Favored SCALPEL's continued progress
07/01/1999
Figure 1. Basic SCALPEL principle of operation showing contrast generation by differentiating more- or les-scattered electrons |
Scattering with angular limitation projection electron-beam lithography promises significant cost benefits over optical lithography for features down to 35nm. This technology was recommended by International Sematech`s next-generation lithography workshop as one of the two technologies to receive continued development funding in 1999. Today, the industry has made significant moves toward commercializing it.
Bell Labs and its partners are working through scattering with angular limitation projection electron-beam lithography`s (SCALPEL`s) proof-of-lithography phase. We are targeting the major engineering issues that affect extensibility, cost of ownership, and commercialization. Beyond this phase, we anticipate that development of a high-throughput tool (SCALPEL-HT) will begin shortly under the aegis of a commercial partnership with several IC companies and a major tool vendor.
Ongoing development has focused on simultaneous evolution of the three components of SCALPEL: the exposure tool, the mask, and the resist and process. Work on the exposure tool includes techniques for stitching the small parts of the pattern exposed at one time into larger features. The goal has been to demonstrate SCALPEL`s viability for production feature sizes from 130nm down through 35nm. We have also begun to develop detailed error budgets, which, together with comprehensive system modeling, drive the design effort.
Market entry costs should be minimal, in part, because the mask, resist, and processing technologies are similar to those used in optical lithography and should follow the same incremental development path. The SCALPEL mask represents a straightforward extension of current binary optical masks and can be manufactured with essentially the same tool set or its probable descendants. Because of the linear nature of printing with SCALPEL, mask technologies such as optical proximity correction and phase shifting are not required. The SCALPEL mask will show a strong mask cost advantage over advanced optical techniques. SCALPEL also uses the same types of single-layer chemically amplified (CA) resists as DUV lithography.
We believe the technology can be extended over several generations, while maintaining an economically viable throughput with only evolutionary advances in electron optics, stages, alignment systems, and processes of the same magnitude that have occurred in optical lithography.
The choice of the next-generation lithography (NGL) technology is likely to be based on the cost of printing a lithographic pattern on a wafer rather than on strictly technical ground, since there are several alternatives that can achieve similar results at different degrees of difficulty and cost. Considerations include:
- the cost of operating the exposure tool, which is proportional to its price divided by its throughput;
- the mask cost, which is the price of the mask divided by the number of wafers to be printed with it; and
- the cost of resist materials and development of the image.
Comparing estimated costs of SCALPEL to advanced optical lithography, the throughput of an optical tool may be as much as twice that of a SCALPEL tool, but is likely to cost twice as much, making operational costs roughly equal. Resist and processing costs slightly favor SCALPEL, since some complexities such as antireflective coatings under the resist are not needed. The most significant difference is in mask costs: SCALPEL operates in a linear printing regime and therefore uses a true 4:1 representation of the circuit pattern on the mask. In subwavelength optical lithography, the masks must be much more complex to compensate for diffraction effects in printing. We believe that mask cost will be the dominant factor in overall costs; thus, SCALPEL technology will have a significant advantage over optical lithography in the subwavelength regime.
Technology progress
There has been significant progress in important areas during the past year. For example, at Bell Labs we have demonstrated a successful writing strategy with stripe stitching, continued commercialization efforts on the mask, evaluated promising resist formulations, and begun work on a SCALPEL-HT system design and large-field mask formats.
SCALPEL mask technology was the first key breakthrough in making projection electron-beam lithography feasible, solving one of its two key problems. The common absorbing stencil mask is susceptible to heating, thus limiting the accelerating voltage that can be employed. In addition, it cannot support patterns containing closed curves unless complementary mask pairs are used.
The SCALPEL mask concept eliminates heating problems with stencil masks. The mask is a low-atomic-number material membrane covered with a layer of a high-atomic-number material; the pattern is delineated in the latter. The mask is almost completely electron-transparent at the energies used (100keV); contrast comes from the membrane, which scatters electrons weakly and to small angles, while the pattern layer scatters them strongly and to high angles. An aperture at the back-focal plane of the projection optics blocks the strongly scattered electrons, forming a high-contrast aerial image at the wafer plane (Fig. 1). The functions of contrast generation and energy absorption are thus separated between the mask and the aperture, so very little incident energy is absorbed by the mask, minimizing thermal instabilities.
