Expert Panel Sees History Of Continuous Photomask Innovations As Key To The Future

The eBeam Initiative conducted its 14^th annual eBeam Initiative Luminaries survey in July and reported the results on September 23, 2025 to more than 200 attendees at its annual meeting during the BACUS SPIE Photomask Technology conference. Industry luminaries representing 51 companies from across the semiconductor ecosystem—including photomasks, electronic design automation (EDA), chip design, equipment, materials, manufacturing and research—participated in this year’s survey. A panel of experts provided insights during the meeting on some of the key technologies covered by the survey, including high-NA EUV, EUV pellicles, and curvilinear masks. The panel was moderated by Aki Fujimura, CEO of D2S – the managing company sponsor of the eBeam Initiative, pictured on the far right below, and included (pictured left to right) Jim Wiley, Wiley Strategic Solutions, Chan-UK (CU) Jeon, CTO of Tekscend Photomask, and Harry Levinson, HJL Lithography. In this blog, I’ll report key insights shared by the panel regarding the findings in the 14^th annual eBeam Initiative Luminaries survey. The full video will be available at www.ebeam.org in November.

Photo: eBeam Initiative panel discussion at the 2025 BACUS SPIE Photomask Conference.

eBeam Initiative (Fujimura): Most of the luminaries in 2025 are positive that photomask revenues will increase in 2025. SEMI reported $5.561B for 2024 photomask revenues and predicts a 5% year-over-year growth for 2025 in their annual report available for purchase at www.semi.org. It’s hard to put a number on the photomask revenues in captive mask shops where most of the more expensive EUV masks are manufactured. Is it possible that the photomask revenues are underestimated?

Levinson: Initially, single exposure EUV replaced multiple exposure optical layers, so there might have been a slight dip in the number of masks made when EUV lithography was first introduced. However, since EUV masks cost around 6x that of optical, the photomask market should be up in terms of dollars. We’re now seeing significant increases in the number of EUV masks made due to a) increases in the number of layers exposed with EUV, b) need for multiple EUV masks because of the need to clean and qualify, and c) short EUV mask lifetime.

Jeon: I used to manage a captive mask shop. [Mr. Jeon was previously Vice President and Team Leader of the Mask Development Team at Samsung Semiconductor R&D Center.] The majority of EUV masks are manufactured by the captive mask shops. Captive mask shops don’t approach mask pricing like the merchant shops. Nobody knows the mask price for super-advanced development products. However, I do feel the captive market is somewhat underrepresented as there is strong growth happening at the leading-edge nodes which are using EUV masks.

Wiley: As wafer steppers and scanners became the primary pattern transfer tools, photomask sales equated to about 1-2% of integrated circuit (IC) sales. EUV masks have raised costs dramatically, however, the 1-2% ratio barely budges because mature node fabs still use relatively inexpensive masks. At the bleeding edge, mask costs are now a dominant fraction of non-recurring engineering (NRE) costs and can distort economics for new entrants, but the macro ratio doesn’t change much.

eBeam Initiative (Fujimura): For EUV photomasks, pellicle adoption has been slow for multiple reasons. But without pellicles, the lifetime of an EUV mask is less than the lifetime of a 193i mask with a pellicle, according to the survey results in Figure 1 below. Further, 57% of the survey participants say pellicles would increase EUV mask lifetime up to 3X. What do you think the ramp for EUV pellicles looks like?

Wiley: EUV pellicle adoption seems to be increasing with the news that TSMC is converting one of its old 8-inch fabs to produce their own EUV pellicle membranes.

Levinson: EUV pellicle transmission is a problem. Wafer throughput is affected by the loss of energy with today’s EUV pellicles. Loss in wafers per hour (WPH) isn’t acceptable for these expensive EUV machines. It will be even worse for high-NA EUV machines. So significant improvement in transmission will be required for pellicle use to expand as fast as it did for optical lithography.

eBeam Initiative (Fujimura): CU, you’re unique in having headed up mask shops for memory as well as logic. What do you think?

Jeon: First, memory and logic have different pellicle strategies. Memory is clearly less aggressive because productivity is the most important metric. Pellicles improve mask lifetime, as the survey says, but also have unintended risks to manufacturing today. When we adopt EUV pellicles, even more development items will follow. Carbon nanotube (CNT) pellicles may be the turning point because productivity loss is minimal, especially for memory makers.

Fig. 1: 2025 eBeam Initiative Luminaries survey – Pellicles impact EUV mask lifetime.

eBeam Initiative (Fujimura): Let’s talk about high-NA EUV and the survey result pointing to the need to produce sub resolution assist features (SRAFs) of 15nm or less in Figure 2. What’s driving this, and is this even possible today?

Fig. 2: 2025 eBeam Initiative Luminaries Survey – Minimum SRAF size for high-NA EUV.

Levinson: The first thing I heard at the SPIE Advanced Lithography conference back in February was the challenge to achieve depth of focus (DoF) for the early installations of high-NA EUV machines. I expect that curvilinear inverse lithography technology (ILT) will be a practical necessity for high-NA EUV to solve the DoF challenge. Curvilinear ILT leads to extensive use of SRAFs so we’re going to need tiny SRAFs for sure. It becomes an engineering problem. We can solve this one, too.

Jeon: The resolution limitation for masks comes from the eBeam mask writer and the mask resist characteristics. Both have continuously evolved. Combing today’s 4^th generation of multi-beam mask writers with the latest resists, we can make 15nm SRAFs. We can push the resist sensitivity to the extreme, but it takes a lot of time to expose the mask. That ultimately comes down to a trade-off between productivity and performance.

Wiley: Assist features went from a clever trick to an industry standard within a few years. Their adoption forced upgrades across the entire chain, most dramatically in mask writers, and set the stage for today’s computational lithography and curvilinear mask era. Given this history of mask innovation, I’m sure we’ll be able to deal with smaller assist features for the next generations of EUV.

eBeam Initiative (Fujimura): In the rapidly changing photomask industry, the shift from conventional Manhattan masks to curvilinear masks marks a pivotal development. In the eBeam Initiative Luminaries survey, we’ve asked participants to rank the challenges to manufacturing curvilinear masks and if they can be manufactured (Figure 3). Luminaries have consistently said for five years that at least a limited number of curvilinear masks can be manufactured. What are the benefits of shifting to curvilinear masks?

Fig. 3: 2025 eBeam Initiative Luminaries Survey – Confidence remains high in manufacturing curvilinear masks.

Levinson: Well, it’s process window on the wafer. Day one with curvilinear inverse lithography technology (ILT), early papers showed that when you go from just Manhattan shapes to more curvy shapes, you get a bigger process window. Even with dry ArF lithography, curvilinear ILT is of interest for extending technology nodes.

Jeon: Curvilinear ILT gives an advantage to chip makers that have a captive mask shop, because OPC/lithography and mask communication is a very important part of this technology. That’s why hotspot-focused curvilinear ILT is already used in captive mask shops and seems to be spreading to the merchant side as well. Of course, freeform full-chip curvilinear ILT takes a bit more time for both.

eBeam Initiative (Fujimura): Jim, what’s the difference between hotspot-focused curvilinear ILT and what CU called freeform full-chip ILT, i.e., Entirely Curvilinear ILT?

Wiley: There’s a good analogy here. Comparing hotspot curvilinear OPC/ILT to full chip is like building or fielding a sports team with hotspot correction. You’re taking out your worst player, the one who’s clearly costing you winning the games and replacing them with someone who’s excellent in just one part of the field. That move keeps you from losing the big games, but the rest of your team is still average, and the weaknesses remain. If you go to entirely curvilinear OPC/ILT, it’s different. You’re upgrading every player on the team. You’ll be sure there are no bad players. Every position is stronger in the team. It’s balanced. Performance becomes consistent. Hotspot correction fixes the glaring problems, but entirely curvilinear OPC/ILT systematically raises the baseline everywhere, leading to a broader process window and higher yield.

eBeam Initiative (Fujimura): Thanks to our panelists this year and to all who took the 2025 eBeam Initiative Luminaries survey. You can find the complete results at www.ebeam.org.