Removing sub-50nm particles during blank substrate cleaning
04/01/2006
The current International Technology Roadmap for Semiconductors (ITRS) requires that all defects ≥32nm on the EUV mask blank and ≥27nm on EUV substrates (i.e., mask blank before the multilayer [ML] coating) be removed by 2010. To ensure the availability of the required technology to satisfy these requirements, Sematech and its member companies have established the EUV Mask Blank Development Center (MBDC) in Albany, New York, with world-class blank inspection, cleaning, and deposition capabilities. This article describes promising results of blank substrate cleaning that demonstrate EUVL is one step closer to realization.
Abbas Rastegar, Sean Eichenlaub, Mask Blank Development Center, Sematech, Albany, New York; Yoshiaki Ikuta, Asahi Glass,Tokyo, Japan; Helmut Popp, Schott Lithotec, Meiningen, Germany; Kurt Goncher, Steag-Hamatech, Austin, Texas; Pat Marmillion, IBM assignee at Sematech, Austin, Texas
Extreme ultraviolet lithography (EUVL) is thought to be the main candidate for next-generation lithography (NGL) for the 32nm node and beyond. A typical EUV mask blank is made of a low thermal expansion material (LTEM) substrate, which is coated with ~50 bilayers of molybdenum and silicon (MoSi), and a capping layer of Si or ruthenium (Ru). The backside of the EUV substrate is coated with chromium for electrostatic chucking (ESC) during the multilayer deposition and exposure processes. Single-phase LTEM materials are very similar to fused silica as far as cleaning and surface preparation is concerned; however, they are much more expensive than fused-silica substrates. Therefore, fused-silica substrates are used for process development for economical reasons, and the optimized process is then tested on LTEM substrates.
Cleaning challenges
Analysis at Sematech [1] shows that ~94% of EUV mask blank defects originate from the substrate, while only 2% are caused by the deposition process and 4% come from plate handling. Hence, the substrate contributes the most blank defects, and the greatest improvement in mask blanks can be achieved by removing these defects. Defect characterization studies at Sematech [2] on fused-silica substrates provided by different glass suppliers indicate that 85% of defects 60nm and larger are easily removable from the substrate (i.e., soft particles) and about 10% are hard to remove (i.e., hard particles). The remaining 5% of defects are pits or scratches on the surface of the glass. Although detectable pits account for only 5% of the total incoming substrate defects, after ML deposition, the percentage of pits may increase to 30-80% of the total detectable defects on the mask blank [3]. The large variation in the percentage of pits post-ML deposition depends on both the incoming substrate quality, which may have many pits smaller than 60nm that are undetectable, and on the ML deposition process parameters, which may increase the detectability of these small pits. Other paths in addition to cleaning must be followed to reduce the number of the nanometer-sized pits on the incoming glass substrate.
Most available cleaning tools are able to effectively remove particles that are ≥100nm when the density of these particles is high (~100 particles/cm2) [3]. However, when the defect density on the incoming plates is low (~0.05 particles/cm2), most cleaning tools fail to clean these plates and, in fact, add to the number of defects on the plate.
Mask cleaning requires not only tool capability to remove particles >50nm but also a tool and cleaning process that does not add any particles >50nm during cleaning. The latter condition is very challenging because different tool parts, chemicals, ultra pure water (UPW), cleaning processes, and airborne particles all contribute to defects added during cleaning. Extremely tight controls on the quality of the UPW, chemicals, cleaning tools, and airborne particulates, as well as fully automatic handling processes, are required to achieve sub-50nm particle removal capability.
Particle removal
The MBDC has successfully improved its inspection capability down to 43nm defects for a fused-silica mask blank substrate. This improved capability is a result of both hardware modifications and development of complementary defect review methods that improve capture rate and reduce the false defect count of the inspection tool.
