ASML Holding N.V. (ASML) Earnings Call Transcript & Summary

Welcome. You're live.

Thank you very much, operator. Good afternoon, everybody. My name is Andrew Gardiner. It's my pleasure here at Barclays to cover the European technology hardware space that, of course, includes ASML. And we have ASML as a regular participant at the Barclays Tech Conference. And I'd just like to thank Skip Miller, Global Head of Investor Relations for ASML for joining us once again.

Thank you, Andrew.

So Skip, if we could dive straight into some of the key topics for ASML at the moment. With your third quarter results, you gave us a fairly clear outlook into 2021, the low double-digit revenue growth you've talked about. Could you start by just walking us through the key end markets and what growth you expect within those end markets? And why sort of -- how next year is shaping up between logic, memory, installed base management?

Yes. Yes. As you noted, we -- for 2021, we did state in our October call that we expect another year of low double-digit growth. We also said that we expect EUV revenue -- system revenue growth around 20% next year. I think if you look at the ongoing transformation in the digital infrastructure, and I think you could argue, it's accelerated even as much as we dislike it and are so ready to get -- move on this COVID world as we're talking, I think you can see that it clearly has accelerated. And you see this in addition to some of the secular end market drivers such as 5G, AI and high-performance compute. They'll continue to fuel this demand for advanced process nodes. And this impacts both logic and memory, which obviously drives demand for our products. So if you look at how we see -- obviously, it's too early to provide too much detail on our guidance, but we expect logic demand, as it looks today, will remain healthy. We expect memory recovery will continue. As we mentioned in October, we're already starting to see -- we expected that to begin in the second half of this year, we could say, a bit stronger in Q4. And we expect that will continue into next year. So I think you put that together with the addition of the continued growth of the installed base, which the service portion of the installed base grows in a pretty predictable fashion with growing installed based tools into the field, the upgrade component is a bit more variable. And so that one, I think, we'll have to see the strength of the upgrades, again, when customers are willing and able to take the machines down to do the upgrades. And we'll get a bit more knowledge of that as we get closer to the year. So I would say we'll provide probably a bit more color, not only on the 2 market segments, logic and memory, but also our installed base view as we get to January next year.

Okay. The -- also, you've talked about -- I think Peter, your CEO, talked about this idea of communicating vessels between the different buckets of spending from your customers. Clearly, one customer in Intel is having a bit of difficulty in terms of its process node migration. Can you just sort of help us understand sort of what you guys have been saying around this idea of sort of derisking certain elements of the outlook into next year? And what might change depending on some of the decisions within your customer base?

Yes. So yes, we did have a -- we don't like to talk customer specifics, but we did have a customer that delayed the timing of their road map, which translated to systems that were needed later in time. So this reduced some demand next year. Again, I don't -- we don't want to get into speculation. There's different scenarios that could play out. They have indicated that they will talk more about this in Q1 in terms of their strategy as relates to external versus internal manufacturing. But regarding the different scenarios, the end market doesn't change in terms of the demand. So the wafers have to be built somewhere. I think Peter in the call was referencing this communicating vessels in relation to wafer demand, meaning that if there were to be the case where there were lower demand internally or lower manufacturing internally, that wafer demand is still being in place would have to pop up somewhere. There are probably some separation in time, because true communicating vessels happen real time. So his point was, it could be a few quarters gap between the lowering of the internal manufacturing versus the increasing external, if that were to be the scenario. So that was the point there is the wafer demand doesn't go away. Where and how it's built may change depending on the strategy. And if so, you have this communicating vessels scenario at play.

Okay. And in terms of the foundry side of the business, there's -- so a question over the sort of rate of node migration. What are you seeing at the moment into '21 and '22? Is that -- are you seeing any let up in terms of the intensity with which that part of the customer base are approaching the technology migration?

