Photonics Spectra BioPhotonics Vision Spectra Photonics Showcase Photonics Buyers' Guide Photonics Handbook Photonics Dictionary Newsletters Bookstore
Latest News Latest Products Features All Things Photonics Podcast
Marketplace Supplier Search Product Search Career Center
Webinars Photonics Media Virtual Events Industry Events Calendar
White Papers Videos Contribute an Article Suggest a Webinar Submit a Press Release Subscribe Advertise Become a Member


How Low Can You (Optically) Go?

Hank Hogan, Contributing Editor

The $140 billion-a-year semiconductor industry depends on Moore’s law, the decades-old observation that device density doubles every 18 months or so. Today, excimer lasers power optical photolithographic tools that help drive this march toward smaller features. These tools routinely print millions of polymer resist lines that are hundreds of microns long and less than 0.2 μm wide.

The conventional wisdom in the industry indicates that fundamental changes loom in photolithography. The most common steppers use lasers operating at 248 nm. By the end of the decade, state-of-the-art tools are expected to use 13.4-nm sources. Dubbed extreme-ultraviolet (EUV) technology, this projection print approach applies UV radiation, but unlike its predecessors, it must operate in a vacuum and requires reflective, not refractive, components.


The semiconductor industry must reinvent everything to perform EUV lithography, from the light source to the optics to the photoresist. Although the industry is well on its way to developing EUV technology, the remaining challenges have engendered discussions about how far optical techniques can be extended. Courtesy of Sandia National Laboratories.


“EUV is more of a revolution from optical lithography as we know it today,” noted Tony Yen, director of lithography at International Sematech in Austin, Texas. Sematech is shooting for the first EUV application in 2007, but Yen said that it may not occur until 2008 or 2009.

Not everyone agrees that EUV, which some claim is closer to x-ray than to optical technology, will be deployed by then or that it is even necessary. Optical lithography advocates point to published research describing the printing of lines narrower than the wavelength of the source. This occurs to some extent every day in microchip manufacturing, but research suggests that the limits of subwavelength lithography could be extended. At Massachusetts Institute of Technology’s Lincoln Laboratories in Lexington, staff member Mike Fritze printed a 9-nm-wide line using a 248-nm source. By contrast, today’s most advanced commercial products have minimum feature sizes of 70 nm or more.


Phase-shift technology typically enables lithographic processes to produce features with dimensions measuring half the source wavelength, but research suggests that the technology can enable features 1/10 of the source wavelength. The gate sizes shown are, from left to right, 90, 25 and 9 nm. All were produced using a 248-nm light source. Courtesy of Lincoln Laboratory.


First reported in 2001, Fritze’s results were for an isolated line and for experimental purposes, but they indicate what might be accomplished with the application of resolution-enhancement technology. “In my own opinion, for this decade, we’re likely to use optics and extensions,” said Fritze, who researches resolution-enhancement technology as part of Lincoln Laboratories’ Advanced Silicon Technology Group.

On the other side of the continent, Eli Yablonovitch, an electrical engineering professor at the University of California, Los Angeles, had a bolder response when asked how long optical techniques can continue to power semiconductor manufacturing: “The simple answer is there is no limit. What people previously believed was the limit, which is the Rayleigh criterion, was misinterpreted. The Rayleigh criterion is a limit on density, not a limit on linewidth.”

The Rayleigh limit, he said, applies to pitch, which is line density or linewidth and spacing. The limit is half the wavelength of the light source, divided by the numerical aperture of the optics.

Challenging reflections

The semiconductor industry has determined that the best way to surmount the pitch constraint is to cut the wavelength of the source. Tools based on 193-nm ArF sources are replacing those based on 248-nm KrF lasers. By the end of 2005, according to Yen, F2 lasers at 157 nm will define state-of-the-art manufacturing. Lithography tools based on these sources all use transmissive optics, firing the beam through a photomask and various lenses to image circuit elements onto a target wafer.

At 13.4 nm, however, EUV tools represent more than a tenfold reduction in wavelength over their immediate predecessor. That will drive pitch down correspondingly. The problem is that transmissive optics are no longer an option at EUV wavelengths.

