Higher NA, complex RET are stretching optical lithography to the limit
06/01/2002
Overview
Forecasting which technology will be used to accomplish feature shrinks is a challenge that has humbled many of the best minds in our industry. Still, a trend is emerging: Stretched scanners at the 248nm and 193nm wavelengths, along with increasingly higher numerical aperture lens systems and resolution enhancement techniques, are clearly the wave of the future.
by Griff Resor
 A progression of lens sizes from Carl Zeiss, including the 193nm version. |
While optics continues to deliver production solutions, just-in-time, next-generation technologies are gaining attention. Just four years ago, 0.25μm features were going to be done with 248nm-wavelength lithography. Two more nodes, 0.15μm and 0.13μm, and six more years would bring us to the death of optical lithography. A next-generation lithography (NGL) would be needed to print 0.10μm features. The choices were electron projection lithography (EPL), ion projection lithography (IPL), x-ray proximity, or extreme ultraviolet (EUV). This roadmap didn't last long.
The 0.15μm node was simply skipped; now, 248nm tools have captured the 0.13μm node, enabled by advances in resist performance and mask resolution enhancement techniques (RETs). Based on papers presented earlier this year, some manufacturers will stretch 248nm all the way to 0.10μm and tools at the 193nm wavelength are ready with very high numerical aperture (NA) (0.78) to print 100nm and 90nm features. Hot on the heels of 193nm systems, the industry now has 157nm tools — a wavelength that International Sematech had deemed undoable just four years ago. If technology progresses as many believe possible, 157nm light will be used to print 45nm features, thus pushing NGL out even further, to 35nm features, as the market entry point.
Dropping out of the NGL race
NGL technologies have been dropping out of the race as the reign of optics persists. In the past two years, IPL and x-ray have been winnowed from the pack. SCALPEL, one embodiment of EPL, was dropped last year, a factor that appears to have strengthened the remaining EPL contenders. EUV continues to gain momentum. Clearly, the industry roadmaps are only a rough guide and not a reliable prediction of the future of microlithography.
Paolo Gargini, Intel Fellow and chairman of the International Technology Roadmap for Semiconductors (ITRS), recently researched Moore's Law and gave the industry a brief history lesson [1]. In 1964, Gordon Moore put together his observation to convince the world that 1000 transistors on a single chip would happen. Conventional belief put the upper limit at 64 bits. Gargini reported that progress for the next decade was made by circuit cleverness, shrinks, and wafer-size increases. This initial push was launched by Intel to move MOS technology ahead of bipolar. The world became accustomed to this pace, however, so shrinking geometry, increasing chip sizes, and increasing wafer sizes provided a 4x growth in transistors every three years for more than 20 additional years.
The great wealth and improved standard of living generated by these gains provide the incentive to continue chasing Moore's Law. Gargini noted that chip sizes can no longer grow, due to lens limitations, and wafer sizes should not increase for many years, since the transition has become increasingly expensive and difficult. The preferred route for Intel is to increase the pace of geometry shrinks. Transistor counts must now grow 2x in two years. This implies that features need to shrink 1.4x every two years.
Gargini also showed that CMOS devices as small as 25nm could work, so the basic device and circuit technology will work at very small sizes. His terse message was, "No more excuses, just execute [shrink features 1.4x/2 years]." This is easier said than done.
 Figure 1. 110nm line/space patterns made by an ASML 0.80NA, 248nm lens [12]. Reprinted with permission of SPIE |
Stretching 248nm tools has also been a popular topic this year. In March, ASML introduced a 0.80NA lens that prints 110nm dense patterns (Fig. 1). RETs continue to be improved in many incremental ways. With increasing accuracy, image transfer is being understood as a diffraction-limited data-transfer task. Masks contain increasing levels of sophisticated optical proximity correction (OPC) bars, attenuating films, and phase-shifting structures, many smaller than the resolution of the printing lens.
Also in March, Numerical Technologies Inc. introduced a software tool called Full-Phase that extends the company's phase-shifting technology from the gate region to the full mask layer. It appears likely that RET methods will be used to stretch existing 248nm fabs to 110nm dense features, but beyond this, 248nm processes appear too marginal. Many in the industry expect the switch to the next wavelength, 193nm, to be a strong trend between 2002 and 2005, as leading-edge production moves to the 90nm node.
 Figure 2. 90nm line/space patterns made by ASML's 0.75NA, 193nm lens. |
Leading-edge production is expected to shift to 90nm in 2003. The suppliers of 193nm scanners have all introduced high-NA lenses. Canon and ASML have shown results using their new 0.75NA lenses for 193nm. Dense patterns printed with the ASML lens (at 90nm) are shown in Fig. 2. Nikon has a new 0.78NA lens for 193nm, the S306C (Fig. 3). Results from all three suppliers have so far been impressive. Clearly both the exposing tools and 193nm resists are ready for production. Expect 193nm to be the mainstream technology for the 90nm node.
