-4.50 range--or slightly ahead of the spot price, Parker says. That's three times the $1.30-to-1.40 price tag for the same chip in early December.

(See Feb. 26 story.)

. . . but adds not enough money is
being spent to maintain capacity

The massive drop in DRAM demand last year is resulting in the DRAM industry not spending enough to maintain global memory capacity this year, according to Micron Technology.

Dave Parker, Micron's investor relations manager, estimates that $13 billion in capital spending will be needed this year just to maintain existing capacity. But, he says, only about $6.5 billion is expected to be spent industry-wide this year.

Micron itself plans to spend $1 billion this year and is leading the charge to 0.13-micron process in the global DRAM business. Already, the Idaho DRAM vendor is building 10% of its wafers with the 0.13-micron process. Some 70% of all wafer starts are in 0.15-micron while the remaining 20% is in 0.18-micron. The company is now moving its overseas fabs to the 0.13-micron process.

The number of players in the global DRAM business continues to shrink. In 1996, there were about 28 DRAM suppliers, according to Parker, but that number had shrunk to only 14 by the end of last year.

Last year, the Micron executive says, the top five suppliers accounted for 80% of the DRAM market last year, and he predicts that "we could have three suppliers with 70-80% market share as we exit 2002."

(See Feb. 26 story.)

Intel working hard on chips
to dominate 2.5G cell phones

Looks like Intel has more than one strategy for developing a new growth market to replace its slowing PC processor market. Add the cellular world to the list.

A new family of parts, being designed with Intel's 0.13-micron process by its R&D lab in Israel, is aimed at recreating in the cellular world the kind of high-volume platform that Intel and Microsoft forged with the PC. Intel is counting on their relatively high speed, performance, and integration to help it grab the market for handhelds that need greater oomph to deliver more applications or bandwidth.

Integrated baseband chips are being developed for 2.5G and 3G cell phones that merge Intel's 400-megahertz Xscale processor, 400-megahertz Micro Signal Architecture digital signal processor and sufficient on-board flash memory to handle all DSP and applications code.

And it's not too late, the processor giant claims. Although as many as 50 General Packet Radio Services systems already have been deployed in North America and perhaps even more in Europe, Intel insists its 2.5G part, which is coming later this year, is not too late. "This market is still in its infancy," declares Ron Smith, who heads Intel's wireless communications group. "Anyone who says this all happens at once and the door slams shut is wrong."

Intel plans to deliver later this year a family of full reference designs for PDAs and smart phones using Microsoft's CE.Net. But RF transceivers for those designs will come from a partner to be named.

(See Feb. 28 story.)

Intel's Barrett is confident
new chip boom is coming . . .

It may sound like coming out for motherhood and apple pie, but Intel CEO Craig Barrett this week was urging the global chip industry to take the same kind of steps to pull out of the current downturn that have done the job for it in past recessions.

Although he says the semiconductor industry is mired in the "deepest trough" it has ever faced, Barrett is urging chip makers and systems houses to develop new products and services to fuel new growth over the next decade. "You need to continue to invest," he says. "We have to innovate out of the recession with new products."

Barrett believes that IC makers and systems houses are on the verge of what he called a "technology break away," which will bring a wave of new products driving new growth.

One major candidate to relaunch the industry is left over from the last boom, he notes. The Internet "has just started its build out," Barrett says. "That's why I'm confident about the future." Other market drivers, he says, are broadband, optical networking, security, wireless, and the PC. "The computer and communications convergence is emerging."

(See Feb. 25 story.)

. . .with 30-GHz MPUs
on 18-in. wafers due by '17

Intel CEO Craig Barrett is looking for a steady evolution of current semiconductor technology for the next 15 years.

This means he's "looking forward to the next-generation" of wafer size, or to 18-inch silicon wafers, by then. Intel has just become the first chip maker to ship production processors made on 12-inch wafers using 0.13-micron technology.

In 15 years, he expects those big wafers will be filled with 30-gigahertz microprocessors. They will be fabricated with a 10-nm (0.010-micron) process. Intel is now starting to fab its chips with a 0.13-micron process to reduce overall chip costs.

