High-throughput e-beam stepper lithography
05/01/2000
Kazuya Okamoto, Kazuaki Suzuki,* Nikon Corp., Tokyo, Japan, Hans C. Pfeiffer,* IBM Microelectronics, Semiconductor Research and Development Center, East Fishkill, Hopewell Junction, New York, Michael Sogard,* Nikon Research Corp. of America, Belmont, California
*Additional authors are listed in the Acknowledgments.
OVERVIEW
Collaboration between Nikon and IBM has demonstrated the feasibility of an electron optical system for an electron-beam stepper based on the PREVAIL technique. This system differs from conventional direct-write electron-beam because it projects and demagnifies reticle patterns while scanning the reticle and wafer, very similar to optical exposure lithography tools. The promise here is throughput at >40 (200mm) wafers/hr while manufacturing ICs with <100nm minimum features.
 Figure 1. Curvilinear variable axis lenses (CVAL). |
Researchers at Nikon have been investigating new lithography systems suitable for manufacturing high-performance microprocessors with 100nm design rules and 16Gbit DRAMs that we believe will be required by 2002 or 2003. While we have been investigating F2 laser-based stepper lithography (i.e., 157nm lithography) [1] and extreme-ultraviolet lithography (EUVL) [2], which are extensions of existing optical techniques, our discussion below focuses on electron-beam (e-beam) lithography.
While high-resolution e-beam direct-write systems have been commercially available for some time, they have all suffered from low throughput, perhaps the only issue preventing early adoption of e-beam lithography for advanced production applications. Nikon's e-beam stepper concept is a reduction-projection e-beam exposure system designed for high-productivity manufacturing as a commercial successor to ArF-laser exposure systems.
E-beam stepper based on PREVAIL technique
Various e-beam exposure tools have been proposed for next generation lithography (NGL), including SCALPEL (scattering with angular limitation projection electron-beam lithography) [3] from Lucent Technologies. In 1992, IBM first proposed, then developed PREVAIL (projection reduction exposure with variable axis immersion lens) [4, 5]. PREVAIL overcomes field-limiting off-axis aberrations through use of an innovative curvilinear variable-axis electron lens (CVAL, Fig.1) [6]. As an alliance partner of IBM, Nikon has endeavored to convert the PREVAIL concept into a complete "e-beam stepper" exposure system.
Table 1 lists the main differences between conventional direct-write e-beam and the e-beam stepper. Accelerating voltage, projection pattern, magnification, sub-field size, and beam current are substantially different, resulting in dramatically improved throughput. Concretely, the 100kV acceleration voltage and large sub-field size (250µm at the wafer) reduces degradation in resolution due to Coulomb effects at large beam current since beam blur scales as [7]:
Beam blur ~ (I5/6 L5/4 M)/(NA3/5 SF1/2 V3/2)
where I is beam current, L is Coulomb interaction length, M is system magnification, NA is numerical aperture, SF is sub-field size at the wafer, and V is beam energy.
Further, large beam deflection capability permits high speed writing, and the use of a reticle means simultaneous exposure of many patterns.
The writing strategy of the e-beam stepper is based on scattering contrast similar to SCALPEL, but it differs in detail (Fig. 2): A 1mm square e-beam generated by an electron gun illuminates the reticle, and an electron lens produces a 4:1 demagnified image. One shot exposes a single pattern within a 250µm square on the wafer. The complete circuit pattern is exposed and stitched by scanning the reticle and wafer where the reticle and wafer stages move continuously in synchronism along a serpentine path (similar to optical scanning steppers). The e-beam must be deflected electro-magnetically by more than ±10mm on the reticle (±2.5 mm on the wafer) to achieve high throughput.
The e-beam stepper can use both stencil and membrane reticles with the imaging contrast produced by scattering. Figure 3 shows throughput as a function of beam current on the wafer for both reticle types. A single stencil reticle cannot fabricate a pattern with ring-shaped structures (for instance, the interior region of the letter "O" will fall out). Thus IC patterning with stencil reticles may require multiple exposures with complementary reticles, as in some optical systems. Even so, a resist sensitivity of 5µC/cm2, and a chip size of 25x40mm equates to a throughput >20, 300mm or 40, 200mm wafers/hr at resolution <100nm, even with complementary reticles. With simpler patterns not requiring complementary reticles or membrane reticles, higher throughputs should be possible.