Since thin-membrane masks are not sufficiently stable over large areas to meet error budget requirements without a supporting structure, a "grillage" of struts supports the mask. This design is compatible with the need to move away from full-field optics: SCALPEL optics do not scale with increasing die sizes and decreasing minimum feature sizes. They require the use of such small numerical apertures to control optical aberrations that, at the beam currents necessary for acceptable throughput, space-charge effects can limit resolution. The small electron optical field - 1mm ? 1mm at the mask - is consistent with the strutted mask design and step-and-scan writing strategy. The electron optical field is the same width as the patterned area between the mask struts.
Our original mask blank work was done using a 100mm-dia. format where wet etching formed the grillage. To achieve chip sizes consistent with the SIA roadmap, a 200mm format is required. The wet etching results in a trapezoidal cross section for the support struts and an inefficient use of mask real estate. In going to the larger 200mm mask format for larger chip sizes, we will also make the step to dry-etched grillage structures with nominally vertical profiles for more efficient use of mask area for circuit patterns. Use of dry etch means the construction of the grillage is easier: a dry etch requires no particular alignment on the backside pattern, so struts with an essentially rectangular cross section can be used. We are also re-evaluating materials choices for process compatibility and simplification.
Imaging and writing procedures
Figure 2. The results of a) butting and b) blending when maintaining CD uniformity over a stripe seam region. |
In the SCALPEL exposure tool, the mask is uniformly illuminated by a parallel beam of 100keV electrons over a 1mm x 1mm area. A reduction-projection optic, in a telecentric doublet arrangement, produces a 4:1 demagnified image of the mask at the wafer plane. Because the features being printed are much larger than the 3.7pm wavelength of the radiation used, the full benefits of the reduction ratio are realized, especially in terms of the mask; imaging is aberration limited, not diffraction limited as it is with conventional optical lithography.
The illumination system is incoherent, so there are no interference effects. This, combined with the absence of diffraction effects and the high ultimate resolution (~35nm) means that our current tool design will operate almost linearly down to feature sizes of at least 70nm. If the k1 factor is reduced to extend the minimum feature size at a given tool resolution, mask feature biasing is simple and effective, again because of the simple, incoherent nature of the imaging process.
The SCALPEL step-and-scan approach confers advantages other than a simplified optical design. The die size that may be printed is not limited by electron optics, but only by available mask size and stage travel. This is different from optical step-and-scan systems, where optics must be large enough to illuminate a slit the width of the entire die.
Another advantage, particularly for potential mix-and-match applications, is that achieving good overlay is easier; since the image is assembled from many small pieces, magnification errors or trapezoidal distortion errors on previous levels can be matched exactly. Because of the way the image is assembled, we must consider the difference between random and systematic errors in placement of image pieces. SCALPEL is simply not the same as step-and-repeat systems in this respect, and we include this difference in our error budget analyses.
The device pattern is segmented on the mask in two dimensions by the struts and must be reassembled, or stitched, to form a continuous image on the wafer. It is crucial that the CD of any feature crossing a stitching boundary is maintained to within tolerances specified by the error budget. The ease with which this is accomplished depends on the technique of joining the two parts of the feature.
Figure 2 shows the results for two different approaches when there is an error in the placement of the two parts. The two image parts can be simply butted together or blended. Blending is achieved by tapering the dose profiles at the edges of the blended regions. This yields minimal variation in intensity distribution and almost no change in CD. Seam blending decouples the effects of dose and position errors, as well as making objects crossing seams much less sensitive to errors in the relative positions of the two stripes.
System components
Shortly after it was constructed two years ago, we used the SCALPEL proof-of-lithography (SPOL) system to print a 1cm x 1cm field comprising 200 stitches; this was an important program milestone. Since then, we have improved the control system to print large arrays over an entire wafer and have extended our analysis of the results to improve system calibration and understand image placement repeatability of the current tool. The SPOL tool is still being used for experimental data and can now expose 200mm wafers.
Source
The ultimate source requirements are driven by error budget analysis. The determining factor is the interaction of dose and stitching errors at a stripe boundary; the analogous case in optical lithography would be the interaction of dose and focus errors.
It is relatively straightforward to design a source that performs well. It is also important since improvements in the source provide increased latitude in the required stitching accuracy and minimize dose-error-related CD variations. We are therefore investing considerable effort to develop the optimum design.
Previous analysis has indicated that the existing 100keV electron source used with our early tools may not meet the specifications needed for a SCALPEL-HT system. For example, we have replaced the original large-area LaB6 cathode with a commercially available Ta disc emitter and have modified the tetrode elements. The first change substantially improved stability, lifetime, and uniformity of the beam; the second permits better emittance matching. At present, however, even with these improvements, the electron-beam uniformity remains unacceptable for use in a SCALPEL-HT system.