Fifteen years from now, processors and other chips will be fabricated at the 10-nm (0.010-micron) technology node, said Barrett during his opening keynote address at the Intel Developer Forum (IDF) here today.

(See Feb. 25 story.)

Intel accelerates launch
of its mobile Pentium 4

The laptop market should get a big boost very shortly. Intel has accelerated the launch of its Pentium 4 mobile processor and will unveil it officially within two weeks.

But "we didn't just take a desktop Pentium 4 and put it in a mobile platform," claims Don MacDonald, marketing chief for Intel's Mobile Platform Group. He says the design of the notebook version of the P4 has been modified to meet the needs of laptop PCs, specifically in the areas of battery life and heat dissipation.

The mobile P4 will match the same 3.5-hour average battery life of its lower-power Pentium III predecessor and will run with the same 24-to-30 watt thermal envelope as the PIII, MacDonald says. "We modified the design to reduce the leakage current--just the opposite goal of the desktop Pentium 4, which wants higher leakage current to increase the frequency."

Despite its lower power, the mobile P4 will be offered at 1.5 gigahertz or higher. A 2-gigahertz version should be on the market by year end. The mobile P4 will be built with Intel's 0.13-micron process technology, the same design rules used for making the new desktop Northwood P4 processor introduced in January.

A ramp up of wireless mobile PCs will be the next major change in the mobile market, MacDonald says. Intel currently is drafting a Protective Access Architecture to cover a variety of wireless PC security requirements. As access increases for wireless connections in airports, convention centers, hotels, and other public places, "we are going to see an entirely new demand for mobile PCs," he predicts.

Also needed will be greater security of wireless transmission, to prevent interception and eavesdropping, which will require the addition of software encryption.

(See Feb. 25 story.)

That buzz you hear is
Itanium's 3 successors

Market reception for the Itanium, Intel's first 64-bit central processor, has been--to put it bluntly--lousy. Systems houses reportedly shipped only 2,500 Itanium-based servers since the big chip was rolled out last year.

At the same time, competitors Advanced Micro Devices, IBM, and Sun Microsystems, are rushing to get out their own 64-bit processors.

Even though Intel and Hewlett-Packard dumped as much as $1 billion co-developing Itanium, the chip giant has also been working hard to come up with a line of replacements to replace the disappointing Itanium. This week Intel outlined its new 64-bit microprocessor launch strategy for the Deerfield, Madison, and Montecito chips.

"McKinley is on track for platform release this summer," says Mike Fister, vice president who heads up Intel's Enterprise Platforms Group. "But you haven't seen anything yet," he says.

Next year, Intel is expected to roll out its next 64-bit Deerfield chip, which is designed as a low-cost processor based on 0.13-micron process technology, Fister says. At the same time, Intel expects to ship another 0.13-micron chip called Madison, which is a higher-end CPU. Coming along in 2004 will be a 0.09-micron processor called Montecito.

(See Feb. 25 story.)

Why Intel's prices
come down so fast

Intel's production technology certainly is a moving target.

It hit a real milestone this week. "Intel is the first manufacturer to ramp up production of 0.13-micron technology on 300-mm wafers," says Sunlin Chou, SVP who heads up manufacturing.

Only a handful of chip makers are now in early production with 300-mm wafers, and those products are being produced with 0.18- or 0.15-micron design rules.

The plant making these production wafers was Fab D1C in Hillsboro, Ore., Intel's fifth wafer fab to begin processing ICs with 0.13-micron technology, the other four are using 200-mm wafers.

There are important reasons for pushing hard to get moving with 300-mm production. "Microprocessors produced on 300-mm wafers cost 30% less than those made using the smaller 200-mm wafers," Chou says. "And by combining the larger wafers with our advanced 0.13-micron process, we are able to quadruple the output per wafer compared to that of the prior process generation."

On top of that, 300-mm production will uses 40% less water and energy per IC than a 200-mm wafer fab, Intel claims.

