SPECIAL REPORT: State-of-the-art processing
Christian Wagner, Winfried Kaiser, Carl Zeiss, Oberkochen, Germany
Jan Mulkens, ASML, Veldhoven, The Netherlands
Donis G. Flagello, ASML, Tempe, Arizona
It is commonly accepted that optical lithography, including 157nm wavelength exposure, will enable manufacturing at the 70nm node. Translating the industry's official roadmap to an exposure tool roadmap, it is obvious that new wavelengths and optics with extremely high numerical aperture will be necessary. With the advance of the latter, depth of focus decreases and focus control becomes more critical. Polarization effects on both the reticle and the resist film must be critically reviewed for numerical apertures exceeding 0.8. Nevertheless, optical extension should preferably be supported by reduced field and large magnification scanners and is expected to take next-generation lithography to at least the 70nm node. Beyond this, next-generation lithography must be both technically feasible and cost-effective.
The ongoing acceleration of the semiconductor roadmap forces the industry to extend optical lithography much further than ever expected. To assure sufficient process latitude, volume manufacturing requires the implementation of resolution enhancement methods, advanced mask technology, as well as layer-tailored illumination schemes and processes. With the use of phase shift masks (PSMs), wavefront aberrations become increasingly important in both critical dimension (CD) and overlay budgets. Advanced illumination technology is essential to control the polarization level and to support low-k1 illumination schemes.
The choice is not whether to use extreme numerical aperture (NA), low k1, or lower wavelength, but rather how to use all these features at once; to reach the 50nm node the various components have to be driven to limits simultaneously (Fig. 1). In Fig. 2, we summarize resolution nodes down to 50nm, showing applicable wavelengths, exposure-tool NAs, and process K1 factors. (Lens and machine performance requirements have been discussed separately in detail .)
Considering the issues and requirements of extending lithography by wavelength (Table 1), we see that down to 157nm, lasers will be available as well as optics designs and materials for building the lenses. Technology for 248nm resist has allowed resolution to be scaled from 0.25mm to <0.13mm. At present, 193nm resist technology is rapidly improving. Although high-contrast resists for 0.13mm-resolution volume production are within reach, validation of production-worthy 0.10mm resists is necessary. For 157nm resist, absorption is a challenge. For mask technology, attenuated phase shift material for 193nm, and 157nm, especially, needs to be developed. At 157nm, pellicle and contamination issues remain.
At 157nm and below, purging of the optical paths will be required. Extending optical lithography further to 126nm does not seem likely because with pure mirror designs ("catoptric") as applied in extreme ultra violet lithography (EUVL) for example, it is improbable that the necessary NA level (e.g., >0.7) will ever be reached. To make 126nm exposure an option, only catadioptric designs seem applicable. MgF2 and LiF are the only candidates for a refractive material, but MgF2 is birefringent on the level of 0.01 and there is no data on laser hardness or the feasibility to grow large crystals for LiF. In addition, mask and resist materials must be developed together with establishing proof of the feasibility of a 126nm laser source. While we cannot say that optical lithography can never reach 126nm, at this time it does not appear possible.
NA and residual aberrations of refractive designs have been greatly improved over the last decade (Table 2). The bandwidth of line-narrowed KrF lasers allows refractive optics with only fused silica.
Figure 2. Wavelength, NA, and K1 for 250nm to 50nm resolution nodes.
Emerging 193nm refractive designs need CaF2 to achromatize lenses to today's state-of-the-art <0.5pm bandwidth lasers. Currently, the most likely solution for 157nm is a catadioptric design, preferably without a beam splitter and obscuration. All design types must be extendible to >0.8NA to support at least the 70nm node with optical lithography. Refractive designs might again have a chance with a CaF2-BaF2 or CaF2-NaF combination. Success, however, depends on the availability of the crystals and minimized laser bandwidths that might be achieved in the timeframe of 157nm production tools. The challenge for catadioptric designs without a beam splitter is either a central obscuration or an off-axis field. Off-axis fields would lead to changes in exposure tool layout and new calibration and optimization strategies.
Figure 3. a) Laser bandwidth vs. NA for 248nm and 193nm refractive designs; and b) optics technology effort vs. NA.
For 248nm, we use an all-fused-silica refractive design and for 193nm, a typical achromatization with CaF2, assuming a slit height of 26mm (Fig. 3). The bandwidth of leading 157nm lasers is about 1pm, which is consistent with catadioptric design solutions. For refractive solutions, much still depends on material dispersion, which is still not known exactly for all crystal types, and design layout. For existing 1pm single-line 157nm lasers, refractive designs with a CaF2-BaF2 combination exist. The large volume of material needed to build these lenses, however, makes this solution only feasible for a reduced slit height (i.e., <22mm). To reduce material volume to a realistic level, we will need a <0.5pm laser bandwidth. In any case, the diameter of the lens image-field height (by track length and thus optical path) affects the laser bandwidth in a linear fashion and a reduction in field size will thus help relax laser bandwidth requirements and laser cost.
