Contact: Paul Preuss, [email protected]

It has been an exciting horse race. But thanks in part to help from Berkeley Lab’s Center for X-Ray Optics (CXRO), researchers dedicated to making extreme ultraviolet lithography (EUVL) the tool of choice for the next generation of semiconductor chip manufacturers have left virtually all competitors in the dust.

Neville Smith, scientific director of the Advanced Light Source, makes a point of taking visitors to see the beamlines where optics for the prototype EUVL chip printer are rigorously tested. The “at-wavelength” EUV interferometer at beamline 12.0.1 is arguably the most accurate wavefront measuring device in the world; lately, this device has morphed into a mechanism that can print test wafers by itself.

The extreme-ultraviolet interferometer at the Advanced Light Source’s beamline 12.0.1 was converted to a subfield exposure station to test printing with EUV optics.

“EUVL research is a perfect example of the maxim that ‘for every $100 invested in research there is a return on average of $30 per annum in perpetuity,'” says Smith, quoting David King, chief scientific adviser to the British government. “Many research projects lead nowhere, of course, so it’s the really big winners like this one that keep the average high.”

Smith estimates that in terms of ultimate impact on the U.S. economy, EUVL research and development alone will have “paid for the ALS and maybe the entire U.S. synchrotron program.”

The micro-crunch
For thirty years the industry has been packing ever smaller, ever more numerous electronic devices into microscopic integrated circuits, doubling chip density roughly every eighteen months by using photolithography. But no printer can craft features smaller than half the wavelength of its light, and at wavelengths shorter than deep ultraviolet, refractive lenses don’t work: instead of bending light they absorb it.

In 1994 a consortium of semiconductor manufacturers launched a search for the best way to use shorter wavelength radiation in chip lithography. Four technologies entered the race: x-rays, electron beams, ion beams, and extreme ultraviolet. In October 2001 these manufacturers winnowed the field, choosing EUVL as the most likely technology to be able to reach the 32 nm “node” in the industry’s technology roadmap — which in practice means the ability to manufacture finished chips with individual lines as narrow as 13 nanometers (billionths of a meter) by 2009.

EUVL has been championed by a group of companies first assembled in 1997, now including Intel, Motorola, Advanced Microdevices, Micron Technology, Infineon Technologies, and IBM, who partnered with the Department of Energy’s “Virtual National Laboratory,” consisting of teams from Lawrence Berkeley, Lawrence Livermore, and Sandia National Laboratories.

Sandia is responsible for assembling the Engineering Test Stand (ETS), the collaboration’s “alpha” camera and chip-printing tool. Livermore employs layers of precision coatings to create optics and masks and uses advanced optical testing methods and defect-inspection technologies.

Berkeley Lab’s role, spearheaded by EUV program manager and former CXRO head David Attwood, has been to use ultraviolet light from the ALS in applying the world’s highest standards for measuring reflectivity and uniformity to test components of the ETS camera.

Multilayer mirrors
Because the chosen 13 nm wavelength EUV is absorbed by glass or quartz lenses (indeed by all materials, including air), the beams must be focused by curved mirrors instead. The mirrors, made by Livermore and Tinsley Laboratories, are built up of dozens of layers of silicon and molybenum, each a few atoms thick.

A small fraction of the light reaching each layer is reflected, but each little reflection constructively interferes and adds up to some 70 percent of the total falling on the mirror. Their critical uniformity is measured by Eric Gullikson and his CXRO colleagues at ALS beamline 6.3.2.

Not only the focusing optics but the flat masks themselves must be made of multiple layers, with opaque regions on the reflective surface forming the printing patterns. Even the tiniest flaw in a mask can damage a circuit printed from it, and flaws can occur on the substrate or in any of the mask’s dozens of layers. Masks are inspected for defects at ALS beamline 11.3.2.

Focusing mirrors are tested with the at-wavelength interferometer at beamline 12.0.1, whose coherent radiation, like light from a laser, is ideal for measuring optical components. The incoming beam is split in two, with one beam passing through and acquiring the aberrations of the optical system, while the other beam passes through a small pinhole, forming a nearly perfect spherical reference wave. The interference pattern of the two beams is recorded on a CCD camera.

