As the number of transistors on a chip continues to increase, the industry’s shrinking feature size is outstripping even the best efforts of optical engineers. Extreme ultraviolet lithography can lead to a more-than-tenfold decrease in wavelength, translating to a startling leap in performance.
Semiconductor lithography is the art and science of printing integrated-circuit chips. The central processing unit in a desktop computer was printed using a wide array of capital equipment in a semiconductor fab belonging to a company like Intel or IBM. For leading-edge circuits, the linchpin of manufacturing equipment is a photolithographic wafer scanner, which projects a circuit-pattern image to a silicon wafer.
Operating a scanner is a bit like using an old-fashioned photographic enlarger, with a 6-in. square glass slide (called a reticle) taking the place of the negative, a 12-in. photoresist-coated silicon wafer acting as the photographic paper, and a massive projection lens between them. The scanner projects a reduced image of the circuit pattern on the reticle using ultraviolet light and scans the wafer under the projection lens in order to print features on the wafer that are 65 nm or smaller in size.
The next generation of semiconductor microlithography may well be executed by a scanner with even more sophisticated optics—one that operates within the extreme ultraviolet (EUV) range.
Description of a wafer scanner
In the past, the standard for optical lithography was a machine called a wafer stepper, which holds the reticle still during exposure and moves the wafer in steps to produce multiple images on it. With a scanner, which is the current standard, the reticle and wafer are scanned in constant motion during exposure. This is done in order to use only a small portion of the imaging field of the lens, where lens aberrations are at a minimum.
An EUV scanner. A DUV scanner is similar except that a laser is used as the light source and the system uses mostly refractive optics.
Several companies, including Nikon, currently manufacture both wafer scanners and steppers operating in the deep ultraviolet (DUV) range. A variety of EUV “microsteppers” (steppers with a very small imaging field) and an alpha-version EUV scanner have been built. Finally, Nikon is building a beta-version EUV scanner, called EUV1, which will be delivered to its first customer in 2007.
Let’s look at the components that are common to both DUV and EUV tools. First, there is of course an ultraviolet light source. A DUV scanner has either a 248-nm KrF laser or a 193-nm ArF laser. EUV scanners use incoherent 13.5-nm light from either a xenon or a tin plasma produced by an electrical discharge or by a laser.
Next is an illuminator, which creates a uniform light distribution at the reticle plane and also constructs a well-defined partial-coherence pattern projected to the reticle. That is, it controls the angular distribution of light on the reticle. One could use 100 percent coherence, i.e., a point source with light incident from only one angle, or completely incoherent light, in which light is equally intense from all incident angles. Typically, neither of these extremes is used, so the illuminator uses partial coherence. The partial-coherence pattern is selected by the user in order to enhance printing of particular circuit features.
Next is the projection lens, which typically has a magnification factor of 1/4. For DUV lenses, a variety of all-refractive and catadioptric lens designs have been used, with fused-silica and calcium fluoride lens elements. In the EUV range, no transparent lens materials exist, so the projection optic is fully catoptric.
Finally, a scanner uses moving reticle and wafer stages controlled by interferometric systems with precision of a few nanometers. Additional systems are included to align the wafer and reticle and perform automatic focusing.
Sub-resolution imaging and the importance of EUV
To understand the importance of EUV lithography, it’s necessary to first look at the recent history of lithographic production. Over the past 20 years, the number of transistors on a single chip has grown exponentially, with a correspondingly rapid shrinkage in the size of the transistors and other printed features. In 1997, a characteristic feature size on a chip was 250 nm. The current state-of-the-art is below 65 nm, and features of a mere 32 nm are just around the corner. None of this would be possible without significant advances in optics.
Resolution scaling of all wafer steppers and scanners is summarized in this equation:
resolution R = k1(λ/NA) ,
where λ is the wavelength, NA is the numerical aperture of the lens, and the “k1 factor” describes the amount of enhancement to the basic resolution. A factor of k1 = 0.5 corresponds to the Rayleigh criterion of barely resolved imaging of nearby features. Wafer scanners now routinely operate below k1 = 0.5 and therefore use sub-resolution imaging. The wizardry by which this is done fills many books and articles. Over the years, enabling technologies have included extreme refinement of lens design, introduction of narrow-band lasers, and the development of high-contrast resists.
The equation shows that, to reduce the resolution of the printed features, optical engineers must increase the lens NA and/or reduce the optical wavelength. In addition, the lithographers who use the scanner must stretch the inherent resolution of the system and lower the k1 factor. The top table on the right provides typical values showing the progression in the industry.
The table summarizes the reasons for the introduction of EUV lithography. Lens NA has continually increased over the years, recently breaking the NA = 1.0 barrier with the introduction of immersion lenses. At the same time, wavelength has been decreased by changing the light source.