IBM's PREVAIL proof-of-concept system
Within IBM's PREVAIL proof-of-concept (POC) systems, the electron optics in the e-beam stepper incorporate several innovations:
- The high-emittance electron source uses a low work-function tantalum single crystal (~10mm diameter) electron source [8] heated by electron bombardment on the side opposite the emitting surface. Servo control of the bombardment power results in <1% long-term stability.
- The reticle illumination system achieves 3.2% uniformity (3s) due to its large beam angle and a low-aberration lens for large sub-field and large deflection optics.
The imaging system uses deflectors, stigmators, and dynamic focus lenses in addition to general focus lenses. The CVAL (see Fig. 1) permits shifting of the electron optical axis along a predetermined path while simultaneously deflecting the e-beam to follow precisely the curvilinear axis [9]. CVAL gives extremely low-aberration, low-distortion performance and enables imaging to be done with very large e-beams compared to conventional exposure systems. Moreover, these larger e-beams greatly reduce the Coulomb effect, allowing exposure of large areas at high current while maintaining good resolution. Rotation, magnification, and focusing errors due to the e-beam itself can be completely corrected [5, 10] using three elements with optimized positions and excitation currents.
 |
The IBM PREVAIL POC system has been used at 75kV to print 80nm lines and spaces over a 250µm sub-field in 300nm-thick positive resist; the SEMs in Fig. 4 show there is no image degradation between a 2.5mm deflected beam and the undeflected beam position. By optimizing the excitation currents of every stigmator and dynamic focus lens, the x and y distortions at the deflected position were reduced to 12nm and 14nm (1s), which was almost the same as at the undeflected position (6nm, 1s) [5]. The targeted specification in the commercial model of this system will be <10nm (3s). The Coulomb effect was suppressed to the point where 80nm lines and spaces could be resolved, even with 7.5µA at 75kV corresponding to 12.8µA at 100kV.
In parallel with development work on the IBM system, Nikon developed a 100kV projection e-beam column to test stencil reticles, develop subsystems, and evaluate e-beam resists. With the Nikon system, beam current at the reticle is estimated to be 50µA maximum and magnification four with a symmetric magnetic doublet (SMD) lens.
E-beam resist development
 IBM PREVAIL proof-of-concept system installation |
We have been investigating, in collaboration with resist vendors, new e-beam resists suitable for the e-beam stepper. Since e-beam resist is already commercially available for reticle fabrication and direct-write systems, we are confident that the requirements outlined in Table 2 (see p. 120) are achievable, in fact more achievable than resist technology for 157nm lithography. The three most critical issues in Table 2 are sensitivity, and control of pattern collapse and outgassing during exposure.
Figure 5 (p. 121) shows newly developed positive e-beam resist profiles exposed by Nikon's 100kV column. We estimated the resist sensitivity at 10µC/cm2, contrast 7.5, and dose latitude ±10%.
E-beam reticle developmentrements outlined in Table 2 (see p. 120) are achievable, in fact more achievable than resist technology for 157nm lithography. The three most critical issues in Table 2 are sensitivity, and control of pattern collapse and outgassing during exposure.
Figure 5 (p. 121) shows newly developed positive e-beam resist profiles exposed by Nikon's 100kV column. We estimated the resist sensitivity at 10µC/cm2, contrast 7.5, and dose latitude ±10%.
E-beam reticle development
Both scattering stencil reticles and scattering membrane (SCALPEL type) reticles (Fig. 6, p. 121) will be used in the e-beam stepper. The stencil reticle has a 2µm-thick silicon membrane with openings representing the pattern to be exposed, all supported by a minor-strut structure (Fig. 7, p. 121). During exposure, the struts, which occur every 1.3mm, are skipped and sub-fields on either side are stitched together. Since little energy is deposited in these scattering reticles, thermal effects are less an issue than for other technologies.
We have developed two types of stencil test reticles at Nikon [11, 12] — a 75mm test reticle for "early learning" and a 200mm test reticle for system adjustment and evaluation, which will be commercialized in 2002 or 2003. We have also developed some fabrication processes using an SOI wafer and have fabricated a 75mm test reticle complete with support frame (the minor strut and membrane area measures 40x40mm).
