Behind the Mask

Comparing software and hardware simulation tools on an embedded-attenuated PSM

Eric R. Poortinga, Rochester Institute of Technology; Justin W. Novak and Benjamin G. Eynon, Photronics; and J. Tracy Weed, Linyong Pang, and Linard Karklin, Numerical Technologies

Virtual Stepper and aerial image measurements systems can analyze defect-induced CD error as a function of various isolated defect sizes on a production mask.

Market demands are continually pushing IC features farther and farther into the subwavelength regime, where device feature sizes are significantly smaller than the wavelength of light with which they are manufactured. This applies to any device below the 248-nm technology node and soon any device below 193 nm.

The fact that manufacturing in the subwavelength environment is possible at all is because of the dramatic improvements in IC processes such as CMP and the introduction of chemically amplified resists. In addition, the increasing acceptance and implementation of resolution enhancement techniques (RETs) such as optical proximity correction (OPC) and phase shift masks (PSMs) have extended the useful life of exposure tools while driving k1 levels lower than previously anticipated. Enhancement techniques work in concert with traditional techniques of lowering wavelength (KrF, ArF, F2) and increasing numerical aperture to achieve better process control and expand the available process window. As a result of their nonlinear behavior, these techniques have led to a tremendous increase in the interdependency of the processes used, from design all the way through silicon fabrication.

Over the next five to eight years, the number of wafers manufactured in the subwavelength environment will increase exponentially, with estimates placing the compound annual growth rate at between 140 and 150%.^1,2 During this time, chip performance must also increase, while size and power usage decrease. These technical challenges are coupled with the need to hit narrowing market windows and meet the resultant aggressive schedules for product introduction, while maintaining or improving cycle time.

With few exceptions, semiconductor manufacturing in the above-wavelength regime is quite straightforward. The design layout looks like the photomask, which in turn looks like the printed pattern on the wafer. A far more likely scenario in the subwavelength regime—thanks to OPC and the other techniques—is that the pattern design looks slightly different from what is written on the mask, which then looks even less like what is printed on the wafer. This transition from above-wavelength to subwavelength processes means that a fundamental change must occur in the way the semiconductor industry addresses the entire design-to-silicon process flow. Any company doing things the old way will no longer maintain a leadership position.

No area has felt the impact of the transition to the subwavelength realm and the need for change more keenly than photomask fabrication. Long a commodity governed by well-established rules, masks are rapidly becoming the enabling technology critical to the success of subwavelength manufacturing. With RETs driving tremendous levels of complexity into leading-edge masks, one formidable task facing the industry is how to guarantee mask quality in a cost-effective manner. To address this challenge, new approaches and ideas must be explored and implemented if a viable solution is to be found that meets customers' increasing demands.

To this end, mask quality has long been approached as an "inspect all­repair all" proposition. Although this brute-force methodology worked well at mask feature sizes >0.5 µm, it is impractical for smaller-sized geometries. It has become critical to be able to separate true defects (those that affect wafer performance and yield) from nuisance defects (those that if left uncorrected will not affect the quality of the wafer outcome for which the mask is intended). The benefits of this more-differentiated approach are higher mask yields, fewer repairs, and better turnaround time for the customer. Millions of dollars are wasted each year remaking or repairing photomasks that do not contain true defects. Fixing all defects is time intensive, costly, and often unnecessary. It may not be wise to repair all defects if some of them do not have an adverse impact on device performance. New approaches to assessing mask quality must be implemented—"inspect all­repair what prints."

There are many commercially available tools and methods that can be used to analyze mask defect printability. Many of these tools are employed every day for mask quality and defect dispositioning but not necessarily in a production mode. Most of the work done has also focused on binary intensity masks (BIMs), not on PSMs. Efforts have increased to develop PSM algorithms to capture the associated phase, transmission, and resulting edge definition. This article explores and discusses the capabilities of two printability simulation tools and their analyses of defect-induced CD error as a function of various isolated defect sizes on a Photronics production PSM.

