Behind the Mask

Examining the challenges posed by NGL photomask fabrication

Michael Lercel, IBM; and David Walker, Photronics

Revolutionary changes in next-generation lithography masks are being driven by the need for small CD and image-placement accuracy, novel substrates, and the control of subresolution defects.

As semiconductor device manufacturers continue to follow the specifications contained in The International Technology Roadmap for Semiconductors (ITRS), feature sizes are becoming smaller than the wavelengths of light used to print them.¹ The key mask specifications supporting lithography nodes for 100 nm and below are listed in Table I. Ingenious techniques such as partial and off-axis illumination, optical proximity effect correction, and phase-shift masks have been developed to extend the production capabilities of deep-ultraviolet (DUV) steppers. These techniques employ phase information to engineer the wavefront of the radiation that exposes the resist on the wafer. However, with limited choices of optical exposure wavelengths below 157 nm, these techniques will soon be insufficient to ensure robust lithographic processes for device nodes below 70 nm. To meet the needs of new feature specifications, a class of next-generation lithography (NGL) technologies is being developed that utilizes extreme ultraviolet (soft x-ray) and electron beams, which will reduce exposure wavelengths well below 35-nm dimensions, the smallest node in the roadmap.

Technology Node 100 nm 70 nm 50 nm 35 nm Mask minimum image size (nm) 260 180 120 80 Image placement (nm) 21 15 12 9 CD uniformity—isolated lines (nm 3x) 10 7 5 3 Defect size (nm) 80 55 40 28

Table I: Key 4x NGL mask requirements as specified by the ITRS.

The fabrication of NGL masks presents old and new challenges that stem from both current and upcoming photomask manufacturing technologies. The transition to NGL will see a shift in emphasis from wavefront engineering with state-of-the-art photomasks to membrane processing, materials damage, and chromatic-aberration control.

Small allowances for critical-dimension and image-placement accuracy; novel substrates; and the inspection, detection, and repair of subresolution defects will force revolutionary changes in the infrastructure of mask technology. In many ways, the experience acquired in developing 1x x-ray membrane masks paved the way for the difficult development of the new NGL technologies.^2,3

This article examines the critical issues posed by the fabrication of NGL masks, including mask formats and new processing technologies. The article also concentrates on the work of the NGL Mask Center of Competency (MCoC), located at IBM's Burlington, VT, manufacturing site. A Photronics facility run in cooperation with IBM and supported by the Defense Advanced Research Projects Agency (DARPA), the MCoC is using the extensive experience gained by IBM's x-ray mask facility to accelerate the development of NGL mask-manufacturing technology. A mask facility independent of lithography tool development activities that addresses the mask fabrication challenges for NGL, the MCoC has fabricated a range of masks for a variety of applications, including proximity x-ray lithography (PXL), electron projection lithography (EPL), scattering membrane and scattering stencil lithography, and extreme ultraviolet lithography (EUVL).

The MCoC's strategy is to apply cross-cutting process technologies to all NGL mask formats. Although there are many differences among the various mask formats, much process knowledge is common to all of them. While electron-beam patterning of small-dimension images with chemically amplified resists, stress control in thin absorber films, E-beam inspection of masks, and etching of high-aspect-ratio absorber structures were techniques developed to produce 100-nm (on-mask) image features for x-ray masks, they are used to fabricate the other types of NGL masks as well.

Mask Formats

The various NGL masks bear certain similarities to one another and share many common technology issues, but they also have significant format differences. (Moreover, the imaging techniques of NGL masks are much different from those used for optical lithography masks, where light is transmitted through a clear quartz substrate and absorbed by a thin chromium layer.) To be transparent to exposing radiation, x-ray masks and both types of EPL masks are fabricated on thin membranes, while EUVL masks use a reflective layer and are fabricated on a solid substrate. Figure 1 presents the four main NGL mask formats, and Table II contrasts NGL exposure radiation types, mask formats, clear materials, and imaging materials.

Figure 1: The four main NGL mask formats.

 

Technology Radiation Mask Format Clear Material Opaque (Imaging) Material Binary optical Photons 157 nm 152 mm sq Quartz Chromium X-ray Photons = 1 nm 100-mm wafer with support ring 2-µm SiC membrane 500-nm TaSi SCALPEL EPL 100-keV electrons eff ~ 10–3 nm 200-mm wafer with membrane array 100-nm SiNx membrane 30-nm W/Cr PREVAIL EPL 100-keV electrons eff ~ 10–3 nm 200-mm wafer with membrane array Stencil—etched regions in membrane 2-µm Si membrane EUVL Photons = 13.4 nm 152 mm sq Reflective Mo/Si multilayer 50–100-nm Cr, TiN, or TaN

Table II: Mask formats and representative materials.