“You have to switch from refractive optics to reflective optics,” said Bob Sell, manager for marketing and strategy development for semiconductor materials at Corning Inc. in Corning, N.Y., which supplies optical blanks to the semiconductor industry for use in photolithographic tools. “No material is going to transmit the 13.4 [nm radiation]. In fact, it has to be done in a vacuum because it won’t even go through air.”


Despite the challenges that calcium fluoride presents to 157-nm lithography, it is still a transmissive and, therefore, familiar technology to semiconductor manufacturing. The next step after 157-nm processes could be EUV lithography, which requires all-reflective optical components. Courtesy of Corning Inc.


The change from refraction to reflection has a number of implications. For one thing, refractive transmission can be higher than 99 percent. Thus, nearly all of the fluence ends up at the target. Such is not the case for reflective processes, particularly at 13.4 nm, in which the radiation reaching the target drops to about 70 percent. That reduction affects throughput of the tool, the lifetime of the optics and the rate at which the source is consumed.

Sell said that Corning plans to offer the semiconductor market an ultralow expansion material called ULE, currently used in the reflective optics of large telescopes. ULE has not found use in microelectronics manufacturing before, but Sell is confident that it can work.

However, not everyone believes that reflective optics operating in a vacuum with a 13.4-nm source is the wave of the future. Some argue that Fritze’s work offers a potential alternative, if aided by resolution-enhancement techniques such as optical proximity correction.

This approach anticipates printed distortion in corners or at the end of bars. The layout of the circuit on the mask is altered in such a way as to remove some of that distortion and to allow the pattern to print much closer to its intended shape and size.

Because a photoresist captures radiation, it also captures interference patterns. That can be exploited, Fritze said, to produce a dark line whose width is half that of the exposing wavelength. This is done today on critical layers that have the smallest feature sizes and that typically define the electrical width of the transistors.

With feature slimming, researchers can make resist lines even smaller by uniformly etching them away. These and other techniques enable an isolated line to be narrowed down to virtually nothing — or until it becomes mechanically unstable.

“That’s very useful, of course, for doing research when you don’t have to pack things very densely,” Fritze said.

Although he has achieved this in a research lab, such subwavelength tricks are also finding a home in large-scale manufacturing of the latest microprocessors, in which features are crammed close together. As in Fritze’s work, these techniques use interference-producing phase-shifted masks, designed with software tools and technology from Numerical Technologies Inc. of San Jose, Calif.

However, although isolated lines can be made arbitrarily small, lines embedded within a dense thicket of neighbors cannot. One solution is to apply phase-shift techniques in a double exposure to pattern one layer of the resist. In this method, half the features are on one mask, and the other mask holds the rest. Dense structures would be interdigitated. This divide-and-conquer approach pays dividends in terms of minimum achievable pitch.

“A double exposure is equivalent to cutting the wavelength in half,” Yablonovitch said.

This could be carried even further with three or more exposures. With a higher numerical aperture and double exposure, the achievable density would more than double. The problem of registering one half of the layer to the other would be no worse, Yablonovitch noted, than aligning separate layers to each other. In an advanced semiconductor process, both would be equally challenging.

Multiple exposures present potential problems for the photoresist, said Mark Slezak, a technical manager for JSR Micro in Sunnyvale, Calif. The difficulty arises because the features are split among two or more masks. Often, one mask will image critical but isolated structures, and the other will image larger features with tighter pitches. During processing, both masks go through the same tool and are exposed to the same source to produce the final image in the resist. Thus, the resist can register not only imaging differences between the two masks, but also any subsequent development and processing differences related to the dissimilar feature sets.

“This makes the resist vulnerable to bias issues that can be enhanced by not only the aerial image of the exposure tool, but also the acid diffusion properties inherent to the material,” Slezak said. “A well-balanced photoresist is important in these situations.”

Another concern arises from resist-slimming techniques, which often are chemical in nature. This is independent of the source, but the interaction of the fluence and the chemistry can be complicated and must be addressed.

Hence, phase-shifting and optical proximity adjustments require interaction with and information from the manufacturing process, said J. Tracy Weed, a senior director of marketing and business development for Numerical Technologies.

“There’s a very intricate feed-forward/feedback loop that has to be maintained in order to create manufacturable features,” he said.

Weed thinks optical processing will continue to be the semiconductor workhorse for the foreseeable future. He noted that the lifetime of optical lithography will be extended by chemical mechanical polishing techniques, which use a slurry of small particles to grind wafer surfaces flat and allow imaging with smaller depths of focus.

Chemical mechanical polishing techniques are important to the discussion of overcoming pitch limitations. Numerical aperture is part of the formula that specifies minimum pitch. In standard photolithographic tools, air or an inert gas occupies the optical path, and filling that path with the right fluid can boost numerical aperture and cut the minimum pitch. Optical microscopists have used this approach successfully for years.

However, increasing numerical aperture decreases the depth of focus. Hence, planarization becomes more important and innovations such as chemical mechanical polishing more vital.

Applying fluid to the optical path raises other concerns for tool manufacturers. “You might have currents and eddies, and you’ll have to deal with those types of distortions, but it sure beats making a multilayer reflective mask that has to be accurate to atoms in terms of planarity and uniformity,” Weed said.

The question of when optical techniques will no longer represent the cutting edge of lithography is one of economics. Both sides’ arguments boil down to the cost of multiple exposures vs. the cost of new light sources and infrastructure.

Multiple exposures mean that the number of masks increases. This may be done only on critical layers, but it still has an impact. For phase-shift masks, the number of critical mask layers will increase fourfold or more. Each mask requires separate exposures, requiring separate trips through the tool. That cuts into the tool’s effective throughput and makes it less economical to operate.

Final analysis

“First, you have to have a number of these masks, and masks are very expensive,” said Sematech’s Yen. “You get expensive masks, and you lose the throughput. So you are being hit twice. That may not be a practical approach.”

Conversely, radical changes in the source and the infrastructure also are expensive. New optical components have to be developed and new resists perfected. Also to be considered are the cost of the source itself and the possibility of process changes necessitated by the new resist.

“Now, do you think that something that has half the wavelength is only going to cost twice as much? It’s going to cost a lot more than twice as much,” Yablonovitch said.

It’s too early for the industry to make a final decision. However, the latest microprocessors are emerging from 248-nm photolithographic tools, suggesting that these sources are producing commercial devices with features at least 95 nm in size. That partly is because more advanced tools are either not available or too costly, which suggests that EUV may take longer to arrive and be deployed than current plans demand.

“The development costs are going up,” said Corning’s Sell. “I think [that] people work very hard to stretch the existing technology. So we’re already seeing 248 [nm] being used today for a node that, five or six years ago, no one thought 248 [nm] would be used for.”



Other contenders

While EUV lithography is the designated heir to current semiconductor manufacturing methods, others could also be contenders. Some of these may even be successors to EUV.

Direct-write electron-beam technology, for instance, offers small linewidths and tight pitches. It currently is used to create the photomasks employed in today’s optical techniques. However, it writes each wafer and every line individually, in contrast with optical methods, wherein photons flood the surface and transfer the layer information all at once from the mask to the wafer.

“No serial processes can be faster than that,” said Tony Yen, director of lithography at International Sematech in Austin, Texas. “You — boom — hit it and then you get the pattern.”

Another possibility is the use of x-rays. In this approach, a mask is brought close to the surface of a silicon wafer coated with resist. The mask is exposed, and the pattern transferred to the resist. This process suffers from a number of problems, including the need to research and develop resists, masks and x-ray sources.

At the University of Texas in Austin, Grant Wilson’s research team is working on another technique, step-and-flash, which applies a quartz mask carrying high-resolution relief images directly to the surface of a resist-coated wafer. After photopolymerization of the resist through the back of the quartz template, the mask is removed, leaving a UV-cured replica of the relief structure on the planarization layer. Step-and-flash techniques have reliably produced features smaller than 60 nm.

However, this and other nano-imprinting methods require the construction of a mold, which sounds no more difficult than making an optical photomask. But optics reduce a photomask’s pattern to the right size on the wafer, allowing mask makers to reliably and inexpensively produce high-quality products. That approach isn’t possible when using a mold, Yen noted.

“If you want to write the features on a mask the same size as the features on a wafer, [then] that mask is really tough to make,” he said.

For these and other reasons, Sematech and its member companies have decided that EUV represents the best choice for what has been called next-generation lithography — for now.

Explore related content from Photonics Media




LATEST NEWS

Terms & Conditions Privacy Policy About Us Contact Us

©2024 Photonics Media