A Revolutionary Chipmaking Technique?

A method to remove the masks—and millions of dollars in costs—from some optical chipmaking processes is under development at a pair of European companies, and while it's not "smoke and mirrors," the idea is, in fact, to do it all with mirrors.

Photolithography uses a mask, or reticle, with the pattern for a computer chip etched into it. Laser light shining through the mask exposes a light-sensitive material on the surface of a silicon wafer, inscribing the lines that make up the transistors on the chip, much like a negative in photography. Each reticle contains the pattern for only part of the semiconductor wafer, and usually several reticles are used for different layers of circuits.

A practical problem arises as chipmakers try to make the inscribed lines smaller and the patterns more complex. A set of 25 to 30 reticles for today's chips, which are based on 130-nm-thick lines, can cost US $500 000. For the next generation of devices, the 90-nm node, a reticle set is expected to cost more than $1 million, and for the 65-nm node, scheduled to be introduced in 2007, some people are projecting a cost of $3 million per set.

Now Micronic Laser Systems AB of Täby, Sweden, and ASML of Veldhoven, the Netherlands, have teamed up to build a lithography system that replaces the reticles with arrays of millions of microscopic mirrors, which would direct the laser light to the wafer according to patterns fed to them by a computer.

Micronic makes laser pattern generators for producing photomasks, and ASML is one of the world's largest manufacturers of lithography equipment, such as steppers. Their system would use several arrays totaling approximately 13 million aluminum alloy mirrors, each perhaps 8 µm on a side. The exact design will depend on how the companies decide to balance various tradeoffs—larger mirrors are easier to build but have a shorter field of view, for instance.

Micromirrors are key

In the microelectromechanical system devised by Micronic and ASML, the mirrors will work with lasers emitting at193 nm, currently used to make 130-nm features [see illustration, below]. The mirrors will use interferometry to vary the intensity of pixels on the edges of features to make the lines as smooth as required.

ILLUSTRATION: STEVE STANKIEWICZ

In the lithography system being developed by a Swedish and a Dutch company, an array of micromirrors [inset, upper left] will replace the reticles used today.

The mirror-array system should be ready to market by year-end 2006, says Jorge Freyer, Micronic's senior vice president of marketing and business development. Though researchers are pursuing other approaches to maskless lithography, including using electron beams or nanodroplets, Freyer argues that the mirror system would have the advantage of fitting seamlessly into current production processes.

The technique should be scalable to 45-nm features, either with 193-nm or 157-nm lasers, says Karel van der Mast, vice president of strategic business development at ASML. It could even be usable in the extreme ultraviolet, if that's introduced in semiconductor manufacturing, he says.

With a projected throughput of only five wafers per hour, the mirror-array technique can't compete with current processes that provide 100 or more per hour. But van der Mast says that for makers of such low-volume products as application-specific integrated circuits (ASICs), the reduced costs will more than make up for lower throughput. And time to market will be quicker because chipmakers will not have to wait for the manufacture of several sets of reticles as they design, test, and then redesign their chips.

"Some of the ASIC companies are really hurting because of mask costs, so they're pushing hard for somebody to come up with something," says William Oldham of the University of California, Berkeley.

Oldham sees designing a mirror system as a major engineering challenge. The other formidable challenge, he says, is the bandwidth required, with computers having to produce large amounts of data to regulate the micromirrors in just a few minutes. The peak data rate in such a system could be a terabit per second, he says. Oldham directs a research group that is evaluating new approaches to lithography for the Defense Advanced Research Projects Agency (Darpa).

In September Darpa cosponsored a workshop on low-volume patterning to examine the concept of mirror-array lithography and to consider whether it's time to sponsor a demonstration project. Sherry Gillespie, a technical management consultant who chaired the meeting, says that while mirror-array lithography with a throughput of one to six wafers per hour is appealing, questions still need to be addressed.

"Is it going to be viable in the 45-nm node and beyond, and is it something that's extendable?" she asks.