X-Ray Masks. X-ray masks are defined in a 100-mm silicon wafer base substrate and consist of an ~2-µm-thick membrane that is typically made of silicon carbide. The wafer is bonded to a support ring that forms a rigid mechanical interface for processing and mounting in the x-ray stepper. A single large membrane, which is usually ~35 x 35 mm in size, contains the patterned regions. These opaque regions with the desired device patterns are formed from a heavy-metal layer, such as tantalum silicon or tungsten.

While optical lithography and most NGL technologies use some means of pattern reduction to print the mask features on the wafer, x-ray masks with feature sizes as small as 100 nm are difficult to manufacture because their patterns are printed while the mask is in close proximity to the wafer, requiring that the patterns be the same size as the desired wafer images. Nevertheless, x-ray masks with feature sizes as small as 100 nm have been fabricated in pilot lines at the MCoC for both internal and external customers, making this the most mature of all NGL mask technologies.

EPL Masks. There are two types of electron projection masks, and in both cases the electron is not absorbed into the imaging layer, but only scattered out of the optical path of the stepper. The first type of EPL mask, the stencil mask, has clear regions that are openings etched completely through the membrane, resulting in 100% transmission. In these masks, the membrane material itself acts as the scattering layer. As used in the SCALPEL systems, the second type of EPL mask, the continuous-membrane mask, has a thin layer, usually made of silicon nitride, that is mostly transparent to high-energy electrons. The imaging layer consists of a high-atomic-number material, such as tungsten or tantalum, which scatters the electrons.

EPL masks are 4x reduction reticles that require a 200-mm mask format with multiple membranes to contain the entire printed die size. However, many early EPL mask demonstrations at the MCoC have been conducted on 100-mm masks because of the availability of tools optimized for this size. EPL masks have a membrane support structure defined into the substrate to support the less rigid membranes, which are ~1 mm wide; ~0.2-mm struts separate the membranes. Because the E-beam can be rapidly deflected, these membrane fields can be easily stitched together on the wafer by the E-beam stepper. However, this process requires a stitching strategy on the membrane boundaries that involves subresolution seam-blending.

The EPL mask blank is fabricated in a high-rate silicon etch process to form the strut-supported membranes. Although this is being done with a wet chemical etch, the trapezoidal strut profile that is formed does not provide sufficient packing density for the membranes to support the die sizes needed for manufacturing. Production mask blanks will be fabricated with either reactive ion etch (RIE) or Si(110) wafers to achieve adequate rectangular strut profiles.

EPL stencil masks have 2-µm-thick membranes, which are similar to those of x-ray masks. However, EPL stencil mask membranes are normally made of silicon, and the imaging-layer features are etched completely through the membrane itself. SCALPEL masks, on the other hand, have much thinner membranes (typically ~100 nm) that are made of silicon nitride. SCALPEL masks have a very thin scatterer layer, which is made of a heavy metal such as tantalum silicon or tungsten/chromium.

EUVL Masks. The EUVL system uses reflective optics and masks in which the image is formed in an absorbing metal. Because EUVL masks have a multilayer reflector, they do not require membranes. However, patterning the absorber layer of EUVL masks poses a similar challenge to patterning EPL-mask imaging layers. In addition, accurate control of the chemically amplified resist process, E-beam writing, and reactive ion etching of the layers are accomplished in a similar fashion as on EPL masks. EUVL masks typically consist of an ~50–100-nm-thick absorber composed of titanium nitride, chromium, or tantalum nitride, which is deposited on top of a stop layer for both the absorber etch and absorber repair steps. This stop layer is usually a buried silicon oxide that is ~50–100 nm thick. A multilayer reflector of alternating ultrathin molybdenum and silicon layers lies beneath the oxide. To prevent the intermixing and silicidation of the multilayers, EUVL masks have thermal processing limits requiring that the mask temperature remain <150°C. Like EPL masks, EUVL uses 4x reduction masks fabricated on 200-mm round substrates. However, future masks will be fabricated on 6-in.-square substrates of ultralow-expansion glass.

Advancing Process Technology

The MCoC plans to move NGL mask manufacturing into pilot production when needed to support lithography development programs. The first step in this plan is the development of functional and practical processes for mask manufacturing that meet the specifications of the ITRS. This plan includes a broad strategy of applying materials and process learning to all mask formats, as well as exploring process and materials options for mask fabrication. The MCoC has developed the manufacturing technology to produce all of the major NGL mask formats. Despite their different formats and absorber aspect ratios, as shown in Figure 2, NGL masks with identical patterns can be fabricated and compared on a common basis.

