157nm takes optics to 2010
06/01/2003
It is clear that optical lithography will be used for IC production for the rest of this decade, but which optical technology will be used at each step of this journey is not. Progress with 248nm resists and reticle enhancement techniques has kept this wavelength in production far longer than most thought possible. This was fortunate because 193nm resists were not ready in time.
Now, however, it seems the technology is ready to meet the challenge, extending optics to the 90nm node, and probably to the 65nm node. Lithography at 157nm is needed to sustain progress to the 45nm node and beyond.
The opening gambit
Work on 157nm started late. It didn't get serious consideration until early 1999, when Asahi and Corning announced a transparent mask material. At the same conference, Cymer and Lambda-Physik reported success with F2 excimer lasers at 157nm and interest in this optical wavelength surged. Within three months, Ultratech Stepper, Exitech, and SVGL announced plans to introduce research tools for imaging 157nm resists. Both Selete and International Sematech (ISMT) took delivery of their tools in late 1999. By May 2000, at the Dana Point Conference, Sematech's members announced they wanted 157nm for the 100nm node in 1Q03, a date just 18 months away. By July 2000, ASML, Canon, and Nikon had clear plans to furnish 157nm machines.
But the early wishes expressed at Dana Point were unrealistic and didn't happen. With the benefit of hindsight, it can be said that the most difficult challenges were properly identified from the start: resists, pellicles, and contamination. There were also concerns for calcium fluoride (CaF2), a critical lens material, although early progress reports were very optimistic. Finding real solutions, however, has taken longer.
MIT reported that contamination was a problem in late 1998. Photons at 157nm are energetic enough to break most molecules into substances that stick to surfaces, blocking 157nm light. But gas purging of optical systems has been necessary since 365nm, so solutions were quickly found for this problem. Three years of work with the first 157nm exposure tools have shown that contamination issues can be managed and this issue is off the list.
null
Concerns about CaF2 are also abating. There was a major scare in 2000 when the polarization effects in CaF2 were discovered to be 4x worse than first thought and the first lenses didn't work. Lens design software was updated and, now, new designs that use the 111 and 100 crystal orientations promise workable lenses; the industry will know for sure this year.
Meanwhile, ASML, Canon and Nikon all anticipate adequate quality and quantity of CaF2 to meet their delivery plans in 2004 and 2005. Schott, Corning, Canon, and Nikon have all built large factories to assure an adequate quantity of CaF2, even at low yields. "We expect CaF2 will be more an issue of cost than supply," says a source at ASML. Lenses for 157nm use are going onto existing step-and-scan platforms, so the lens had been the gating item for machine delivery.
Canon plans to have its first 157nm machine ready for joint development work with Infineon in Japan during the second half of this year; the tool will be ready to ship by year's end. ASML plans to ship a research machine to IMEC in the first half of this year — a tool based on the former SVGL (now ASML) Micrascan VII platform. ASML expects to deliver its first production machine, a TwinScan unit, in mid-2004. Nikon plans to ship its first 157nm step-and-scan system at the end of 2004.
Pellicles: Still serious
Pellicles remain on the list of serious concerns. By mid-2002, no useful soft pellicle material had been found, so a decision was made to switch to hard pellicles while the search for soft pellicles continued. A 0.8mm-thick fused silica window will be attached to a 3.3mm high spacer frame that will in turn be attached to the mask. A 0.8mm-thick window becomes a part of the lens, and must be built to lens tolerances — a very difficult challenge. Extremely tight local tilt, thickness, and flatness tolerances must also be met. For example, the local tilt needs to be <25μrads, or 25nm/1.0mm. The sag due to gravity is 4.0μm (up to 80nm/mm). At these tolerances, handling and mounting introduce significant errors. Nevertheless, good progress has been made.
Asahi, a major supplier of hard pellicles, has learned how to make weakly bowed substrates, exactly offsetting the gravity-induced sag. Experiments are now underway to mount hard pellicles to frames and masks without introducing significant added error.
