Given enough time and money, engineers can solve almost any problem. But when it comes to next-generation lithography (NGL), time and money are running out.
In the early 1990s, the semiconductor industry expected to choose an NGL system by 1998 and have a production system ready by 2003. This NGL system was able to create microchips with critical dimensions (CD) of 130 nm, far beyond the capabilities of optical lithography steppers built around excimer lasers. In response, vendors placed several NGL systems on the high-tech alter, each capable of resolving features down to a few nanometers. These systems included extreme ultraviolet (EUV) sources using laser-produced plasmas, electron beam, and x-ray lithography, among others.
By 2003, the semiconductor industry had both beat and missed its own forecasts. Microchips had CDs of 100 nm - better than expected just eight years before - but the industry's International Technology Roadmap for Semiconductors (ITRS) had slowly pushed NGL's debut from 2003 and the 100-nm node to 2009-2013 and the 32-nm node. Electron beam and x-ray were essentially relegated to short-volume application-specific integrated circuit production, while the EUV question was confused by new variants that used tin or lithium to produce more power at 13.5 nm compared to supersonic xenon jets. Adding pain to pressure, the (almost) clear winner, EUV, was joined recently by a resuscitated method called imprint lithography, which is far cheaper but very different from existing complementary metal oxide semiconductor processes; and the champions of the low- volume e-beam lithography systems aren't going quietly into that good night.
So what's the answer to the $40 billion annual semiconductor manufacturing equipment question? While the NGL winner may be unknowable at this point, the drivers behind the final solution are easily identified, says to Michael Lercel of IBM Microelectronics Division (East Fishkill, NJ) and co-chair of the Emerging Lithographic Technologies IX conference at SPIE's Microlithography 2005 symposium held this month in San Jose, CA. "It will be industry momentum and economics that carry the final solutions, whether it's e-beam or EUV or something else," Lercel says. "They are all capable of meeting the requirements, but they also take a lot of infrastructure development to get them to a useful state, and that is affected by economics more than anything else. It's too expensive to implement multiple NGL solutions, but there could be applications for maskless lithography [such as e-beam and imprint lithography] that target different market segments than volume lithography." Bits on hard drive surfaces and other memory cells do not have the strict alignment requirements of microprocessors, for example.
Economic concerns about the cost of developing NGL are likely to be acerbated by recent predictions of a 9 to 15% drop in semiconductor equipment sales in 2005, and the most recent announcement from Semico Research Corp.'s (Phoenix, AZ) Inflection Point Indicator (IPI) that a rebound in the equipment industry may not occur until Q4 2005, a quarter later than previously expected.
Despite these challenges, however, the future is far from bleak. "EUV has definitely had a lot of challenges all along, although significant progress has been made in the last year or so. Everyone's pretty excited to see what the current status of the technology is at this year's [SPIE] microlithography show, but we know the [NGL] systems are not remotely close [to being ready for production]."
Although the semiconductor industry and its lithography stepper suppliers in the optoelectronics industry have used a variety of resolution enhancement techniques to extend the lifetime of today's excimer-laser-based steppers, these optical magicians are quickly running out of tricks. The good news is recent findings indicate significant improvements to both of the critical challenges facing EUV. Development of new tools to create defect-free mask blanks is on schedule, and EUV sources have improved to 6 W for laser-produced plasma and 40 W for discharge-produced plasma using tin instead of xenon or lithium.
The bad news is there aren't enough metrology systems available that can detect defects smaller than 80 nm, much less the 25-nm defect-free requirement for NGL blanks. In addition, a production light source will need to deliver 115 W, not 6 or 40 W.
Just a little more time and money. oe