Figure 2: SEM images of masks fabricated at the MCoC with widely varying aspect ratios: (a) x-ray mask; (b) stencil mask; (c) EUVL mask; (d) scattering mask.

Critical to the success of NGL technologies is the production of masks that guarantee reasonable yields and help to control costs. Improving control over critical dimension (CD), image placement, and defects is essential to producing high-yield masks. In order to determine the factors that contribute to image-size uniformity and image-placement accuracy, the MCoC has conducted extensive partitioning and optimization experiments.

Critical-Dimension Uniformity. CD uniformity will continue to be a key quality specification for NGL masks. Although the wafer lithography mask error factor (MEF) is expected to return to near unity for NGL options because of the significant reduction in exposure wavelength, CD uniformity requirements will become vanishingly small (7 nm for the 70-nm node). Consequently, significant process, materials, and metrology improvements will be required.

The MCoC has partitioned image-size uniformity into controlling parameters for all NGL masks by adopting a first-principles approach. Image size generally is controlled by E-beam profiling, delivered dose, resist activation, pattern transfer, and metrology. Each of these techniques can be partitioned into physical elements. For example, delivered dose is based both on tool control (shot time, beam current, etc.) and proximity effects.

When chemically amplified resists are used, another large contributor to image size is resist baking. In such cases, two approaches have been explored to improve CD uniformity while maintaining the advantages of chemically amplified resists. First, temperature profiles have been modeled on NGL membrane masks in order to find ways to minimize the large temperature gradients that these masks exhibit because of the different heating rates of the membrane and the wafer.⁴ Second, resists that are less sensitive to bake temperatures have been explored. Recent results with IBM KRS-XE resist suggest that this low-activation-energy resist can improve CD uniformities on all NGL masks, as demonstrated in Figure 3, which compares CD profiles on EPL stencil masks fabricated with two different resists.⁵

Figure 3: Comparison of CD profiles on EPL stencil masks: CD uniformity >20 nm (3x) on mask produced with a commercial chemically amplified positive-tone DUV resist (left), and CD uniformity of 9 nm (3x) on mask produced with IBM KRS-XE resist (right).

Etching is another step that affects CD uniformity. Because pattern transfer achieved by RIE influences image-size uniformity, this step has been optimized for x-ray mask production. In addition, etching for both stencil and continuous-membrane EPL masks is no more difficult than high-aspect-ratio etching for 1x x-ray masks, because stencil etch, although it involves 4x features on the mask, is also a high-aspect-ratio step. While the scatterer layer on all-membrane masks is very thin, thermal control during etch is needed to maintain image-size uniformity, and high selectivity to the underlying membrane is needed to maintain uniform transmission across the mask. EUVL mask absorber etch also requires high selectivity to the multilayer reflector, but more research on the absorber layer is needed before comprehensive etch development can take place.

Image-size uniformity must be approached differently for EPL masks than for other masks. The array of membrane subfields allows for image-size analysis both across the mask and within individual membranes. This analysis is necessary to partition CD variations into cross-mask and intramembrane components.

While long-range cross-mask variations are typically associated with the lack of process uniformity in such steps as baking, developing, and RIE, intramembrane variations may be caused by pattern density, membrane-related bake phenomena, or localized variations in resist thickness. Figure 4 shows the image-size distribution on a completed 100-mm SCALPEL mask. Detailed mapping across a single 12 x 1-mm membrane shows image-size uniformity of 11 nm (3x), while measuring sites in each membrane across the entire mask demonstrates image-size uniformity of 16 nm (3x).

Figure 4: Image-size uniformity on 100-mm all-membrane (SCALPEL) mask: detailed CD uniformity across a single 12 x 1-mm membrane (top), and CD uniformity across the mask (bottom).

The two industry proponents of NGL technology, Agere Systems (Austin, TX) and Nikon Precision (Belmont, CA), offer different format and process options for EPL masks. The MCoC has fabricated stencil and SCALPEL masks in the nominal PREVAIL and SCALPEL formats. Both EPL mask types operate in fundamentally the same way in that the mask serves as a scatterer layer for the electron beam, and both types are, in principle, capable of using either mask format. Similarly, the membrane layouts of the small-square PREVAIL format (~1 x 1 mm) and the rectangular SCALPEL format (~12 x 1 mm) are not fundamentally different; they merely result from different stepper-specific writing strategies.