The first tool, the Virtual Stepper System from Numerical Technologies (San Jose), takes the optical image generated by an inspection system or metrology tool and applies an optical stepper model to produce fast and accurate simulations of an intermediate aerial image pattern. A proprietary algorithm simulates the final wafer pattern and allows the prediction and analysis of defect printability and, ultimately, mask quality. The simulator considers the exposure system wavelength, numerical aperture (NA), coherence factor (), illumination modes (conventional, annular, quadrupole), and lens aberrations. Contour plots based on resist exposure and etch process simulations can also be applied to the aerial image pattern to further refine the simulated wafer image.²

The stepper system's analytical capabilities include:

• CD versus exposure level.
• Process window (PW) analysis, including multiple overlapping PW.
• Exposure dose latitude trade-off.
• Cross-sectional aerial image intensity (at different focus planes).
• CD versus defocus (Bossung plots).

The Aerial Image Measurement System (AIMS) MSM-100 from Carl Zeiss (Thornwood, NY) can also be used to perform physical aerial imaging of masks prior to wafer printing. AIMS software, which was jointly developed by Zeiss and IBM, can display postlithographic images by simulating the stepper conditions to be used for a particular mask; thus, the settings of NA, , magnification, and wavelength are adjusted. Because the tool utilizes a mercury-xenon lamp, it can image patterns at wavelengths of 436, 365, and 248 nm using transmission filters. A 193-nm system is also available.³

Experimental Methods

The sample used in this evaluation was a 6 x 6 x 0.25-in. Ulcoat PSM with a 6%-transmission MoSiON absorber. The mask was built utilizing Photronics's proprietary ePhaseII PSM process, and the SEMI standard programmed defect test pattern for photomasks was used. Figures 1 and 2 show the specific pattern analyzed.

 

Figure 1: Background pattern overview.

Figure 2: Programmed isolated defect.

Once the mask was processed, an LWM-250 white-light CD metrology tool from Leica (Deerfield, IL), set at 150x, measured clear-space CDs of nondefective areas of the SEMI standard pattern. Since the target feature size was near 1.0 µm (at the mask plane), monochromatic light was not necessary. The widths of nondefective lines were measured to provide a baseline from which to set the threshold values for the stepper system and the AIMS. Seven locations were selected, with defect sizes ranging from 80 to 200 nm at the wafer plane. Another set of reference data were captured with a defect measurement system from AVI (Santa Clara, CA). This tool, which correlates well with a CD SEM, was used to make quick, accurate measurements of the isolated defects themselves.⁴

For image analysis, bitmaps were taken of each defect under investigation on a KMS-400 CD metrology tool from Zygo (Middlefield, CT). The imaging wavelength was g-line (436 nm) and the magnification was 100x, with a pixel size of 52.6 nm. These images were fed into the stepper system's software for printability analysis. Examples of the KMS-generated defect images are shown in Figures 3 and 4.

Software Stepper Simulation

This study marked the first time an enhanced prototype version of the existing stepper system software was used. Parameters were set to match the LWM CD values as closely as possible. Since there was no phase information available from the KMS measurements and the mask transmission retrieved by the g-line KMS systems was different from the mask transmission at actinic wavelength (248 nm), all necessary phase and transmission corrections were done at the stepper system simulation.

 

Figure 3: Small programmed defect.

Figure 4: Large programmed defect.

Simulated wafer images from these input bitmap mask images were initiated by the generation of an optical stepper model, which was done by the application of Numerical's ModelGen software. This software creates accurate, user-specific stepper models. The user then customizes the system for this particular application—the wafer exposure conditions from the tool in which the mask will ultimately be used. The actual wafer image calculation can be performed on any portion of the input bitmap mask image. Once completed, contour plots were generated and overlaid onto the aerial (or resist) image pattern. From these plots, a mask inspection engineer or operator has enough information regarding the defect performance at the wafer level to make reliable pass/repair/reject decisions.

