Lithography: 0.18 µm and Beyond
Optical lithography must face many challenges if it is to be extended to 100 nm geometries.
Staff -- Semiconductor International, 2/1/1998
 |
At a Glance | | |
|
| ||
| |||
Track systems such as the Polaris must control air, water or temperature to within 0.1°C for CD control needed to achieve <= 0.18 µm geometries. (Source: FSI International)
Resolution
Inherent in decreasing linewidth is improving resolution. The place to start is with wavelength (l) reduction that may coincide with an increase in depth of focus (DOF) if the numerical aperture (NA) is reduced and DOF is proporational to l/NA2. The Raleigh equation shows a dramatic decrease in DOF with reductions in feature size. A shorter wavelength can partially compensate for this. The transition from g-line to i-line was driven by a 19% reduction in the wavelength. The industry is currently in the midst of realizing a 47% decrease in wavelength from i-line to DUV, and in the relatively near future, another 28% to 193 nm.NA is a second way to improve resolution. NA, the sine of the angle between the most oblique ray and the normal incident ray along the optical axis, has a maximum of 1. There are fundamental DOF limits that occur when NA is increased, in order to increase resolution. For example, at 0.7 NA, the DOF of a minimum resolution feature is roughly two wavelengths of light; therefore, at 193 nm, there is ~400 nm total DOF. Planarization technologies such as CMP eases the severity of such limitations.
Step-and-scan technology, pioneered by SVG Lithography (Wilton, Conn.), has been implemented in virtually all DUV lithography systems to effectively enhance NA optics. Unlike steppers, which require the chip to lie fully within the lens field, scanners use a narrow slit that is swept across the field. With this technique, a smaller lens can be used and better focus control can be achieved by following the topography during the scan. With sub-0.18 µm features there is less tolerance to lens aberrations; therefore, limiting aberration control to the area of the slit provides an additional advantage with scanning technology. Many believe that the limit for large field, high-throughput optics is between 0.7 and 0.8 NA.
The last way to improve resolution is with the k1 factor, a measure of the degree of difficulty in printing. For a given linewidth (w), k1 is the scaled linewidth, k1 = w x (NA/L). Over the past 20 years, the k1 factor has decreased somewhat linearly. The theoretical limit is about 0.25. Conventional imaging using simple binary masks and stepper optics with some standard NA and sigma is sufficient in the high k1 regime. For k1 <0.65, however, life gets complicated logistically, and designers and lithographers must work closely during process development. Here, OPC, PSM, improved resists and high NA equipment with off-axis illumination become important. Extending optical lithography to 100 nm is clearly in the low k1 realm.
Exposure tools
Imaging at 0.18 µm and below further extends the challenges of resolving feature sizes below the illumination wavelength. The drive for higher NA, the unavoidable reduction in DOF, resolution enhancment techniques and stringent CD and linewidth uniformity requirements are pushing lithography tool manufacturers toward greater optimization of their large-field, high-throughput imaging systems. Suppliers, including Nikon (Belmont, Calif.), are implementing programs to maximize imaging from the reduction optics and to minimize aberrations. At 248 nm, companies such as ASM Lithography (Tempe, Ariz.) will be improving projection lenses and optimizing off-axis illumination to image 0.18 µm features. NAs will be as high as 0.68 to 0.7, as in Canon's (Irving, Texas) FPA-5000ES2.
The sheer cost of new DUV scanners adds to the value of mix-and-match schemes. Mix-and-match for 0.18 µm products requires a wide-field stepper to match to a DUV step-and-scan for critical layers. By defining a larger field size of 44 x 26 mm, Ultratech's (San Jose, Calif.) Saturn family of steppers allows a 1:1 match with a typical step-and-scan field size of 33 x 26 mm and maintains a 2:1 field match to traditional i-line steppers. Improvements in mix-and-match schemes for these fine linewidths include i-line resolution capabilities down to 0.55 µm and mix-and-match overlay of 150 nm, mean +3s, noted Robbyn Culver, senior product manager, semiconductor systems at Ultratech Stepper.
Reticles
Driving optical lithography to 0.18 µm and beyond poses special challenges for the photomask manufacturer across the board in design and production. Mask writing technologies are being developed to increase resolution and uniformity. Etec (Hayward, Calif.), for example, has a multipass gray (MPG) writing strategy, introduced in the MEBES 4500S, which provides a >50% improvement in uniformity and a reduction in writing times by up to a factor of 4 and delivers a fourfold increase in the maximum electron dose. This electron-beam pattern generation tool can take advantage of dry etch resist technologies that have previously prosed a problem because of their low sensitivity to e-beam radiation.
