After more than a year of peer reviews of submitted articles, we will publish the JM3 special issue in July, with 23 papers covering such topics as laser produced plasma (LPP) and discharge produced plasma (DPP) sources, mask metrology sources, modeling source components (debris mitigation, spectral purity filter [SPF] and lasers) and papers on many alternate concepts for EUV sources. Although all of them are well worth reading, I want to highlight papers that analyze technology limits, offer solutions on how to advance current technology, and provide alternate EUV source concepts.
DPP sources have some definite advantages. They are simpler and cheaper than LPP, and I can personally testify that they can run continuously for eight hours and more. However, since 2007, the power for installed DPP sources has remained in the 7-10 W range, although their reliability has improved. My guess is that thermal mitigation is still the issue. Over the last five years, I have seen convincing data on power scaling potential for these sources, and I see even more conclusive scaling data now. However, I will withhold judgment on how well these sources can scale in power, until end customers report on their long term operation at higher power.
A paper by Koshelev et al. on metal jet offers an interesting approach to DPP power scaling. It was shown to be more than a concept when he provided experimental results at the 2011 Source Workshop. Koshelev expects to soon scale his source power to 800 W at source (80 W at IF) by using 32 kW input and 2.5% conversion efficiency (CE), and details his new DPP approach in this issue. As with all new concepts, it will need engineering work to become a commercial product.
LPP source power has improved from the mW range in 2007 to ~10 W today. The physics of scaling seems straightforward for these sources as well. Now that many LPP sources are in the field, much more attention is paid to them today. But in fairness, I must withhold judgment on the power scaling potential of LPP sources as well, until I see more field data from customers.
Papers from Toshio Tomie and Gerry O’Sullivan provide excellent reviews of LPP technology and offer opinions on its limits. Also worthwhile is a theoretical paper from Koshelev et al. on distributed targeting for LPP (Akira Endo expanded on this concept in his talk at the 2011 Source Workshop). The distributed target approach is interesting and promises higher CE and better debris control, but it needs to be brought into practice and then into manufactured products.
Alternate concepts for EUV sources are explored in papers on EUV lasers, the Laser Compton effect, tabletop synchrotron, inverse laser Compton effect, EUV lasers, electron cyclotron resonance (ECR)-based plasma and free electron lasers (FEL) lasers. Of these approaches, FEL, tabletop synchrotrons, and ECR plasma papers claim the ability to scale up for high EUV source power requirements. Other alternate concept papers focus on metrology applications. All of these papers provide experimental proof at some level.
An FEL paper delivered by William Barletta at the 2012 SPIE Advanced Lithography conference claimed a 500 W source with an estimated cost of $100 million. I support serious examination of these concepts, most probably by a group of experts that include end users as well as scanner makers. Even so, it’s difficult to find funding for alternate technology development when conventional suppliers are claiming that such high source power capability lies within reach of their technology in the near future. But if we do not see significant power scaling results this year, a serious review (not development) of alternate technology would be desirable. We owe it to ourselves, and it is very much worth the effort, to at least do a serious assessment. In any case, I encourage you to read the papers on current and alternate concepts and share your comments with me.
- When will these HVM-level EUVL scanners be used to make products?
- What will trigger the industry-wide insertion of EUV scanners in HVM production lines?
- Will it be a certain source power level, throughput, yield, or cost that raises the confidence of users?
- Who will be the first users of EUVL in HVM, and for what products and which node?
Similar questions can be posed for EUVL mask defect metrology tools. With scanners, we have an approximate throughput model that is widely known, so we can estimate throughput and cost of ownership at a given source power level. This is not quite so for mask defection inspection tools, as they are still being designed and need brighter EUV sources. The leading option for sources for defect metrology tools (and the only one for a 24 x 7 operation) is the Energetiq Xe discharge-produced plasma (DPP) source. However, current performance levels for this source allow only development of prototypes.