Figure 3 also shows the relative optics technology effort as a function of NA, independent of wavelength-specific technology. To first-order approximation, lens element diameters are proportional to the tangent of the aperture angle (for 26mm slit height). Material volume follows the second order, when it assumes that the track length of the lens stays constant because of design innovations, technology advances, and continuous improvement processes. Reduction to a 22mm wide field will, by scaling, reduce lens material at least by the diagonal of the field squared.
Low-k1 imaging only works when the mask layout, illumination conditions, and resist process are optimized simultaneously. With alt-PSM entering the scene, chip design rules will have to be integrated into this global optimization strategy. Considering the overwhelming cost of NGL techniques, low-k1 imaging may be explored as a comparatively low-cost application, even if it is common knowledge that rising mask costs and double exposure will lead to reduced throughput (Table 3).
Exposure latitude for dense and isolated lines depends significantly on the distribution of diffracted light in the entrance pupil of the lens. For K1 factors between 0.35 and the theoretical limit of 0.25, the interfering rays are roughly located on opposite sides of the entrance pupil. This situation is created by either using alt-PSMs or chrome masks and dipole illumination [2-4]. If one approaches the 0.25 K1 factor limit, small s-settings (partial coherence factor) become extremely helpful. Although the theoretical resolution limit does not depend on s, the minimum resolution that gives a certain level of contrast (i.e., exposure latitude) may be reduced by decreasing coherence.
For NAs exceeding 0.7 and K1 factors approximately equal to 0.3, polarization effects at image recombination in the aerial image will play a major role. At first, dose effects must be controlled. Depending on whether the light is polarized in or out of the plane of the incoming rays, Fresnel equations predict dose differences >20%. To limit horizontal and vertical (HV) linewidth differences, the residual polarization should be below 10% . Thus, in addition to intensity uniformity, polarization uniformity must be established. The second effect is that light being polarized in the plane of the incoming rays does not fully interfere anymore (the extreme case of a 90° angle is prevented in the resist for typical refractive indices). To optimize contrast, polarization vectors should be orthogonal to the plane of incoming rays. For dipole illumination using double exposure, this may be easily realized using linear polarization parallel to mask features for each of the two masks.
Technology supporting optical extension
It is instructive to link formal solutions of optical lithography's extension with CD and overlay performance requirements as given by the 1999 International Technology Roadmap for Semiconductors (ITRS). Figure 4 shows simplified CD and overlay budget trees. The energy budget is an extended dose concept that summarizes all contributions to CD error in best focus. Higher-order lens aberrations are grouped by focus (even aberrations) and overlay (odd aberrations). For a first estimate, we assume that all sub-budgets will scale down with resolution.
The imaging requirements of the technology nodes must be broken down to the level of optical component specifications. First, we must determine the dependence of CD uniformity and overlay on lens aberrations. The absolute sensitivity depends on feature type, mask, and illumination. For example, in a K1 = 0.3 process with 5% CD uniformity due to lens aberrations, a wavefront RMS of 0.02l is necessary. Stringent overlay requirements come from printing isolated lines with alt-PSMs because two-beam imaging that uses a small s is extremely sensitive to odd aberrations. To minimize pattern shift, both lens distortion and pattern-specific specification of single odd Zernike coefficients will be necessary.
For all types of lens aberrations, we must break down the error budget and calculate component specifications (Fig. 5). The optics technology roadmaps are driven by requirements that come from imaging, overlay and throughput, and the increase in lens NA to support future resolution requirements.
Core competencies in building high-NA optical systems may be described along the process chain, with incoming materials, polishing and coating of the lens elements, mounting, assembly, and adjustment and metrology capabilities (Table 4). To take care of requirements for low-k1 imaging, all aberrations must be controlled extremely well. For example, odd aberrations come from homogeneity and surface errors and from tilt and de-center errors in the lens at assembly. The difficulty in reaching these specifications increases along with NA because the sensitivity of the higher-order odd aberrations to lens element tilts and de-centers increases. To limit these errors, all mounting and assembly processes have sub-micron tolerances and push mechanical fabrication to its limits.
Extending optical lithography with high-NA optics and decreasing wavelengths also sets challenging requirements on step-and-scan system performance where focus and machine dynamics are especially critical. Other items are related to the implementation of 157nm, which requires a nitrogen-purged optical path, and implementation of low-k1 imaging schemes such as multiple exposures.
Figure 6. DOF roadmap for isolated lines referenced to ITRS DOF values.
Depth of field (DOF) is an obvious issue related to high-NA imaging. Traditional DOF scaling, derived for dense lines and three beam imaging, leads to extremely low DOF for high-NA optics. On the other hand, low-k1 imaging enhancements offer a possible increase in focus latitude . Consequently, it is difficult to predict what the exact workable DOF will be at a certain technology level. Figure 6 shows a tentative DOF roadmap for isolated lines based on assumptions of applicable wavelength, NA, and enhancements. Both the introduction of new wavelength technology and the use of imaging enhancements are assumed to be conservative. Thus, these DOFs represent a worst-case scenario.
Double exposure technology, using complementary-PSM  or dual illumination, is expected for extremely low-k1 applications. The step-and-scan machine must support this enhancement method with throughput and accuracy. Image placement errors of sub-images, which can be optics and machine induced, are critical.