To address the increasing number of available light sources from 365 nm to the anticipated 193 nm, pellicles -- the transparent covers used to keep photomasks contaminant-free -- must be engineered to be compatible with a given light source without significant degradation. Ideally, these products must be engineered for greater versatility to be cost effective. To address this issue, suppliers such as DuPont Photomasks (Round Rock, Texas) manufacture pellicles made from a patented Teflon AF material, designed to work with either i-line wavelengths or DUV.
Inspection and repair
With reduced feature size and enhancement techniques, mask-related errors have a greater impact on yield. At 0.18 µm, the target CD error is 12 nm. Current production linewidth measurement tools have a 3s precision of 6 nm, and CD errors previously too small to print on semiconductor wafers are undetectable by current inspection equipment. A PSM with a small 100 nm pinhole in the open field, for example, is inconsequential even if it prints. A defect between a subresolution assist bar, OPC serif or near a gate, however, will modify the CD. Incorporating technologies such as atomic force microscopy (AFM) into a reticle production environment is one way to boost precision (Fig. 1), according to Andy Zanzal, director of marketing at Photronics (Brookfield, Conn.).
Fig. 1. An AFM image of a quartz etched Levenson-type DUV phase shift mask has three different phases (stair step surfaces). Two higher thin lines are chrome lines. (Source: Photronics)
Mask manufacturers typically inspect masks and repair them. Suppliers such as Applied Materials (Santa Clara, Calif.) and KLA/Tencor (San Jose, Calif.) are developing detection systems that promise 100% capture rates of 150 nm defects. These tools have been shown to capture at these dimensions on chrome on quartz reticles. Still in development are detection capabilities on embedded molysilicide phase shifter coatings and in detecting phase defects. To do this, at-wavelength inspection tools become critical, noted Wolf Staud, reticle inspection tool marketing manager at Applied Materials. I-line inspection tools with higher sensitivity and speed are planned for mid-year release, and DUV tools will follow in 1999.
With the finer design rules and the addition of enhancements, the repair process itself may cause more damage than the original defect, noted Ed Grady, vice president and general manager at KLA-Tencor. One solution being considered is to analyze each reticle defect and run simulations identical to printing conditions. In this way, a successful reticle repair of each defect can be designed.
To overcome the damage often caused by laser repair of reticles with fine, complex features, the photomask industry is turning toward focused ion beam (FIB) technology, Zanzal said. FIB tools can effectively be used to reconstruct complex images by adding or removing masking material to reticles that are unrepairable with current laser technology. By eliminating the need to rewrite, cycle times are reduced. A concern remains among reticle manufacturers that FIB removal of masking material may result in some transmission loss that could surface as printable defects. Further study in this area is ongoing.
Challenges of 193 nm
The many challenges facing 193 nm technology are being addressed internationally. SEMATECH and ASET are playing important roles in coordinating and moving 193 nm lithography forward. Studies are under way to investigate new resist materials and subsequent etch and process integration. One obstacle has been in the availability of necessary tools for running experiments, but this is changing. A 193 nm stepper, ISI Lithography's (Tewksbury, Mass.) ArF MicroStep, is currently used for research and resists, such as one supplied by Olin (Norwalk, Conn.), which will be released for laboratory testing this quarter.
Another area of concern is the lens material that was changed from glass to fused silica with the arrival of DUV lithography tools. While this will continue for 193 nm imaging tools, there will likely be some limited use of calcium fluoride in sections of high optical power, particularly at the illuminator. Currently, fused silica lens systems can be built for limited throughput pilot line tools; learning is still needed for full-production tools with the necessary throughput and lifetime. One possible problem is the potential need to replace a $1.5 million lens after three years' use.
Beyond 193 nm
What is the likelihood of optical lithography below 193 nm? Seeking to extend optical lithography, labs such as IBM Semiconductor Research and Development Center (SRDC, East Fishkill, N.Y.) have been investigating even shorter optical sources such as argon fluoride and fluorine laser lines and Ar2 lamps. The limiting resolution for various wavelengths is shown in Table 11 using a k1 value of 0.4 and a NA of 0.7. Technically, g-line lithography can be used to produce quarter-micron lines. Given a choice, a shorter wavelength is preferable.