A serious issue in reticle technology is image placement accuracy (IP); this requires membrane stress control. In the stencil reticle, the acceptable range of tensile stress is estimated at <10MPa. Measurements and finite element anal-ysis show impurity doping holds membrane stress in this range. Sub-field pattern displacements are <30nm, agreeing with model analysis results. Improving membrane stress distribution uniformity should reach <20nm IP requirement in time for making 100nm features.
Another reticle issue is critical dimension (CD) control. The reticle pattern CD control level of the 75mm test reticle easily is <35nm, mainly determined by e-beam pattern writing and trench dry etching. The requirement for the e-beam stepper is 14nm for CD control on the 4x reticle for the 100nm-node features. Some optimization of the reticle fabrication process is still required, including the development of new pattern writing tools. However, this requirement may be easier to fulfill for e-beam stepper reticles than for optical technologies subject to the mask error enhancement factor (MEEF).
 |
Another issue is the support frame; it may be needed to handle membrane reticles easily and to maintain reticle-flatness on the e-beam steppers. Bonding the e-beam reticle wafer to the support frame is a key to producing highly accurate e-beam reticles. We are using silicon as the support frame to avoid reticle deformation due to the thermal expansion differences. Bonding remains an issue, but one-point bonding shows good results (static DIP<10nm) on x-ray masks.
Since it is difficult to use a pellicle on an e-beam reticle, cleaning is an issue. In situ reticle laser cleaning is a powerful method for stencil reticles, but works poorly on heavy metal particles, so we are investigating other cleaning techniques like plasma and laser processes.
 Figure 4. 80nm lines and spaces in positive resist for undeflected and deflected sub-fields. |
Focused ion beam (FIB) technology, widely used to repair reticles, also works for stencil reticles. FIB can remove <0.1µm patterns and deposit <0.5µm patterns. We are now looking at deposited pattern materials to see whether we are able to obtain good scattering contrast with the 100kV e-beam Reticle inspection seems possible with high-throughput optical and e-beam die-to-database inspection systems. Overall, we do not foresee any unreasonable technical risks, except possibly CD control, for the e-beam scattering stencil reticle. Our 200mm-test reticle process development is in progress.
Thermal loading and distortion
With the e-beam stepper, the temperature of scattering reticles does not rise from absorbing e-beam power, as much as would a conventional cell-projection reticle consisting of thick silicon electron absorbers and apertures. However, we have investigated thermal characteristics [13] and deformation of the silicon membrane and bending from gravity. The e-beam heating power (Q) is expressed as Q=IxVxa where I is e-beam current (100µA at reticle), V is 100kV acceleration voltage, and a is 2.3% absorbed energy in the 2µm-thick membrane, according to Bethe's stopping power equation. This equation yields Q as 230mW. For the purposes of MSC/NASTRAN simulation, we assume a shot time of 62.5µsec with 10µC/cm2-resist sensitivity and 25µsec between exposure shots.
Figure 8 (p. 122) shows quarter symmetry modeling result of nine (3x3) grids with 1, 2, and 4µm membrane thicknesses. Deformations of the sub-field due to thermal expansion after one shot and of bending by gravity are illustrated. Here, it is assumed that the center sub-field in the nine sub-fields is illuminated.
 Figure 6. a) Scattering stencil and b) scattering membrane reticles for the e-beam stepper. |
Under the conditions of Fig. 8, the displacement of a 1µm-thick membrane is almost entirely due to gravity after a single shot. The stiffer and more absorbing 2µm-thick membrane experiences a 4.32K temperature rise and a maximum sub-field deformation of 6.2nm vertical and 2.0nm horizontal, of which about half is gravitational. A 4µm-thick membrane distorts even less in our simulations, but the fabrication process appears more difficult. We have also simulated thermal deformation with a 2µm-thick membrane after 85 shots. Even in this case, the maximum temperature rise is estimated as 3.46K, and deformation values are also minute (~1nm horizontal and 6nm vertical). Thus, these results show that a scattering stencil reticle is acceptable for an e-beam stepper.