In my opinion, memory makers probably will be the first adopters of EUVL technology. Throughput as high as 40 wafers per hour (WPH) will convince chip-makers that EUVL is a viable technology, and sales of EUVL scanners and associated tools and products will start soaring. Mask defect metrology tools will remain unready for HVM until a brighter source becomes available. We can expect a déjÃ vu: low throughout metrology tools waiting to be upgraded.
In the case of high-power sources, we have three major suppliers with the commitment and resources to continue development. Not so for metrology sources, and the end customers will have only themselves to blame this time. I have read that end users are spending up to $150 million on metrology tool development, but not a dime on developing EUV sources for metrology. A source supplier told me that promises of support for metrology source development have not materialized, even though EUV sources for metrology are the weakest link in the chain. With virtually all of the money going to engineering development of tools and none to this weakest link, the results are very predictable.
As the industry Roadmap moves to smaller nodes of resolution (10 nm and below), will we choose EUVL with double patterning, or change the wavelength once again and move toward Beyond EUV (BEUV)? Gadolinium (Gd) at 6.8 nm is the current leading option for BEUV source material, as we saw from the latest development results on source and optics in last year’s EUVL Source Workshop. Increased optical proximity correction (OPC), off-axis illumination (OAI), and double patterning may require more power at these nodes than 13.5 nm sources can provide, so it might make sense to move to BEUV. For BEUV, we have a leading source material for multilayer (ML) optics and resist development has already started. We also can apply lessons learned from 13.5 nm to BEUV tools, although more infrastructure work will be required.
These and related topics will debated by panelists from Intel, GlobalFoundries, Toshiba and Applied Materials (AMAT) at the 2012 EUVL Workshop being held June 4-8 in Maui, Hawaii. The panel will be moderated by Sushil Padiyar of AMAT. The Workshop also will feature many papers on BEUV and EUVL R&D from some of the world’s leading researchers. I will be blogging here about these new developments after the EUVL Workshop.
Status of EUVL
The EUVL conference started with an invited talk from IMEC, which presented process development data from ASML’s NXE3100 beta level EUVL scanner. The scanner’s throughput is now 5 wafers per hour (WPH) for ~7 W of power. They are patterning at 16 nm half-pitch (HP) using dipole illumination, and can get 5 nm machine-to-machine overlay in order to mix and match with another ASML 193 immersion (193i) scanner.
ASML now has six NXE3100 scanners operating in the field, five of them in process development and one in qualification. Line and space (LS) of 16 has been achieved with a 33 mJ resist. These scanners currently add 0.05 particles per reticle pass (or 1 particle in 20 passes). This recent 20x improvement is critical, since an EUVL mask does not have a pellicle.
ASML plans to ship the HVM version of their NXE3300B scanner in the second half of 2012. For this tool, numerical aperture (NA) will go to 0.33 from 0.25, with 4% flare optics. This tool has specs of > 69 WPH, and I am sure source availability will drive this number. ASML also presented their EUVL roadmap to 10 nm HP and below, utilizing double patterning (DP) or a switch to 6.8 nm.
EUV source remains the key driver of scanner throughput and EUVL’s introduction into HVM. Current power reported for NXE3100 in the field is 7W for Xtreme and 10W for Cymer. Gigaphoton has a 7W source in their lab and plans to ship a higher-power source later this year for integration. Xtreme, which has shipped a 20W source for NXE3100B and is working on 50W-100 W sources, described in detail their new electrode design for high-power sources.
Cymer is working to upgrade its current sources in NXE3100 to 20W and plans to have 50W sources in Q3 this year. The 20W upgrade is based on a double pulse approach, which has been proven effective for many years. So this upgrade should be successful, although perfectly aligning two lasers, both pointing at 30 microns droplets moving at the speed of 50 k Hz with 100% accuracy is no small task.
Gigaphoton plans to ship 50 W sources in Q4 this year. They presented 4% conversion efficiency (CE), the highest so far for laser-produced plasma (LPP), and progress on their debris mitigation design. I am looking forward to their sources in the field this year.
In terms of delay in source technology development, the main issues for LPP seems to be debris mitigation and droplet stability, plus engineering to ensure 24 x 7 operation. For discharged-produced plasma (DPP), the issue is thermal mitigation.
Two years ago, ASML introduced “exposure power” as a new way to describe source power. The term is based on the concept that due to plasma stability and losses from spectral purity filter (SPF) use, only half of EUVL source power at intermediate focus (IF) reaches its target. In other words, 100W of source power at IF produces only 50W of usable “exposure power.” I believe that as sources make progress, the gap between IF and exposure power will get smaller. However, the difference was not made clear in presentations, leaving us to guess whether the 20W or 50W that suppliers plan to have in the second half of 2012 is exposure power or IF power.
EUVL and the Art of Auto Repair: EUVL Success Stories
Last year my friend Bob tried to fix a small issue with my Suburban, the auto with the now famous “EUVL” license plate. Bob is always working on cars and expected this repair to be a piece of cake. But after several tries he gave up, saying: “Nowadays cars are too complex, now you need a computer to fix anything.” If only it had been an older model, he would have been able to help.
Not too long ago, the EUVL alpha scanner was criticized by an expert as looking “like you rolled an electromagnet through an automotive junkyard,” while others called it a “science experiment.” Now that early prototype has evolved into a smooth machine, although still a slow one. Stepping away for a minute from the issue of low throughput due to source power, I do not hear any complaints about the scanner, which is a very complex machine in itself. Lithography in vacuum seemed impossible to many not too long ago. This is an incredible machine: a scanner that was never supposed to work at all now performs so well that the tool itself is apparently not an issue.
Unfortunately, I did not hear any praise of ASML (who integrate sources but do not build them) for all that they have achieved so far. Maybe that’s because this year’s SPIE AL seemed pretty tense. One of my colleagues, who has been coming to these meetings for years, said he did not hear anybody laughing or joking over the entire three days. Instead, people were walking around as if the world was going to end in 2012, per the Mayan calendar.
When I started in the EUVL business, sources were not yet the main challenge, but optics contamination, optics quality and masking without a pellicle were said to be limitations that would doom EUVL. Now these former threats have been tamed. Yes, the technology is lot more complex, but increasing complexity is part of both life and a next-generation, better machine.
Comparing Apples to Oranges: Looking Beyond Throughput to System COO
EUVL has attracted a fair bit of criticism, and I believe that comes from its being the front runner of NGL technologies. One of most frequent criticisms of EUVL is throughput. While 193i scanners can process 200 wafers per hour, EUVL is at 5 WPH and is expected to climb to just 40+ this year. So how can EUVL ever compete with immersion Lithography?
The answer lies in the simple fact that we need to consider the throughput and cost of ownership (COO) of the entire lithography module, rather than just looking at the exposure tool. Double patterning has more than doubled the cost and processing steps, resulting in some tools with low throughput. A COO model accounts for this. And even ASML, which sells lots of 193i scanners, has said that it is much more cost-effective to go with EUVL. So the real question we should ask is, at what throughput will the COO of EUVL become the same or less than 193i? Although I have not seen an analysis of this question, I have heard that 40 WPH throughput will be a favorable turning point for EUVL, and at 60 WPH most chip-makers will adopt it in HVM. We are not at these throughputs yet, but how far along are we really? Answers will depend on our assessment of rate of progress of EUV Sources.
Critical Issues: Do We Know How to Make HVM EUV Sources?
The most positive aspect of EUV source status is supplier commitment. Gigaphoton and Cymer, who also supply light sources for 193 nm lithography, are investing heavily in EUV source development. Ushio, another light source supplier, is also very committed and investing in the source business. However, to move to next generation sources, we may need innovation and new technology in addition to supplier commitment.
At the November 2011 Source Workshop, EUV experts gathered to discuss how we can get to higher power levels for EUV sources. It was time for honest introspection. Konstantin Koshelev, whose lab built original prototypes for Sn DPP, delivered a plenary talk entitled, “Do We Know How to Make HVM EUV Sources?”
The current design of DPP introduces high thermal load on electrodes, and droplet-based tin delivery systems introduce debris and stability issues which create great engineering challenges. Koshelev’s proposals for new DPP and LPP design, along with many other ideas presented in the workshop, need careful consideration and support for further exploration. Based on current publications, I see only Gigaphoton actively involved in R&D to continue looking for new ideas for their next generation sources.
How Do We Address the Critical Source Power Gap to Bring EUVL to HVM as Soon as
EUVL has attracted attention as it is a multi-node technology, but it will need more and more power at coming nodes. I can get behind 100W or even plans to get 150W. However, I cannot see how the 250W or 350W levels on EUVL roadmaps can be achieved without new technical approaches.
Leading consortia have left source development to the suppliers, which are great engineering firms but not necessarily places for developing new technology. Even after years of delay, there is no change in policy to address this critical gap. Lots of money gets spent every year to help suppliers make progress on various EUVL-related issues, but almost nothing goes to develop EUV sources which are #1 issue for EUVL.
This is surely insanity. You will not hear a supplier say, “I do not have the technology for next generation sources and need help," but many will claim, "I have the technology and just need more money and time.” How can a supplier do otherwise, when their competitor is promising "lots of power tomorrow with proven technology?"
Our industry needs to step up and say, “Enough of this. Let’s generate more technical solutions and make them available to the source suppliers so they can increase source power.” With a steady stream of funded innovation flowing from Universities and National Labs, I am confident existing suppliers will be able to take care of source power needs.
Here are my ideas on what needs to be done to increase and sustain EUVL technology development:
(1) Create a consortium that focuses on development of next generation EUV source technology. We need innovative solutions to address debris, thermal management, low conversion efficiency, out-of-band (OOB) radiation, and plasma stability to take us to 500W source designs. I believe that a $10M annual budget for three to five years is needed for such studies. We can have Small Business Innovation Research (SBIR)-type programs where solutions that meet technical criteria move on the next stage to receive additional funding from the consortium and industry. Which existing consortium can take on this challenge?
(2) Although EUV double patterning combined with optical proximity correction (OPC) is suggested for taking EUVL scanners to the end of the Roadmap, the power required is too high at 13.5 nm but can be reduced by changing the wavelength to 6.7 nm. Work has already started on 6.7 nm optics, sources and resists (see workshop proceedings at www.euvlitho.com), but we need to put a lot more emphasis on this wavelength switch and be ready to provide solutions that can produce commercial products.
(3) Perhaps most important is that Nikon has slowed down its EUVL development work. They, like ASML, developed two alpha level scanners but chose Xe DPP sources over ASML’ choice of Sn DPP sources. Xenon plasma source size is greater than a tin plasma source; hence, Nikon tools hit power limits sooner than ASML’s alpha demo tool (ADT).
Currently, Nikon has programs in optics and contamination, but they are not developing an EUVL scanner. They are very capable of doing so, but will not be able to catch up with ASML if they wait any longer. I have been known to say that you do not get a second chance in this business. Is it good for the computer chip industry to have only one supplier for its most critical tool? As this question directly relates to corporate bottom lines, chip-makers should be concerned about answering it.
Lotus Bet Update
During the conference, a couple of colleagues congratulated me after they incorrectly assumed that I already have the Lotus keys from my wager with Litho Guru Chris Mack. (Chris took the dubious side of a bet that there would be no papers presented on EUVL in 2011, or that EUVL would not be in HVM by 2013-14). There appeared to be no shortage of EUVL papers in this year’s conferences, and audiences spilled over from EUVL sessions until they were moved to larger rooms. I am still confident that EUVL will be in HVM soon. By the way, the Litho Guru was praised highly by at least one supplier for his contributions to the development of EUV resists, and for his help in reducing line edge roughness (LER), a critical challenge of EUVL.
As a leading litho expert, I am confident that Chris and many others in the 193nm litho community can continue doing much to help EUVL, especially in the areas of LER and EUVL resist development. I look forward to our becoming “one big, happy family” of litho experts as we work together on the latest optical lithography that will take us to the end of the Roadmap.