Kelly stressed a need to further research copper, looking into impurities and grain structure. The industry will be dealing with copper for a decade or more, he predicted, and these problems have to be solved or "they will stop us."

Design complexity is an increasing technical issue, becoming a massive problem for the industry, said Kelly, giving rise to an interesting new class of problems that exist in the boundary between design and process. These problems include signal integrity/noise, yield, power, and test.

Issues of signal integrity are becoming a larger issue with some of IBM's leading-edge customers, he noted, "particularly on critical timing paths."

Yield problems come at the interaction of a design and the process. Because the technology has advanced so quickly, designers don't understand well enough the issues and techniques used in deposition, printing, etch, etc., and the problem will worsen as the process accelerates. Design tools, methodology, and debugging haven't kept up with the process technology, he asserted.

The future winners will be companies that reach out to design houses and offer design support - something Kelly said he's seeing customers request more often.

Other winners will be companies that focus a business plan on one of two things - either providing a product differentiated by high customer value and high technological profit, or those who provide a commodity product, at low customer cost and low technological profit. The industry, he said, will bifurcate along those lines, and woe be to the company that tries to cover all the bases.

"If a company chooses to exist anywhere but in the upper right corner or lower left corner, you will be challenged significantly for existence [see illustration]," he said.

Successful companies will be tightly coupled with their customers in growth segments, and there will be a growing need for R&D partnerships.

"The largest companies in this industry can't solve these problems alone - we need more and more collaboration," he said.

Kelly also gave some mention to industry drivers, pointing mostly to grid and autonomic computing as key technologies. Grid computing refers to using the Internet backbone to link computational power of various machines. Work is being done on this in Great Britain, and the US national labs are also driving research in this area. Autonomic computing refers to systems in the grid that can self-heal, and monitor other computers in the grid and help them heal when issues arise, too.

There's also a lot of activity in pervasive computing, said Kelly, wherein a home will be full of devices that connect over the Internet and share resources.

But, he cautioned, a lot has to happen to see these things come to pass. In particular, he mentioned continued work is needed on:

• LINUX/open source
  
• Open middleware standards
  
• Wireless standards
  
• Fiber bandwidth evolution
  
• Powerful devices linked to backend infrastructure
  
• Security
  
• The "last mile"
  
• Data storage/processing standards (XML).
  

APC Gains Traction; e-Manufacturing Next
For years chipmakers have talked about automatic process control (APC). Now that a wafer may cost more than a new BMW, they are putting their money where their mouth is. Evolving control strategies for all levels, from individual processes to fabwide optimization with fault detection at each step, are being explored at the AEC/APC Symposium in Snowbird, UT, this week.

Keynoter Brian Harrison, Intel VP and GM of fab/sort manufacturing, said that every wafer Intel processes now has feedforward and feedback control of CDs (critical dimensions) based on CD/SEM data. Dynamic adjustment of etch time compensates for measured variation, and feedback is used for dose control in lithography for the next wafer. CD variation in high-performance microprocessors can cause some chips to run slower, making them worth hundreds of dollars less.

Harrison cited some other major benefits of APC at Intel. Rework has been reduced over 50%, and it takes half the headcount to maintain the same quality for 130nm production as it does for 180nm wafers. He added a long list of other benefits, including productivity, better Cpk, device performance, speed matching, and more precise control of processes. Wafer values shoot way up going to 130nm and beyond, Harrison pointed out, so a major increase in APC applications is planned.

Fault detection classification and analysis is a major target for future APC applications, but the problems are complex. Some variations can be tracked to a single cause, but many are dependent on a wide range of other variables. Multivariate analysis is done successfully in other industries. Joe Qin, U. of Texas, Austin, did a survey for AMD and found 4,600 applications, mostly in the chemical, refining, and other process industries as well as aerospace, with the most complex involving a matrix of 64X25 variables.

"In a refinery a defect is more forgiving," Qin said, since the misprocessed fluid can be used for something else. "But in a wafer fab, you throw the wafer away."

In the push for quick results, some fabs are tackling the univariate problem now while working on multivariate solutions for the future. At Infineon's fab in Regensburg, Germany, for example, Peter Hoehmann reported on a pragmatic approach to managing APC for fault detection control (FDC). He estimated more than a billion data points are gathered each month in the fab, which makes logic and power devices. A covariance matrix is set up for each tool, with correlations and deviations, resulting in a single number. Setting up the matrix isn't easy, Hoehmann explained. Decisions have to be made about which deviations are relevant or significant, and this takes engineering judgment. A tool is observed for about a week to build an effective model.

"About 80% of process deviation is univariate," he believes, so major advances can be made without dealing with the 20% that are multivariate. These are complex and hard to interpret.

Interest is rising in e-Manufacturing, an approach to integrated fabwide control, reported Harvey Wohlwend of International SEMATECH.

