Peter Silverman, Intel's director of lithography capital equipment, said that to achieve commercial viability, EUV (extreme ultraviolet) scanners must reach throughput levels of about 120 wafers per hour, a goal that left many in the audience shaking their heads in disbelief. Today's scanners have a peak throughput of about 90 wafers per hour, but at the factory floor typically achieve 60 wph productivity levels, with down time and servicing figured in.

Silverman also mentioned a per-scanner target cost of $20 million, which prompted one symposium participant to say he was "surprised the audience didn't break out laughing, it was so unrealistic." Some experts have predicted that EUV scanners will cost $50 million or more, with mask costs also hitting new levels of sticker shock.

About 250 people are attending the International Sematech-sponsored event, which supersedes the Next Generation Lithography (NGL) Workshops organized by Sematech in previous years.Silverman said that for commercial EUV scanners to be ready for production at the 45-nm process technology node, which moves into commercial production in 2007, "Intel and other companies that are rapidly pushing logic need beta EUV tools by 2005."

Optimism about the long-term potential of EUV lithography, which uses 13.5 nm wavelength light and thus can be extended for several process generations, always has been balanced by fears that the solution may exceed the budgets of all but a few manufacturers of microprocessors and memories.

With much of the semiconductor equipment and materials industry facing "very hard times," Silverman said the commercial evolution of EUV requires that the major semiconductor companies "step up and show their interest" by placing orders for EUV beta tools, he said. "I'm sure that there are people here who would be glad to take orders" for beta EUV scanners, Silverman said.

ASML is currently the only company offering to produce beta EUV systems, but the two major Japan-based lithography vendors, Canon Inc. and Nikon Corp., are expected to be ready with prototyping tools by 2005, said Chuck Gwyn, program director of the EUV LLC (Livermore, Calif.) Silverman said at the 45-nm node a logic chip might require four mask levels of EUV lithography for critical layers, another 11 mask layers supplied by 193-nm lithography, and the rest with 248-nm tools.

Striking gold

For the 32-nm node expected in 2009, EUV scanners and masks will be needed for nine layers. Asked after his speech if Silverman was expecting the industry to jump over the 157-nm lithography generation, Silverman said "the issue for 157-nm lithography is purely a calcium fluoride issue: high-quality calcium fluoride is more expensive than gold."

For the 193-nm scanner lenses now coming to the market, about 10 percent of the lens material comes from calcium fluoride. For the 157-nm scanners, all of the lenses are made from the CaFl material.

"The yield for 157-grade calcium fluoride is quite, quite low. Several new factories are needed, and while some investments are being made, nothing big enough is being built to supply the industry. So it is an open question whether 157-nm systems will be available in volumes, in the time frame that industry needs them. It also depends on whether EUV will be ready in time," Silverman said in an interview following his keynote address.

Silverman said that Intel's "primary goal" is to make EUV work for critical dimensions at the 45-nm node. "The default is to make 157-nm work, and if we can't do that, then we will make 193-nm lithography work by using more reticle enhancement techniques."

To get high-throughput EUV systems, the laser source must be improved to generate more of the extreme ultraviolet radiation, or light. Today's best YAG lasers hit a xenon target and generate only about 10 Watts of radiation, a level that must be boosted to 100 Watts or more for high-throughput commercial production.

Gwynn, the EUV LLC director who plans to retire from Intel at the end of this year, said there are several ways that the throughput of EUV scanners will be improved. The most obvious is to boost the output power of the source by an order of magnitude, and Gwyn said over the past two years source output has improved by an order of magnitude, albeit only to 10-20 Watts.

Improved resists needed

The sensitivity of the resist must be improved, and in his keynote address Silverman mentioned a target of 2 milliJoules of sensitivity for commercial EUV resists - another goal considered lofty by participants at the symposium.

Gwyn said throughput can be increased as the reflectivity ratio is increased for the 11 multi-layer coated mirrors, which form the condensor and reflective optics subsystems. A 1percent improvement in the reflectivity of each mirror will boost the power of the light at the wafer by 10 percent, he noted. And much work remains to be done on condensor design, collecting more of the photons generated by the laser source.

The symposium comes as the EUV LLC winds down much of its early development and proof-of-concept work. Since 1997, more than 250 scientists from three national labs - Lawrence Livermore, Lawrence Berkeley, and Sandia - have worked within a "virtual national lab," which worked in cooperation with the EUV LLC. A much smaller number of engineers from the EUV LLC member companies, including the founding members, AMD, Intel, and Motorola, and more recently Micron Technology, IBM, and Infineon, also worked to develop the EUV engineering test stand at Livermore.

With EUV moving to the commercialization phase, International Sematech will handle much of the vendor coordination effort, Gwyn said. And EUV development efforts are picking up momentum in Europe and Japan, he added. Since the United States lacks a major lithography vendor that would commercialize the EUV technology, the EUV LLC has worked closely with ASML, which acquired the Silicon Valley Group lithography division two years ago.

"Europe is leading in EUV commercialization, but Japan also is scaling up rapidly," Gwyn said. If progress continues, the industry will meet its goal of using EUV initially for commercial production in 2007, with volume scanner shipments by the end of the decade.