Behind the Mask

Studying chemically amplified resists and top antireflective coatings in photomask fabrication

Charles Howard, DuPont Photomasks

Device specifications for critical dimensions (CDs) and CD uniformity have driven laser pattern generators to shorter wavelengths in order to enhance productivity and shorten cycle times. Current laser pattern generators operate in the range of 248–257 nm. Exposure doses at those wavelengths require the use of chemically amplified resists (CARs). However, positive and negative CARs are relatively new to photomask manufacturing and present a host of new process challenges. For example, they are extremely susceptible to environmental amines. Parts-per-billion levels of amines are sufficient to adversely affect the photochemistry of the resist system. One method for protecting CARs is to apply a top antireflective coating (TARC) to act as a diffusion barrier.

This article presents a defect partition study of two CAR candidates for photomask fabrication. The two resists are compared with a mature electron-beam resist system, and the nature of the defect populations produced by the two CAR candidates is contrasted.

Test Background

The tests described in this article were prompted by the findings of a previous study, which compared ZEP7000, a mature E-beam resist from Zeon (Tokyo), with another material known as resist A. In that study, AR8 chrome-coated photomask blanks from Hoya (Tokyo) were coated with 4975 Å of resist A. The same inspection tool, inspection algorithms, and sensitivity settings were used for both resists.² ZEP7000 has demonstrated good defectivity performance over the lifetime of its use and performed better than resist A.

Figure 1: General process flow of the photomask partition test.

That study attempted to partition photomask processes to determine the source of defects.^1,2 Previously, the standard industry practice had been to perform defect inspection after processing had been completed, making it extremely difficult to determine the sources of defects. By performing inspection steps after each principal process, the sources of the final photomask defect count could be identified for each process and eliminated. The general flow of the partition test is mapped in Figure 1.

The original study used an Alta 4300 deep-ultraviolet (DUV) laser reticle writer from Etec Systems (Hayward, CA), an Applied Materials company, which required a CAR with a top antireflective coating. The primary purpose of this coating was to suppress CD variation during exposure. Typically, uncoated test masks have a distinct top-to-bottom CD signature: depending on tone, the CDs exposed first are either larger or smaller than the ones exposed last. The initial challenge was to achieve a defect-free topcoat with sufficient thickness to retard CD drift.

Test Protocol

The study described here is based on the earlier partitioning test of ZEP7000. Two different resists were tested: resist A and resist B. Five masks were coated with resist A and one mask with resist B.

First, each mask was exposed on an Alta 4300 DUV reticle writer, which uses an exposure wavelength of 257 nm. After the writing (exposure) step was complete, there was a delay of one hour before the masks were baked. The masks were baked at 70°C and developed. An ASP5000 from Steag HamaTech (Sternenfels, Germany) was used to develop three of the test masks, while an ASP500 was used to develop the remaining two. Each tool used a different formulation of 2.38% tetramethyl ammonium hydroxide (TMAH) photoresist developer. Two PlasmaTherm VLS700 tools from Unaxis (Pfäffikon, Switzerland) were used to etch the masks. Each mask was etched in a Cl₂/O₂ environment.

Test Results for Resist A

After developing and plasma etch steps, a generic line/space pattern was printed and the masks were inspected. Partition results for the top-coated prototype resist A are summarized in Table I. The defect contribution for postdevelop and postetch inspection steps listed in the table refers to the final defect count on the mask. For example, plate 1 had a total of 76 defects, 37 of which had been discovered after the develop step. Postdevelop inspection captured more than 37 defects, but only these 37 were transferred to the final mask pattern.

While resist A demonstrated acceptable lithographic performance, it exhibited a time-dependent CD range variation. Depending on the thickness of the coating, variations between 5 and 20 nm over a 120-minute exposure time were noted. Time-dependent CD nonuniformity was ameliorated by applying a TARC sealant.

Figure 2: SEM image of common missing-chrome defect located in the resist.

Coating techniques can have a significant impact on defects. One unexpected result of using the topcoat/resist combination was the presence of missing-chrome defects. Of all defect types, missing-chrome defects are among the most difficult to overcome and are therefore particularly vexing to photomask manufacturers, since they can make the proper placement of deposition repairs extremely challenging. On some mask features, improper placement of a deposition repair can lead to a catastrophic failure on the wafer. Moreover, some deposition repair processes leave residues that can induce local transmission gradients on the mask. The SEM images in Figures 2 and 3 show common missing-chrome defects. While Figure 2 presents such a defect in the resist, Figure 3 presents the defect after chrome etch and resist strip processes have been completed.

Figure 3: SEM image of common missing-chrome defect after the completion of chrome etch and resist strip processes.

The presence of missing-chrome defects may have indicated the topcoat-induced local breakdown of a coformulated additive in resist A known as a latent image stabilizer (LIST). Breakdown of the LIST would allow the sulfonic acids generated by photolysis during exposure to migrate into unexposed areas of the resist. In the test discussed here, several process modifications were attempted to eliminate these defects. However, while performing additional rinses before postexposure bake reduced the magnitude of the missing-chrome problem, no method was found to completely eradicate it.

Test Results for Resist B

Since various process development efforts proved unsuccessful in eliminating the missing-chrome defects from the initial test resist, the focus shifted to finding an alternate CAR. After some research, an alternate CAR was found that demonstrated lithographic performance similar to that of resist A. Time-delay tests indicated that the alternate resist exhibited a flatter time-dependent CD variation than did the coated version of the initial test resist. Figure 4 shows the difference between the first top-coated CAR (resist A) and the uncoated one (resist B).