To help provide data for possible EPL mask standardization, the MCoC has successfully fabricated masks with various combinations of membrane layouts and mask types. In addition to demonstrating the feasibility of producing masks in either format, mask-characteristic data confirm that mask format differences do not fundamentally limit the standardization of EPL mask formats. Figure 5 compares the image-size uniformity between EPL stencil masks fabricated with 1 x 1-mm membranes and masks fabricated with 12 x 1-mm membranes. In this example, the CD uniformity of the 12 x 1-mm membrane layout is slightly worse than that of the 1 x 1-mm membrane because of the larger mask area covered, but CD is not affected by membrane size.

Figure 5: Image-size uniformity on EPL stencil masks fabricated with two different membrane layouts: 1 x 1-mm membrane (left), and 12 x 1-mm membrane (right).

Image Placement. Another critical specification of NGL masks, particularly of membrane-based structures, is image-placement error. Images placed on or etched through a thin membrane are distorted as a result of the magnitude and uniformity of the layer's latent stress; external out-of-plane distortion (OPD) caused by chucking in the mask write, metrology, and lithography tools; and the spatial characteristics of the image pattern when it is transferred into the stencil or image layer.

X-ray learning has developed successful techniques for controlling stress through deposition and annealing processes, as well as product-specific emulations (PSEs), in which empirical metrology data from test masks are used to create a placement correction calibration (predistortion) for the write tool. The PSE technique has achieved <20-nm (3x) image placement on production x-ray mask membranes, although additional mask writes are required.

For silicon wafer–based EPL masks in particular, the fixture that holds the mask in mask-manufacturing equipment such as E-beam writers, metrology tools, inspection tools, or lithography tools themselves defines the mask's OPD and, hence, its image-placement accuracy. To meet the required overlay and placement specifications, it is critical that all EPL-mask-related tools have identical mechanical interfaces, requiring that mask and lithography tool vendors standardize their products. Although mask-material stress magnitude and uniformity can be controlled, etching the desired pattern in the imaging layer causes in-plane distortions (IPDs) from the material stress gradients.

Stress-control and stability measures have been developed for the amorphous tantalum silicon film of x-ray masks, which has aided the development of low-stress materials for EPL and EUVL masks. However, the different substrates, starting materials, and film thicknesses of the different types of masks require that further research be conducted. In addition, modeling results have demonstrated that a mask's pattern type, pattern density, and pattern density gradients appear to be large contributors to process-induced IPD.^6,7 In collaboration with the University of Wisconsin, these modeling results are being compared with experimental data from the MCoC for EPL masks. Both modeling and preliminary experimental results indicate that pattern density gradients can contribute substantially to IPD. Results of this work, funded by International Sematech, will be published in the near future.

Although NGL mask materials are fairly well established, the MCoC is conducting exploratory studies on new materials that may provide process and yield benefits. Extensive material evaluations have been conducted for x-ray masks, which established important guidelines for stress control, radiation damage, cleaning durability, low defect deposition, and RIE compatibility. Uniform, controllable, and stable stress has been achieved for a tantalum silicon absorber by controlling deposition and metal-annealing processes. Radiation damage of such membrane materials as diamond has been found to vary significantly with different deposition parameters, leading researchers to experiment with EPL stencil masks containing diamond membranes.

Defect Control. Manufacturing defect-free NGL masks will be challenging. The MCoC has fabricated defect-free x-ray masks using a SEMSpec inspection tool from KLA-Tencor (San Jose) for die-to-die E-beam imaging inspection and a focused ion-beam tool with ion milling and gold deposition capabilities. The SEMSpec is nominally capable of detecting 70-nm defects.

In addition, Photronics at its Allen, TX, manufacturing plant has worked with Agere Systems on basic SCALPEL-process defect-reduction learning that will be continued and extended to all NGL technologies at the MCoC. The major thrust of this work will be to isolate such defect sources as mask blanks and materials, the lithography process (resist, E-beam tools, and development), pattern transfer, and cleaning processes. Laser-scattering inspection tools and the SEMSpec perform in-process inspection procedures to determine and minimize the root causes of defects and minimize mask defects.

For the 70-nm lithography generation, a defect-detection sensitivity of 55 nm is required. Although the SEMSpec is almost capable of this sensitivity, it cannot function as a manufacturing inspection tool because of its inadequate throughput. Unfortunately, such a manufacturing tool does not yet exist. The MCoC is using the SEMSpec for low-defectivity process development, material evaluations, and initial prototype mask builds. While this defect-inspection system cannot function in a manufacturing setting to produce defect-free masks, it has aided in the development of inherently low-defect-level processes. KLA-Tencor and a consortium of partners, including Photronics, are developing production-capable inspection tools to meet both the size sensitivity and throughput requirements of NGL mask fabrication. The micrographs in Figure 6 are from programmed-defect NGL masks supplied by the MCoC to aid in the development of new manufacturing tools.