 Figure 3. Nikon's S306C, 193nm lens (0.78NA). Reprinted with permission of SPIE |
After 90nm, the next node is 70nm, scheduled for 2005, but the picture is even less clear. John Petersen, president of Petersen Advanced Lithography, in three papers with ASML MaskTools and ASML, has presented creative ways to use chromeless phase features to print sub-wavelength images at 70nm using 193nm light [2-4] (see "Manufacturing at k1 = 0.2 with chromeless phase lithography" on p. 67). Some chrome features are maintained until the masks are inspected; then the chrome is removed, leaving the mask as a sophisticated array of just phase features. The chromeless phase features can create the full gamut of lines, pads, and OPC features. Through a segmentation of the main feature, called "half-toning," the critical CD can be biased as needed to hit the target in a real process. In more general terms, RET mask technology is expected to stretch 193nm tools to the 70nm node, perhaps even to 65nm, the present official designation of that node. This work is expected to be completed by 2005.
Lenses, lasers, resists
A critical component in 193nm lenses is CaF2. About a year ago, the intrinsic birefringence (IBR) of CaF2 was discovered to be 2x worse than previously documented, producing a flurry of activity. Recently, all three lens makers discussed IBR issues in depth [5-7]. By rotating lens elements and tweaking their designs, all now adequately control IBR aberrations in their 193nm lenses.
The question of CaF2 supply, however, remains open, in spite of announcements from Canon and ASML saying they have adequate supplies. Canon is bringing up its new CaF2 plant; this added capacity, along with additions at Corning, should provide adequate supply. But there are also issues with optical quality and polishing results. Many remain concerned that there may be a shortage of CaF2, and hence, a shortage of 193nm scanners in 2003. This fear may drive planners to use a greater mix of stretched 248nm scanners with fewer 193nm scanners in their early 90nm fabs.
 Figure 4. Cymer's MOPA laser architecture narrows bandwidth and doubles power. Source: Cymer |
Meanwhile, a new laser architecture from Cymer has generated a flurry of interest among industry experts, and will help defuse some concerns regarding lens materials. Cymer observed that demand for increased power levels at all three wavelengths (248, 193, and 157nm) requires a new arrangement of the laser's internal subsystems. By separating the master oscillator (MO) from the power amplifier (PA) (Fig. 4), a much narrower bandwidth pulse (0.1pm FWHM [full-width, half-maximum]) can be delivered without sacrificing laser power.
Until now, line narrowing has always reduced laser output and throughput. Cymer's experimental data show that acceptable laser tolerances for divergence, pointing, jitter, and wavelength accuracy are maintained. At-wavelength power up to 70W can be provided without pushing flash rates above 4000Hz. The lifetime of components within the laser is also expected to improve. A pulse-stretching module helps lower damage to optical components in the illuminator and main projection lens. With its much narrower bandwidth, this laser configuration can probably be used to reduce the amount of CaF2 in future 193nm lens designs [8].
Researchers in Japan, working on a line-narrowed 157nm excimer laser, have also moved to a MOPA- (master oscillator/power amplifier) type configuration. While an F2 laser has a narrow bandwidth (1.0pm) with good power output (20W), it does limit lens design choices. Two years ago, ASET (Association of Super-Advanced Electronics Technologies, Japan) launched a line-narrowing project, which, while it was generally believed to be possible, was expected to cut light output by at least 2x. By adopting the MOPA-type architecture, this trade-off was avoided. Currently, ASET reports that they have achieved the project's goals, namely a 0.2pm bandwidth (FWHM), 6mJ/pulse, and 5kHz pulse rate, for an average power of 30W of useful light at 157nm [9].
Two years ago, champions for 157nm thought this technology might be ready for the 65nm node in 2005, and while this remains the target for all three scanner suppliers, resist has been a major concern, especially when it is remembered that 193nm resist took longer to develop than originally planned. The first work on 157nm resist was not promising (too opaque). This year, there have been many papers that reported useful 157 resists with absorption around 2.0/μm. Asahi Glass found a novel polymer that may be used as the base in 157nm resist systems. Drawing upon the company's fluorine chemistry know-how from glassmaking, Asahi researchers have formulated a base compound that has an absorbance <0.1 for a typical 0.25μm resist thickness. This chemically amplified resist has a sensitivity of 8mJ/cm2, which is competitive with DUV resists. Etch resistance needs improvement, though, a common theme for 157 resists. Still, these results represent significant progress in just two years, with resists for 157nm exposures having moved from the transparency of aluminum (none) to these results.
