Disruptive Changes Ahead For Photomasks?

Experts at the Table: Semiconductor Engineering sat down with four experts to explore the current state and future direction of mask-making, with insights from Harry Levinson, principal lithographer at HJL Lithography; Aki Fujimura, CEO of D2S; Ezequiel Russell, senior director of mask technology at Micron; and Christopher Progler, executive vice president and CTO at Photronics. What follows are excerpts of that conversation. To view parts one and two of this discussion, click here and here.

L-R: HJL’s Levinson; D2S’ Fujimura; Micron’s Russell; Photoronics’ Progler.

SE: Mask cost has historically been a major concern, but it seems like it’s now more accepted at the leading edge. Will that trend continue? And what can companies do to manage or reduce mask costs going forward?

Russell: To be honest, I don’t know anyone saying masks are cheap or insignificant. At the leading edge, the cost of wafer manufacturing has skyrocketed because of the increasing number of process steps, tool complexity, and advanced equipment. Maybe in relative terms masks look like a smaller portion of the total cost now. But the number of masking layers per device is going up a lot. And with EUV, you also have shorter reticle lifetimes, so you have to replace masks more frequently. That makes them a recurring cost, not just a one-time investment. So while the cost of an individual mask might look smaller compared to the rest of the process, the total cost of a full mask set — and the need to replenish it over time — is still very significant.

Progler: I don’t think anyone has really accepted today’s mask costs. It’s still a huge concern, especially in EUV. The problem is that there just aren’t many options. That said, for our customers, speed often takes precedence over cost. If you’re going to spend that much money on a mask, you want to get it quickly. And it better work right so you can get your chip to market faster. So we’re having more conversations about speed than sticker price. But yes, there are strategies for reducing cost. That includes improving yield, driving down materials cost (especially for EUV blanks), and leveraging computational tools to reduce experimental waste and converge solutions more quickly. These are small optimizations, but collectively they help. It’s the same pattern we saw in earlier nodes — 45nm masks were over a million dollars per set when they launched. They’re far less now. EUV will follow that trajectory over time, but we’re still early in that curve. And one observation on EUV, nothing drives costs down like options and competition. So to the extent that cultivating a viable merchant EUV mask supply is supported by the user base, many of whom have captives, it will have the effect of driving the costs down across the ecosystem for all users. It has happened with every prior node, and EUV would be no different.

Levinson: A lot depends on what kind of product you’re making. Moshe Preil, who’s now at Zeiss, made a good point. He said that for a highly complex logic design, your design costs might be hundreds of millions of dollars. In that context, a $10 million to $20 million mask set is just a tax on bringing the product to market. Now, no one likes paying taxes, so people will always complain. But if the product has high enough value, the mask cost is tolerable. It’s a different story at the low end, especially in automotive or consumer markets where price sensitivity is extreme. There, every cost matters, including masks. The design costs are lower, so the mask costs represent a much larger portion of the total investment. We’re always working to reduce costs wherever we can. But right now, I wouldn’t say mask costs are crippling the industry. They’re significant, but manageable, depending on the use case.

Fujimura: I’ll offer a slightly different perspective. Eighteen years ago, when I first started in this space, saying ‘we should increase mask cost to improve wafer quality’ was practically taboo. You’d get attacked for even suggesting it. But that mindset has changed. For example, I recently heard someone say the cheapest and most effective way to improve wafer CD uniformity is to improve mask CD uniformity. That makes sense. Any variation on the mask gets repeated across every wafer, so improving the mask improves all downstream performance. We’re also seeing this with curvilinear ILT. Yes, it increases mask cost because it increases computational load, but it expands the wafer process window, which improves yield. Frank Abboud at Intel has said many times that the mask community should never stand in the way of better wafer manufacturing. If better wafer results require a new mask technology (which will likely be higher cost, at least in the beginning), then the mask shop will figure it out. One example is large-format masks — 6 x 12-inch masks for high-NA EUV. These are expensive and require massive retooling, but they’re getting serious attention. If they help wafer performance, then they may be worth it.

SE: High-NA EUV will introduce new capabilities, but also new requirements — including a potential shift to 6 x 12-inch masks. How disruptive would this be, and what impact could it have on the broader ecosystem?

Progler: If the industry adopts a 6 x 12-inch mask format, that’s going to be a significant disruption. It affects everything, from the way you manufacture the mask substrate to the mask writers, etchers, and inspection tools. Basically, every tool involved in making masks would need to be redesigned or replaced to varying degrees. If that change happens, it’s going to ripple through the entire supply chain. Aside from 6 x 12, high-NA also drives new mask requirements for anamorphic exposure, stitching and specifications on pattern fidelity, films and substrates such as flatness.

Levinson:High-NA brings with it a few challenges. One of them is sub-resolution assist features (SRAFs). Recent work has shown that with high-NA, SRAFs could start printing at around 4.5nm on the wafer, and printing an SRAF is not acceptable. To prevent that, we may need tighter control on mask resolution — below what most mask shops can currently deliver. That’s one aspect where high-NA may force the industry to upgrade capabilities. Another is customization of the absorber stack. If we need to optimize different blank materials for line-space versus contact layers, that would be very disruptive. Right now, we rely on a single blank for most layers. Moving to a multi-blank strategy would complicate everything, from blank manufacturing to etch and metrology.

Fujimura: Curvilinear is already a big change, and if the large-format masks come into play, that’s going to be even bigger. It’s not just the cost, although the cost is very high. It’s that the entire mask infrastructure would have to change. So yes, if the industry decides to go that route, it will be a major shift.

Russell: Let me add something that raises the stakes even further. ASML has announced that their new scanner platform will be a unified platform, meaning that if they adopt large-format masks for high-NA EUV, they’ll likely back-port that format to low-NA EUV, as well. That would have huge implications. Instead of having a split infrastructure — 6 x 6 masks for low-NA and 6 x 12 for high-NA — you would have to move everything to the larger format. From a mask shop’s perspective, that means outfitting the entire line for large-format tools, doubling up on equipment, and taking on a massive capital burden. But it also makes sense economically. If you’re going to invest in that level of tooling, you want to amortize the cost across more masks, not just the high-NA ones. Using large format masks will improve the EUV machine productivity (wafers per hour) for both high-NA as well as 0.33NA EUV tools by doubling the exposed field in one scan, resulting in lower cost and increased throughput for high-volume manufacturing. This productivity increase will come at the expense of more complex and expensive masks. Large format EUV masks could create a real divergence in the industry. For companies not ready to adopt high-NA soon, the avoidance to switch to large-format could be a dead end. They might not get access to the latest scanner generations if they can’t justify the infrastructure changes in the mask shop.

SE: Looking ahead five years, what do you expect to be the most disruptive change in mask technology?

Progler: We already discussed 6 x 12 so I’ll not repeat that here. There’s an interesting debate right now about mask resolution. We probably have more resolution than we need today. The question is whether future patterning modalities will require more, or if the mask making resolution roadmap stalls out, similar to when twin stage and 193i shifted scanners to productivity as the differentiator. If higher resolution no longer adds value at the wafer level, that would be a disruptive mindset shift. It means mask suppliers will need to find other ways to differentiate beyond just driving resolution. Also, the ability to accurately predict and optimize lithography outcomes in the absence of detailed experimentation, from design to final wafer, has the potential to disrupt how we do everything. We’re getting closer, but we’re still not fully there. With this, injecting AI into the mask flow has high potential. Masks have always lagged wafer and design on new developments, but we can learn from them and embrace opportunities. With that, broader adoption of high performance computing enables a model drive photomask ecosystem. If we can convince customers it’s secure and it reduces risk, then that unlocks enormous potential.

Levinson: If the 6 x 12-inch mask format gets adopted, that absolutely will be disruptive. Every part of the mask-making process would be affected, from growing the larger mask blanks to redesigning all the tool platforms. Widescale use of curvy features could be disruptive, because it affects many parts of the mask-making process — data formats, data preparation, defect inspection, and metrology. I hope that we aren’t just seeing the tip of the iceberg. And if we end up needing to customize the absorber stack — say, different materials or thicknesses depending on whether we’re printing lines, spaces, or contacts — that would disrupt the standardization we currently rely on. So there are definitely major changes ahead.

Fujimura: Curvilinear is going to continue being a big change, and it’s not done disrupting us yet. If the industry really does adopt large-format masks, that will be a very big deal. It will change everything, from the economics to the tool chain. It’s not a question of whether it’s technically feasible. It’s whether the benefit at the wafer level is large enough to justify the massive investment in the mask infrastructure.

Russell: I agree with everything that’s been said so far, and I’ll add one more key point. ASML’s next-generation scanner platform will use a unified format. So if they adopt large-format masks for high-NA EUV, they’ll also apply that same format to low-NA EUV. That creates a huge impact for everyone, not just the early adopters of high-NA. Suddenly, everyone using EUV would need to transition to large-format masks. For mask shops, that means duplicating your entire line. It’s not just the capital cost. It’s the operational complexity, the inspection tools, the etch tools, the writers, the metrology—it all changes. It also creates a potential divergence. Companies that aren’t ready for high-NA may find themselves stuck. If they can’t support the large-format mask infrastructure, they may be locked out of the next generation of scanners. That’s a big risk, and it’s something the industry will have to navigate carefully.