In addition to improving defect inspection capability, the MBDC has developed a single defect tracing method (SDTM). This method enables tracing of single defects as small as 43nm during different characterization steps with different analytical techniques and tools. After inspection, the inspection tool creates a set of fiducial marks on the surface of the plate. The fiducials are used to align the plate in the atomic force microscopy (AFM), scanning electron microscopy (SEM), and energy-dispersive spectroscopy (EDS) tools. Next a punch mark (~5µm in size) is created within ~5µm of the sub-100nm defect. This punch mark is easily detectable by analytical tools and is used to find the desired defect.
To study defect removal mechanisms for sub-50nm defects, ~50nm-sized particles before cleaning were selected and marked using size data obtained by the inspection tool. Defect morphology was determined by using AFM imaging before and after cleaning; composition and geometry were determined using SEM and EDS after cleaning. The AFM images were used to determine a sphere equivalent volume diameter (SEVD), which was used as a standard to define irregularly shaped defect sizes. The SEVD is defined as the diameter of a sphere whose volume would be equivalent to that of the defect. SEVD is calculated based on the defect volume determined by AFM software (Veeco Nanoscope software). To determine the accuracy of the SEVD approach, the SEVD for polystyrene latex (PSL) spheres was calculated from AFM images. An AFM scan of two 50nm and two 100nm PSL spheres was used. The calculated SEVD values for the PSL spheres were compared with the expected values. The SEVD values were ~5% larger than the PSL diameter supplied by the manufacturer. This is reasonable considering the AFM tip shape and AFM imaging artifacts and the small, expected variation in the PSL size distribution.
There are several mechanisms by which sub-50nm particles are removed from the surface. A few sub-50nm particles are completely removed by the cleaning process. One such evaluation resulted in the complete removal of two sub-50nm SEVD particles that were weakly adhered to the surface and relatively easy to remove. However, many sub-50nm particles can adhere strongly to a surface and are very difficult to remove completely.
 Figure 1. AFM image of a particle a) before, and b) after cleaning. |
Figure 1 shows an AFM image of a particle before and after cleaning. In this case, the particle has a large contact area with the surface and, therefore, is hard to remove. Figure 1b shows the remaining trace of the particle after cleaning. This and other similar data indicate that for a particle that has a large contact area and that is mainly organic, chemical dissolution is the main mechanism of particle removal on the glass substrate.
Most theoretical analyses show particle rotation due to drag forces from the fluid to be the main physical mechanism of particle removal. These models are validated based on experimental removal data using PSL spheres deposited on the surface of interest [5-8]. However, chemical dissolution seems to be a dominant mechanism for small native glass defects that have large contact areas or that are strongly bonded to the surface.
In addition to particles that are attached to the surface by adhesion forces, other particles may be embedded in the glass surface and/or chemically bonded to the glass surface. Removal of such a hard particle will deform the surface deformation and create pits.
The cleaning process can damage the glass surface, increasing the surface roughness or creating and/or enlarging the pits. To examine whether the cleaning process damages the surface by increasing the size of pits or surface roughness, pit dimensions and the surface roughness on the substrates were measured with the AFM before and after cleaning. The surface roughness measurements indicate that the increase in surface roughness (rms) is 0.004Å/cleaning cycle when the optimized cleaning recipes are used. This is negligible and within the error of the measurement.
AFM cross-section analysis of nanometer-sized pits before and after cleaning of a quartz substrate indicates that there is no change in the pit size due to the optimized cleaning processes. The results for these individual pits were confirmed for all pits on the mask through statistical analysis of pit sizes measured by our defect inspection tool.
Based on these results, cleaning recipes that are able to shrink the size of particles by 20% without adding additional particles or pits on the surface of the plate have been developed.
 Figure 2. Reducing the particle size by cleaning. |
Figure 2 shows the relationship between the size of particles before (x-axis) and after (y-axis) cleaning. The data points are the SEVD of particles as measured by AFM. The light blue data points show that the size of the particles was reduced by ~20% when using an optimized cleaning recipe. Extrapolation of the data suggests that particle dissolution should be the dominant mechanism for residue removal down to a few nanometers.