Hardware Aerial Image Simulation

Before measurements could be taken, the lamp had to be precisely aligned with the optical column (consisting of NA, sigma, condenser, and field stop). Once aligned, the apertures were then focused on the mask plane. A clear reference image was then taken in an open quartz area (2 x 2 mm) to normalize the forthcoming measurements. The stepper simulation parameters chosen for this study were:

• Wavelength = 248 nm.
• Magnification = 4x.
• Numerical aperture = 0.5.
• Partial coherence factor () = 0.6.

Using the defect inspection report from a KLA 353 (KLA-Tencor, San Jose), the stage received a reference point that allowed the MSM-100 to accurately move (within 3 µm) to each defect location. Once at the defect site, a preview option using UV light allowed the defect to be centered if it was large enough to resolve. A through-focus image was then taken using 11 2-µm focus increments. Image analysis helped to accurately focus the image at the defect location. The numerical aperture was adjusted using micrometers to precisely align the through-focus images. The stage was also adjusted in the z-axis to find the best focus setting.

Once an aligned and adjusted through-focus image was captured, defect analysis could begin. An intensity profile plot shows changes in intensity across an x or y swath of a preset width. By setting a resist threshold (i.e., the amount of intensity to activate resist), a two-dimensional contour plot can be extracted showing the resist image on the wafer level. A linewidth versus defocus plot shows the change in linewidth at defect locations compared to defect-free locations.

This study used both the contour plot and simulated linewidth plot to compare the results after employing the stepper system and AIMS. A resist threshold of 30% (AIMS) was used to determine printability on the wafer, which accurately correlates to the above-mentioned LWM measurements. Taking the percentage change in CD at each defect location removed possible image bias, allowing a greater degree of comparison between the stepper system and AIMS.

Results and Discussion

The CD measurements from the LWM-250 ultimately determined the threshold values used for the AIMS and the stepper system analyses, so a zero printed bias for this reticle was assumed. This is not normally the case for embedded-attenuated PSMs, but it was done here because printed wafer data were not available. As Figure 5 shows, a mean clear space CD of 0.288 µm at the wafer plane was targeted based on an average of the seven sites evaluated. The LWM had a mean of 0.288 µm with an 8-nm range, while AIMS simulated a mean of 0.287 µm (at a 30% threshold) with a 10-nm range, and the stepper system showed a mean of 0.290 µm (at a 37.6% threshold) with a 14-nm range.

 

Figure 5: Comparison of nondefective clear-space CD measurements.

Using the AIMS linewidth versus defocus plots as a guide, the best focus at clear-space locations correlated to approximately a 30% threshold. Examples of the 0.118-µm defect location are illustrated in Figures 6 and 7. At this threshold, the target CD of 0.288 µm was produced. For the stepper system, an exposure threshold of 37.6% was found to correlate to the targeted CD. Figures 8 and 9 are contour plots from both AIMS and the stepper system that show the effect of a 0.118-µm defect.

 

Figure 6: AIMS linewidth versus defocus plot.

Figure 7: AIMS intensity plot.

Figure 8: AIMS contour plot.

Figure 9: Stepper System contour plot (pixel size = 13.16 nm).

A linewidth of the space across the defect was also measured at each site. By taking the difference of this measurement and the nondefective space measurement from the same region, a change in linewidth was determined, also called CD error. The CD error was then compared to the defect sizes measured by the AVI tool. As Figure 10 reveals, the CD error does not correspond linearly with actual defect size. This well-known phenomenon is the result of a defect­main feature interaction caused by an optical proximity effect.⁵ The virtual stepper system found a larger CD error than the AIMS tool did for defect sizes >0.100 µm. That finding agrees with what had been seen previously, where the stepper system demonstrated that it can approach defect printability from a more conservative side than AIMS.⁶ The <0.100-µm defect produced a lower CD error using the stepper system than that found with AIMS. Interestingly, AIMS predicted the same CD error for a 0.095-µm defect, while the stepper system showed a higher sensitivity for sub-0.1-µm defects (see Figure 10). Moreover, the stepper system predicted readily seen resist lines bridging at a 0.20-µm defect size, and AIMS simulation showed fully separated lines with a delta CD of 69%.

