SOI wafers will be a boomer, if cost drops

TOKYO--Silicon-on-insulator technology is really neat and proponents have long predicted that widespread usage would be coming. But today, even its strongest backers admit that high wafer costs are still holding SOI back.

Cost is the only remaining frontier, declares Motorola staff scientist Michael Mendicino. "Five years ago, I could show you road maps that said costs would be competitive with bulk at the volumes we have today," he points out. "We have the volumes now, but we don't have the prices. Cellular phone makers would love SOI's 20% performance benefit and dynamic power, but they can't afford the price."

"We think SOI is a strong candidate for the future, but these days the most important point is the cost of the substrate," agrees Yoshio Miura, retired R&D director at Nippon Motorola. "When I was working at Motorola, "We used to say that if the price could drop to three times that of bulk, SOI use would increase dramatically. But today that price is still between five times and 10 times."

Motorola has shouldered most of the cost to develop this technology and wants SOI in place for the next upturn, Mendicino says. The chip maker intends to develop a future generation of SOI-based DSPs, he says.

Motorola isn't the only chip company pushing SOI. IBM, Advanced Micro Devices, and Texas Instruments also are heavy backers for 0.13-micron and 0.10-micron logic. But some people have been sending mixed signals on the merits of SOI vs. bulk CMOS. Sun Microsystems and TI decided not to use SOI for Sun's next-generation microprocessor. But IBM will still use SOI to make its upcoming 0.13-micron Power4 processor.

SOI got a thorough airing at the recent solid-state devices and materials conference, with 16 SOI-related papers. SOI wafers were riddled with as many as 10,000 defects as recently as 1997, but have been reduced to a fraction of that level and now nearly match those of bulk silicon wafers, Mendicino said.

SOI's reduced junction capacitance, lack of reverse body effect and radiation hardness all indicate a bright future for the technology in limited applications, Mendicino said. He was bullish about its prospects for high-end devices, but bearish on its use in memory, ASIC and low-power communications devices. SOI costs will never be low enough to compete in the memory sector, and the ASIC market moves too fast for SOI-based designs, he said. "But everyone is agreed that for high-performance logic, SOI is top of the list," he said.

What DRAM biz needs is
for major vendor to pull out

TOKYO--The global DRAM business is going to end up looking like the U.S. airline industry if Dataquest's latest market forecast holds. But even if two suppliers end up owning 90% of the market, would they still make any profit? Probably not.

Even though the DRAM market has gone through a heckuva lot of consolidation in recent years, current market trends could accelerate this concentration even more, Dataquest says.

The market researcher now figures that worldwide DRAM revenues will decline by 67% in 2001 and an additional 19% in 2002. That means that sales will hit $10.5 billion this year and decline to only $8.5 billion next year. That would amount to just 20% of the record year of 1995 when DRAMS sales reached $41.8 billion.

"The omens are bad for the DRAM industry next year," says Dataquest analyst Andrew Norwood. "All DRAM companies are now losing money and will do so until the fourth quarter of 2002." In fact, he predicts, "when the market recovers, we may find that there are only four major DRAM players and not the current six."

"Some vendors may not even survive that long," he adds, because things are bound to get worse. While "some companies announced small production cutbacks in the past few months, we had hoped to see substantial cuts by the major companies," Norwood notes. But it didn't happen. "No company wants to take the initiative for the fear of losing market share," he notes, "so now they are all going to suffer in 2002."

Some companies already are on life-support systems. South Korea's Hynix Semiconductor is being kept afloat by its creditors in the hope that the market will recover in 2002. Others already are looking at how to exit the DRAM business. "Toshiba is investigating a possible memory joint venture with Infineon," Norwood notes.

"It's not too late for a possible upside to 2002, but it will be consolidation that makes this happen," Norwood says. "If we saw a major manufacturer go under, things would change fast. Then we'd see market revenue grow significantly in 2002."

Will Strauss cuts his 2001
DSP forecast even more

TEMPE, Ariz.--Growth of digital signal processor business continues to be "in lock step" with overall semiconductor market "train wreck," says analyst Will Strauss of Forward Concepts here. He's not only lowering his forecast for the global chip market from a 25% decline to a 30% fall, he's also dropping his DSP forecast to a 30% decline to $4.3 billion in 2001.

Strauss also lowered his DSP recovery next year from a 35% to a 32% growth rate. Still not bad.

The major disappointment this year was wireline communications, including telephony infrastructure equipment, and this business will continue to be a drag on DSP sales through the second quarter of next year, Strauss predicts. An exception to this forecast, he adds, appears to be Internet Protocol lines and PBX systems.

But he is lowering his 2001 forecast of cellphone shipments to 390 million units, down from 416 million shipped last year. "It now appears that only about 380 million actually reached customers," he says, "and this problem was compounded by DSP chip shipments that overshot the 416 million mark by another 20 million or so."

The only other DSP market with more than 100 million units shipped annually is for hard disk drives and it is running flat to slightly positive this year, Strauss says. That's because disk drive manufacturers had their inventory glut last year.