Comparison of images produced using EUV vs. ArF imaging. The top image is produced by 0.25-NA EUV optics, while the bottom was made with 1.30-NA ArF (193 nm) optics. Scales are in nm. The contrast in the EUV image is much higher. One could still make a printed image from the ArF optics, but it would require enhancements to the reticle and much more work to optimize the process.
Unfortunately, even these increases in NA and decreases in wavelength are not enough. The k1 factor shows that the industry’s shrinking feature size is outstripping even the best efforts of optical engineers. In other words, we are pushing the resolution limits of the tool harder and harder. This means that all the clever tricks that a lithographer has to play are becoming increasingly difficult to pull off.
Clearly the need is here for a significant change in the printing systems. This is the purpose of EUV.
An EUV lithography tool
The major change from DUV to EUV technology represents a more-than-tenfold decrease in wavelength. EUV lithography uses a group of spectral lines in the xenon atom (or a similar set in the tin atom) at 13.5 nm. Although the all-reflective design of an EUV projection optic means that the NA of the optics is reduced, the wavelength change yields a startling leap in performance, as shown in the second table on the right.
Not only does the system achieve lower resolution, but it does it with a k1 factor comparable to that of ten years ago. Since many of the resolution-enhancement schemes devised in the past few years would have to be reinvented for the new wavelength, this is a great advantage. The optical engineers have given the lithographers some breathing room!
We can see this by performing a litho simulation, as shown in the figure on the right. A 45 nm line end close to another 45 nm line is shown (transistor gates look something like this). The EUV system makes a clean image, while the ArF (193 nm) system produces a fuzzy one. To be sure, this is a somewhat misleading comparison, as the industry can and does make sharp developed resist images from 193 nm systems. However, it requires more work to do that. The point of EUV lithography is to offer extendibility into a new range.
EUV projection optics design
The EUV projection optic embodies several interesting design decisions that had to be made to meet challenges in NA, lens aberrations and scattered light or flare. As we have already stated, DUV lenses use both refractive and catadioptric designs. Current lenses have the illuminated part of the imaging field on axis, but this was not always the case. History has somewhat repeated itself in EUV projection optics.
In microlithography, pupil obscuration is a handicap for some partially coherent illumination and imaging conditions, especially when features are imaged close to the resolution limit. As a result, off-axis, unobscured mirror systems were investigated in the early days of EUV research at AT&T Bell Laboratories and Brookhaven National Laboratory.
Optical designs were required with wavefront aberrations on the order of thousandths of the illumination wavelength. At the same time, distortion over a relatively large field had to be fractions of a nanometer. These challenges led to a rediscovery of the benefits of step-and-scanning a narrow annular field across the reticle and wafer, with optical forms reminiscent of early DUV 1x all-reflecting concentric designs. In fact, one of the first laboratory prototype EUV projection systems at Bell Laboratories used a pair of mirrors from an early DUV Perkin-Elmer Micralign scanner, re-coated to operate at 13.4 nm.
Reduction variants of Offner’s three-mirror design lose 1 x symmetry before and after the aperture stop and, as a result, lose concentricity. Instead of being zero by symmetry, coma and distortion must be corrected by means of off-axis segments of rotationally symmetric aspheric mirrors. Early three-mirror 4 x reduction designs were investigated at Tropel Inc. and Sandia National Laboratory. However, it was the optical designers at Bell Laboratory who figured out that, at NA 0.1, four mirrors would be needed.
An even number is preferred to make the object and image planes separated and accessible for the scanning stages, and the extra mirror gives sufficient design freedom to arrange for the aperture stop to be located at a mirror; this is desirable so that all parts of the ring field see exactly the same pupil. Similar designs were proposed by Lawrence Livermore National Laboratory, which, together with Sandia, went on to build the first prototype scanning EUV projection tool, funded in part by several interested chip manufacturers.
Researchers at Lawrence Livermore developed new optical alignment methodologies and had considerable experience and expertise with both laser plasma light sources and EUV multilayer coatings. Close collaboration between them and Lawrence Berkeley National Lab also led to developments in at-wavelength metrology of mirrors and systems on a synchrotron beam line.
Unfortunately, during this work, due to continued improvement in DUV lithography, the target moved to higher lens NA. Of course, increasing NA leads to an increase in design complexity and size. With a relatively low mirror reflectivity of less than 70 percent, it would be highly desirable to avoid increasing the mirror count, but higher-NA four-mirror designs with sufficiently small aberrations have remained elusive.
However, 0.25-NA designs that have six mirrors and an aberrated intermediate image allow improved simultaneous correction of coma, distortion, astigmatism and oblique spherical aberration. Such designs have been reported since the late 1990s by Silicon Valley Group Inc. and Lawrence Livermore National Laboratory. In the current decade, all of the major lithographic tool manufacturers have been busy developing polishing and coating technologies for such designs to consistently meet the very demanding optical tolerances required to reduce aberrations and scattered light to acceptable levels.