 Figure 3. Defect maps of the fused-silica plate a) before, and b) after cleaning. The histogram shows the defect count for different defect sizes. |
null
Mask substrate cleaning
Figure 3 shows defect maps for fused-silica glass before and after cleaning. The upper square shows the defect map, which indicates defect location on an area of 142×142 mm2 (5mm exclusion zone from glass edge). The lower portion of each figure is a histogram of defect counts where bins represent defect sizes. For a quartz surface, bin one and greater represents defects >43nm and bin 4 and higher represents defects >53nm. No defects >53nm are left on the glass surface after cleaning; this is the first data reported on total particle removal down to this size. The surface roughness is ~1Å (rms) before cleaning and remains the same after cleaning. Optimized conventional cleaning processes based on a sulfuric acid/hydrogen peroxide mixture (SPM) and an ammonium hydroxide/hydrogen peroxide mixture (APM) were used for cleaning these plates.
 Figure 4. Defect map of a cleaned fused-silica plate. All particles >43nm are removed. |
The MBDC’s latest achievement in total particle removal with current detection limits is shown in Fig. 4. These results were obtained using the optimized conventional cleaning processes. This illustrates that all particles >43nm were removed from the plate, and the only defects remaining after cleaning are four pits that were on the incoming substrate.
Conclusion
Surface defects are one of the most critical issues in the development of EUVL technology. Substrate defects are the main contributor to total mask blank defects. Glass supplier companies have made impressive progress in reducing defects and surface roughness of fused silica and LTEM substrates. Sematech’s MBDC has demonstrated it is possible to further clean these already rather clean substrates and remove all particles >43nm far ahead of the ITRS requirement. Current activities are focused on demonstrating similar results on LTEM substrates.
References
- A. Rastegar et al., “Recent Advances in Cleaning EUV Substrates,” International Symposium on Extreme Ultraviolet Lithography, San Diego, California November 2005.
- A. Rastegar et al., “Progress in Mask Cleaning in Sematech,” Sematech Mask Cleaning Workshop, Monterey CA, October 2005.
- P. Kearney, et al. “Overcoming Substrate Defect Decoration Effects in EUVL Mask Blank Development,” Proc. SPIE, Vol. 5567, p. 800-806, 24th Annual BACUS Symposium on Photomasks.
- A. Rastegar, “Cleaning of Clean Quartz Plates,” Surface Preparation and Wafer Cleaning Workshop, Austin, April 2005.
- K.L. Johnson, K. Kendall, A.D. Roberts “Surface Energy and the Contact of Elastic Solids,” Proc. R. Soc. V 324, p. 301, 1971.
- G.M. Burdick, N.S. Berman, S.P. Beaudoin, Journal of Nanoparticle Research 3, p. 455-467, 2001.
- K. Cooper, S. Eichenlaub, A. Gupta, S. Beaudoin, Journal of the Electrochemical Society, Vol. 149(4), 2002.
- M.A. Hubbe, Colloids and Surfaces 12, p. 151-178, 1984.
Abbas Rastegar received his PhD in physics from Ljubljana U. in Ljubljana, Slovenia, and is project engineer at the Mask Blank Development Center, Sematech, 255 Fuller Rd., Ste 309, Albany, NY 12203; ph 518/956-7106; e-mail [email protected].
Sean Eichenlaub received his PhD in chemical engineering from Arizona State U. and is postdoc at the Mask Blank Development Center, Sematech, Albany, NY.
Yoshiaki Ikuta received his PhD in material science from Tokyo Institute of Technology in Tokyo, Japan, and is manager of advanced research at Asahi Glass, Tokyo, Japan.
Kurt Goncher received his masters in engineering from Colorado State U. and is a process engineer at Steag-Hamatech, Santa Clara, CA.
Helmut Popp received his physics diploma from Friedrich-Alexander U. in Nuremberg-Erlangen, Germany, and is a process engineer at Schott Lithotec, Meiningen, Germany.
Patricia Marmillion received her MS degree in materials science from the U. of Vermont and is an IBM assignee at Sematech, Austin, TX.