Some smaller DSP markets are showing positive growth this year. This includes digital still cameras, a 17-million unit market this year that is growing 57% over last year, and MP3 Internet audio, which is now going into portable CD players, camcorders, digital still cameras, cell phones, and flash memory players).

Embedded SRAM continues
to kill discrete SRAM market

SCOTTSDALE, Ariz.--Even when you take a longer look at the static RAM market to get away from the short-term carnage now going on in the memory business, the picture is not a pretty one.

Revenue for SRAM chips over the next five years will continue a slow decline that began in the mid-1990s, predicts Cahners In-Stat Group. That means this market will drift from $6.5 billion last year to $4 billion in 2005, according to a new forecast.

"The biggest factor contributing to the decline is an ongoing migration from SRAM to embedded SRAM," says analyst Jerry Worchel. This is going on in high-complexity, customer-specific, ASIC applications, and application-specific standard product (ASSP) designs, he notes.

When a sagging economy that's hurting all memory products is added to the equation, the SRM market "is unlikely to stabilize for another 8-to-12 months," Worchel predicts.

Over the next five years, the SRAM business will be dominated by asynchronous memory designs, with more than half of the revenues during this period going to these designs. Nearly two-thirds of the SRAMs sold during these five years will be asynchronous memories, according to In-Stat.

Some 80% of SRAMs shipped during the first half of this decade will go into communications systems, In-Stat estimates, with more than 40% of the total static RAMs turned out to be used in the Americas.

IDC says contract manufacturers
will zoom to $230 billion in 2002

FRAMINGHAM, Mass. --Here's some unbridled optimism for you. While contract systems manufacturers such as Solectron, Flextronics, and Celestica are going to show another sequential drop in revenues in the third quarter, International Data predicts a blow out for this business next year!

IDC already sees "light at the end of the tunnel" and predicts that worldwide revenues for contract electronics manufacturing will surge next year, doubling 2000 revenue of $103 billion to $230 billion.

Just a little on the optimistic side, I'd say. But there's no doubt that this class of customers is becoming increasingly important to the chip industry. Many contract manufacturers have expanded their design and semiconductor purchasing services.

"The current move to vertical and virtual integration in the contract manufacturing industry will allow the CM vendor to further penetrate the product development cycle and, ultimately, strengthen the relationship between the CM vendors and the branded system suppliers," says IDC analyst Kevin Kane

Many of the contract manufacturers also are using the downturn to "accelerate their long-term plans of targeting specific growth areas, such as wireless and optical segments or to position themselves for the economic recovery," says Kane.

IDC figures that Solectron continues as the world's largest contract systems manufacturer and holds about 21% of the market. Flextronics International is second largest with 12% of the business, while Celestica accounts for 10% of the business, according to IDC.

Eighteen of top 21 fabless
IC suppliers are U.S. based

SCOTTSDALE, Ariz.--It always makes me feel good to find a chip market where the U.S. is kicking butt! That certainly was the case in fabless IC suppliers, where U.S. companies grabbed 18 of the top 21 positions last year.

Zilinx and Altera, the major suppliers of PLDs, continue to lead the rankings, according to IC Insights here. Other top 10 U.S. companies were Qualcomm, Broadcom, Cirrus, nNvidia, PMC-Sierra, SanDisk, and Lattice Semiconductor.

Three Taiwanese companies were the only foreign vendors to make the Top 21 list. VIA Technologies was number three, while Sunplus Technology and Realtek Semiconductor were back in the pack.

Top 10 fabless chip suppliers

Global sales in $ millions

	Country	2000 sales	Growth
1. Xilinx	U.S.	$1,560	74%
2. Altera	U.S.	$1,377	65%
3. Qualcomm	U.S.	$1,250	9%
4. Broadcom	U.S.	$1,096	114%
5. VIA Technologies	Taiwan	$909	160%
6. Cirrus Logic	U.S.	$729	37%
7. Nvidia	U.S.	$699	97%
8. PMC-Sierra	U.S.	$695	165%
9. SanDisk	U.S.	$602	144%
10. Lattice Semiconductor	U.S.	$568	76%

Source: Company reports, IC Insights

iSuppli cuts 2-week-old
global forecast for ICs

EL SEGUNDO, Calif.--Another market researcher has gotten cold feet and has cut back its two-week-old forecast for the global chip market this year and next. "Decreasing consumer spending and lower projections for IT investments by enterprises will significantly reduce semiconductor revenues below our recent forecasts for the next few quarters," admits Greg Sheppard, head of market research at iSuppli. "The chip market will still experience a small gain in the fourth quarter thanks to traditional seasonal factors, but overall revenues for 2001 will decline by 30.9%--down 2.6% from the 28.3% decline we were forecasting just two weeks ago," Sheppard says. Next year's forecast was cut back as well. "With OEMs and contract manufacturers beginning to stretch out deliveries, we are also lowering our forecasts for semiconductor revenue growth from 9.9% to 4.1% for 2002," Shepard says. "We anticipate little positive momentum in the marketplace until well into the second quarter of next year."