Figure 1 shows a late 2002 result from ISMT. Local tilt errors are <50µm-radians, over about a third of this blank. Most of this error results from the mounting of the pellicle, not the blank itself. In early January of this year, ISMT received a special measurement tool from Zygo that can provide individual surface data even when internal reflections are present and which should accelerate development of better mounting techniques. Hard pellicle technology has been dubbed a "just-in-time solution," and 18 months of additional experimental work should provide useful hard pellicles just in time for 2005.
Lens designers are also helping meet the hard pellicle challenge. They have found ways to add adjustments within their lenses to compensate for some hard pellicle fabrication errors — such as average thickness errors — that will make hard pellicle manufacturing easier. In fact, some of these adjustments exist in today's 193 and 248nm lenses. The real challenge is to develop in situ metrology improvements to find the added hard pellicle errors and compensate for them. Real experiments however, cannot begin until the first imaging machines are completed and near-spec hard pellicles are available. The work to integrate hard pellicles with optical systems will therefore extend through 2005.
null
Resist challenges the hardest
Resists for 157nm lithography remain the number one challenge. In 2000, John Petersen, president and Microlithography Fellow, Petersen Advanced Lithography Inc., said, "Resist progress will be the key to 157nm insertion timing" [1]. His prediction now appears to be correct. Within 18 months, the world will have 157nm machines, excimer laser light sources, masks, hard pellicles, and good lenses for 157nm imaging. The resists, however, probably will not be ready.
Clariant reported the first good 157nm results in March 2001. Transparency of the base resin, though, is still quite poor, forcing users to image in very thin layers, generally 0.1&μm thick. DuPont is supplying a base resin to Shipley and TOK. The DuPont material is significantly improved, but is still quite opaque at 157nm. Approximately 40% of the light is absorbed in a 0.1μm-thick film.
In March 2002, Asahi and others announced base resins that are 3–5x more transparent. These materials come closer to meeting imaging requirements, but early tests show they do not stand up to subsequent etch steps as well as the less transparent materials. thus, for single layer resists, the search for an etch resistant, transparent base resin continues.
Resist makers are exploring bi-layer resists to create a more robust thin resist technology. In this kind of system, a silicon-containing (silated) top layer of resist is imaged, developed, then hardened using an oxygen plasma to convert the silicon to glass. Dry etching then transfers the hardened pattern into a thicker underlying layer. The thicker layer controls etching of the wafer. Shipley has a full R&D team working on both a good single layer solution and a good bi-layer solution (Fig. 2). The company is optimistic that it will have useful 157nm resists ready for production at the 65nm node (2005–2007).
Resist papers presented at this March's SPIE Microlithography Conference showed a distinct difference between 157nm and 193nm resist status. The 157nm community is still searching for useful materials, while the 193nm community is optimizing for production at the 90nm node. If 157nm resists improve as slowly as those for 193nm, it will be at least five more years (2008) before they will be ready for production.
On a positive note, development of 157nm resists may move faster than 193nm did. This is the fifth wavelength in 20 years and resist developers are getting better at finding and optimizing the material combinations that make a good resist. One researcher, however, has said that, "Computer models lag too much to help." This effort really needs more teams working on different combinations and more exposure tools are needed for this to happen, but they won't be ready until 2004–2005.
Because it's going to be a stretch for 157nm resist suppliers to be ready by 2005, many IC makers expect that 193nm systems will be called upon to print the tough layers at the 65nm node. This puts insertion of 157nm at the 45nm node in 2007–2008. As one observer noted, "This has been the history of optical lithography: the older wavelengths last longer than planned, the newer ones come into production later than planned, while NGL continues to be pushed into the distant future."
Using 157 at the 45nm node in 2007 or later does not seem to alarm 157nm champions, but it does raise the bar on specifications, and it does delay the return on all this investment. As usual, getting the timing right in this business is the real challenge.
Griff Resor III, Resor Associates, Boxborough, Massachusetts
Acknowledgments
MicraScan VII and TwinScan are trademarks of ASML.
Reference
1. Lithography Review, Semi, Vol. 2, Issue 6, p. 9, 1/30/00.
Griff Resor III is president of Resor Associates, 629 Massachusetts Avenue, Suite 200, Boxborough, MA 01719-1528; ph 978/263-7826; fax 978/263-1011; email [email protected].