Especially noteworthy is the fact that the number of research groups working on 157nm resists has now grown to 10 within the last two years. These groups are testing many formulations to improve sensitivity, etch resistance, and other properties. At this pace, it seems risky, but now possible, that a 157nm resist will be ready by 2005. If not, 157nm technology will have to shift its insertion target to 50nm in 2007.
 Figure 5. 55nm line/space patterns printed on an Exitech microstepper at Selete, using a 0.85NA, 157nm lens from Corning/Tropel [10]. |
Selete presented imaging results for its 157nm Exitech microstepper. This machine has a Corning/Tropel 0.85NA lens, which is corrected to a 0.05 wave over a 0.7mm-dia. field; reduction is 15x. Images at 55nm were clearly printed in an experimental fluoropolymer resist using strong phase-shift (alt-PSM) masks (Fig. 5). This work demonstrates the feasibility of using 157nm light for the 50nm node [10].
Lenses for 157nm do suffer from IBR about 4x more than lenses used at 193nm. All three suppliers demonstrated that their lens design software now incorporates the latest modeling needed to control IBR problems. Use of <111> and <100> crystal orientations will be required, and CaF2 suppliers are rushing <100> samples to lens makers to speed the learning curve. IBR issues have clearly delayed 157nm lens availability by about one year; however, this does not appear to block completion of the first machines, which are expected by 2005. But there is added risk; polishing the <100> crystal orientation to 1/50-wave tolerances is considered to be a serious challenge.
Masks
At 157nm, mask protection remains a critical issue. DuPont Photomasks has presented data showing progress on soft pellicles for 157nm; the company has extended lifetime to several wafers, which is significant progress, but a 100x gain is still needed [11]. More than 58 polymers and 300 formulations have been explored in the past 2.5 years. DuPont remains optimistic that it will find ways to prevent darkening of its soft masks. The slow rate of progress to date, however, puts use of 157nm in 2005 at risk.
In a parallel effort, lens designers have begun looking for ways to accommodate hard pellicles. A hard pellicle is basically a fused silica window. Each window becomes a significant part of the optical path. Sag, thickness errors, and deformation due to mounting forces all add significant imaging errors. Every element, other than the hard pellicle, is now fabricated to a tolerance of 1/50th of a wave. It appears unlikely that this tolerance can be reached on a hard pellicle window as well. ASML's adaptive optics concept, presented more than a year ago, had shown great promise, but was not presented this year at the SPIE conference.
Mask protection remains an important concern for 157nm technology, particularly by 2005, so it seems that pellicle and resist issues may delay use of 157nm until the 50nm node in 2007. Expect technology at 193nm to be stretched to 70nm to fill the gap.
EPL
Nikon's PREVAIL project is making steady progress; key components have been individually tested and are ready for integration. New vacuum-compatible air-bearing stages for both the wafer and the mask have also been built and tested. These stages will soon be placed into their vacuum chambers (Fig. 6), and integrated with the electron projection optics.
 Figure 6. Nikon's PREVAIL stages being prepared for integration this year [13]. A progression of lens sizes from Carl Zeiss, including the 193nm version. |
The tool is being optimized for contact hole lithography. The pitch between neighboring contact holes limits memory density (contact hole pitch lags in all optical roadmaps). Completion is planned for late 2003, with delivery in 2004 to a real customer. To speed insertion into chip production by the member companies, expect the first PREVAIL deliveries to go to Selete and International Sematech. EPL could be the surprise technology in the battle for microlithography market share.
EUV
The interest in EUV has grown significantly over the past two years and many commercial players are now in the game. ASML, by acquiring Silicon Valley Group, has the lead, but Canon and Nikon plan to be players, too. The EUV workshop sponsored by International Sematech last fall generated enough commercial interest to merit one workshop each year.
Meanwhile, the EUV-LLC team continues to make impressive progress: the engineering test stand (ETS) had its laser source upgraded to 500W, and now uses a liquid xenon source. The #2-projection optics has been fully tested on the synchrotron at Berkeley and looks very good. Correction is better than 1/20th of a wave; flare, however, is 40%. The optics will be integrated into the ETS this year and scanned exposures are expected by year's end. In 2003, the team will continue to add source power while using the ETS for process development work. International Sematech has purchased an Exitech tool for EUV exposure. Zeiss is making a two-element 0.25NA EUV microfield lens that should have only 5% flare, with completion of the lens planned for this year. Delivery of the Exitech tool to International Sematech is set for late 2003.
EUV still has several potential showstoppers to conquer, though. Source power, lens flare, mask defects, and mask protection continue to be critical issues. The most difficult appears to be source power. This challenge has stimulated a great deal of EUV point-source research. Several papers describing work on dense plasma focus and capillary sources were presented this year. This work is being done to lower source cost and improve source-conversion efficiency. But 30 years' worth of work on point sources, launched years ago to help x-ray lithography, have all hit the same problems. The hot plasma destroys the electrodes too quickly, and the plasma cannot be controlled adequately. So far, there have been no changes in this pessimistic outlook for EUV point sources in spite of the added effort.