A one-year program has been established to modify the source design to limit the amount of cathode detail transferred into the illuminating beam, and to identify commercially available cathodes that have highly uniform emission properties across their entire surface over the anticipated full range of normal operating conditions.
Optics
The electron optics consist primarily of magnetic lenses to focus electrons, along with deflectors or correctors that are magnetic or electrostatic depending on speed and stability requirements. Because of difficulties in correcting aberrations in large magnetic lenses, we have chosen to use compact, inexpensive lenses that illuminate a 1mm x 1mm portion of the mask at a time. In the current system, the complete pattern is exposed by mechanically scanning the mask and wafer synchronously through the illumination. In the next system, electronic scanning of the beam will be implemented to supplement the mechanical motion.
We do not anticipate major modifications to the projector lens design for SCALPEL-HT; principal changes will be the addition of large-throw deflectors (±6mm at the mask, ±1.5mm at the wafer or more). Since large-scale deflectors can result in aberrations, correction elements will also be added. Dynamically corrected deflection fields several millimeters at the wafer have been achieved in direct-write, cell projection, and shaped-beam systems.
Space-charge effects
We made an initial attempt at measuring the space-charge effect last year. Although inconclusive in terms of determining the functional dependence of blur on beam control, these preliminary results did indicate that the magnitude of the effect was no larger than theory predicted, suggesting we had a good understanding of the dynamics of the effect. It appears that when models are applied appropriately and scaling factors taken into account so valid comparisons can be made, most of the predictions agree quite closely.
Throughput, extensibility
We have conducted detailed analyses of the factors affecting throughput in a SCALPEL system (Fig. 3) and have shown that it is intimately linked to resolution by space-charge effects common to charged-particle systems.
FIGURE 4. Projected throughput vs. CD extensibility for SCALPEL. |
Our initial analyses indicated that sufficient throughput could be achieved with an acceptable cost of ownership for the 130nm and 100nm device generations. However, it is important to understand how extensible the technology is. Resolution is not an issue. For our first-pass design, it is apparent that 50nm features can be readily printed (Fig. 4).
Our most recent analysis, which assumes gradual learning about optical design and resist processing, shows that the throughput can be maintained at acceptable levels down to 50nm. The model results shown in Fig. 4 also assume the availability of both positive and negative resists. In this way, the appropriate resist tone can be chosen for the pattern to be printed to minimize the total beam current requirement. For example, a critical gate level could be printed with negative resist using a minimal beam current density (i.e., <50%). The model shown in Fig. 4 assumes that the beam density will always be <50%. This is certainly true for critical gate and contact levels, and analysis of current mask set data indicates that all levels in a CMOS process, with the possible exception of the isolation level, will meet this criterion. Therefore, with modest improvements in system and process, SCALPEL can deliver throughputs of at least twenty-four 300mm wafers/hr down to the 50nm technology generation.
Stage technology
The step-and-scan writing strategy is similar to that used in the optical Micrascan system, except that the mask and wafer stage positions do not need to be controlled accurately, just known very accurately. Stage positions are monitored by laser interferometry, and any relative positional error between stages is corrected by an electrostatic image deflector, an option only available with charged-particle techniques.
Alignment, overlay
The image deflector is also the key component in the fine alignment of the system. (Coarse alignment is obtained with a finely focused probe produced by the mask illumination system.) Fine alignment is accomplished using the system with the standard flood illumination and scanning the image of a grating or similar mark over a corresponding mark on the wafer. The signal generated is at its maximum when the two marks are aligned. We have recently begun investigating alignment mark types in more detail and are considering which alignment strategy is most effective.
The SCALPEL writing strategy is compatible with conventional mark acquisition strategies. It is possible to use global, intermediate global, and fine alignment marks. Topographic marks can also generate good contrast and are suitable. Optical alignment systems used with conventional steppers are also compatible with the SCALPEL tool, and might be appropriate in mix-and-match applications.
Wafer heating
Heating in the mask, optics, or wafer is becoming increasingly important for all types of lithography. For SCALPEL, although mask heating is not a concern because so little energy is deposited in it, the reverse is true of the wafer.
We have embarked on a very detailed investigation of wafer heating in SCALPEL, and regard it as one of our critical issues.
The effects of wafer heating fall into three basic categories:
- pattern placement effects occurring at the instant of exposure, which are determined by peak temperature rise;
- large-length scale, slowly varying placement errors, which are controlled by the gradual increase in the average wafer temperature; and
- intermediate time and length scale effects resulting from the interaction between the region being exposed and those regions immediately adjacent that have just been exposed.
|
If the dose of illumination is delivered in a single pass, the maximum temperature is higher and the temperature profile is strongly peaked around the beam location. When the dose is delivered in several passes, the temperature profile is much less strongly peaked. The number of scans required in the latter scenario does not affect throughput.