 

Figure 10: Correlation of simulated CD error with the defect size.

The CD error of each tool was then taken as a percentage of the defect size. The difference in CD error could then be taken between simulation tools. Figure 11 illustrates those findings.

 

Figure 11: CD error difference versus defect size.

Conclusion

In the case of attenuated PSMs, the stepper system predicts higher defect printability compared with the AIMS estimate, results that are consistent with a previous study that looked at binary defect printability.^6,7 These findings also highlight the importance of correctly positioning the prediction of defect printability. A conservative estimate is thought to be most appropriate for a leading-edge production environment.

Both AIMS and the stepper system are simulation tools and should be calibrated and verified against the only objective judge—printed silicon. The fact that even without third-party calibration both tools demonstrated consistent defect printability results is quite encouraging. As shown in Table I, if one uses the common practice of setting a 10% CD tolerance threshold, the AIMS and the stepper system tools agree on which defects matter and which do not. It is also important to note that the algorithm used by the stepper system is an enhanced PSM prototype and will need additional investigation to better correlate the system's results with printed wafer data. A mask inspection engineer or operator should be able to separate defects that need repair from those that do not. While this has been evaluated on binary masks in the past, this study shows that on this data set, either AIMS or the stepper system could be used to accomplish the same task on MoSiON PSMs.

 

AVI Defect Size (µm) Delta CD (%)--AIMS Delta CD (%)--Stepper System 0.080 6.9 1.7 0.095 6.9 8.8 0.118 13.8 17.7 0.130 14.3 18.9 0.145 20.7 26.0 0.163 32.1 39.8 0.200 69.0 100.0

Table I: Inspection cull at 10% threshold

While the existing stepper system has shown good agreement when used on binary intensity masks, the prototype virtual stepper algorithm for use with attenuated PSMs will continue to be modified to correlate even more closely with wafer data. Images were collected for the stepper system tool using g-line illumination. The simulation accuracy can be greatly improved by collecting mask data at actinic wavelength. In the future, i-line, 248 nm, and even SEM images should be studied for their effects on improving these errors. Wafer prints of the phase-shift mask will provide the final say on what threshold values or calibrated resist model should be used, thereby validating or disproving the use of nondefective lines as a reference.

Acknowledgments

The authors wish to thank Suresh Biligiri of Photronics in Milpitas, CA, for collecting the KMS bitmap data, and Darren Taylor of Photronics in Allen, TX, for collecting the AVI defect size data.

References

1. Information found on Web site of VLSI Research, www.vlsiresearch.com
2. "A Perspective on the EDA Market: Technology Forces Driving SOC Design" (Santa Clara, CA: Collett International, 2000).
3. Numerical Technologies, Virtual Stepper System User's Manual 1.1 (San Jose: Numerical Technologies, 1999).
4. R Budd et al., "Development and Application of a New Tool for Lithographic Mask Evaluation, the Stepper Equivalent Aerial Image Measurement System, AIMS," IBM Journal of Research and Development 41, no. 1­2 (1997): 119­129.
5. P Fiekowski, "The End of Thresholds: Subwavelength Optical Linewidth Measurement Using the Flux-Area Technique," in Proceedings of SPIE Symposium on Photomasks—Japan (Bellingham, WA: SPIE, 2000).
6. L Karklin, "A Comprehensive Simulation Study of the Photomask Defect Printability," in Proceedings of SPIE 2621 (Bellingham, WA: SPIE, 1995), 490­504.
7. D Taylor and R Eandi, "Advanced Mask Printability Analysis Using TINT Virtual Stepper System," in Proceedings of SPIE 3546 (Bellingham, WA: SPIE, 1998), 466­476.
8. K Phan et al., "Comparison of Binary Mask Defect Printability Analysis Using Virtual Stepper System and Aerial Image Measurement System," in Proceedings of SPIE 3873 (Bellingham, WA: SPIE, 1999), 681­692.