Table 1. Wavelengths for Optical Lithography
| l nm |
Dl/l % |
Wmin nm |
DOF nm |
|
| G-line | 436 | - | 311 | 850 |
| I-line | 365 | 19 | 260 | 730 |
| KrF | 248 | 47 | 175 | 500 |
| ArF | 193 | 28 | 140 | 400 |
| F2 | 157 | 23 | 112 | 320 |
| Ar2 | 126 | 25 | 90 | 257 |
Researchers at MIT are investigating the chlorine dimer source, an excimer source at 157 nm. There is, however, a severe optical material problem. Fused silica for the lens system is essentially useless at this wavelength, requiring calcium fluoride instead. One desirable quality of fused silica is a small thermal expansion coefficient of half a part per million per degree centigrade. Thus, small thermal fluctuations are tolerable when writing the mask and using it inside the stepper. If, however, a 157 nm stepper were designed and a transmissive mask with a calcium fluoride substrate used, a 40X increase in the thermal expansion coefficient would create thermal stability problems. Other issues related to resist processing arise because of the need for a dry, oxygen-free environment.
One 126 nm Argon lamp source is even more speculative. Currently, it needs to be ~100 times stronger to be useful and the materials require all reflective optics, which have challenges of their own. At this point, a non-optical approach may be more likely to meet the SIA Roadmap for 100 nm in the 2003 time frame.
Enhancements
For a wavelength to be useful at a smaller linewidth, OPC and PSM technologies are needed. The technologies are currently used in IC manufacturing in a limited capacity. As device features shrink, the very ends of device feature lines can be very different from in the middle. By placing serifs at the ends of the lines as in OPC or eliminating interference effects along an edge with PSM, geometries that are smaller than the wavelength of the source illuminator can be accurately imaged.
To achieve 100 nm resolution with optical lithography, there is also a need for improved mask technologies. Normally, a 40 nm linewidth variation on a 4X reticle translates to a 10 nm linewidth change on a wafer. This mask error factor, the differential between simple optics and the printed feature, becomes nonlinear at very low k1, noted Dr. Tim Brunner, manager of advanced lithography and metrology at IBM SRDC. If the mask error factor is 2, for example, then a 40 nm reticle linewidth will cause a 20 nm linewidth change on the wafer. Table 2 indicates the maximum feature size at which an enhancement becomes essential.
Table 2. Technology Threshold
| Technology | Enhancement | Enabler | ||||
| i-line | DUV | 193 nm | i-line | DUV | 193 nm | |
| Var. NA | 0.50 µm | 0.35 µm | 0.30 µm | 0.35 µm | 0.25 µm | 0.18 µm |
| Var. sigma | 0.50 µm | 0.35 µm | 0.30 µm | 0.35 µm | 0.25 µm | 0.18 µm |
| Off-axis illum | 0.35 µm | 0.25 µm | 0.18 µm | 0.30 µm | 0.20 µm | 0.15 µm |
| AR coating | 0.50 µm | * | * | 0.25 µm | * | * |
| CA resists | - | * | * | - | * | * |
| OPC | 0.35 µm | 0.25 µm | 0.18 µm | 0.30 µm | 0.20 µm | 0.15 µm |
| PSM | - | - | - | 0.25 µm | 0.18 µm | 0.13 µm |
* = Needed for all feature sizes.
Phase shift mask technology is generally applied to memory devices that have regular features. To increase its versatility, several groups have been working on automatically generating the phase shift layer to address the irregular patterns of logic devices. IMEC (Leuven, Belgium), for example, is currently evaluating software that uses two photomasks and two exposures: one a regular mask for larger features, and one with phase shifters for smaller lines. The impact on throughput is 1.5 times longer, however the advantage is that the PSMs become less critical because they contain only for the small features, noted Luc Van den Hove, director of advanced semiconductor processing at IMEC.
Currently, photomasks represent 1-1.2% of IC production costs. Because of the increasing costs of building and equipping next-generation wafer fabs, the percent cost of adding OPC and PSM reticles to mask sets is expected to only increase slightly to the 1.5-2% range.
