Will the next generation of ICs be built using optical lithography? "It is worth noting that this density of components can be achieved by present optical techniques," said Gordon Moore, "and does not require the more exotic electron beam operations which are being studied for the possibility of making even smaller structures."The legendary Intel co-founder made that comment in 1965, referring to transistors on 2 mil centers. Optical lithography has been buried more often than Dracula, but like that fabled creature it

keeps defying attempts to put a stake through its heart. For more than three decades, conventional wisdom has predicted that shrinking line widths would soon spell the demise of optical techniques in favor of electron beam, ion beam or X-ray lithography. Yet that day of reckoning continues to be postponed, leading some to question whether optical will ever meet the scrap heap.

"Even the greatest visionaries had no idea optical lithography would last as long as it has," said Dan Hutcheson, president of VLSI Research (San Jose, Calif.), who uncovered Moore's quote. "Part of it goes back to the philosophy underlying the scientific method, in which you only deal with facts and don't consider unknowns."

Another observer compared optical lithography's longevity to climbing a steep hill: "You can't see past the hill until you get to the top of it," said Gil Shelden, optical extension manager for Sematech (Austin, Texas). "Smart people found smart ways around problems."

The technical shortcomings of optical are usually apparent, but those of next-generation lithography often are not, Hutcheson said. At the same time, the current infrastructure is being sustained with the help of considerable resources: "Every engineer on the line is trying to make the existing lithography tools better," Hutcheson said. Next-generation lithography, meanwhile, doesn't enjoy the same scale of support.

"Analyses of when optical runs out of steam usually assume that only wavelengths will improve, and everything else is constant," Hutcheson said. "But what has kept optical going is that everything improves in time-photoresists, reticles, software."

For example, the ultimate barrier to using optical lithography was once thought to be the wavelength of light, said Lloyd Harriott, who heads the advanced lithography research department at Lucent Technologies' Bell Labs (Murray Hill, N.J.). "Using normal rules, you could resolve something equal to the wavelength of light, but if you went to a shorter wavelength the defraction lowered the contrast of the image, so it was blurred.

"The chemists saved the day. Improvements in deep UV photoresists were so high-contrast you could pull out images better than people thought," Harriott continued. Bell Labs recently fabricated a DSP with 120-nm features using a 248-nm deep UV stepper with phase shifting. "That's less than half the wavelength of light, well beyond what people thought possible," Harriott said.

"Momentum is the strongest factor in technology," said John Warlaumont, director of silicon technology for IBM (East Fishkill, N.Y.). When contact printing became problematic, the industry just moved onto projection alignment. And when the 1X projection mask proved a deterrent, chip makers embraced g-line and then i-line steppers in succession, he said.

Over the years, new terms have been entering the vocabulary-off-axis illumination, phase shift, optical proximity correction-as ways to defeat the perceived limits of optical. Another one could join them: imaging interference lithography, which is currently being researched at the University of New Mexico. "Some consider it virtually impossible to build, but that doesn't mean someone won't build it," Sematech's Shelden said.

Sematech is actively pursuing a 157n-m optical program, and equipment vendors are starting to add that to their own road maps. But the drumbeats for nonoptical lithography are growing louder. Recent optical technologies, such as deep UV, tended to take up to a decade to develop and market commercially, but the life span of those technologies is shrinking to only a few years, according to IBM's Warlaumont.

Phase shift has some serious technical and economic drawbacks that will open the door to technologies such as the Scalpel e-beam, predicted Bell Labs' Harriott. "Phase shift allows you to make smaller gates, but doesn't allow you to fully scale to increase packing density. Phase-shifting masks are also much more expensive than conventional masks."

E-beam systems used in the Scalpel program do not need special masks, and work with photoresists is already being used with deep UV, Harriott said. In addition, an infrastructure of optics and components already exists because of electron microscopes. With commercial vendors such as Applied Materials and ASM Lithography starting to embrace Scalpel, Harriott predicted Scalpel tools will begin making production chips by 2004.

IBM has been experimenting with nonoptical lithography for decades. It used e-beam for customization of bipolar logic for awhile during the 1980s and has also worked on X-ray lithography. The company has suspended development work on X-ray, however.

"We don't think we need to do a lot more on X-ray to demonstrate its potential," Warlaumont said, noting that IBM has fabricated complete 64-Mbyte DRAMs using X-ray lithography. E-beam technology, though, still needs to prove its viability, he said. IBM is working with Nikon and Darpa on the Prevail e-beam program, which is a "close cousin" of Scalpel, he added.

While some claim the 1x mask used in X-ray lithography is more difficult to make than the reduction mask used with e-beam, "I don't think anybody has demonstrated that," Warlaumont said.

IBM has no clear favorite between the two technologies, he said. But Warlaumont doubted that direct-write e-beam will ever be widely used for production because of throughput issues. So far, it has only been adopted with low-density gallium arsenide circuits, he said.

Ion beam has proved impractical so far, because it requires expensive reticles which are eroded by the abrasive nature of the radiation, VLSI's Hutcheson said. "Potentially it can also damage the wafers."

Hutcheson is predicting that optical's life could be extended to 2015 or beyond, at which point further scaling may become impossible, no matter what type of lithography is used. "Very few people agree with me," he conceded. "The only thing I can say is that I've been right for the past 15 years, and everybody else has been wrong."

"I've been hearing that the sky falling for the last 30 years," said Sematech's Shelden. "I'll retire before optical lithography ends."

The Century of the Engineer: Misunderstood Milestones