Extreme ultraviolet (EUV) lithography, showcased at a press conference at Lawrence Livermore National Laboratory on Wednesday, will allow chip manufacturers to "draw" circuits as small as 10 nanometers wide. That's 1/18th the size of the features on today's most advanced chips, or 1/18,000 the diameter of a human hair.
PC processors made with this equipment, expected to hit the market in 2005, will eventually run at 30GHz and contain billions of transistors. Moore's law states that the number of transistors on a chip, and hence its power, doubles every two years. Increasing the number of transistors comes from shrinking them.
Just as important, the new process will make it easier to boost chip power than in the past.
"Lithography has been one of the limiting factors of our industry," said Craig Barrett, chief executive officer of Intel, one of the primary members of the EUV LLC coalition.
Still, a lot of work remains. To commercialize the process, the industry will have to adjust to a completely new way of making chips, said Sunlin Chou, senior vice president of the technology and manufacturing group at Intel.
With current lithographic techniques, circuit patterns are reduced through a series of lenses and then get printed onto a silicon wafer. In EUV lithography, the pattern bounces off a series of highly polished mirrors during the reduction process.
Mirrors present a whole new cavalcade of scientific challenges. "The surface of the mirrors has to be polished to atomic dimensions," Chou said.
Techniques for keeping dust out of the manufacturing process also have to be developed. Currently, anufacturers put a reticle, or membrane, on top of the wafer to capture debris. The reticle, though, interferes with the reflecting images. One possible solution now, he said, is thermothoresis, which involves creating an envelope of heated air.
The conversion will also cost a lot of money. EUV LLC has spent approximately $250 million to get to this point. To develop machines for the commercial market will take another $250 million to $750 in research, Chou said.
Semiconductor companies are also going to have to invest in new, more expensive equipment. Machines that scan semiconductor images for EUV lithography, for instance, will sell for around $20 to $30 million, twice as much as today's scanners.
Converting to new forms of lithography, though, is inevitable. Current optical lithography methods are close to their physical limits. The images simply can't be made much sharper, which limits how small manufacturers can shrink chip features.
Lens development has also become a major chipmaking challenge. Initially, it was easy. Intel founder Gordon Moore, Barrett recalled, went down to a camera store to select lenses. Now, companies are forced to test out new materials and coatings all the time.
"That's why the current extensions of today's technology have been so slow," Chou said. "With EUV, there does not need to be material changes along the way."
Manufacturers will begin to install test EUV equipment in late 2003, Barrett said. Chips produced with the new technique will come out in 2005. Features on these chips will measure 70 nanometers. The same basic equipment, however, will be useful for producing chips with 10-nanometer elements years later.
The success of the project will likely prompt further government-business cooperation. In 1996, EUV was only one of several potential options for becoming the standard for the future in lithography. Now it is the leading contender, thanks in large part to the development work of government and industry researchers participating in EUV LLC.
"In the old days, you might find a Bell Labs or an IBM doing this kind of research, but those days are gone," said Richard Stulen, chief operating officer at Sandia National Laboratories. "This is one of those large projects that a single company couldn't have undertaken."
"Then when the problem is solved, we will go out and beat each other over the head in the industry," Barrett added.