(See Feb. 25 story.)

Taiwan chip makers get go-ahead
to invest in China fabs, paper says

The word in Taipei is that the Taiwanese government has decided to give the island's chip makers the go-ahead to build 200-mm wafer fabs on Mainland China. Executives from the nation's IC companies, including TSMC chairman Morris Chang, have complained harshly about the government's delaying tactics here.

According to Taipei's Economic Daily News, Taiwan's cabinet will give island semiconductor companies approval to build 200-mm wafer fabs in China by the end of March. The Chinese-language paper reported that the cabinet and parliament agree on how to go here, but the government's priority is to set up a system of "effective management" for chip production investments in Mainland China.

Under the planned policy, the paper said that only two-to-three 200-mm wafer fabs will be allowed to move to China, according to Chen Ruey-long, a vice economic minister. And the total investment cannot exceed $5.7 billion, he was quoted as saying.

The government has been waffling over its election campaign pledge to lift the ban on chip manufacturing investments in China. Restrictions already had been eased for a number of industries, but not for plants for semiconductors or laptop computers.

(See Feb. 27 story.)

Intel says recession isn't over
in communications IC market

Intel is still describing the communications equipment market as dismal and says this business will not show any signs of a recovery until the second half of this year.

"The recession is not over," declares Sean Maloney, EVP running the chip makers Communications Group." We don't see a recovery until the second half of this year."

You remember the communications market--it took off like a rocket in the latte 1990s but ran into trouble at the end of 2000. Chip makers claimed the market was only experiencing an "inventory correction," but a glut of inventory and collapsing demand turned this market into a disaster.

The dynamics of the market, however, have changed in recent weeks. "It is not an inventory issue right now," Maloney says. "Everyone is pushing out capital spending. That's the real issue." He is talking about a lack of capital spending among the major carriers and equipment vendors.

But Intel does see some bright spots. The Gigabit Ethernet market is making a rapid transition from the enterprise to the high-volume desktop market--a move that will spur demand for LAN cards and chips, Maloney says. Intel is one of the world's leading suppliers of LAN cards and chips.

Another budding market now is metropolitan-area networks, he says. Equipment for 10-gigabits-per-second applications, including SONET and Ten Gigabit Ethernet, are beginning to move from metro nets to the enterprise, Maloney says.

Wireless local-area networks are also taking off, especially products employing the 802.11 standard. "802.11 is a clear winner," Maloney says. "My believe is the market will move to an 802.11ab combination. 802.11g is a distraction," he notes.

(See Feb. 26 story.)

Some see chip gear
sales getting better

I count myself in the group that sees another bad year for the semiconductor production equipment industry, but there are optimists out there. One of them went public this week.

While she still sees semiconductor equipment revenues in 2002 running flat to down 10% from last year, Robertson Stephens analyst Sue Billat is now looking for strong sequential quarterly growth this year as chip makers begin raising their capital spending budgets.

"We believe things are getting better," she reported at Robertson Stephens conference in San Francisco. Bookings in North America are improving and positive signs are showing up in the assembly and test, lithography, and photomask segments, she points out.

Transition to 300-mm wafer fabs will require additional capital equipment spending this year, Billat expects. She estimates that eight pilot lines and 11 volume production fabs will be running with 300-mm wafers by the end of the year. Spending for 300-mm fab equipment will hit $10 billion this year, and will shoot up to $15 billion next year, she says.

But here may be a fly in the ointment. Some chip equipment suppliers could see their 2002 numbers hurt by the introduction of new accounting rules. SAB 101 will delay the recognition of tool sales from the ship date to when the customer gives its final acceptance. Switching to the new rules may cause delays in the overall recovery, she says, as equipment makers adjust their accounting methods to comply with these new rules.

(See Feb. 27 story.)

HP counting on ZX1 chip set
to get McKinley gear rolling

Hewlett-Packard had great hopes for the IA-64 Itanium processor line that it and Intel had spent years designing. No wonder--they reportedly spent $1 billion getting the job done.