The result of advanced surface polishing techniques. Low-, medium- and high-region spatial frequency roughness (LSFR, MSFR and HSFR) of a mirror in the Nikon EUV1 projection optics (pm=picometers). These contribute to aberrations, flare and reduced reflectivity, respectively. In addition, the smoothness of the power spectral density curve demonstrates high precision in all areas.
EUV tool construction
Now we turn to building some hardware. There are several experimental tools with printed features that are very challenging for DUV tools.
For the next stage, Nikon has developed a beta EUV scanner for imminent delivery to semiconductor manufacturers. It has these basic characteristics:
• A Xe-discharge EUV source;
• An illumination system that allows user-settable partial coherence in the reticle illumination;
• A six-mirror ring-field projection optic, with 0.25 lens NA;
• Aberration levels of less than 1 nm RMS (0.074 waves) and flare of 10 percent or less; and
• Full-field imaging with a scanned slit yielding 26 x 33 mm images on the wafer.
The aberration content, which is low in comparison to previous EUV tools, was made possible partially through the development of new high-precision test optics. It is now customary to evaluate the aberration content of lithographic lenses interferometrically several times in the construction process, and to make adjustments based on the results. EUV tool development required improved Fizeau-type interferometers (operating with visible light), where even the deformation of glass surfaces due to gravity had to be taken into account. The new interferometers have a repeatability of 0.032 nm RMS.
One might think that the all-reflective optics, with no chromatic aberration, could be evaluated without the use of 13.5 nm light. Due to the specialized coatings used, this is not completely true. Therefore, we used both visible-light interferometers and EUV interferometers, using synchrotron radiation, to check for chromatic effects in the coatings.
In addition to aberrations, scattered light or flare is a significant challenge for EUV optics, since its scaling with wavelength is faster than linear scaling (i.e., the problem is more than ten times worse). Flare is a deleterious effect because it reduces optical contrast. Flare in microlithography is measured in percent and is defined as the amount of light scattered into dark regions at a specified distance from a bright region.
Advanced polishing techniques are required to reduce flare, and these methods have been much refined in the past few years. For instance, a previous prototype Nikon optic, with two mirrors, had about 7 percent flare, and an alpha tool with a six-mirror projection optic has been estimated by its manufacturer to have 16 percent flare. Since flare increases with each mirror, 16 percent with six mirrors is better than 7 percent with two mirrors. Nikon expects EUV1 to have 10 percent flare with six mirrors—a significant improvement in the state of the art.
So far, we see that designing a catoptric lens to work at EUV wavelengths is possible, that we have light sources available at those wavelengths, and that the resulting image from such a lens is considerably better than that from a DUV lens. It all looks very good. Oddly enough, most of the challenges in EUV lithography are not from the optics, but rather from the light source at the beginning of the system and the photoresist at the end.
Stable EUV light sources exist, but operate at a power level about one-tenth of that needed for volume semiconductor production. Scaling up the power will require changing from the current generation of sources using a discharge in a xenon plasma to a new generation that uses a tin-vapor plasma. This results in exotic designs ranging from an electrical spark between rotating tin cylinders to vaporization of tin from discs rotating in a molten tin bath. Sources like these will need more engineering time before they are ready for production.
On the resist side, very few formulations exist now that can simultaneously resolve small printed features at the photospeeds needed. Indeed, because the printed features are so small, some have hypothesized that we are seeing image-degradation effects due to the length scales of the chemical reactions in the resist. Again, engineering work is progressing, but more time is needed.
In a sense, the EUV optics were the easy part of this challenge. Simply stated, we now have a wonderful camera lens, but are shooting in dim light with slow fine-grain film.
For some time now, EUV lithography has been waiting in the wings as the potential successor to DUV lithography. The benefits of using its short wavelength have long been obvious, but technology development for the optics and the associated infrastructure was quite challenging.
Optics technology has proceeded to the point where projection-optic construction is feasible for beta scanners, and we fully expect that little modification will be required for production scanners. Currently, the most important challenges are in the light source and in the photoresist. Once those are solved, EUV lithography will be ready for the next stage of development.
The authors are grateful for the considerable support for the research and development of EUV optics provided by Japanese government/industry consortia in the Extreme Ultraviolet Lithography System Development Association (EUVA) project via the New Energy and Industrial Technology Development Organization (NEDO).
[ Stephen P. Renwick is with Nikon Precision, Inc. in Belmont, Calif. David Williamson works for Nikon Research Corp. of America in Tucson, Ariz. Kazuaki Suzuki is with Nikon Corporation, Kumagaya, Saitama Prefecture, Japan. Katsuhiko Murakami is with Nikon Corporation, Sagamihara, Kanagawa Prefecture, Japan. ]