Figure 6: Micrographs from programmed-defect NGL masks used to aid in the development of new manufacturing tools: (a) wiring corner pinhole, (b) wiring opaque pindot, (c) contact clear extension.

EPL membrane masks must undergo inspections that are sensitive to E-beam–printable defects (scattering). If optical inspection is performed, the back side of the membrane must be inspected as well. The inspection of EPL stencil masks must be sensitive to defects at any level through the membrane, and clear defect repair requires a bridging process to span the defect with sufficient electron opacity at lithographic tool acceleration voltage.

EUVL masks present other challenges; they require an inspection system that generates sufficient contrast between the absorber metal and both the silicon dioxide and the EUV reflective multilayer stack when a wavelength other than EUV is used. If EUV inspection prequalifies substrates as defect-free, engineers must ensure that no new reflective defects are created during subsequent repair steps.

Conclusion

To meet the challenges of NGL mask fabrication, the MCoC is implementing cross-cutting process technologies. The center is the only facility that has produced masks for all NGL options. The improvement of mask parametrics, such as CD and image placement, is occurring through error-source partitioning; early results indicate that similar image-size uniformity and image placement can be achieved on different mask types. Defect detection and reduction are being targeted through internal efforts and collaboration with inspection tool vendors. Meeting aggressive NGL mask specifications and schedules will require continued emphasis on the development of technologies for all NGL mask applications.

Acknowledgments

The authors would like to acknowledge Kevin Collins, Monica Barrett, Michael Trybendis, Chris Magg, Mark Lawliss, Neal Caldwell, Ray Jeffer, Dave Muzzy, Ken Racette, Carey Williams, and Louis Kindt, all of IBM, for their contributions to this article and to the work at the MCoC. They also wish to thank D. P. Mathur of Photronics.

References

1. The International Technology Roadmap for Semiconductors, Tables 41A and 41B (San Jose: Semiconductor Industry Association, 1999); available from Internet: http://public.itrs.
   net/Files/2000UpdateFinal/Lithography2000final.pdf.
2. D Walker, "Current Status of NGL Masks," in Proceedings of SPIE 4066, Photomask and Next-Generation Lithography Mask Technology VII (Bellingham, WA: SPIE, 2000), 94–104.
3. M Lercel et al., "Next-Generation Lithography Mask Development at the NGL-MCoC," in Proceedings of SPIE 4066, Photomask and Next-Generation Lithography Mask Technology VII (Bellingham, WA: SPIE, 2000), 105–115.
4. M Lercel and C Magg, "Image Size Control in Next-Generation Lithography Masks," in Proceedings of SPIE 3997, Emerging Lithographic Technologies IV (Bellingham, WA: SPIE, 2000), 266–275.
5. C Magg et al., "Evaluation of an Advanced Chemically Amplified Resist for Next-Generation Lithography Mask Fabrication," in Proceedings of SPIE Photomask Symposium 2000 (Bellingham, WA: SPIE, to be published).
6. GA Frisque, EG Lovell, and RL Engelstad, "Pattern Transfer Distortions in IPL and EPL Masks with Pattern Density Gradients," in Proceedings of SPIE 3997, Emerging Lithographic Technologies IV (Bellingham, WA: SPIE, 2000), 568–577.
7. P Reu et al., "Modeling Mask Fabrication and Pattern Transfer Distortions for EPL Stencil Mask," in Proceedings of Micro and Nano Engineering 2000 (to be published).

Michael Lercel, PhD, works for IBM at Photronics' Mask Center of Competency in Burlington, VT. His experience at IBM includes work on selective dielectric etching for advanced CMOS technologies, and x-ray and NGL mask development (RIE and process integration). Lercel is a member of the American Vacuum Society and SPIE. He has published widely and holds several patents in the microelectronics and lithography fields. He has a BS in physics from MIT in Cambridge, MA, and a PhD in physics from Cornell University in Ithaca, NY. (Lercel can be reached at 802/769-8474 or lercelm@us.ibm.com.)

David Walker is general manager, next-generation lithography, at the MCoC. He has been involved in electron-beam mask making since 1986, working for RCA, Harris Semiconductor, Lepton, and, since 1995, Photronics. His initial field of work included fabrication of novel electron optical components, including pioneering development of silicon membrane structures for electron optical lenses and calibration plates. Walker has published extensively and holds numerous patents in the areas of silicon microstructures and electron-beam lithography systems. He received his BSEE from MIT (Cambridge, MA) in 1976. (Walker can be contacted at 802/769-1885 or dwalker@ngl.photronics.com)

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