Readying Resists for the 90 nm Node

The industry's realization of 90 nm devices will depend on the IC manufacturers' ability to bring 193 nm lithography into production. At the heart of this accomplishment is the combined performance of 193 nm photoresists, high-NA ArF scanners, masks and tracks (Fig. 1). While the 193 nm photoresists appear to be capable of meeting the resolution and process windows for the 90 nm technology node (expected to enter production in 2004), they missed the introduction envelope for 130 nm technologies.

The performance criteria for 193 nm resists are continually being improved — reducing line-edge roughness (LER), improving etch resistance and preventing pattern collapse. For some 193 nm resists, line slimming, where resist dimensions shrink during CD-SEM measurements, is still an issue, but methods are available to manage the problem from the SEM tool side. Dual damascene, copper and low-k dielectric structures introduce new challenges for the lithography process. It appears that bilayer schemes will finally enter the production environment for damascene application, addressing issues of via poisoning, varied topography and low aspect ratio (AR) imaging.

To some extent, tool availability has hindered 193 nm photoresist progress. However, it is a type of chicken-and-egg problem, where full-field, high-NA scanners are needed to fully optimize resist performance, but state-of-the-art resists are needed to determine the limits of patterning without resolution enhancements, so that engineers can decide where and when enhancements must be applied.

Engineers also face more compressed timelines for technology development. DUV resists took more than 10 years to develop, 193 nm materials took about eight years and 157 nm materials will only be allowed four years if they are not to impede 157 nm adoption.

	1. The Clean Track ACT 12 300 mm coater and developer track features small footprint, airborne chemical filtration, reduced chemical usage, improved thermal uniformity on PEB hot plates and throughput compatibility with scanners. (Source: TEL)

A year ago, resist developers were striving to rapidly introduce, qualify and extend DUV (248 nm) resists in production, while evaluating the merits of 193 nm resist platforms.¹ At that time, methacrylate and COMA (cyclic olefin maleic anhydride) type materials were available, each offering specific advantages, but neither meeting all performance needs at 193 nm. The polymer backbones are significantly different than the hydroxystyrene-based resists that companies have become comfortable with at the 248 nm wavelength. Meanwhile, the first work on 157 nm technology indicated serious issues associated with identifying appropriate 157 nm photoresists, which, together with pellicles, were flagged as showstoppers to the practical use of 157 nm (F₂) vacuum UV lithography.

Much has changed in 12 months. IC manufacturers continue to enhance DUV performance, allowing all of the critical layers of the 130 nm node to be patterned with 248 nm, including sub-100 nm isolated lines and 140 nm contacts. But 193 nm lithography will be the workhorse of the 90 nm node because even strong enhancement technologies like Levinson phase-shift masks (PSMs) combined with 0.85 NA DUV scanners (0.8 NA available today) cannot pattern contact holes smaller than ~130 nm.

Resist suppliers have considerably improved 193 nm resists. By taking new hybrid approaches to resist synthesis, they have devised creative ways of getting the best properties from different platforms. With the maturing of 193 nm resists, and the recent installation of higher-NA (0.75-0.8) scanners for volume production over the past six months, 193 nm lithography is on the bridge of adoption. It will be the first lithography technology to use an exposure wavelength that is twice the size of the targeted feature.

One hybrid approach, a result of addition polymerization, combines the resolution and good process windows of acrylate-based systems with the etch resistance and low surface roughness associated with norbornene derivatives. Mark Slezak of JSR Microelectronics (San Jose) explained the advantages: "There's so much that is known about the methacrylate systems with regard to flexibility of design. We can use different protecting groups based on size and activation energy to get specific properties. We can also control, for instance, iso/dense bias by controlling the type of PAG (photoacid generator) and loading level."

	2. 70 nm gates patterned with 193 nm resist and an alternating phase-shift mask. (Source: IMEC)

The first companies to use 193 nm lithography will be those producing a limited number of wafers per reticle set. The companies pursuing 193 nm lithography are device manufacturers that make ASICs, embedded memory, analog devices, and other lower-volume devices. On the other hand, companies making DRAMs or high-volume microprocessors, for instance, run enough wafers to sustain the cost associated with million-dollar reticle sets using phase shifting as well as advanced optical proximity correction (OPC) methods.

	3. 150 nm 1:1 and isolated contact holes imaged in 650 nm oxide over SiON, after oxide etch. The contacts were patterned using 400 nm Epic V VEMA 193 nm resist. (Source: Shipley and IMEC)

The 193 nm resists must meet a variety of performance criteria including high-resolution patterning, adequate process window (exposure latitude and depth of focus), good etch resistance and adhesion to a variety of substrates. Mature resists feature little LER and good profile control.

Though the raw performance of many 193 nm resists appears to be sufficient for the 90 nm node, integration issues remain. "The contrast is approaching the contrast of 248 nm processes, and the process windows are getting better, but improvements are still needed in the areas of etch resistance and line-edge roughness," said Kurt Ronse of IMEC (Leuven, Belgium).

Users need similar process latitude to that of 248 nm processes — in the range of 0.5 µm DOF and 20% exposure latitude for 1:1 120 nm lines. Process windows continue to improve. "Engineers and researchers are continually optimizing the resist formulations and processes, trying to tweak out another 50 nm or 100 nm of focus," said Will Conley, project manager of the University Photoresist Project and Motorola assignee to International SEMATECH (Austin, Texas).

LER correlates with the quality of the aerial image. "In essence, you need to improve the quality of your aerial image, whether that's through a higher-NA lens, optimization of the partial coherence and your reticle to the geometry you're building, by using phase-shifting masks, or all of the above," Conley said.

Though the correlation between LER and the mechanics of the lithography system are fairly clear, the relationship with photoresist chemistry is less understood. Trends have been identified, such as reduced LER associated with slower resists. "Whereas three years ago, most of the 193 nm resists had sensitivity in the 5-15 mJ/cm² regime, most products today are between 25-30 mJ/cm²," Ronse said.

Photoresists contain an imageable polymer backbone, photoacid generators (PAGs), dissolution inhibitors, etch barrier, and an acid labile, base-soluble group. Different constituents can be used to optimize resist performance. For instance, adding low molecular weight constituents tend to reduce LER. "Chemically, we are addressing line-edge roughness through different PAG/polymer combinations, PAG sizes, types and loading levels, and the addition of low molecular weight additives," Slezak said.

	4. 120 nm 1:1 lines patterned with an early 193 nm resist (left) vs. the more mature AR237J resist (NA=0.63) shows a significant reduction in line-edge roughness. (Source: JSR Microelectronics)

The challenge of controlling LER will only get worse as the industry continues to scale, added Mark Thirsk, of Shipley (Marlborough, Mass.). "At 70 nm, the targets are much more demanding, and we're beginning to get close to the size of the polymer molecules in the resist," he said. Slezak added that the surface roughness of the resist is also becoming important. "People focus on line-edge roughness, but surface roughness is going to be a bigger issue down the road," he said.

With closer features and higher-AR resists, the issue of pattern collapse becomes more serious. "A capillary effect together with the mechanical rigidity of the resist and the resist profile all contribute to pattern collapse," Slezak said. He commented that COMA-based systems tend to have higher mechanical rigidity than methacrylate systems.

Ralph Dammel, R&D manager at Clariant AZ (Somerville, N.J.), explained that cleaning methods also have a bearing on pattern collapse. "On high-resolution features we do see collapse above some critical aspect ratio, which also depends on the pitch," he said. "However, new supercritical CO₂ cleaning techniques, which lower the surface tension to essentially zero, allow patterning of features with ARs as high as 8:1 without collapse."

Outgassing from the resist is considered a controllable problem for 193 nm resists. "You can live with a little bit of outgassing because scanner manufacturers are introducing air flow under the projection lens, but this approach will be insufficient for 157 nm resists," Ronse said. One solution at 157 nm involves the use of CaF₂ plates. "Some of the new mid-field tools have field-serviceable CaF₂ plates as part of the optical path." However, Dammel suggests that, even with these plates, the resists must be optimized for very low levels of outgassing.

With the thinning of resists needed to attain high resolution at decreasing wavelengths comes an increasing need to use hard masks on select layers to compensate for limited etch resistance. Hard masks can also be optimized for antireflective qualities.

Another way to address limited etch resistance is by using two layers of resist — a thin top imaging layer and an underlying, etch-resistant bottom layer. Drawbacks include the lower throughput associated with two coating steps as well as the need for a dedicated etcher for the silicon-containing resist. Though there have been significant improvements in material quality, processability and resolution of bilayer resists, "outgassing remains a key concern," Conley added. "The silicon-oxygen bond is one of the strongest we know of, and if it is able to deposit on the lenses, it is nearly impossible to remove." Slezak argued that, when silicon is incorporated into the main polymer chain, as opposed to protecting groups, outgassing can be virtually eliminated.

Despite increased cost, dual-damascene applications are bringing in bilayer resists, particularly in via-first schemes where a thin single-layer resist cannot effectively pattern the thick oxide layer. Also, the problem of resist poisoning (interaction of basic materials with acid in the resist that renders the resist insoluble) introduces an additional requirement because the resist cannot contact the low-k dielectric. "With bilayer approaches, you improve planarity by optimizing the underlayer to fill the vias," explained Murrae Bowden, R&D director for Arch Chemicals (Norwalk, Conn.).

193 nm bilayer approaches are benefiting from 157 nm resist learning using silicon-containing materials. "Some of the materials in our university project have very low absorbency at 193, actually better than the existing 193 nm platforms," Conley said. "Some resist suppliers are already taking advantage of this."

Controlling the reflective properties in thin resists, antireflective coatings have become a necessary underlayer and sometimes top layer for sub-200 nm patterning. "Antireflectant technology is surprisingly dynamic," Thirsk said. "We have moved beyond simply controlling the optical constants and adding planarizing or conformal characteristics, to optimizing the antireflectant chemically for compatibility with certain resists."

For instance, the change to an antireflective coating with optimized formulation gives higher contrast and finer profile control (Fig. 5). Figure 6 shows the patterning of 110 nm 1:1 lines and spaces using a 193 nm resist with a bottom antireflective coating (BARC) of almost equivalent thickness. In such cases, the BARC must feature much higher etch rate than the resist to minimize resist loss or profile damage.

  
5. Thickness contrast for Epic V4 resist on different organic antireflective coatings. Sharper turn-on (1) leads to a flatter top profile; sharper turn-off (2) gives a straighter sidewall and less footing; higher contrast (3) results in higher resolving system. (Source: Shipley)

An entirely different phenomenon than LER, CD slimming refers to the shrinking of resist features that takes place during CD-SEM measurements. Though some chemical modifications are helping to produce resists that are less susceptible to line slimming, the answer to the problem has come from the SEM tool side.

	6. 110 nm 1:1 lines/spaces patterned on 0.63 NA scanner using 350 nm TOK resist and 387 Å ARC. (Source: Brewer Science)

Some resists are more affected by electron bombardment in the SEM than others. "We compared our 193 nm methacrylate platform with our 193 nm VEMA resist and our mature DUV product, and saw a 20 nm change with the methacrylate but only 6-7 nm change for VEMA and DUV," Thirsk explained. Bake temperatures and solvent content in the resist can also affect slimming. Some process steps such as flood e-beam or UV stabilization can reduce the magnitude of the slimming but, more importantly, recent data indicate that it might reduce LER.

"The bigger question is whether the single-layer resist shows sufficient stability for a given application, or whether companies have to incorporate additional process steps," Bowden explained. Despite the usefulness of stabilization techniques, engineers must consider the additional costs of implementation associated with a process step that may not be performed in-line.

The introduction of the first 157 nm commercial resist, made by Clariant AZ only about three years into development, represents an enormous breakthrough for the industry. It was facilitated by International SEMATECH members and several universities in the United States, as well as a pivotal decision to make all 157 nm resist developments widely available for the first time in industry history.

"In the past, people were working behind closed doors in secrecy, trying to get control of the most valuable piece of intellectual property, with huge redundancy and a slow learning process," said Grant Willson of University of Texas in Austin. "Everyone recognized that, if 157 nm was going to be valuable, the resist work had to be done much faster than before."

Willson added that, although many challenges lie ahead in 157 nm resists — to further improve the transparency, manage outgassing, etch resistance, pattern collapse, process window, etc. — he revels in the accomplishments to date. "Our students were able to improve the transparency of this material by a factor of a million, and now we can image 40 nm features — 40 nm in optical, I never thought I'd see that in my life."

"Indeed, the progress has been much faster than we anticipated back in 1999, when the first work on 157 nm systems really began," Dammel said. Clariant's experimental 157 nm resist is a trifluoromethacrylate material, which provides very low absorbency to 157 nm radiation. Dammel expects the chemistry of most of the fluorine-containing resists to be somewhat similar. "Probably all these fluorocarbon-based polymers will contain some form of norbornene or cyclic systems with a pendant alcohol group, so they will be similar in a way that all DUV resists are similar because they all contain benzene with hydroxy groups on it somewhere," he explained. "I doubt that they will be as different as the COMA and acrylate systems are at 193 nm."

The task of designing transparent materials for 157 nm photoresists is especially difficult because of the small number of materials exhibiting transparency to this high-frequency wavelength. Between the two most promising platforms, silicon-containing and fluorocarbon-based polymers, the fluorocarbon approach is prevailing due to the unacceptably low glass transition temperatures of siloxanes and silsesquioxanes. One problem with fluorocarbon-based resists is the high fluorine content. "That fluorine ends up in the etch gas, whether you like it or not, so that will require changes in the etch process," Willson said.

The industry knows of only two experimental 157 nm tools in the field today, at International SEMATECH and the Japanese consortium SELETE. Conley claims that the tool in Austin, a 0.6 NA mini-stepper manufactured by Exitech Ltd. (Oxford, UK), "operates around-the-clock, testing experimental and first-generation 157 nm resists."

Results to date are extremely encouraging (Fig. 7). But some companies have reached the point where they need higher-NA tools to further advance the formulations. "From the chemistry side, we've made good progress, but we're running up against the limits of discrimination," Thirsk said. "We cannot tell if we've made an improvement or not because the aerial image that we're getting from the optics is not good enough to tell whether a small change in formulation is advantageous."

    
7. Results from International SEMATECH Universities 157 Resist Project are 80-100 nm lines/spaces (1:3 AR) imaging using a 0.6 NA microstepper and 180 nm resist on silicon. (Source: International SEMATECH)

Meanwhile, 157 nm progress needs to be further accelerated. Ronse commented on the likelihood that 157 nm lithography will be ready to implement at the 65 nm generation: "Most people are starting to believe that 157 nm lithography will come too late for 65 nm devices, not so much because of the resists but of tool delivery schedules," he said. The problem of CaF₂'s intrinsic birefringence has already caused delays of six months, Ronse estimated. CaF₂ lens designs had to be modified to fix the problem (see "157 nm Optics Demand a Bag of Tricks"). "I think the earliest timeline for production use of 157 nm lithography is in 2006," Ronse commented.

The choice of material for 157 nm pellicles remains an issue. Companies are investigating inorganic films because traditional organic material will not work at 157 nm.