Technology extensibility
 Figure 7. Fabricated silicon stencil reticle. |
Could an e-beam stepper produce ICs at technology nodes after 100nm? We think the answer is yes. Although we cannot disclose the details at present, we have already studied ways to reduce the Coulomb effect and geometrical aberrations that are dominant factors limiting resolution. Our modeling suggests that a CVAL, which uses an optimized NA, short column length, unique illumination systems, etc., can give beam-blur <35nm at high-beam current. Under such conditions, we believe throughput >20wph for 300mm wafers will be obtained for the 50nm technology node.
Conclusion
 Figure 8. Quarter symmetry modeling result of nine grids with membrane thickness of 1, 2 and 4?m. |
The feasibility of the PREVAIL electron optical system for an e-beam stepper has been shown. This system uses a high-emittance electron source, a uniform reticle illumination system, and a low-aberration projection lens with large sub-fields and large deflection along with membrane and stencil scattering-reticles with low thermal loading. The remaining resist and reticle issues are not showstoppers. The e-beam stepper promises to make ULSI chips with minimum feature sizes <100nm at 15-20 wafers/hr (300 mm) using complementary stencil reticles. The development of the e-beam stepper system is in progress and Nikon intends to manufacture and market such a system in 2002 or 2003.
Acknowledgments
 Figure 5. Representative lines and spaces from the Nikon 100kV exposure column using newly developed positive e-beam resist. |
Additional authors are: Teruaki Okino, Shintaro Kawata, Kenji Morita, Kazunari Hada, Kiyoshi Uchikawa, Akikazu Tanimoto and Takeshi Yamaguchi of Nikon Corp.; Rajinder S. Dhaliwal, Sam K. Doran, Steven D. Golladay, Michael S. Gordon, Timothy R. Groves, Rodney A. Kendall, Jon E. Liebeamerman, David J. Pinckney, Christopher F. Robinson, Robert J. Quickle, James R. Rockrohr, Werner Stickel and Eileen V. Tressler of IBM Corp.; and Gil Varnell, and W. Thomas Novak of the Nikon America Research Corp. The authors thank S. C. Suzuki, H. Shimizu, A. Yamada, K. Nakano, K. Kamijo, S. Shimizu, T. Fujiwara, N. Hirayanagi, H. Yahiro, H. Yamamoto , S. Kojima, T. Irita, K. Hasegawa, and others at Nikon.
References
- T.M. Bloomstein et al., J. Vac. Sci. Technol., B 15, 2112, 1997.
- C. Gwyn, et al., J. Vac. Sci. Technol., B 16, 3142, 1998.
- S.D. Berger, J.M. Gibson, Appl. Phys. Lett., 57, 153, 1990.
- H.C. Pfeiffer, Jpn. J. Appl. Phys., 34, 6658, 1995.
- H.C. Pfeiffer et al., Abs. 43rd Int. Conf. Electron, Ion, Photon Beam Technology, and Nanofabrication, paper NGL2-1, 1999.
- H.C. Pfeiffer, G.O. Langner, J. Vac. Sci. Technol., 19, 1058, 1981.
- W. Stickel, J. Vac. Sci. Technol., B 16, 3211, 1998.
- S.D. Golladay, R.A. Kendall, S.K. Doran, Abs. 43rd Int. Conf. Electron, Ion, Photon Beam Technology, and Nanofabrication, paper SO4, 1999.
- W. Stickel, G.O. Langner, Abs. 43rd Int. Conf. Electron, Ion, Photon Beam Technology, and Nanofabrication, paper SOP6, 1999.
- S. Kojima, et al., Ext. Abs. 60th Autumn Meeting of Japan's Applied Physics Society, paper 4-a-E3 (in Japanese), 1999.
- S. Kawata, et al., Abs. 43rd Int. Conf. Electron, Ion, Photon Beam Technology, and Nanofabrication, paper MP2, 1999.
- K. Uchikawa et al., Abs. 43rd Int. Conf. Electron, Ion, Photon Beam Technology, and Nanofabrication, paper NGL6-3, 1999.
- K. Morita et al., Jpn. J. Appl. Phys., to be published.
For further information, contact Kazuya Okamoto, Nikon Corp., 1-6-3, Nishi-Ohi, Shinagawa-ku, Tokyo 140-8601, Japan; ph 81/337-731-111, e-mail: [email protected].