"We've had standing room only sessions with 50 plus attendees at our meetings at recent SEMICONs," Wohlwend said.

SEMATECH and Selete in Japan have worked jointly on equipment engineering guidelines, which are now on the SEMATECH website, along with the e-diagnostics guidebook 1.4 (www.sematech.org/public/resources/ediag/index.htm). There will be a vote in October on security interfaces for e-diagnostics, he noted.

Much work remains to be done, Wohlwend suggested, not just in agreeing on standards, but also in getting interoperable software and in getting better, more accurate, and complete data from tools. While standards committees are working toward agreement on interfaces and frameworks, chipmakers such as Intel and Texas Instruments already have developed their own fab-level architectures, he agreed in response to a question. He said that different architectures were expected, but emerging standards will help provide more commercial solutions that are interoperable. Right now, one speaker suggested, 80% of the money spent on fab software is for proprietary systems, which slows development, creates duplication of effort, and makes interfacing more difficult.

A major driver for APC is 300mm wafer processing. Intel's second 300mm commercial production plant, FX11, is ramping up in New Mexico and two more are being constructed, Intel's Harrison reported. A 300mm plant provides Intel with 2.4X die/wafer, 30% lower cost/die, and 40% less water and energy consumption. To make them work efficiently, Harrison said, "e-Manufacturing is a must."

"We need delivery of some lagging 300mm standards," he urged, citing ones for process management (E40), control job (E94), and substrate tracking (E90).

The difficulty of establishing an automated FDC system in existing fabs, using SECS/GEM interfaces, was cited by Elfido Coss of AMD. The company's vision for a FDC system is not one database (DB), but many DBs in a data storage component managed by a data broker. An analysis engine, possibly with several external analysis engines, would have links to a classification engine. Built in would be business rules in the fab, such as who should be contacted, whether there is a need to wake someone up, or whether work on a lot should stop immediately.

The FDC will act like a traffic cop, Coss explained, linking users and suppliers of data through filters. Both synchronous and asynchronous communication will be needed, but commercial data brokers don't allow asynchronous links, he complained.

Many added sensors and additional capabilities will be needed to make such an automated FDC feasible, so AMD needs to build the infrastructure from the ground up, Coss explained. This can't be imposed on an existing fab so it is a goal for new fabs, he said.
--Bob Haavind, Editorial Director
(October 31)

Semico: Mild US Recession, Still Positive on China
Semico Research Corp. started its recent forecast workshop emphasizing that it is not in the double-dip camp and it projects 30% revenue growth (17% unit growth) for semiconductors in 2003.

Another downturn is expected in 2005 due to 300mm capacity coming on board in China and a lull in PC and cell phone upgrades, since the replacement cycle will have taken place in the 2003 to 2004 timeframe.

Comparing economic data from previous US recessions, Semico notes that the current US recession is mild. The worst recession since 1960 occurred from 1973 to 1975 - it lasted 16 months and the real GDP was -3.40% with a peak unemployment rate of 9%. The next worst recession was in 1981 to 1982 - it too lasted 16 months; the real GDP was -2.90% and the peak unemployment rate was 10.8%. The 2001 recession was nine months long, had a real GDP of 0.3% and a peak unemployment rate of 6.1%.

Semico sees some indicators in Japan improving. Thanks to its auto industry selling more to Europe and the US, and providing electrical devices and general machinery to Asia, the country has an export surplus. Korea also has a better economic outlook, driven by a 20% increase in exports during the first seven months of 2002, especially in consumer goods and communications. Europe, however, is still going through difficult times with unemployment in Germany hovering at 10%, and in the UK unemployment is at 5.5% for 2002.

Looking at the differences between IDMs and foundries, Semico Senior Analyst Joanne Itow notes that because most IDMs will not build 300mm or 100nm-capable fabs, by 2006 foundries will provide over 40% of the wafers manufactured at 180-, 130-, and 100nm technologies.

Itow does not believe that China's foundries will pose much of a threat to those in Taiwan for several more years. She also predicts that foundries will see a healthy growth from the communications market between now and 2006: from a 16.3% year-over-year growth rate in 2002 to a projected 22.5% by 2006. In contrast, the computers segment represents a 41.7% year-over-year growth rate in 2002 and is expected to decline to 29.7% by 2006.

In marked contrast to the gloomy outlook on China presented at last month's SEMI forecast luncheon, Semico is still keen on the country. Jim Feldhan, president of Semico, notes that there is an underlying fundamental that cannot be stopped - the priority the country places on education. As an educated populace becomes more aware of the possibilities at its feet, its wants and desires tend to increase. He expects the trend to pick up momentum as more and more people return to China - part of the country's re-patriation program. "This is a normal progression from an agrarian to an industrial culture," explains Feldhan. "This movement has started and is now nearly impossible to stop."