Another possible application for multiple exposure schemes can be found in so-called image field extensions. One might consider reduced field optics, especially for hyper-NA optics where costs and focus budget become critical. For large chip sizes, however, fields must be enlarged using either a two-mask field-stitching approach or a raster scan approach.
We performed some simple dual exposure experiments to evaluate current sensitivity for placement errors. We exposed 0.30mm lines with conventional illumination as the first layer and 0.15mm lines with annular illumination as the second layer (Fig. 7). Line sizes were butted together resulting in a fork-shaped pattern. Line placement errors in this experiment can come from odd aberrations in combination with stage positioning errors. Figure 7 shows measured pattern shift as a function of the slit height. The maximum observed line shift is 6nm, well below the allowed value of 10nm in a field-stitching budget for 0.15mm lines. A dedicated machine design for stitching should make seamless stitching possible.
low-k1 imaging imposes stringent requirements on mask technology. While supporting the resolution roadmap, additional subresolution features (serifs, assist lines, etc.) are used to preserve DOF and exposure latitude requirements. Table 5 lists the requirements for 100nm and 70nm nodes, with comments on the feasibility . Compared to today's best performance data, we believe improving mask parameters by >50% will be extremely difficult, with CD uniformity being an especially critical issue. A change of reduction ratio to 5x or even 6x would significantly relieve this problem. We foresee that this change will be necessary for the 70nm node.
Table 6 combines the issues discussed here and outlines the most likely lithography roadmap. At 50nm, reduction in DOF and increased complexity in lens technology may enforce a further reduction of field size in combination with higher reduction ratio. The latter is necessary to alleviate pressure on mask CD requirements, which becomes increasingly difficult because of high mask error factor (MEF). Depending on the performance of the emerging EUV beta-systems, EUVL might be inserted in this same time frame.
With the current issues, we do not see further extension of optical lithography below 50nm. The 126nm node lacks infrastructure on all the important subsystems (i.e., lasers, materials, resists, and masks). At this point, EUVL will completely take over.
We shouldn't forget that the semiconductor industry is also driven by cost-effective technology. Manufacturing cost will increase per technology node. To minimize the cost increase for exposure tools and masks, one should consider a paradigm shift in chip size developments and reduction ratio. According to the 1999 ITRS, 22mm x 22mm field size machines are suitable for both DRAM and logic manufacturing. For larger chips (e.g., system on a chip), one should consider stitching. It seems obvious that a 22mm slit combined with 5x reduction is a better choice than today's "standard" of a 26mm slit and 4x reduction.
Figure 7. Critical pattern extension experiment with double exposure. Pattern shift for 0.15mm dense lines relative to 0.30mm dense lines, obtained with a double exposure using annular and conventional illumination settings, respectively.
We conclude that optical lithography can go well into the 70nm node, and, for most applications, can be extended to 50nm as well. The latter step requires 157nm with extreme NAs, PSMs, or dipole exposure. Using optics at the 50nm node is technologically challenging. To optimize yield and reduce technical risks of very high NA lenses as well as to remain cost-effective, field sizes of 22mm or smaller and reduction ratios of 5x or larger are recommended.
This article is adapted from a presentation at the Optical Microlithography XIII Conference of the SPIE Microlithography 2000 Symposium, February 27, 2000, Santa Clara, CA.
- D.G. Flagello, et al., "Optical Lithography into the Millennium: Sensitivity to Aberrations, Vibration and Polarization," Proc. SPIE, Vol. 4000, in press.
- J. Finders, et al., "DUV Lithography (KrF) for 130nm Using Off-axis Illumination and Assisting Features," presented at Semicon Japan 1999.
- M.E. Kling, et al., "Practicing Extension of 248 DUV Optical Lithography Using Trim-Mask PSM," Proceedings of SPIE, Vol. 3679, pp. 10-17.
- C. Wagner, et al., "Advanced Technology for Extending Optical Lithography," Proceedings of SPIE, Vol. 4000, pending publication.
- J. Mulkens, et al., "Challenges and Opportunities for 157nm Mask Technology," Proceedings of SPIE, Vol. 3873, pp. 372-384.
Christian Wagner received his PhD in quantum optics from the University of Munich. Wagner is responsible for systems engineering and optics technology roadmaps at Carl Zeiss, D-73446 Oberkochen, Germany; ph 49/7364-20-2997, fax 49/7364-20-4509, e-mail email@example.com.
Winfried Kaiser earned his diploma in physics from the University of Tübingen. Kaiser is responsible for product strategy and development of the Lithography Optics Division of Carl Zeiss.
Jan Mulkens graduated with a degree in solid state physics from Eindhoven University. Mulkens works on i-line, KrF, ArF, and F2 exposure systems at ASML.
Donis G. Flagello received his BS in photographic science from Rochester Institute of Technology and his MS and PhD in optical science from the University of Arizona. Flagello is a fellow at the ASML Technology Development Center, 8555 S. River Pkwy., Tempe, AZ 85284; ph 480/383-4329, fax 480/383-3978, e-mail firstname.lastname@example.org.