Keep it (Relatively) Simple, Keep it (Relatively) Cheap

Trying to rest my weary feet after a week of rushing from appointment to appointment and event after event, I'm reflecting on a SEMICON West that had no shortage of talk about the current and future state of lithography affairs. Immersion lithography, for example, has been at the forefront of ongoing research and dialogue over the past couple years, and it was the theme again at a Tuesday morning breakfast meeting. Design for manufacturing, another subject that's moving into the limelight, was the topic of Wednesday's keynote address. After-hour gatherings were also abuzz with conversations about the inevitable demise of 157 nm lithography, as well as increasing interest in immersion techniques for 248 nm lithography in addition to the much talked about 193 nm immersion.

Design for manufacturing (DFM) is certainly a subject to which lithographers — and their counterparts in the design world — are going to have to start paying more heed, and is a concept that must become a mainstay in this industry, according to Aart de Geus, chairman and CEO of design software provider Synopsys Inc. (Mountain View, Calif.), who presented Wednesday's keynote. A key to DFM is design intent, which could significantly improve mask yield by indicating to the maskmakers which design features are critical or non-critical. By taking information from the design phase and giving something useful to the maskmaker, designers can reduce considerable cost and pain in the manufacturing process, de Geus said.

Looking back at the past four years, there have been some pretty significant changes in this industry, not least of which have included a move to 90 nm technology, and a move from 200 to 300 mm wafers — all during an economic downturn that was twice as deep and twice as long as any previous downcycle. In contrast to the preceding 10 years, 0.13 µm was rather painful, de Geus said, with the introduction of copper and low-k dielectrics compounding some serious yield issues. "The downsides came in as a big surprise to the designers," he said.

Along the way, success in the lithography arena has become more and more difficult. The number of chips that are meeting specifications is declining, the number of chips operating as expected is declining, and the number of full mask set respins is on the rise.

Traditionally, design and maskmaking were very separate; it was a hand-off, de Geus noted. "If things didn't quite work out, you'd add a few rules, and just make sure you stayed within parameters. Then everything would be OK, as long as you followed the rules." But the set of rules has grown massively (by a factor of 5-10×), and some of the rules actually conflict. As technology transitions from 130 nm to 90 nm and beyond, mask designs involve a rising number of resolution enhancement techniques, including increasing degrees of optical proximity correction (OPC) and progressively more complex phase-shift masks. Through all of these tricks has come an explosion of data and an increase in the time it takes to get ready for the mask.

Enter design intent, which could be applied to the mask to simplify the actual amount of data being manipulated. It could significantly impact yield, which has previously been the domain of the manufacturing side.

Of course, there are ways to stall the data explosion. Immersion lithography, for one, obviates the need for so many resolution tricks. During Tuesday's Lithography Breakfast Forum, as industry players explored cost issues as they relate to immersion development and use, one note that was hit on more than once is that immersion lithography is significantly cheaper and less complex than the alternatives.

Patrick Martin of DuPont Photomasks Inc. (Round Rock, Texas) gave an example of printing a simple right angle. With a binary mask without OPC, the pattern would require two shots to print; rule-based OPC would require 12 shots; and model-based OPC would take about 25 shots. Immersion lithography, with its effective shorter wavelength, gives lithographers an opportunity to go back to fewer OPC tricks, thereby simplifying masks.

While discussing the cost points of immersion lithography, Gene Fuller, principal engineer for Nikon Precision Inc. (Punta Gorda, Fla.), touched on another idea that has been floating around the lithography community: Changing print magnification from 4× to 8× would also reduce mask complexity and cost by letting the maskmakers go back to larger feature sizes. A switch to 8× magnification would make lens design easier as well, Fuller noted, but would hit throughput by 30-40% because of the reduced field size. "So, to keep cost under control, it's important to stick with 4× for as long as possible."

Of course, at some point there may be no choice but to convert to the increased magnification, noted Phil Ware, senior fellow, lithography, at Canon USA Inc. (Irving, Texas). The leading alternative at 32 nm and beyond — EUV lithography — presents an outrageous jump in cost and complexity, not to mention a significant throughput hit if the source developers cannot figure out how to increase power levels sufficiently.

For additional information on lithography, go to www.semiconductor.net/lithography