Regarding China's lagging edge semiconductor technology - now at 0.25-micron and larger - Feldhan believes the situation will change. "TI will be sending partially processed 0.13-micron wafers to China for final processing - this will be a learning experience," says Feldhan. He does allow that China needs to do a better job at protecting IP, and until that happens, he thinks the flow of IP to the country will probably be slower than it could be.

Feldhan further notes that the Chinese government is spending billions to upgrade its infrastructure. One such example is that of ON Semiconductor - the plant had a sub-standard road leading to its plant and the government built a six-lane highway to it. "As the Chinese economy develops, there will be growing pains," states Feldhan. "Corporate and personal failures will occur, but for every failure you can find several success stories."
--Debra Vogler, Technical Editor
(October 24)

China Gearing Up for IT Manufacturing
China is becoming a better market for selling high tech gear, but at the same time it is building a foundation to become a formidable competitor to the world's leading high tech nations, especially in IT manufacturing. These were the major messages of Dan Brody, China representative for the US Information Technology Office, at a special program on China markets at SEMICON West.

The value-added tax (VAT) has dropped from 17% to 6% as China enters the World Trade Organization, Brody pointed out. Its foreign trade of about $550 billion a year makes it seventh among trading nations, with a GDP of about $1.1 trillion compared to $10 trillion for the US. In 2001, China took the world lead in cell phone users at 166 million, but that growth is now slowing, reaching 176 million users in June, 2002, Brody said, as the market begins to saturate. He predicted that the cell phone market there is poised for overcapacity and price wars.

Meanwhile, however, China is becoming a major assembler of high tech products. IT manufacturing should grow 20% per year for the next five years, Brody predicted, rising from $181 billion in 2002 to $302 billion in 2005. PC manufacturing is rising in China, and this year the Chinese expect to export 750,000 PCs, when none were exported last year. He suggested that the world's PC vendors should start to be afraid, "... very afraid!" he added.

Television manufacturers in China are building about 40 million sets/year, 30 million for the domestic market, and 10 million to be "dumped" in the US, since the European Union has very high tariffs, Brody stated. There are hundreds of thousands of workers in this industry, Brody said, and if the trade were shut down, almost all jobs in entire cities would be lost.

Out of some $16 billion of semiconductors used in China, more than 80% were imports, with only about $2 billion worth made in China. That should change dramatically over the next few years, he suggested, with domestic production reaching 30% by 2005 and 50% by 2010. One sign of the impending change is the rise of IC design firms.

"A year ago there were 30 to 40 design houses," Brody explained, "Now there are 240 to 250 of them, many with managers from Taiwan or returnees from other foreign countries."

Right now, he added, the firms are mostly very small, with only five companies having revenues of more than $12 million.

At SMIC, with 2,400 employees, about a third are from overseas, with Fab 1 and Fab 2 in production and Fab 3 being built. SMIC plans to take advantage of government efforts to bring more chipmaking to the Beijing area by borrowing up to $600 million for a factory there. The government will give them $15 million, and other companies will put in another $200 million, with SMIC investing over $300 million, he said.

Foundry services are also rising in China, Brody said, with NEC Hua Hong in Shanghai, for example, going from 30% foundry operations currently to 50% by the end of the year as it exits the DRAM business.

While COCOM is gone, licensing of exports of process tools and equipment to China are still required in the US under the Wassenaar multilateral agreement among 33 nations, according to Bernard Kritzer, director of the US Office of Strategic Trade & Foreign Policy Controls. While semiconductor manufacturing offers a big growth market for industry, there is concern about diversion to military uses, he explained. Last year, of about 1,300 license applications, some 72% were approved, he said.

A lack of enough information was often the reason for disapproval. National Security Export Controls apply to semiconductor manufacturing equipment, and licenses are considered case-by-case without any predetermined limits on feature sizes or wafer diameters. Requests to export equipment aimed at military end uses will be denied, he indicated, and sometimes the vendor may be unaware that intelligence throws suspicion on the parties buying the equipment.

Each vendor is responsible for knowing the regulations, Kritzer said, and the rules are updated constantly. They are on the office's website at www.bis.doc.gov. While our trading partners in Europe and Japan also follow the same procedures, their licensing is much faster than in the US, he said, so efforts are being made to shorten the application review, which now averages about 74 days for China versus an average of 44 days for other countries, such as India.

There is an annual review of the Wassenaar rules, and semiconductor manufacturing equipment should get much higher priority, Kritzer pointed out, as the fab/foundry market grows in China.
--Bob Haavind, Editorial Director
(October 17)

| Add RSS Feed

---

More Industry News Articles >

Search Archives >

Advanced Packaging Article Categories:

Industry News		SEMICON News
New Products
Advanced Packaging Current Issue Table of Contents

---

Search Industry Specs >

Search Industry Jobs >

Magazine & E-Newsletter Subscriptions >