Innolite AB in Sweden has developed a novel laser-pumped xenon source for EUV. It seems to be a breakthrough technology that may solve the EUV source problem. By cooling the xenon almost to its solid phase, then squirting it from a fine nozzle, a solid filament of xenon is formed, strong enough to permit laser pulsing 50mm from the nozzle. This may be far enough to protect the nozzle parts.
Before doing extensive lifetime tests, Innolite examined the EUV light output. The laser pulse converts the frozen filament to EUV radiation with 0.75% efficiency, a result competitive with other laser-pumped xenon approaches. Pulse-to-pulse jitter control is adequate, and high-speed photos show the filament is strong enough to permit high pulse rates without being destroyed by the hot plasma flash. More work remains to scale up the power and test lifetime, but this is the first pulsed plasma source architecture that promises a real solution to the nozzle-damage problem.
To compete with optics, an EUV scanner that costs $40 million and prints 80wlph (300mm), must aim at the 25nm node. Even at Intel's fast pace, however, such capability is not needed until 2011; EUV may be ready by then. However, this scenario leaves no solution for the 35nm node in 2009, so EUV may be needed by then to fill this gap. No other high-speed alternative is getting serious attention.
Conclusion
Data presented to date indicate that the 90nm node will be met with a mix of stretched 248nm scanners and new high-NA 193nm scanners in 2003. In 2005, the most likely technology for meeting the 65nm node will be stretched 193nm scanners, using novel RET techniques and even higher-NA lenses. By 2007, the most likely candidate for the 45nm node is 157nm, again with very high NA lenses and even more complex RET.
Mask protection is now the gating item for 157nm technology. Resists for 157nm are not a done deal, but progress has been very rapid. EPL will be developed in parallel with these optical technologies, and may take over more of the contact hole imaging in memory products, where the benefit gained by a denser pitch can pay for the technology. EUV probably will be needed in 2009 for the 35nm node, but significant issues remain in source power, lens flare, mask defects, and mask protection.
Acknowledgments
Full-Phase is a trademark of Numerical Technologies Inc. TWINSCAN is a trademark of ASML.
References
Note: Each reference is from SPIE's 27th Annual International Symposium and Education Program on Microlithography, March 2002.
1. P. Gargini, "Worldwide Technologies and the ITRS in the Current Economic Climate," paper #4688-01, Emerging Lithographic Technologies VI.
2. D. Van Den Broeke et al., "Complex 2-D Pattern Lithography at λ/4 Resolution Using Chromeless Phase Shift Mask (CLM) Technology," paper #4691-19, Optical Microlithography XV.
3. J. Petersen et al., "Sub-0.10μm Imaging with Chromeless Phase-Shifting Imaging with Very High-NA KrF," paper #4691-50, Optical Microlithography XV.
4. J. Petersen et al., "Developing an Integrated Imaging System for the 70nm Node Using High Numerical Aperture ArF Lithography," paper #4692B-39.
5. Y. Chiba, K. Takahashi, "ArF Projection Optics for New Generation Lithography," paper #4691-68, Optical Microlithography XV.
6. T. Matsuyama et al., "High NA and Low Residual Aberration Projection Lens for DUV Scanner," paper #4691-69, Optical Microlithography XV.
7. R. Rubingh et al., "Performance of a High Productivity 300mm Dual Stage 193nm 0.75NA TWINSCAN AT:1100 System for 100nm Applications," paper #4691-70, Optical Microlithography XV.
8. E.D. Onkels, "Performance Evaluation of a Prototype Microlithography VUV F2 Laser in MOPA-Topology," paper #4691-203, Optical Microlithography XV.
9. T. Ariga et al., "Development of 5kHz Ultra Line Narrowed F2 Laser for Dioptric Projection System," paper #4691-65, Optical Microlithography XV.
10. T. Suganaga et al., "157nm Lithography with High Numerical Aperture Lens for 70nm Technology Node," paper #4691-58, Optical Microlithography XV.
11. R.H. French et al.,"157nm Pellicles: Polymer Design for Transparency and Lifetime," paper #4691-57, Optical Microlithography XV.
12. K.V.I. Schenau et al.,"System Qualification and Optimization for Imaging Performance on the New High NA 245nm Step and Scan Systems," paper #4691-64, Optical Microlithography XV.
13. T. Miura et al., "EPL Tool Development Summary," paper #4688-62, Emerging Lithographic Technologies VI.
Griff Resor is president of Resor Associates, 629 Massachusetts Ave., Ste. 200, Boxborough, MA 01719; ph 978/263-7826, e-mail [email protected].