X

A fab construction job

When IBM wanted to start making tiny chips on 300-millimeter wafers, the first thing it had to do was raise the roof on an existing facility--all 4 million pounds of it.

Michael Kanellos Staff Writer, CNET News.com
Michael Kanellos is editor at large at CNET News.com, where he covers hardware, research and development, start-ups and the tech industry overseas.
Michael Kanellos
5 min read
When IBM wanted to start making tiny chips on 300-millimeter wafers, the first thing it had to do was raise the roof on an existing facility--all 4 million pounds of it.

The fab plant in East Fishkill, N.Y., had a 10-foot ceiling, which was sufficient to produce the 5-inch wafers it had been built to make. But equipment in 300-millimeter fabs needs 14 feet of clearance.

Big Blue's newly retrofitted East Fishkill, N.Y., chipmaking plant And that was just the roof. Getting the equipment into the factory took roughly five months and extensive planning. One of the lithography systems used in the facility weighs 40,000 pounds and took seven specially equipped big rigs to deliver.

"You don't have any perception of what you are spending sometimes," said Rich Brilla, director of 300-millimeter operations at IBM Microelectronics. "There are a lot of zeros."

IBM's retrofitting of its Fishkill fab exemplifies the challenge facing semiconductor makers today. The exploding costs of such fabrication facilities are widening the gaps that separate a few giants, which can afford to build their own fabs, from the rest of the semiconductor manufacturers that must share facilities or hire companies such as IBM to make chips for them. While outsourcing has its advocates, it creates risks and costs that can put a company at a disadvantage.

These new fabs are necessary to process wafers with 300-millimeter diameters, larger and more complex to make than today's less efficient 200-millimeter wafers. But even for a giant such as IBM, building a fab is an enormous task.

"There's more power in the servers room than NASA has for the space shuttle," Brilla said.

Construction zone
Here's some of what makes up IBM's revamped 140,000-square-foot East Fishkill, N.Y., facility.

• 200 miles of pipe and tubing

• 600 miles of cable and wiring

• Two million pounds of ductwork

• 50,000 gallons of paint

• Server room holds over 1,700 1GHz-plus processors, 1.5 terabytes of memory and 110 terabytes of storage. Runs on Linux.

• Capacity: over 500 wafers a day

• Cost to rebuild: over $2.5 billion  

Source: IBM
In the end, the final cost came to over $2.5 billion. By 2007, the price of building a fab is expected to reach $6 billion.

By contrast, in the early days, building a manufacturing plant for chips was sort of like shopping for art supplies. Intel founder Robert Noyce bought some of the company's first lithography lenses, used to focus light for "drawing" circuits on a chip, in the bargain bin at Adolph Gasser's camera store in San Francisco.

Circuit patterns were drawn by hand on wax, said Dan Hutcheson, CEO of VLSI Research, which studies the semiconductor-equipment industry.

The increase in cost and complexity is encapsulated in Rock's Law, which dictates that fab construction costs double every four years. That law is itself tightly connected to the better-known Moore's Law, which dictates that the number of transistors on a chip doubles every two years.

The vast majority of a fab budget goes to equipment that can make today's chips, which have more than 250 million transistors. About 80 unique tools--ranging from furnaces and dicers to esoteric equipment such as ion implanters and chemical vapor deposition machines--had to be installed in IBM's Fishkill facility before the first piece of silicon could be produced.

Fabs that make 300-millimeter wafers also require far more automation. For the first time, wafers are housed in something known as a "FOUP," or front opening unified pod, hermetically sealed housings that are shuttled about by robots.

Such automation is driving down personnel costs. Previously, two or more workers operated each piece of machinery, but only a few employees can manage entire production bays in 300-millimeter manufacturing.

"Operators are there for error correction and problems. The software skills of the operators are significantly greater," Brilla said. Breathing apparatuses and classic full-body "bunny suits" are also no longer needed because the FOUPs protect the wafer from environmental contaminants.

A world of building sites
Employing fewer workers will also reduce the advantage of building fabs in countries with low labor costs. Such costs will still be important for testing, packaging and assembling facilities, which depend more on manual labor, but not for fabs or semiconductor development labs, which employ far fewer but better-paid white-collar engineers. Most of the work is accomplished in fabs or development centers.

"Your equipment costs are the largest part of your operating budget. Your labor costs are immaterial," said Chuck Byers, director of worldwide brand management at Taiwan Semiconductor Manufacturing Co., likening semiconductor manufacturing to "doing the pyramids backward."

In China, for example, labor costs are low, but companies mostly tend to cite local demand, tax incentives and cooperation from local universities in development and recruitment as the three most significant factors for moving to the People's Republic.


Special report
Semi survival
The soaring costs of chipmaking
promise to recast an industry.


Even then, Western companies building in China, such as Intel, don't locate fabs there. Instead they opt for test facilities for the lower labor costs and research centers to capitalize on the growing engineering base.

Some countries are not pursuing manufacturing but are trying to participate in the semiconductor industry by becoming a research center for chip companies. France, for instance, is counting on its universities to make the country a center for analog chip development instead of focusing on manufacturing contracts, said Alex Pepe, vice president and director of strategy for technology and manufacturing in Motorola's Semiconductor Products Sector. Spain is pursuing similar goals.

By contrast, Mexico is mired in a difficult transition. For years, it was a manufacturing center. Now, however, it's more expensive than China, though it doesn't have the university or industrial effort in place to become a design center, Pepe said.

Advanced Micro Devices CEO Hector Ruiz, meanwhile, has his eye on Eastern Europe, citing the availability of engineering talent in Hungary, Poland, the Czech Republic and Russia. Complex math "is one area that Russia really put a lot of effort into, and it is paying off," he said. The company built a factory in Dresden, formerly part of East Germany.

The United States gets mixed reviews. Three of Intel's first four 300-millimeter facilities are in the United States, where the company is often the recipient of business-friendly tax or bond laws.

"You've got a number of governments. One is the sovereign state of New Mexico," Intel President Paul Otellini said. "I don't think you have to look over the ocean to find a cooperative government."

Others, though, say political and business considerations often don't make it easy. Despite years of effort, Virginia's semiconductor manufacturing base remains relatively small.

"It is much easier to sell the public to issue bonds to build a football stadium than to build fabs," said AMD's Ruiz. New York State, he added, "is the only place I know in the country that is doing it."