Erik Anderson, director of the Center for X-Ray Optics, with the extreme ultraviolet "at-wavelength" interferometer at the Advanced Light Source.

The extreme-ultraviolet "at-wavelength" interferometer at the Advanced Light Source's beamline 12.0.1 is arguably the most accurate wavefront measuring device in the world. Operated by the Center for X-Ray Optics (CXRO), this device has lately morphed into a mechanism that can print semiconductor wafers by itself.

"The interferometer is wonderful for testing optics, but the proof of what the optics can do is in the printing," says CXRO's director, Erik Anderson. In its new configuration, the interferometer delivers the printing proof.

For decades the semiconductor industry has been packing ever smaller, ever more numerous electronic devices onto chips, doubling density roughly every eighteen months by using photolithography. In 1994 manufacturers began searching for ways to use even shorter wavelengths.

Extreme ultraviolet lithography is the technique championed by a group of companies including Intel, Motorola, Advanced Microdevices, Micron Technology, Infineon Technologies, and IBM, working closely with the Department of Energy's "Virtual National Laboratory" consisting of teams from Lawrence Berkeley, Lawrence Livermore, and Sandia National Laboratories. Berkeley Lab's participation was spearheaded by EUV program manager and former CXRO head David Attwood.

The wavelength chosen by the EUVL collaboration, about 13 nanometers (13 billionths of a meter), is readily absorbed by glass lenses--indeed by all materials, including air--so the beams must be focused by curved mirrors instead. These are built up from dozens of layers of silicon and molybdenum, each a few atoms thick. A small fraction of the light reaching each layer is reflected, but the reflections constructively interfere, adding up to some 70 percent of the total falling on the mirror.

To test the mirrors, an incoming beam of coherent EUV radiation is split in two. One beam acquires the aberrations of the optical system while the other forms a nearly perfect reference wave. If the optics were flawless, 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 tested two sets of four mirrors for the prototype EUV printer built by Sandia, measuring tolerances so fine they were less than the radius of a hydrogen atom. One set of mirrors was used when the prototype printer went into service in 2001. The second set was not immediately needed, so the team modified the at-wavelength interferometer to print test patterns from them.

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

The optics performed even better than their design specified. Mirrors designed to make features with resolution better than 100 nanometers readily created 70 nm features. By manipulating parameters like beam angles and exposure times, the CXRO team achieved much smaller features, with a line width just 39 nanometers wide.

(Top) The CXRO team programmed defects into masks, then used coatings optimized for smoothing to reduce the irregularities in printed features. (Above) Researchers easily created 70 nm features on chips, as in these "elbows." By changing parameters, they were able to achieve much smaller features, with a line width of just 39 nm (far right).

The SES also tested ways to get rid of defects in the flat masks, also made of multiple layers. 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.

"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," Anderson says. The team used the Nanowriter, CXRO's ultra-high-resolution electron-beam lithography machine, to create masks with programmed defects.

The SES printed from these masks using different coatings developed at Livermore. Anderson says, "Coatings optimized for smoothing minimized the irregularities, like snow on grass. With non-smooth coatings, the defects stood out like snow on a boulder."
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."

These capabilities were sufficiently impressive that in April of 2002 Intel ordered the first production EUV stepper from the manufacturer ASML, to be delivered in 2005. Meanwhile, manufacturers hope to extend current photolithography techniques to 157 nm wavelengths, making chips with line widths less than 70 nanometers. To make anything smaller than that, EUVL is essential.

By 2007, commercial production is expected to begin, using an EUV stepper to manufacture chips with 18 nm gate widths (after etch). In October 2001, manufacturers chose EUVL as the most likely technology to be able to create chips with gate widths only 13 nanometers wide by the year 2009.

That will be just the beginning of the next generation of superdense integrated circuits.

-- Paul Preuss