No. I think if you look at the advanced nodes pretty -- overall, you'd say they're quite aggressive. Our customers, our customers' customers are moving to these advanced nodes. And I think if you -- we talk about what transpired over the course of the year, we had the impact called the geopolitical impact that reduced the demand from one of our customers' customers, that due to being blocked from being able to take wafers. That created a void, if you will. And so initially, they had to -- because of this had some adjustment to their view in terms of demand as it relates to next year. The void, I would say, at these advanced nodes, if you listen to what our customers are saying, it appears to be backfilling at a faster rate than maybe expected as more customers are moving into these advanced nodes. And I think you could even say, based on the progress that's being made at the next node, industry [ 3 and 3 ] that you see that there could be even further move to these advanced nodes. So I think it's filling up quite fast, which is having a positive impact on the wafer demand viewed at our customers, which could in turn mean a positive impact on EUV demand as it relates to next year. So we'll just have to see how that unfolds. But the momentum is clearly, I would say, headed in a positive fashion at it relates to logic demand in particular for the foundry customers.

Okay. Perhaps we could switch gears and discuss memory, in particular, DRAM, which is, of course, important to you guys a bit more so than NAND. Where do you think we are in terms of the recovery at the moment? What can you guys see within the market in terms of tool utilization that might point to a return to stronger spending for that customer group?

Yes. So I think on memory, again, at the start of this year, we kind of saw that we expected a recovery in the second half of the year. And as I mentioned earlier, I think we're seeing that with the start really strengthening in Q4. I think the other part that as it relates to sustainable recovery, there's a lot of questions around inventory. And I think if you look at what our customers are saying, they're sorting that out now. So we're returning to more, call it, Normal levels. And the other thing is there was also what will next year look like in terms of a bit demand. And I think if you listen to what our customers are saying there, they are expecting to see a strengthening bit growth next year, talking on the order of 20% for DRAM, which obviously is the biggest driver for litho demand for us. And then the components to support what that means in terms of our demand, if we look at the litho tool utilization that peers have been talking about over the course of the year, coming from an underutilized state to increased utilization and starting to hit this, call it, max lever or ceiling that we're starting to see now, it means that any additional demand needs to be created by additional systems out there. So I think first thing you have to keep in mind is the -- what we're shipping this year in Q4, by the time you install, qualify and actually have all 5 bit output, you're talking towards the middle of next year. So you do have that factor to keep in mind. And so if you look at that from our perspective with a net growth of 20% or higher, you can't get there just on technology transitions alone likely. So you'll need some additional capacity. And so we'll expect that we'll continue to build into next year and see how these bit demand growth continues. But we expect it'll be another stronger year in memory. And we'll bit more color around that how strong exactly do we see it when we talk more about this in January.

You mentioned the geopolitical issues that face some of your customers or customers' customers. For you guys, how has the trade war [Audio Gap] sales into China. SMIC, for one, has obviously been in the headlines in terms of sort of NIRA restrictions coming through. But even beyond that, what about the other sort of newer sort of classes startup, foundry and memory customers in China. There have also been some reports about financing difficulty some of them have been having. So I'm just -- how has that changed over the course of 2020?

Well, I think 2020, if you look at what we said at the start and how we finished, I think things have, you could say, surprised or unsurprised, I don't know. But the -- I think have come out even through all the COVID, all through the geopolitical. The year has largely turned out as we saw at the start of the year. And I think that's also true with respect to China. If you look at China, last year, business-wise, the domestic China, we had around EUR 800 million of business. This year, we expect to be over EUR 1 billion. This year, it's more heavily weighted towards logic. Next year, we expect that mix to shift to more of a memory-driven demand. So I think that's one thing to state. And again, you look at these customers in China, and they are just executing to milestones. They continue to follow their long-term plan. They provide purchase orders and then payments are received. Regarding regulations as it relates to China, we can continue to ship DPV systems from the Netherlands. We require an export license for parts that are shipped directly from the U.S. to customers that are affected by some of the rule changes as it relates to the military end use at the end of June. But in general, we've been doing business in China for 30 years, and we continue to support our customers, and we'll continue to follow the regulations that are in place as it relates to shipments. But yes, growing opportunity and continue to be a good business in China.

Okay. Perhaps we could shift to some of the technical side of things, in particular EUV. Can you give us an update as to where things stand from your side in terms of tool development? 3400C, you've had a fairly good year this year, not quite hitting the output, but pretty close in terms of the number of tools. And then what we can expect for next year? And in particular, with the migration to the 3600D in the second half of the year, what is that sort of tool migration giving your customers in terms of improved performance?

Yes. Okay. Yes. So I think confidence is clearly improving as we continue to drive improvements in EUV performance, both the raw throughput of the machines as we go from the B to C and then plan for the D next year, but also availability and there's still a focus on availability as we continue to drive that number higher to the 90% and beyond. I think this confidence is also true as it relates to the ecosystem. When I say ecosystem, I'm talking about photoresists, masks, pellicles, inspection tools. And if you look at the different conferences, the industry conferences where customers have provided their assessment of the ecosystem, you see these stop light charts that are presented that were quite a bit of red in yellow a few years ago are largely green now. So I think that's, again, the important metric, because of how our customers view it. And it's making a great progress there. So there's still -- I think if you look at future nodes, there's still evolutionary improvements required that are happening. I think today, the industry views at a pace that will meet these future nodes. And I think, as I mentioned, the availability piece, I think, our focus is driving on availability to not only the average up, but also tightening the distribution such that we go to 90% and beyond. As a reminder, availability on DPV systems are more towards the 95% type range. So we still will make some progress there. And we have a road map that we work with our customers on to drive that number. With respect to the 3600D improvements, the value that the 3600D provides over the C is improvements in customer value along the lines of imaging overlay and productivity. And we're still planning to ship that machine starting in the middle of next year.

Okay. And that in terms -- to you, in terms of the ASP uplift, that's on the order of 10% to 15%?

Correct, depending on -- yes. Exactly, depending on the reference point from the C.

And you mentioned the improvements elsewhere within the ecosystem that is then helping to give your customers increased confidence in adopting EUV. What is the latest in terms of the layer count, both in foundry as we move from industry 7 to industry 5 over the next couple of years? And then with DRAM, as you're moving through the different 1z, 1-alpha. So what the customers are telling you in terms of the layer count improvement?

Yes. I think as we -- as I just noted, EUV technology continues to mature, again, primarily with improvements in productivity and availability. But obviously, we're making improvements in imaging and overlay as well. We continue to see the increasing number of layers adopted on future nodes, both in logic and memory. Memory just recently started to pick up as we go. I think we commented in our Q3 call that if you look at the industry 7-nanometer node with 10 or more layers, that we see the industry 5-nanometer node roughly doubling or, say, over 20 layers as you move to that node. So that's obviously a step in the right direction and continue to be for the logic players. I think if you look at future nodes on ALT, you see that continuing. And then eventually, High-NA comes into play as well. But this obviously help drives the shrink, and it's still doing this at a cost-effective reduction in terms of cost per transistor, cost per bit coming down over time. In memory, I think the first mover to use EUV has been quite public in their communication in terms of what they're planning. They're starting with the 1z node and using it more as one layer and use it kind of as a learning node. And then they'll plan to, as they stated publicly, to fully deploy EUV in high-volume manufacturing at the 1-alpha node. And so if you look at the timing at which they plan to ramp that node, this translates to taking EUV machines starting in 2021, starting next year. And obviously, what drives this is around the value, the value that EUV provides. One, you can just do a cost analysis and see that now that we're at these higher productivity levels, how that compares to multi-pattern, the advantage that comes through there. But also, I think if you look at the value that comes in the form of process simplification, that translates into device performance and yield as well as cycle time. I think that's really the driver from our customers' standpoint and our point of view.

Okay. You mentioned the eventual move on to High-NA, EUV High-NA from the current Low-NA tools. Where are we there in terms of development? And where is ASML's degree of confidence in terms of getting the R&D tools out along the time lines you've talked about and then moving on to high-volume manufacturing around the middle of the decade? It seems like key element, of course, is the new optical system coming from ZEISS. So where are they and yourselves in terms of development? And what's your degree of confidence in getting that on the current time line?

Yes. So I think we've been making good progress on High-NA. I think the -- we've talked about the major areas of innovation required in High-NA. We talked about the source. We talked about stages. And we've talked about the optics of the lens. I think maybe first, if you kind of -- we'll walk through each of these. I think the source -- the part of the source is that different from Low-NA, we had to develop everything from scratch, basically, or working with eventually acquiring Cymer, is that we now are utilizing that same type of device architecture that they call LPP, laser produced plasma, source and just scaling that. So not that there isn't still a lot of engineering ahead in that front, but the scaling capability and the feasibility that we've demonstrated, I think, give us confidence that we know how to scale that source, and therefore, believe we can deliver on that front. On the stage, obviously, we want to continue the productivity at a high level on the High-NA machine, continue to -- where we have left off the Low-NA. And obviously, this is an area of ASML's expertise in stage technology. And we've already started working on protos. It gives us confidence that we can deliver on this aspect of as well. And then maybe lens, like we talked about, is probably the biggest change of, at least, say, the most extensive development required as it relates to High-NA. And again, the front -- the challenge on this is that you have to scale the size of some of these optics on the order of factor of 3 or 4. And you have to do that while maintaining a very, very flat surface of those mirrors. And one, to try to be able to manufacture this, but you have to be able to measure it. And to be able to measure it, we had to develop with our partner, Carl Zeiss, an interferometer that can actually measure at the picometer level, this type of flatness. And so that's something that we have successfully put together. It's working out, measuring. It's still -- obviously, there's room for improvement, but the point being that we demonstrated a milestone last year. We're able to start manufacturing these optics. And then we'll have to start piecing all this together. And we're still confident that we can deliver to the initial development systems in the '22-'23 time frame and then high-volume systems in the 2025 system -- time frame. I think the other area that we talked about also in Low-NA is the ecosystem. And I think that, on the ecosystem front, what's again, call it, you could say, makes High-NA a lower risk introduction is that you are -- basically, you need evolutionary developments in this technology. You can reuse a lot of the resist technology, along with, obviously, improvements. The same goes through with the pellicles and the mask and the inspection arena. And so that takes some of that risk away where as opposed to when you're first introducing a wavelength, you have to develop all this technology, and that takes some time in addition to take the momentum and get the industry behind it. So I think put all that together, we're still quite confident that the timing -- we're on track for the timing that we discussed. And the customer motivation, obviously, is there to keep a void, you say, a multi-pattern in EUV.

So you mentioned some -- being on target to hit high-volume manufacturing or delivering of the High-NA tools in 2025. ASML has clearly given a 2025 model. You first put that out about 2 years ago. I know at the time, I sort of thought the -- some of the scenarios you had given, particularly the low market scenario seems pretty conservative. Of course, at the time, 2 years ago, we were in the flows of, say, a mild inventory correction for the industry. We've come through that. We're now weathering the pandemic. The industry is in a pretty strong state, all things considered. And so do you not think that, that scenario, in particular, at the low end of the range seems pretty unlikely now where you were calling for a 20% decline node on node on node in terms of wafer start?

Well, short answer is probably yes. The -- a bit more detail around what in reference to 2025 scenarios. As you know, we had one sensitivity to technology. We had another set of scenarios that were more, call it, end market sensitivity type numbers. And if you looked at those 2, there's much less sensitivity comparing low or high on the technology or you call it, the EUV adoption scenarios compared to the end market scenarios. I think if you look at the end market sensitivity on logic, for example, it was linked to the assumptions on the node size. And you had -- the low scenario was shrinking node sizes of 20% node-on-node, starting with the 16- to 14-nanometer node. The high scenario was basically keeping it flat. And obviously, the middle would be a 10% node-on-node reduction. I think if you look at where we are, talking earlier about these end market drivers and the strength in demand of these advanced nodes, it probably is a bit conservative to assume that the node sizes will shrink significantly as some of these scenarios would suggest, which is where the low market, I was talking about 20%. We will have or planning to have our Investor Day on June 23 next year in London and where we will expect we will update these 2025 scenarios based on our current view.

Sounds good. Well, here's hoping that's an in-person event, yes, by that point next year. But for now, I can see that our time is up. I have countless further questions, but I think we'll have to call it a day for now. So thanks again. Thanks, as always, for attending, Skip. Much appreciated.

All right. Thank you, Andrew. And yes, look forward to seeing each other in person, hopefully, at Investor Day, if not before then. Take care.

Thank you.