Our approach has been to design the wafer chuck so motion of the wafer without slippage is predictable and known, and can therefore be counteracted by control of the electron beam. This solution also depends on real-time positioning information provided by SCALPEL`s actinic through-the-lens back-scatter alignment sensor, which gives information on wafer-chuck reference anomalies, wafer thickness change in an iterative process, etc.
Although wafer heating is a complex problem, we believe we have identified an appropriate pathway to its solution, and a major next step will be devising the appropriate correction algorithms, along with refining the design of the wafer chuck to ensure that wafer expansion occurs as smoothly as possible.
Mask blanks, patterning
The Microelectronic Center of North Carolina (MCNC) is fabricating 100mm-dia. SCALPEL mask blanks commercially and work on 200mm blanks is progressing. Masks are being patterned by both DuPont and Pho tronics as part of a Sematech program to collect data on CD control and image placement for 180nm features.
|
Like any thin-membrane mask, the SCALPEL mask is subject to stress-induced pattern placement errors. However, its network of supporting struts helps significantly to reduce the magnitude of these placement errors by providing in-and-out-of-plane stiffness to the membrane. To understand how to minimize pattern placement errors, we are working with researchers under contract to Sematech, which is studying the stability of all NGL masks through a combination of finite element analysis and analytical modeling.
Resist, process development
To minimize the entry barriers to SCALPEL, we have endeavored to ensure that resists currently used with 248nm and 193nm DUV systems can also be used with SCALPEL. This is important if introduction through a mix-and-match strategy is considered. Therefore, the main questions are:
- Is the exposure-induced chemistry in commercially available CA DUV materials compatible with the operating conditions of SCALPEL?
- Do these CA resists have the necessary pro cess latitude at doses that do not compromise throughput?
The first issue is an swered if materials are available that are not affected by vacuum exposure and wafer heating during exposure. Vacuum exposure is generally benign for CA resists provided they do not outgas and affect the vacuum. Suitable materials are available. The second issue is an swered if good sensitivity and process latitude can be obtained simultaneously. To address this issue, we are working with resist companies TOK, Sumitomo, and Arch Chemicals, who have aggressive roadmaps (Table 1) for achieving sensitivities ~6µC/cm2 process dose at 100keV for SCALPEL processing (Fig. 5).
|
Conclusion - A number of critical issues in SCALPEL`s development are well on the way to being resolved for commercial readiness. These include stitching of images, mask manufacturing, system modeling of space-charge effects, alignment strategies, and cost-of-ownership and extensibility calculations. Also, we are putting into effect a commercialization plan (Table 2) covering all the basic areas of the technology.
|
The SCALPEL program at Bell Labs and its partners will support the development of the mask, resist, and exposure tool with a goal of a proof-of-concept demonstration of high-throughput operation with in the next three years. The promise of SCALPEL technology is illustrated with 80nm contact holes in 0.75?m-thick resist (Fig. 6).
Moreover, in January 1999, Applied Materials Inc. and ASM Lithography Holding NV announced their intention to cooperate in accelerating the development of SCALPEL; they will participate as members in our Advanced Development Agreement program to develop SCALPEL technology and commercialize the exposure tools in the future.
The overall goal is to have SCALPEL beta-level tools and mask technology available to semiconductor manufacturers by 2002. This will target volume manufacturing for the 100nm technology node and beyond.
Acknowledgments
Work on the SCALPEL program has been supported in part by DARPA and Sematech. SCALPEL is a trademark of Lucent Technologies. Micrascan is a registered trademark of SVG Lithography.
Authors
Lloyd Harriott received his PhD in physics from SUNY, Binghamton, NY. Harriott is head of the advanced lithography research department at Bell Labs, Lucent Technologies, Murray Hill, NJ; ph 908/582-4922, fax 908/582-2300, e-mail [email protected].
Warren Waskiewicz received his BS in biophysics from Penn State University and his MS in materials science from Stevens Tech. He is currently technical manager for SCALPEL exposure tool development at Bell Labs, Lucent Technologies.
Anthony Novembre received his MS in chemistry from Stevens Institute of Technology and his PhD in polymer chemistry from Polytechnic University of New York. Novembre is a technical manager at Bells Labs Advanced Lithography Research Department and is responsible for SCALPEL mask and resist processes.
J. Alexander Liddle received his BA and DPhil in materials science from the University of Oxford. Liddle is a technical manager at Bell Laboratories in the advanced lithography department and is currently responsible for SCALPEL system modeling and engineering.