Figure 4: Comparison between CD drift on photomasks with the top-coated CAR (resist A) and the uncoated CAR (resist B).

The data for Plate 6 in Table I were collected from a partition test performed on a blank mask coated with resist B. As demonstrated in Table II, the results of using resist B on plate 6 compared favorably with the best results from the use of resist A on plates 1 through 5. Based on the test results, process development on actual device patterns commenced using resist B. Test masks processed using this resist did not have missing-chrome defects. As development efforts have proceeded, defect trends have consistently declined.

Determining defect counts is one way to compare processes. Determining relative defect size is another. Defect inspection reports contain information on the extent of x- and y-axis pixels of each defect. By applying simple rules to these pixel extents, the pixels per defect count for each defect on a given mask can be estimated. In these tests, either the extents were summed or the square root of the area described by the extents, depending on symmetry, was calculated. These techniques proved to be 90% accurate in estimating actual pixel counts during test inspection among 80 defects. The accuracy improved to 95% or better for defects smaller than 10 pixels.

Figure 5: Histogram comparing the pixels per defect resulting from the use of resist B and ZEP7000. (The pixels were 150 nm.)

Inspecting the Photomasks

After the completion of the tests involving resist B, the ZEP7000 partition test masks from the earlier study and the resist B development masks were inspected using a TeraStar system from KLA-Tencor (San Jose). The inspection pixel size used was 150 nm. The ZEP7000 study consisted of 10 masks, while the resist B study consisted of 8 masks. Both sets of test masks were compared based on the number of defective pixels assigned by the inspection tool to each defect. The total chrome defect population of all masks from each set was binned according to the defect size assigned by the inspection tool. Using this methodology, a statistically large sampling of defects from both resist types was normalized and compared. The results of this inspection are plotted as a histogram in Figure 5.

The goodness of fit between the two populations had a χ² value of 0.361, with seven degrees of freedom, implying that the defect signature of the resist B process was statistically the same as that of the ZEP process to within a 5% margin of error. The resist B data were skewed toward the extremes of the relative-size distribution, indicating that the resist B process resulted in a slightly higher percentage of single-pixel defects than the ZEP process. Furthermore, the resist B process also had a higher percentage of large (>15-pixel) defects than the ZEP process. It was assumed that defects 15 pixels or larger were generated by other factors unrelated to the resist or resist process.

Conclusion

CARs have been used in chipmaking for several years. Photomask manufacturers are in the early stages of integrating these resists into their production processes. When choosing a potential CAR for DUV photomask lithography, mask makers have a strong financial incentive to choose materials that are already in wide use in the semiconductor industry and whose engineering and development costs have already been borne by wafer lithographers. Finally, choosing such a CAR ensures a stable supply at a reasonable cost.

Using a TARC on a prototype photomask CAR is a reasonable and sound approach. Taken in a broader context, the data presented in this article reflecting the marginal results achieved using the TARC technique serve more to underscore fundamental differences in wafer and photomask lithography than a particular process development shortcoming.

There are two major differences between wafer and photomask applications: they cluster resist process equipment and exposure tools differently, and they involve different exposure time regimes.

In semiconductor fabs, resist and TARC are first applied, and then the wafer moves directly to the exposure tool. Exposure times are short and the various systems are well integrated. Throughput times of commercially available systems can be on the order of 120–140 wafers per hour.

In contrast, coating, exposure, and process equipment in photomask fabs are independent, stand-alone systems. If resist-coated blanks are used (as they were for resist A), applying the TARC requires 30 to 40 minutes per photoblank. Typical exposure times for a photomask laser writer range from 2 to 2.5 hours per mask. The minimum amount of time the TARC is interfaced with the resist, both in an unexposed and exposed state, is therefore 3 hours or more. Hence, there are structural differences between wafer and photomask applications involving the same resist-TARC system. The specifics of the photomask application certainly played a role in the wide CD drift and increased defect counts seen in the resist A tests, providing an object lesson against borrowing too liberally from the wafer world.

When choosing resists, the photomask lithographer faces a number of constraints that should serve to winnow the field of prospective CAR candidates. However, the results of the resist B tests demonstrate that good alternatives are available. Those tests resulted in CD stability during exposure that compares well with the CD stability of the resist A/TARC approach. Moreover, resist B's defect performance was similar to that of ZEP7000. Shelf-life tests of up to 16 weeks have proven that resist B is stable, indicating that blanks can be commercially coated by photoblank vendors.

In conclusion, the semiconductor industry provides manufacturers with a wealth of experience, but it should always be kept in mind that photomask fabrication is a niche business. In the final analysis, judicious screening and vigorous prototyping should guide the CAR selection process.

References

1. C Howard et al., "Partitioning of Photomask Processes for Defects," in Proceedings of the 21st Annual BACUS Symposium on Photomask Technology (Bellingham, WA: SPIE, 2002), 418–429.

2. C Howard and M Lamantia, "Partitioning of Photomask Processes for Defects—Part II," in Proceedings of the 22nd Annual BACUS Symposium on Photomask Technology (Bellingham, WA: SPIE, 2003), 1001–1009.

Charles Howard is an optical lithography development engineer at DuPont Photomasks (Austin, TX). Since joining the company in 1991, he has held several engineering positions. He received an MS in physics from Southwest Texas State University in San Marcos. (Howard can be reached at 512/310-6208 or charles.howard@photomask.com.)