NGL: Forever Next-Generation?

Despite the insistence that the semiconductor industry must make a firm decision about which technique to pursue for next-generation lithography (NGL), the industry just seems to keep going on without ever really deciding. Sure, much of the industry is leaning pretty strongly toward a path of 248 nm to 193 nm to 157 nm to EUV lithography. But not everyone is on that path, and lithographers do not seem entirely confident that the pieces will really all come together.

Semiconductor International decided to conduct its own survey of lithographers to get their perspective on what the future holds for lithography. The results, collected by Reed Research, are based on 100 respondents who said they have buying influence for lithography equipment. Much like the industry as a whole, respondents seem content to continue with optical techniques for as long as humanly possible.

Today, the primary technology used to print critical layers is 248 nm lithography. In fact, more than three-quarters of our survey respondents said they are using 248 nm for most of their critical layers (Fig. 1). At this point, there are only about 10 machines worldwide using 193 nm lithography in a production environment, according to Peter Silverman, an Intel fellow and director of lithography capital equipment development at Intel Corp. (Santa Clara, Calif.). "By the end of this calendar year, there will be lots of 193 machines in use," he said. "By the end of 2005, there will be lots of 157 nm machines in use."

And so time marches on. Or, perhaps more accurately, Moore's Law marches on. Just how long the semiconductor industry can keep following Gordon Moore's increasingly daunting prediction is anyone's guess. In our lithography survey, we also asked people how long they thought Moore's Law would remain valid. Almost 80% of the respondents thought that, come 10 years from now, Moore's Law would be extinct. Some 29% of those people thought it would die out within five years.

Nevertheless, for now it is still a reality. Intel, for one, puts a new technology node into production every two years (generally in January of every odd year), according to Silverman. Last year, the company introduced 130 nm technology, and 90 nm technology should make its entrance next year. Assuming the schedule holds, the 65 nm node will be introduced in 2005, and 45 nm in 2007.

1. With i-line steppers still prevalent, lithographers are largely relying on 248 nm tools for most critical layers. They even figure prominently as the next production-worthy systems of those surveyed, with 193 nm systems next in line. (Source: Reed Research)

At Intel, each of these technology introductions requires a changing of the equipment for critical layers, Silverman said. "Every two years, we have a fundamentally different technology, requiring new equipment," he said. Although 248 nm tools are used now, Intel will use 193 nm lithography for critical layers next year. Accordingly, 157 nm lithography is planned for 2005, and EUV for 2007.

Although these are the years planned by the industry for 157 and EUV introduction, considerable effort remains to get them ready by those dates. And people certainly realize that they might not come in under the deadline. "At this point, it doesn't look like [157 nm lithography] will be ready in time, so we'll start off with 193, then switch to 157 when it becomes available," Silverman said. The same is true for EUV, at which point Intel will rely instead on 157 nm tools.

Some chipmakers may jump directly from 193 nm lithography to EUV, according to Steve Carlson, senior vice president of technology for Photronics Inc. (Jupiter, Fla.). As a maskmaker, Photronics must be prepared to offer masks for whichever technology its customers pick, he said.

All along, the lithography industry has found ways to delay the introduction of new tools. Current optical techniques seem to always be extendible just a little further, as engineers find new ways to make the old tools last. And when the next generation runs across technology problems, the industry really has no other choice but to find a way to extend current techniques.

"In the past, the use of non-optical lithography techniques has always been five to seven years away," said Christophe Pierrat, director of R&D for Numerical Technologies Inc. (San Jose). "But in the five to seven years that non-optical solutions are being developed, the industry is also using the same five to seven years to extend optical lithography. As a result, by the time non-optical techniques have reached viability for a given process node, optical lithography has surpassed it in technical viability."

2. The majority of lithographers surveyed are already working at the 130 nm technology node, but don’t expect to reach the 65 nm node until 2006. (Source: Reed Research)

Phil Ware, senior fellow, lithography, at Canon U.S.A. Inc. (Irving, Texas), agrees that there will be several more iterations of high-NA KrF and ArF before any NGL method becomes mainstream, noting that a wide array of resolution enhancement techniques (RETs) promise to extend traditional optical lithography well into the realm of NGL. "Both 157 nm and EUV still face many serious issues, and it is not at all clear just when solutions will be found."

It's not really major breakthroughs that enable the extension of optical lithography, Silverman said, but just an evolution of techniques. "More and more companies are using optical proximity correction and phase-shift masks, and those things are required to do what is on the roadmap right now." Simpler phase-shift masks (PSMs) are used routinely for contact holes, and the more complex alternating PSMs are beginning to come more into use, he said.

"As has happened with every node transition in the past, lithographers will find ways to extend 248 and 193 nm beyond the originally intended capability," Carlson said. "The cornerstone to this extension capability is engineering the wavefront by providing mask-based solutions." Granted, the more complex masks come at a cost, but maskmakers have been working to minimize those increases, Carlson said. "Of course the enormous increase in complexity of the mask adds additional cost, but balanced against the value being delivered now by the mask, instead of the traditional areas of new projection systems, resist, etc., mask value vs. cost actually shows a much more stable trend. At some point it becomes a game of trade-offs."

Numerical aperture (NA) is also being pushed to extend current techniques, higher-NA optics enabling a higher resolution from current light sources. A few years ago, NA was at 0.5, Silverman said; now it's at 0.75. To effectively push NA higher requires lasers with narrower bandwidths. Always looking for ways to improve bandwidth, Cymer Inc. (San Diego) recently announced a new laser architecture that will narrow the bandwidth of its excimers (see Semiconductor International , April 2002).

Ultrahigh-NA systems under development, along with the various RETs, will allow lithographers to continue to push k₁ factors closer to the theoretical limit, agreed John Cossins, EUV product manager for ASML in Tempe, Ariz. "Combinations of phase-shift masks and ultrahigh NA will take 193 nm well below 100 nm," he said.

A significant hurdle for 157 is the availability of a pellicle, Silverman said. "The machines are designed assuming there will be a pellicle," he said. "If pellicles are not there, it will take about a year to redesign the machines."

Developing soft pellicles for 157 nm photomasks has been difficult because progress has been slow on the development of polymers that maintain their transparency while being irradiated at 157 nm. According to reports at the third 157 nm technical data review, held last month by International SEMATECH (ISMT), ISMT has launched a project involving several universities to study the mechanisms of photochemical darkening. This should ultimately lead to the development of polymer materials that can withstand 157 nm radiation.

Meanwhile, progress is also being made on hard pellicles, which may present an interim solution. Hard pellicles use modified fused silica sheets of ~800 µm that provide ~95% transmission at 157 nm.

Optics are the other prime concern for 157 nm lithography. Projection optics will use calcium fluoride, which — despite advances in solving difficult challenges — still remains of considerable concern in terms of getting enough quality material (see Semiconductor International , February 2002). The quality of the CaF₂ must be some four times better than that required for 193 nm lithography, Ware said, and high-grade CaF₂ is "outrageously expensive." "Also, these very high-NA optics require that physically larger-diameter elements be polished to even tighter tolerances than today's best ArF and KrF lenses, which is a very difficult task."

"EUV's all-reflective optics are even worse, requiring defect-free multilayer coatings, with figure and surface roughness tolerances in the 0.1 nm range," Ware said, adding that the sources present problems as well. "EUV sources are at the top of anyone's hard-to-do list, with 10× to 15× increase in output power required for production tools," Ware said. "Scaling today's leading contenders to those levels could result in a source with an enormous footprint and a $4-7M price tag."

"EUV isn't a slam dunk," Silverman admitted. Intel has been a major proponent of the EUV effort, spearheading the EUV LLC, a consortium of chipmakers that also includes AMD, Motorola, Micron Technology, Infineon and, just recently, IBM. Intel recently ordered an EUV lithography beta tool from ASML (Veldhoven, Netherlands), slated for delivery in late 2005. Getting a source with sufficient power is indeed the biggest problem, with heat dissipation and contamination issues also figuring into the scenario. But the EUV LLC — working with three national labs and other industry groups — has been solving these problems, and has had some success with sources, he said.

Our survey respondents listed system price (68%) and cost of ownership (62%) as top concerns for EUV lithography — a common theme, in fact, for each of the NGL techniques (Table). Cost of ownership (CoO) can include several issues, ranging from system cost to cost of consumables and masks, noted Chuck Gwyn, program director for the EUV LLC. The system source, however, is a major CoO driver, Silverman said. Essentially, the cost of a 10-wafer machine is the same as a 100-wafer machine, he said, and that throughput is tied directly to the source's power level.

There are four or five EUV sources being worked on by various groups, mainly varieties of laser-produced plasma (LPP) and electrical discharge-produced plasma sources. The highest continuous power output has so far come from an LPP variety, but even that is at ~5 W, compared with the 50-100 W that will be necessary for a production tool, Silverman said. It's too early to know which type of source will win in the end, he said, but the EUV LLC hopes to know within a year.

Gwyn agrees that the source is the most challenging hurdle facing EUV lithography because of the target of 80 wph throughput. "At present, the LPP source is the most promising source because the most work has been done on this source to understand problems and resolve scale-up issues," he said. "If a compact illuminator design can be made to multiplex sources, many of the discharge sources have good potential because of the lower cost and better conversion efficiencies of electrical energy to EUV flux."

Including the source, the top five hurdles for EUV lithography are environmental control (to maintain the required component lifetimes, contamination control, and maintenance support); mask fabrication to tight specifications (flatness, defect levels, uniformity, inspection metrology, and repair); mask protection (removable pellicles and storage); and commercialization infrastructure (sources for all components, including masks, metrology equipment, optics, stages, etc.), according to Gwyn.

To overcome mask hurdles, Photronics collaborated with IBM in early efforts to define maskmaking challenges for NGL, according to Carlson. "This three-year program was intended to identify the mask challenges of NGL and narrow the focus," he said. "This was very successful, and paves the way for [Photronics] to initiate the next step, a pilot line for EUV masks in 2003."

Although masks make our survey's list of top hurdles for almost every NGL technique, it is actually a tribute to development progress that they do not rank higher on the list, Carlson said. "This is in stark contrast to just a few years ago, when masks were thought to be the No. 1 showstopper to keeping on Moore's Law. Now it is more of an economic issue."

Though certainly not to the degree of 157 nm and EUV lithography, e-beam lithography is the only other NGL candidate being given any degree of serious consideration at this point. E-beam has been a contender for more than 20 years, Ware noted, and is still not where it needs to be. "To many observers, e-beam's own set of problems seem even more of a challenge than those of 157 nm and EUV," he said. And, unlike the worldwide initiatives backing 157 nm and EUV, maskless e-beam lithography is being pursued primarily by small private efforts, he added.

Maskless — or direct-write — e-beam lithography is simply way too slow for volume production of semiconductors, Silverman said. Even e-beam projection lithography (EPL), which is being pursued primarily by Nikon Precision Inc. (Carlsbad, Calif.), is significantly slower than EUV, he added. But Silverman also has concern with EPL's masks, which he contends will be difficult to make and therefore expensive. "There's absolutely no data on mask defects," he said. "Everybody worries about mask defects on EUV, but nobody seems to worry about EPL masks. I think that's because there isn't any data, so people take an optimistic view."

No matter what the technology, decisions about what technique to pursue at any given time come down primarily to economic drivers. This economic issue is a given as tool complexity continues to grow with each new generation, ASML's Cossins said, noting extreme contamination requirements for 157 and EUV, as well as complex metrology and software requirements. "Nonetheless, manufacturers must be able to make devices in a cost-effective way. Consequently, toolmakers strive to make systems that continue to reduce the cost per transistor or function printed on a wafer. In this way, the increased lithography tool cost can be justified."