EUV continues to soak up the majority of lithography R&D dollars, leaving some potentially viable alternatives on the sidelines.
Over the years, next-generation lithography (NGL) has suffered various setbacks and delays. But until recently, the industry basically shrugged its shoulders and expressed relatively little anxiety about the NGL delays. After all, optical lithography was doing the job in the fab and NGL would eventually materialize.
Today, however, the mood is different. In fact, there is a growing concern, if not panic, among chipmakers about the current state of NGL. The leading NGL candidate, extreme ultraviolet (EUV) lithography, remains late and will likely miss the 10nm node. And the other NGL technologies, multibeam and nanoimprint, are also behind.
So, chipmakers must extend today’s 193nm immersion and multiple patterning to 10nm and perhaps beyond, which could be an expensive solution. Originally, chipmakers were hoping to avoid the multiple patterning era and NGL was supposed to do the trick.
Given the issues with NGL, the key questions are:
Regarding the cost, or R&D dollars, the numbers are difficult to ascertain. But in total, it could take billions of dollars in R&D just to complete or fix the problems associated with the various NGL tool technologies in the market.
As before, EUV is expected to obtain the lion’s share of the future R&D funds, which remains a bone of contention among many in the industry. In total, EUV already has consumed no less than $14 billion in funding over the years, according to estimates from multiple experts. And today, ASML Holding has some 1,000 engineers working on the oft-delayed EUV power source alone. As a point of reference, that may be more engineers than the multibeam and nanoimprint industries have combined. And to date, EUV is still not in production.
Meanwhile, investments in multibeam and nanoimprint have paled in comparison to EUV. Some would even argue that the other NGLs have suffered by not getting a fair share of the R&D pie. “EUV has received a crazy amount of funding,” said Dan Rubin, a general partner for Alloy Ventures, a venture capital firm. Rubin is on the board of Molecular Imprints (MII), a supplier of nanoimprint tools and a potential competitor to EUV.
“Most of the funding (for EUV) comes from companies with vested interests,” Rubin said. “Financial investors would rarely take the EUV bet because of the crazy high number of things that EUV had to do and still needs to do. So I can’t for the life of me figure out why people invested so much in EUV. The evidence that it will work is not very strong.”
EUV: On the hot seat
Still, EUV remains the leading NGL candidate. “If it works, then EUV is still the number one NGL solution. But the big question is if. And whether it is (ready) on time or not, the industry is still not sure,” said Aki Fujimura, chairman and chief executive of D2S.
Originally conceived in the 1980s, EUV has obtained an astronomical amount of funding. To date, ASML Holding itself has invested about $2.8 billion in R&D for EUV. In 2012, ASML also obtained a combined total of $1.9 billion in R&D funds from Intel, Samsung and TSMC. In addition, Intel, TSMC and Samsung also have the option to buy an aggregate equity stake of 23% in ASML for $5.3 billion.
Over the years, there have been other direct and indirect investments in ASML’s EUV efforts. Without counting the equity funding from Intel, TSMC and Samsung, the industry has invested a total of approximately $7 billion in terms of R&D for ASML’s EUV efforts alone, according to estimates from multiple experts.
That doesn’t include the EUV R&D conducted at other companies, national labs and universities since the 1980s. Combining those efforts, the total R&D funding for EUV could be “two to three times” the $7 billion figure, said G. Dan Hutcheson, chief executive of VLSI Research. So, the total R&D funding for EUV could range from a whopping $14 billion to $21 billion.
So far, the funding has produced mixed results. ASML reached a milestone in 2010 when it shipped the world’s first pre-production EUV tool, dubbed the NXE:3100. But the big test for EUV is just getting underway, as ASML late last year began shipping its first production-worthy EUV scanner. The tool, the NXE:3300B, has a resolution of 22nm half-pitch.
The soonest that the NXE:3300B would be ready for mass production is the 7nm node, according to lithography expert Chris Mack. That, of course, largely depends on whether ASML can solve the ongoing issues with the EUV power source.
In any case, the likelihood that EUV succeeds or fails remains a topic of debate and the predictions are all over the map. “It’s the only NGL that works,” VLSI Research’s Hutcheson said. “No one said EUV was going to be easy. We all know they need to get the power source working. I am confident it will happen.”
Others are cautiously optimistic, especially about the oft-delayed EUV power source. To date, the source only generates 10-Watts of power, enabling a throughput of less than 10 wafers per hour (wph). By year’s end, the industry wants an 80-Watt source, enabling an EUV throughput of 58 wph.
“We are close to getting to 70 wph,” said Dean Freeman, an analyst with Gartner. “The 100 wph figure is where it is difficult to tell. If you listen to Gigaphoton, we are approximately two years away from having a production source that can do this, and that is with a lot of ‘ifs, ands and buts.’ I would think by 7nm you would have a 90+% probability of 70 wph, and an 80% probability of 125 wph. This is one of those topics that’s very difficult to tell.”
Not surprisingly, the EUV detractors have a different opinion. “If by some miracle, all the missing pieces—such as the masks, resists and source—come together at the same time, it’s still too expensive for memory production. And the small number of layers that could conceivably be used in non-memory markets makes it seem like a very bad economic bet,” said Alloy Ventures’ Rubin.
Multibeam: Waiting in the wings
Like EUV, direct-write e-beam is also the subject of debate. Originally developed by IBM in the 1980s, direct-write is attractive because it enables fine resolutions without the need of a photomask. But the throughputs for single-beam e-beam are too slow, making it expensive for volume IC production.
To solve that problem, the industry has been developing direct-write technology that makes use of multiple beams. In total, the three major players in multibeam—KLA-Tencor, Mapper and Multibeam—have raised roughly $700 million thus far, according to estimates from multiple sources.
But the production-worthy multibeam systems still require more work and are not expected in the market until 2018. If the multibeam hopefuls decide to finish their respective systems, the total investment required could hit $1 billion or more, according to experts.
In multibeam, vendors have encountered various technical roadblocks and funding issues, which have delayed the shipment schedules. “For direct-write, people are making progress. We know the physics are good. So, it’s a matter of making it work. What I’ve always said about multibeam direct-write is I think it’s really a matter of investment,” D2S’ Fujimura said.
Others agree. “If we look at our experience with Mapper, we can see that some of the delays have been definitively due to the lack of funding,” said Serge Tedesco, lithography program manager at CEA-Leti. “On the other hand, Mapper has always been able to raise new funds.”
Still, CEA-Leti is moving full speed ahead with its multibeam consortium, dubbed Imagine. For some time, the group has been working with an alpha tool from Mapper. CEA-Leti recently installed a new pre-production tool from Mapper. Initially, that tool will consist of 1,300 beams and have a 32nm resolution.
The first exposures for the tool are slated for June of 2014, Tedesco said. The tool is expected to have 13,000 working beams by the fourth quarter of 2015. “The Imagine program is still on track and making steady progress,” he said.
Like EUV, there is uncertainty with multibeam. “Not every e-beam vendor will be successful,” said David Lam, chairman of Multibeam.
Multibeam itself is developing what it calls Complementary E-Beam Lithography (CEBL). Instead of patterning all layers, Multibeam’s CEBL addresses a small but important part of the equation. “CEBL could be used to pattern line cuts in critical layers to complement optical lithography,” Lam said. “The critical layers are difficult to pattern with optical. And, of course, they are the most costly because of multiple patterning.”
Still others remain skeptical about multibeam. “The problem with e-beam is that it violates Moore’s Law,” VLSI Research’s Hutcheson said. “Direct-write has been around a long time, but it hasn’t made much progress. The throughputs have been stuck between one to three wafers an hour for a long time.”
Nanoimprint: A new chapter
Hutcheson also has reservations about nanoimprint for IC production. “I don’t think imprint is the next NGL,” he said. “Imprint has been successful in some markets, but not for semiconductors.”
Meanwhile, MII, the leading supplier of nanoimprint tools, has secured $165 million in funding thus far. MII has gained traction in non-semiconductor applications, such as displays and storage. But defectivity, overlay and throughput have prevented the company’s imprint tools from making significant inroads in chip production.
MII is also beginning a new chapter. Canon recently acquired the semiconductor unit of MII. That group is called Canon Nanotechnologies (CNT). Meanwhile, the non-semiconductor unit of MII will continue to be called Molecular Imprints.
CNT and MII will share the same building in Austin, Texas, but the two companies will be physically separated and treated as independent entities. CNT will pursue the chip business, while the new Molecular Imprints will focus on emerging market applications, such as displays, lighting, biotech and storage, said Paul Hofemann, vice president of corporate marketing and business development at MII.
CNT is developing a new modular, CMOS cluster nanoimprint tool. “Each module is capable of yielding sub-20nm die with a cumulative cluster tool throughput targeted for more than 80 wph,” Hofemann said.
The tool has demonstrated 15nm half-pitch images. “The resolution is controlled by the mask and DNP has recently shown resolutions down to 9nm,” Hofemann said. “Another advantage of nanoimprint over optical is that it is not subject to the same design rules. This means you are not restricted to 1D patterns, but rather are free to develop whatever the device design requires.”
CNT is expected to announce its new tool in 2015. “We expect initial adoption to be in the Non-Volatile Memory segment in 2015, followed by DRAM a couple of years later, and eventually in logic applications,” he said. “Fortunately, the end-demand for innovative consumer electronics is alive and well, leaving the door open for a disruptive NGL technology.”