But, so far, systems based on the 64-bit processor have bombed. HP estimates that only 2,500 Itanium-based systems have been sold--about half of them by HP. The reasons were high cost, unimpressive performance, and frequent rollout delays.

Both Intel and HP are counting on new systems based on the successor McKinley chip to clean up their problems here. To get the job done HP has spent a year testing the new processor with its ZX1 chip set and that should lead to a smooth rollout when Intel starts shipping McKinley by midyear, it says.

McKinley's improved features over Itanium come from a 128-bit-wide system bus that runs three times faster than before, a large on-chip L3 cache, and a higher core frequency of 1-gigahertz. The HP chip set was designed to get the most performance at a minimum cost, so features ordinarily found on a general-purpose chip set were left off.

The chip set was designed to run four processors in parallel, though it can also be scaled down to two-way and single-processor configurations. The set is designed to handle more than 4 gigabytes of double-data rate DRAM running at 266 megahertz.

(See Feb. 25 story.)

TI tries to grab bigger share
fast-growing camera market

Digital cameras are finally doing what promoters have been predicting for more than a decade. About 17 million digital still cameras shipped worldwide last year, up more than 50% from the 11 million sold in 2000. Revenues grew to $8 billion last year, with Sony leading with about one-quarter of last year's $8 billion market. Other digital camera leaders are Olympus, Kodak, and Hewlett-Packard.

The major reason why cameras are flying off the shelf these days are lower prices that make the electronic cameras affordable for the average consumers. Toshiba, for example, now retails a 4-megapixel camera for $399, a resolution that would have cost $699 a year ago.

In a major move to grab some of this fast-growing market, Texas Instruments has started sampling a system-on-a-chip that's designed to provide more bang for the buck than its present chips. The new SOC incorporates TI's C54X DSP, an ARM9 processor running at 160 megahertz, and a peripheral set that supports digital imaging and audio applications, including MPEG-4 decode and encode functions. Volume production will begin about a year from now.

The faster DSP and ARM9 core makes the SOC about four times more powerful than earlier TI products aimed at the digital still camera market, TI claims. It supports both MPEG-2 and MPEG-4 decoding at 640 x 480 VGA resolution, and real-time encoding of MPEG-2 and MPEG-4 video at 352 x 288 common interchange format resolution.

That versatility is important because digital still cameras are not so "still" anymore, according to Ed Lee, analyst at Lyra Research. About half of all digital cameras introduced in the past year support video clip capture of 15 seconds to one minute in length, he points out.

"The fact that the TI silicon is programmable really helps the camera makers adapt to changes like video clips," Lee says. One vendor was able to add the clip function to one of its high-end cameras in just a few months, something that would be difficult to do with non-programmable electronics, he notes.

(See Feb. 25 story.)

Conexant spins off
new kind of foundry

Now here's what I call a real gamble. Conexant Systems is spinning off what will become a U.S.-based foundry to build specialized, high-performance products, including its silicon-germanium BiCMOS technology. This move could lead to an expansion of the pure-play foundry market from primarily a mainstream CMOS business to sophisticated specialty-processes and creating a new market segment of fabless RF chip companies.

To get the job done, Conexant has struck a deal with the Carlyle Group, a private equity firm, which will pay the chip maker $20 million and invest $30 million for 55% of the spin-off venture. Conexant is contributing specialty process technologies and manufacturing equipment in its Newport Beach, Calif. wafer fab.

Conexant is signing long-term supply contracts with the new foundry to build products for its mobile communications, broadband access, and Internet infrastructure businesses. The foundry, which still doesn't have a management team or name, will employ 650 and have estimated revenues of $150 million in its first year of operation.

The new foundry is the latest move by Conexant to spin off operations into new ventures. It is merging its wireless IC business with Alpha Industries to form a new supplier of integrated radio-frequency chips for cellular phones. Last year Conexant spit off an Internet infrastructure chip company called Mindspeed Technologies.

(See Feb. 25 story.)

We want your feedback, comments, criticisms, or questions. E-mail us at bhenkel@aol.com.

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