Eric R. Poortinga is working toward a BS in microelectronic engineering at the Rochester Institute of Technology (Rochester, NY). He is fulfilling his second cooperative education requirement as a development engineer in the Photronics Technology Group in Austin, TX. Poortinga has concentrated on AIMS research while working on other projects such as mask layout design, metrology tool utilization, and mask processing. (Poortinga can be reached at 512/248-6172 or epoortinga@austin.photronics.com.)

Justin W. Novak joined the Photronics Technology Group in 1999 and is working as a development engineer at the company's Austin, TX, facility. He is responsible for equipment and process development of advanced photomask repair and printability programs. Before joining Photronics, he worked as a lithography/etch/diffusion process engineer, technical marketing engineer, and yield enhancement engineer for semiconductor and equipment manufacturers. He received his BS in microelectronic engineering from the Rochester Institute of Technology in 1999. (Novak can be reached at 512/248-6173 or jnovak@austin.photronics.com.)

Benjamin G. Eynon is director of back-end-of-line technology development at the Photronics Technology Group in Austin. Before assuming that position in 1998, he worked as a process engineer, technical marketing engineer, and manufacturing manager for semiconductor and photomask manufacturers. He received a BS in microelectronic engineering from the Rochester Institute of Technology. (Eynon can be reached at 512/248-6169 or beynon@austin.photronics.com.)

J. Tracy Weed, PhD, joined Numerical Technologies (San Jose) in 1999 as senior director of marketing and business development. He is responsible for the semiconductor equipment and mask technologies necessary to support advanced OPC and PSM techniques in addition to product management associated with the Virtual Stepper System and IC WorkBench. Weed worked for IBM Microelectronics from 1984 to 1999, where held a variety of engineering and management positions, largely focused in the area of advanced lithography development specializing in OPC and PSM techniques. He earned his BS and MS in structural inorganic chemistry from the University of Connecticut and his PhD from the University of California at Riverside in the area of synthetic organometallic chemistry. (Weed can be reached at 408/273-4320 or tweed@numeritech.com.)

Linyong Pang, PhD, joined Numerical Technologies in 1999 as the engineering manager of the Virtual Stepper System. He is responsible for the development of this product and advanced mask defect printability analysis software. Before joining Numerical, he worked at Acuson's R&D lab and held senior engineering and management positions, during which time he invented and developed FreeStyle, an extended-field-of-view imaging product. He has eight patents pending in that area. He has a BS and an MS in mechanical engineering from the University of Science and Technology of China in Hefei, Anhui, and a PhD in mechanical engineering and an MS in computer science from Stanford University in Palo Alto, CA. (Pang can be reached at 408/273-4330 or lpang@numeritech.com.)

Linard Karklin, PhD, joined Numerical Technologies in 1997 and now serves as the company's chief scientist. He pioneered the Virtual Stepper concept and has published numerous papers on mask defect printability. Previously, he was vice president of engineering and U.S. operations for Sigma-C, where he conducted the development of SOLID-C, a 3-D lithography simulator. Between 1993 and 1995, Karklin was an independent consultant on advanced photolithography development and simulation. From 1989 to 1993 he was a project manager working on the development of lithography and process simulation tools at Silvaco. Before that, he was a senior scientific staff member at the Latvian Academy of Sciences. Karklin received his BS and MS in electrical engineering and semiconductor physics from the Latvian State University, Riga, Latvia, and his PhD in physical chemistry from the Latvian Academy of Sciences, Riga. (Karklin can be reached at 408/273-4311 or lkarklin@numeritech.com.)

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit