By Kristin Lewotsky
With process technologies that can take 15 years or more to bring to production levels, the semiconductor lithography industry is nothing if not good at planning for the future. Although 193-nm lithography is only just ramping up to broad deployment in 2001, all eyes are turning toward the next critical path on the technology road map: 157-nm lithography with fluorine (F2) excimer lasers. According to Richard Harbison of International Sematech (Austin, TX), 157-nm lithography is slated for insertion at the 100-nm node at the start of 2003 and is set to ramp up to broad deployment by 2005.
Only a few years ago, 157-nm lithography was off the road map, and the industry was expecting to push the reach of 193-nm tools while accelerating development of next-generation- lithography (NGL) technologies. Technological barriers such as the lack of a transmissive mask substrate appeared likely to delay 157-nm lithography to the insertion point of NGL. "The lithography community didn't seem to be interested in 157 nm because of the limited lifetime," says William Arnold of ASML (Veldoven, Netherlands).
In the past couple of years, key breakthroughs have put 157-nm lithography back on the map. Researchers discovered that low-water-content, fluorine-doped silica yields a highly transmissive photomask substrate, and chip makers such as Intel (Santa Clara, CA) came out in strong support of the program. "We have a number of customers directly working with us on trying to push 157 nm so that [our] first tools come out at the beginning of 2004," says Arnold. "On a parallel path, there's been a lot of interest in how far you can push 193 nm. It's not clear which will be used at 70 nm. Clearly, 157-nm lithography would have better imaging contrast at 70 nm, but 157 needs several pieces."
As with any technical undertaking of this scope, the challenges are myriad, but the primary technical hurdles for 157-nm lithography are a viable pellicle, an adequate photoresist, sufficient high-quality calcium fluoride (CaF2) material for projection optics, and contamination control. Transparency at such a high-energy wavelength is an issue that comes up time and time again, whether in optics, photoresist, photomasks, or pellicles.
In lithography, photoresist plays a key role in patterning the wafer by protecting select areas from process steps such as etching, implantation, or metallization. "The big problem is the optical density of the materials," says Grant Willson of the University of Texas (Austin, TX), noting that initially the only materials that appeared to be transmissive at 157 nm were Teflon and an oil called polysiloxane, neither of which were suitable as a starting point for photoresist development. "It was a really daunting issue. The possible polymers were too strongly absorbing, so how in the world were you ever going to design a resist that light can get through, and do chemistry clear through the depths of the resist?" Willson asks.
Typical resists for 248-nm applications have an absorbance metric of 6/µm (arbitrary units), for example, but displayed an absorbance of 0.4/µm at 157 nm. The new 157-nm materials under development by Willson's group are now displaying absorbance of 1.3/µm, which is just sufficient for imaging. Similar progress is being made at DuPont, Inc. (Wilmington, DE) and by groups in Japan (see figure 2). Unfortunately, although such films are transparent enough to generate an image, they are not thick enough to withstand the rigors of etching.
Figure 1 A 157-nm resist from UT generated this 130-nm, 1:1.5 feature.
One solution is a bilayer resist involving the deposition of two separate layers over the substrate. After the top layer gets lithographically patterned, the image is transferred to the second layer via anisotropic, oxygen-based reactive ion etching, which does not affect inorganics such as the silicon substrate. It's not really a practical solution for the long term, however. "You can make good images, but it's at the price of process complexity and cost," says Willson. The group is currently making good progress on the development of a single-layer resist and is producing its first images of reasonable quality (see figure 1).
The technology appears to be past infancy and moving up the development curve, where researchers are settling in for the hard slognot interesting, perhaps, but critical to commercial viability. "To get from the stage where graduate students are coming up with some neat tricks to having a bottle of photoresist available for the third shift someplace is a big jump," says Phil Ware of Canon, Inc. (San Jose, CA).
optics and lasers
Although fluorinated silica displays sufficient transmissivity to be used in 0.25-in. thick photomask substrates, it still induces too much loss to be used for projection lenses, which can have up to 30 elements. The real workhorse at 157 nm is CaF2. Used at 193 nm as a color-correction element, CaF2 has the benefit of the intensive development that has taken place over the past several years. That doesn't mean all the problems are solved, however.
Unlike the amorphous materials that most lens manufacturers deal with, CaF2 is a crystal, with specific planes or directions in which the material is hardest. "It's almost like wood, where you have grain," says John Bruning of Tropel (Fairport, NY). "You use pretty much all of the same equipment [for fabrication], but you have to modify processes, slurries, and liquids accordingly." That makes for difficult processing, particularly when the requirements are for */8 or */10 surfaces (p-v), or */20 rms.
One of the major concerns in the industry is simply ensuring an adequate supply of bulk material. Compared with fused silica, CaF2 takes longer to grow and anneal, as well as to fabricate into components. Manufacturers such as Canon, Bicron, Inc. (Newbury, OH), Corning, Inc. (Corning, NY), and Schott ML GmbH (Mainz, Germany) are expanding their CaF2 production facilities. "You can't buy the quantities you need from anyone on the open market," says Canon's Ware. "The major reason for [our] expansion was so that we would become totally self sufficient."
The main splitting point in the stepper design community is whether to go with all-refractive projection optics or whether to turn to a catadioptric (reflective/refractive) design. Generally speaking, an all-refractive approach needs some sort of compensation for chromatic aberration. One option is the use of a line-narrowed laser (see oemagazine, January 2001, page 12). The other solution is the introduction of a second refractive material.
The problem with a color-corrected refractive design is that currently there is no mature alternative to CaF2, let alone one that sports a sufficiently different refractive index to provide color correction. Magnesium fluoride (MgF2) is birefringent. Barium fluoride (BaF2) presents its own set of problems: It is hydroscopic, prone to thermal shock, and harder than conventional materials.
"It took many years to develop calcium fluoride, which is a much easier crystal to grow and polish," says Larry Thompson of Ultratech Stepper (San Jose, CA). "Most people are concerned that if you had to use barium fluoride there's not enough time [for development]."
His colleague, Doug Anberg, agrees. "Most people are trying to design around barium fluoride so they don't have to deal with it. The most successful design will probably be all calcium fluoride, whether catadioptric or refractive."
"The shortage of bulk material is definitely an issue, and not just in optics," says Vladimir Lieberman of MIT Lincoln Laboratory (MITLL; Lexington, MA), who cites laser windows, prisms, and beamsplitters as problems. "Most everybody planning to build 157-nm steppers is feverishly buying up supplies of calcium fluoride." On the bright side, crystals with too many defects to provide 6-in.-diameter projection optics may still be viable for 1- to 2-in. laser optics. "The quality for laser material is not as high," Lieberman observes. "It's a good entry point for new companies."
An all-refractive, line-narrowed approach makes a certain sort of sense for manufacturers who have historically focused their development funding on all-refractive designs. "When you go to catadioptric, the geometries are fundamentally different," says Thompson. Line-narrowing raises its own set of issues, however.
According to David Knowles of Cymer, Inc. (San Diego, CA), the natural linewidth of the strong output line of an F2 laser is about 1.1 pm, which is compatible with a catadioptric design. A color-corrected all-refractive lens could operate with a 0.5- to 0.6-pm linewidth, but an all-CaF2 design would require a linewidth of 0.2 pm. Line-narrowing in general means sacrificing power, an issue in the case of photoresists, which require a certain optical dose for exposure. In addition, the line-narrowing ups the cost of the lasers.
keeping it clean
The issue of contamination looms large in 157-nm lithography. "With 157 nm, oxygen is just a killer," says Gerhard Gross of International Sematech. "Everything seems to contaminatelenses, masks, pellicles, coatings."
"Small amounts of hydrocarbons can yield a carbonaceous coating on optics," notes Mordechai Rothschild of MITLL, who likens it to laser-induced chemical vapor deposition. "Even very thin layers of this diamond-like carbon would reduce transmission of optics at 157 nm."
Initially, researchers expected to run 157-nm systems in vacuum. More recently it has become clear that purging the optical system with nitrogen will be sufficient, provided hydrocarbons and water vapor are kept to negligible levels (see frontis, below).
"Our experiments indicate that photo-induced outgassing could be a major source of [contamination]," says Rothschild. "At 157 nm, the wavelength is so short you can induce photochemistry into almost anything." The challenge is building a stepper out of material that doesn't suffer from this problem.
The photoresist presents a troubling source of contamination. "A lot of resists outgas like crazy," says ASML's Arnold. "You can almost count on the customer to pick the worst ones," he adds wryly, noting that frequently the resists that outgas provide the best imaging performance.
Ironically, introducing trace amounts of oxygen into the optical path can induce self-cleaning of the system. "Even though 157-nm illumination hates hydrocarbon contamination, 157 aggressively takes care of it, also," notes Jim McClay of the Silicon Valley Group (SVG; San Jose, CA). "The laser itself does the cleaningyou just have to make sure the environment is conducive to it."
is it viable?
There is no question that the 157-nm time line is aggressive. Although 157-nm lithography is scheduled for insertion at 100 nm, it is really slated to be the workhorse at 70 nm. According to the official road map, NGL techniques come on the scene for the 50-nm node, essentially making 157-nm technology a one-node pony. "From a market perspective, the tool generation should be able to do at least three nodes; otherwise it won't make financial sense both from development sense for the vendor and also for the chip manufacturer," says Risto Puhakka, an analyst at VLSI Research, Inc. (San Jose, CA).
Given the development challenges still ahead and the likelihood that the 100-nm node will be primarily the province of 193-nm systems, is there really a cost argument for pursuing 157-nm lithography? "If they put 157 nm on line at the 70-nm node, you're already dealing with extreme numerical apertures and [resolution] factors. Out of the box you're talking about an extremely difficult process," says Canon's Ware. "Does it make sense? Making sense isn't necessarily the right answer. If the other vendors do it, you have to do it too." Canon is aggressively committed to 157 nm, as are the other stepper vendors across the board, whether they are focusing on test and development systems, like Ultratech, or full production steppers, like SVG.
Cost of development and cost of ownership are major issues, but most expect the technology to pay for itself over time. "It's a broader lifespan than most people think," says SVG's McClay. "There's not a single lithographic [generation] that has not been driven beyond the capability it was sold for. If you draw a line in the sand today and say 193 will get to here, it will go past it. If you were to draw one for 157, you would be surprised at how far it will go past it, too."
"The reality is that if you have 28 layers, you use your most advanced steppers on three or four layers. Then you use your previous steppers on 24 layers," says Yan Borodovsky of Intel. "Those steppers pay for themselves over many layers."
As to whether the technology will be ready for deployment on time, Puhakka, for one, remains sanguine. "Technologists have their doubts, but they have had their doubts for 193 nm and 248 nm also," he says. And look at how well they've worked out. oe
193-nm lithography hits the big time
The CaF2 lens for 157-nm lithography must be sealed in nitrogen atmosphere.
The year 2001 is destined to be the breakout year for 193-nm lithography. "193 nm is moving into broad deployment," says Gil Shelden at International Sematech. "As of last summer, each of the major manufacturers had eight to 10 steppers out there. Now we're starting to ramp up production tools."
"On high-end 193-nm machines, we're going to start our shipments in 2001," says Peter Convertito of SVG. "We see a pretty steep ramp in 2002."
The 193-nm story may indicate what's ahead for 157-nm technology. "The planned insertion [for 193 nm] had been the 0.18-nm node," says Shelden. "Now the 130-nm node is the transition node, and the industry plans to extend 193-nm technology into the 100-nm node."
"It's pretty clear that 193-nm machines will pretty much own the 100-nm node," agrees Doug Anberg of Ultratech.
Two issues that remain a problem are photoresist and inspection equipment. "They're still not as good as the current 248-nm resists," Shelden acknowledges. "All resists have to mature."
The inspection issue is a bit more tricky. "The resists we have are damaged by the [critical dimension] SEMs used for inspection. If you image a resist line, you will physically change it." Metrology system manufacturers are working to improve performance, however, and expect improvements as the technology matures.