Converting wireless nets
is going to cost big money

BOSTON -- Cost of converting wireless networks to next-generation services is going to be darned expensive for operators, but they're going to have to do it to stay competitive. As a result, global spending on infrastructure systems for wireless mobile communication networks will jump from $99.4 billion this year to a peak of $120.2 billion in 2004. Then spending will drift down to $114.6 billion by 2006, according to Yankee Group.

Much of the new investments over the next five years will be aimed at overlaying 2.5 and 3G generation technologies on top of existing cellular networks and services. While it's going to be expensive, "operators cannot avoid these costs without the risk of losing market share," says Phil Marshall, Yankee Group analyst. This is due to "the imminent obsolescence of second-generation technologies and the promise of next-generation services."

The lion's share of the new infrastructure investments will go for GSM, GPRS, and WCDMA formats, the analyst says, with an increasing amount of money going for electronics--63% of total capital expenditures in 2005 vs. 53% in 2001.

One chip gear sector is
still growing, thank you

NEW TRIPOLI, Penn.--Believe it or not, there's one sector in the semiconductor capital spending market that's shooting up this year.

Investments in 300-mm wafer processing tools are expected to grow 39% in 2001 to $11.4 billion from $8.2 billion last year, according to a new report from The Information Network here. Next year, the market researcher expects growth in 300-mm equipment purchases will slow to an increase of 16.7%, reaching $13.3 billion worldwide.

Next year, revenues from 300-mm systems will represent 48.2% of all frontend fab tools in 2002, The Information Network says. "Some of this equipment will be bridge tools, able to process 200-mm and 300-mm wafers," says Robert N. Castellano, president of the research firm.

"The strong growth of 300-mm fabs and fab equipment in 2001 . . . is due to the substantial savings of up to 40% that could be realized by 300-mm chip production," Castellano says. "That trend will continue in 2002 as chip manufacturers move to smaller feature sizes to remain competitive and build on a 300-mm platform to further reduce costs." Now, the SEMI trade group is predicting that worldwide spending on chip production systems will fall 35% in 2001 after surging last year by 87%. First-half sales were down 16.1% in chip equipment to $18.3 billon. Next year SEMI predicts semiconductor equipment spending will rise 11% next year.

Look out! 10-Ghz MPU
is "right around corner"

SAN JOSE--It's always fun for a journalist to cover the Microprocessor Forum, held annually in mid-October in San Jose. I call it the "gee whiz" show, where the experts try to out-do one another in predicting future products and technology trends.

The star of tomorrow at this year's conference was the 10-gigahertz microprocessor. Yup, an incredible 300-million transistor processor for PCs is right around the corner, predict the experts.

Right now, Intel claims the world's fastest processor for desktop PCs, which runs at speeds to 2 gigahertz. And by early next year, it will launch the 2.2-gigahertz Northwood processor.

"You can expect to see a 10-gigahertz processor by 2005," predicts Bill Pohlman, chip veteran who is now chairman of Primarion, a Tempe-based developer of optical I/O technology. "Moore's Law is pretty much on track," he adds. "You might even see one earlier," he says.

Such a super processor most likely would be manufactured with 0.05-micron CMOS technology, Pohlman says. The 0.5-to-0.8-volt part will probably carry more than 16-megabytes of cache, offer fault tolerance features, and run up to 30-gigabyes-per-second, he says.

But there's still a lot of work to be done. And there are some major challenges, "namely power and thermal limitations in processor design," Pohlman notes. "Packaging is also behind," he adds.

"There also will be a need for processor and system designers to work together, says Martin Rausch, platform technology manager at Intel. "Ten gigahertz processors pose significant system-level challenges to the industry," he points out. "Cost will also be a major factor in the acceptance" of such monster chips.

12% dip in IT spending this year
will be 1st annual decline in decade

SCOTTSDALE, Ariz.--Don't count on information technology customers to help lift the chip business out of the hole this year or maybe even 2002. Overall business IT spending will fall more than 12% this year from 2000--the first annual decline in overall business IT spending in a decade, according to Cahner's In-Stat Group.

The impact of the September 11th terrorist attacks is expected to drive IT spending down even more, particularly among companies with less than 100 full-time employees, where business failures are predicted to be on the rise. Along with smaller IT budgets in 2001, this guarantees falling IT spending by U.S. businesses this year and a slow recovery in the future, In-Stat says. "This is literally a new economy," says In-Stat research director Kneko Burney.

"The unfortunate reality is that many businesses over-invested in IT products and services in 2000, creating sluggish demand in 2001," he says. This factor alone was expected to lead to a decline in IT spending before the September 11th attacks, he adds. "We believe the fear and economic uncertainty following these events will lead to a freeze in IT investments for the next few months and will continue to negatively impact our economy and IT spending well into 2002."

IT spending segments that will be least affected by the attacks will be communication services, networking, and outside services such as applications integration and hosting, In-Stat predicts.

The average company spending for IT this year is predicted to be $19 million for enterprise firms (more than 1,000 employees), a drop of 18% from 2000; $846,000 for middle market firms (100 to 999 employees), a decline of 13%; $70,000 for small businesses (5 to 99 employees), a drop of 17%; and $6,000 for SOHO businesses with less than five employees, a decline of 27%.