Simulation
What is the best type of PSM needed to obtain the best DOF? How should the design be tweaked? The challenges of low k1 lithography is the difficulty in choosing which enhancement technology to use and how to optimize the design. The process windows are expected to be smaller than they are today. Wafer flatness and process topography will continue to be improved, and sensitivity to dose and focus variations will increase. Process simulation in the low k1 regime is essential. For example, optimizing OPC in production is virtually impossible because of the number of iterations necessary to find the most effective design. By implementing simulation, customers can optimize their processes more quickly and accurately, on the order of days instead of months, noted Chris Mack, president and chief technical office at FINLE Technologies (Austin, Texas). Ultimately, this can result in better yields (Fig. 2).
Fig. 2. CD yields decrease rapidly when not at optimum conditions as simulated by PROLITH/3D and ProCD. (Source: FINLE)
At IBM2, using DUV for gigabit DRAMs required 10,000 simulations to optimize length and width control. DOF at 10% exposure latitude was used as a measure of the process window. Six different types of masks were studied, two bright chrome on glass, attenuating and alternating PSM; different optics including annular off-axis illumination; and the bias in the design, of both the length and width to try and find which printed the best with the highest process window. Finally, various resist processes of varying quality, both positive and negative were factored in. The simulations indicated that a significant increase in DOF could be gained by implementing an attenuating or alternating PSM using a negative resist process. The data confirmed the simulated results.
Proximity correction is already in use in many fabs. The final intended result is affected by various factors such as post-exposure bake, development characteristics, etch selectivity and underlying topography effects, said Dr. Victor Boksha, product marketing manager at Technology Modeling Associates (TMA, Sunnyvale, Calif.). Studies has indentified topography as one of the major influences during an OPC procedure. Joint efforts between TMA and Mosel Vitelic (Hsinchu, Taiwan) examined the rounding effects, which can cause electrical leakage, specifically in the highly non-planar Poly2 layer of a flash memory design (Fig. 3). Through simulation, an effective layout correction was determined and a manufacturable DOF was achieved.
Fig. 3. Top-view SEMs show the Poly2 layer following etch, uncorrected (l) and corrected based on Depict and Terrain simulations (r). (Source: TMA)
With each new generation of devices, the amount of critical information required to control a process more than doubles. On the equipment side, stringent overlay and CD budgets mean stepper-to-stepper and lot-to-lot materials and process variations will be so significant that fully automated feedback and feedforward control will no longer be an option, said Joe Pellegrini, president of NVS (Cambridge, Mass.). Real-time control will soon be a requirement for maintaining a profitable process. Equipment software, such as NVS* ARGUS Advanced Process Control System, becomes essential for minimizing overlay and CD variation particularly at fine geometries.
Limits of optical lithography
What are the likely limits to optical lithography? According to Brunner, the likely tool is a step-and-scan printer with a 193 nm light source, 0.7 NA and 0.40 k1 factor. Assuming a k1 of 0.4 can be achieved, then 110 nm CD with 220 nm pitch is possible.
In order for optical lithography to continue, lithographers will push designers to a more off-pitch feature, even for DRAMs (Fig. 4), according to Will Conley, member technical staff at Cypress Semiconductor, Technology Development Center (San Jose, Calif.). While current logic gates can have a 1:2 or 1:3 pitch, sub-100 nm gates in production may even be possible using optical lithography if the pitch is significantly larger than 220 nm.
Fig. 4. By increasing the pitch to 375 nm, a 150 nm linewidth can be achieved using 248 nm chrome-on-glass lithography. (Source: Cypress Semiconductor)
With the strengths of optical lithography from linewidth and overlay capabilities, proven productivity and mature mass production infrastructure, non-optical techniques must prove comparable. Whether EUV, ion beam projection, SCALPEL or X-ray proximity printing succeed optical lithography, productivity approaching that of optical and a mask technology infrastructure are potential roadblocks, according to Brunner. In each non-optical technology, a distinct type of mask is required, each needing excellent pattern capability, cost control and acceptable turnaround times. In addition, there are no pellicles currently available for protecting these masks from particles to prolong their usable life.
For now, enhancement in tools, resist, masks and device design keep optical lithography the industry's most viable imaging technology.
References
2. A.Wong, et. al. "Deep-UV lithographic approaches for 1 Gb DRAM", Symposium on VLSI Technology, Kyoto, Japan (June 1997).
Talkback
Related Content
Related Content
SPONSORED LINKS