The at-wavelength interferometer tests multilayer coated optics by generating interferograms, which measure departures from the ideal reflected wavefront.

If the optics were perfect, interference between the two beams would constitute a perfectly regular array of fringes; in the real world, aberrations displace the fringes from their ideal location.

A CXRO team led by Ken Goldberg, Patrick Nalleau, and Jeffrey Bokor tested two sets of four mirrors for the Engineering Test Stand (ETS), the alpha printer at Sandia, with the at-wavelength interferometer, measuring tolerances as fine as .04 nm — less than the radius of a hydrogen atom! In April 2001, using the first set of optics, Sandia’s ETS demonstrated its capabilities as the first full-scale EUVL prototype step-printer.

Meanwhile, back at the beamline…
A second set of mirrors designed for use in Sandia’s ETS was of much higher quality. Because these were not immediately needed in the ETS, Naulleau and Goldberg and their colleagues modified the at-wavelength interferometer and used it to print test patterns.

“The interferometer is wonderful for testing optics, but the proof of what the optics can do is in the printing,” says CXRO’s present director, Erik Anderson. He explains that “because high-coherence ALS light is ideal for interferometry but not for printing, Paul Denham and Patrick Naulleau developed a scanning illumination system that could fill the patterned mask area with a wide, programmable range of incident-angle EUV light. Using this scanning illuminator, we can produce almost any possible illumination pattern, which is very important in lithography research.”

The resulting device was called the SES, the “static exposure station” — static because, unlike the full-blown ETS, it projects only a portion of a mask pattern and exposes only part of a wafer at a time.

With optics designed for 100 nm features, the subfield exposure station easily created 70 nm features, as in these “elbows” (left). Exposure process control enabled printing of 39 nm features (right).

The actual chip-making performance of the optics was even better than they had been designed to achieve. Mirrors designed to make features with better than 100 nanometer resolution readily created 70 nm features. And by manipulating parameters like beam angles and exposure times, the CXRO team achieved much smaller features spaced just 39 nanometers apart.

They also tested ways to get rid of defects in masks. “If there’s a defect at the bottom of all those layers, you can’t directly repair it. So the question is how to smooth it out.” The team used the Nanowriter, CXRO’s ultra-high-resolution electron-beam lithography machine, whose operations Anderson manages, to create masks with programmed defects.

With the SES, they printed from these masks using different coatings developed at Livermore. Anderson says, “Coatings optimized for smoothing minimized the irregularities, like snow on grass. With nonsmooth coatings, the defects stood out like snow on a boulder.”

Programmed defects were introduced into masks before coating. Coatings optimized for smoothing reduced irregularities in printed features.

By printing from the newest set of optics and special test masks, says Anderson, “we have verified the interferometry and demonstrated the impressive capabilities of EUV lithography.”

And as commercial chip manufacture draws closer, CXRO’s EUV interferometer has another role to play. “We set the standard in interferometry, but for commercial use, the goal is get visible-light interferometry accurate enough so that at-wavelength interferometry won’t be needed for EUV steppers used in routine manufacturer. In close collaboration with our partners at Livermore, who specialize in visible interferometry, our task is to help insure that the required accuracy is achieved.”

The size of things to come
In April of 2002, Intel ordered the first “beta” EUV stepper from the experienced, Netherlands-based lithography manufacturer ASML, to be delivered in 2005.

Meanwhile current photolithography techniques, which have already used deep ultraviolet light at 248 nm wavelengths to craft chips with line widths of only 80 or 90 nanometers, will be extended to 193 nm and possibly 157 nm wavelengths for making chips with line widths less than 70 nanometers.

To make anything smaller than that, EUVL is essential. By 2007, commercial production of chips at the 45 nm “node” is expected to begin, using Intel’s ASML beta machine. By 2009, capability will be extended to the 32 nm node, resulting in finished chips with 13 nm line widths or less.

Computers using the first EUVL-made chips are expected to have ten times the speed and ten times the memory of today’s machines. That will be just the beginning of the next generation of superdense integrated circuits.


Additional information: