KLA Corporation (KLAC) Earnings Call Transcript & Summary

Great. We'd like to get started. Good morning, everyone. Thank you so much for coming. My name is Toshiya Hari, I cover the semiconductor space as well as the semi cap equipment space. Very honored to have the team from KLA with us this morning. We have Bren Higgins, Executive Vice President and CFO. We also have Ahmad Khan, EVP and President of the Semiconductor Process Control segments. Bren, Ahmad, thank you so much for coming.

Yes. Thank you for having us.

I'll go through a bunch of questions. We'll try to keep it open. So we'll definitely take questions from the audience as well.

Do you want me to just do some level-setting comments here.

Please.

Okay. First, let me just take a second to talk a little bit about Ahmad. I'm happy to have him here. I mean Ahmad had a longstanding history in the company. I know a lot of investors know him. We've recently made some changes in our org design where Ahmad has responsibility for the broader systems business across the company. Really driven by changes in our customers and the markets in terms of how the backend is now merging with the front end, and that translates into advanced packaging and a lot of advanced packaging opportunities, which we'll talk about. And so I think in KLA, we've always been pretty proactive in terms of aligning our go-to-market with our customers' focus and strategy and also people. So we did some things to drive some better effectiveness in our channel, as it relates to our specialty semiconductor business, which is our process business that has meaningful exposure to advanced packaging. So I think those are some of the changes. And of course, now Ahmad has a little bit more of end-to-end ownership across the company as it relates to how we engage in that part of the business, which, as we talked about in earnings, is growing meaningfully over the last couple of years. across the company. It's about a greater than $500 million opportunity in 2024 versus about $300 million last year. So meaningful growth in the overall. And certainly, a bigger factor strategically for our customers as they think about larger die in integrating devices and heterogeneous packaging, particularly to support high-performance compute markets. Overall, 2024 is been a transition year for the industry. We're continuing to see some legacy investment. We're starting to see more leading edge investment here in the second half of the year, and we think that translates into what we think will be a pretty good expansion environment into next year. The company has performed from a relative performance point of view pretty well over the last few years, as we've seen scaling and design starts and its focus on the leading edge of trying to manage through a much richer product mix environment. And then we'll see that. We've seen our customers be pretty -- primary customer be pretty bullish about like the opportunities in the node, the pace of the ramp, the design starts and so on. Memory, I think, is more of a '25 opportunity. In '24, we're seeing our customers' business fundamentals improve, and we think that will ultimately translate into more investment into next year. So WFE levels roughly around mid-90s or so this year. And again, a view that we see that grow into next year. Business model seems to continue to perform well. We talked about a March quarter bottom sequential growth through the year. We still feel very good about that. Gross margins in what has been a flattish year, somewhere in the mid around the mid 61% range. And the operating model focusing on 40% to 50% incremental operating margin leverage on revenue growth is performing in line with the model this year. And certainly, as we see more sustainable growth moving forward, opportunities to continue to drive KLA expected leverage in the model over time. So sure we'll dig into some of these opportunities I talked about, but just want to take a minute and frame the environment.

No, I appreciate that. Thank you. Thanks, Bren. Maybe kicking off with thoughts on WFE. You did go through your '24 expectations. You touched a little bit about -- a little bit on '25 as well. I know it's early, but as you think about '25, I think you've given directional guidance. How are you thinking about the various applications? Again, I know you touched on that, but if you can expand on leading edge where things are healthy, trailing edge, DRAM into '25, that would be helpful.

Yes, sure. I mean one of the good things about over the last few years, we've seen KLA share of the market grow. From 2021, we were about in the low 6%. I mean everybody does the math a little bit differently, but it's a low 6% into the mid-7% range. And over the last couple of years, we've been dealing with that with very little leading-edge focus, right, a lot of legacy investment. So as we move forward, we're really encouraged by opportunities at the leading edge. Our strategic goals inside the company right now are preparing for growth as a leading edge because we know that, that drives a very different process control model for our customers. And so as we move into next year, I think there's some competitive opportunities across different players and how they're investing. But we're encouraged that we'll see some WFE growth across the different players based on sort of bottoms-up and top-down analysis. And so that will be a good driver. I think on the legacy front, I think in the investment levels in China are more or less stable. It could be up a little bit, it could be down a little bit. But in general, I think it's fairly stable as we look at next year. The other non-China legacy is corrected a lot this year. So I'm not sure exactly how to think about where growth is. But hard to see how it's down. So I think we feel pretty good about that overall. I talked about memory. DRAM, there's a lot of focus on utilization improvement across our customers. HVM devices are consuming some of that capacity. And so I would expect to see DRAM invest probably more next year. And then flashes operated at an extremely low levels. And I'm less optimistic about growth there, but given where it's currently at, I think it grows into next year also. So even as you have memory growth, I mean, typically, logic/foundry process control intensity tends to be higher than memory, and you have more memory, which will change the mix, maybe 5 points or so into next year. I think given the focus on leading edge and how customers invest there, some of the supply constraints around key products in the company that we're getting incremental supply. And so strong position in the market. I think part of the market that will grow faster, I mentioned packaging, which is also growing faster. I feel very good about our ability to sustain the, the relative growth that we've been seeing over the last several years into next year despite some changes in the mix. Over the long run, we think the mix is -- will be greater than 60%, 65% logic/foundry given expectations for growth in that part of WFE. And in any given period, if you have memory put some pressure on the process control intensity. But I think over the long run, we're well positioned. But even in the short run, given the dynamics I talked about, I think we're in a pretty good place in terms of our expectations for the business and how we'll perform relative to the overall market.

Awesome. Thank you. I wanted to ask about M2. I think TSMC has made very constructive, bullish comments as it pertains to -- what are your expectations from a design starts perspective? And how do you see that translating to your business?

So why don't I let Ahmad start, and I'll contribute if needed.

Yes. So working with customers is an area which I'm very passionate about. And we use a flywheel in KLA on collaboration, innovation and execution. Collaboration is really about working closely with the customers to understand all the inflections that are going to come into the marketplace. And then we innovate to solve the problems and then execute. So on the collaboration front, we look at all the inflections that are coming in. And as you know, for M2, there's a pretty large inflection in the transistor architecture change. You go from planar or FinFET to gate all around. And these are very difficult transistors to make. Very, very few companies have been able to patent them and make them defect free. It's a stack of silicon, silicon germanium stack. And then after that, you have these recesses where you could have very, very small defects like 20 nanometers to 5-nanometer defects that can kill the entire device. So we worked about three to four years ago, seeing this inflection. We worked with IMAX, we worked with IBM, we worked with TSMC and everybody else to determine what is required, and we developed two new systems to characterize it. One of them is our Gen 4 system. Actually, we released a new version of Gen 4, which is non-upgradable. And one of the reasons is because we needed to add a lot of bells and whistles for buried defects. In this case, get all around it a very defect. So we released a new version of Gen 4 system that came out in the marketplace, and we've been working with customers for a year. And we also released a new e-beam inspection system. As you know, KLA was somewhat absent in the e-beam inspection market, but we invested in this area, looking for the right timing to bring a system in, and we have brought in our first single-beam e-beam inspection system to characterize the get all around. So today, Gen 4 is able to give full wafer coverage for these buried defects. And when the defect size becomes far too small 1- to 5-nanometer, we're able to use our e-beam inspection system and find that defect. So we feel pretty comfortable that we can help support our customers do this interesting patterning. And this gets worse over time. I sometimes joke that even though the things are very difficult, we are in vacation time right now. Things that are going to get harder and harder as CFET and other architectures come where transistors are further layered up. So that's one inflection. The other inflection of course, is in packaging. And in packaging, it's a lot about interconnect density, and we see that very few defects can kill the entire package. So in the old days, your job was done kind of when the wafer was fully processed. And after that, you sell that wafer to a customer, you have to yield well. But today, your job is 70%, 65% then when you're done making the wafer because after that, you have to take these large dies and then integrate them on a silicon interposer and add HBM and there's a lot of losses there. So KLA is releasing a number of systems that were available in the front end and have the sensitivities to be able to catch those critical defects. And now we are making them film frame capable and releasing all those systems for packaging. And we have pretty good traction in logic because interconnect density is very important there and bringing those systems into the marketplace and see that. We see another inflection, which is really in HBM. As you know, the growth in memory in DRAM is primarily going to be in the HBM side. And in HBM, we see more process control intensity. One, of course, HBM DRAM die is larger because you need a TSV connection. So every time the die is larger, your defect density becomes more important. If you had a wafer with a 1,000 chips and one chip die because of one defect versus a wafer that has that let's say 20 chips, one wafer dies, you can do the math. So defect density becomes very important. The second is that the logic die is separate from the rest of the DRAM dies. You have to build your own logic circuitry and because of that HBM process control intensity goes up. And when you stack all of these, there's a lot of losses, so customers want to do a lot of inspections. So these are some of the things that we're doing with our customers working very closely to understand what the inflections are today, what the inflections are in the future to make sure that the systems and their road maps are aligned.

I mean over the last several years, we talked a lot about lithoscaling driving our business, right? And the introduction of more UV layers drive smaller defects, smaller pitches in terms of the distances between lines and spaces in a device. But also architecture changes are also another vector of growth. If you go back to when the industry transitioned from planar into FinFET devices, it was the last local high that we saw in terms of process control intensity. We talked about all the usual sort of small defect litho challenges that you'll still have it into -- from N3 Ahmad talked about some of the defect mechanisms that are now new use case in terms of opportunity that's driving incremental demand. You also have depositing more layers, you have more critical layers in metrology, voltage control becomes a challenge in terms of how they're structuring these devices. So there are lots of [ operating ] change is what's good for us and change in customer process. And as we look at N2, you've got a lot of the challenges we've been dealing with, which has been driving our business, but also these new opportunities related to the architecture change as well. The device environment, the N2 node is very power-friendly node, and we're not going to get into a number of devices, but our customers are clearly pretty bullish on the design activity that they're seeing today versus previous nodes at a similar stage. And expectations have been fairly bullish in terms of the pace of that ramp as they have more designs driving process qualification earlier in the process life. And so that also tends to be an area where they're managing a more variable market or a more variable process. And so given that how design rules are challenged in different ways across different devices. So it does create, we think, opportunities to see the business start to ramp to support that. And then given the incremental demand we see on N3 and the product cadence, particularly around HPC devices, consuming N3 capacity, the ability to take any of that capacity and migrate it to the N2 ramp is also compromised. Customers are always trying to be as efficient as they can with capital. But typically, what we call the reuse factor is likely to be lower given the expectations of the ramp for N2.

Got it. That's really helpful. Just as a quick follow-up. Given what you guys know about N2 and given your positioning as sort of the key provider of process control, I'm sure you've got visibility into what comes next, whether it be at TSMC or other leading edge logic/foundry suppliers. How do you think about process control intensity given N2 and what's to come? Is the slow positive? Is it -- you're already at a pretty high base? Are you thinking flat? Any thoughts on that?

Well, I'll let Ahmad maybe add some detail. I mean I'll start with -- as we see early indications as we do see, and I've said this to investors over the last few months is we are seeing KLA share market improving. Now there's a share element, but most of it related to process control intensity given the issues we've talked about at N2 relative to N3. So early indications are are pretty strong. All the factors we talked about are factors in it. But is there more to it.

So as Bren said. So from 3 to 2, we see intensity increase because the transistor architecture is very different. And to control that transistor architecture, you got -- you have to find these defects that were not very easy to find before. They're buried and they're recessed. So you have to have very innovative technology to be able to do it. We also changed the architecture of the systems to go from x86 to GPUs to be able to process a lot more data because you do defect detection and you get a lot of candidates for a defect, but you don't know which one is. So you have to process a lot of data. So we're changing to the GP architecture. We've been able to process a lot more data. That is one of the reasons why the Gen 4 system is not upgradable. So moving forward, as you go from N2 to A14, you bring in wafer-to-wafer bonding because the metalization layers become too high and you have an RC issue. So you bring a second wafer and you put that wafer and attach it to the front of the wafer to power the transistors from the back of the wafer itself. That is a whole slew of learning and solving new problems with bonding the wafers, the overlay control is critically important. The defect density is important because you have processed the full wafer now and you bond a second wafer and then you get that wrong, it's not very good. So intensity goes up just by adding a second wafer. Of course, if you add a second wafer, our Surfscan business does well because now you have to process two wafers at the silicon supplier, that is happening also in NAND flash, and it's also happening in DRAM in the future with 3D DRAM and an end wafer to wafer to wafer bonding will happen eventually. So silicon wafers would go up. So wafer-to-wafer bonding brings that additional complexity, which drives process control intensity while keeping everything else that we just talked about the N3 to N2. Then after that, you go to CFET. And CFET is basically more vertical transistors in finding those critical defects both optical inspection is going to be very important and e-beam inspection is going to be very important, and we are seeing those systems now to make those capable. While all of that is happening, the number of transistors per device is increasing significantly. If you looked at a GPU about five, six years ago, I don't quote me on the exact numbers, but 20-ish billion transistors. The latest one that is being talked about today is 200 billion transistors. You can't really make all of that very easily unless you stitch multiple devices, you have to do the stitching of multiple devices after you do the silicon processing on an interposer. Then after that, you got to bring HBM in, you've got to integrate that. The sizes of these devices is well above a reticle field. So you have to have multiple reticle fields to be able to stitch all those devices. Now people are talking about silicon bridges to reduce costs. All of this is process control intense outside of the fact that you still have to make the wafer. So process control intensity in the wafer is high and then all of those. Last thing I would say is in order to pack more transistors, you have to reduce the reticle pitch, you reduce reticle pitch. It becomes there's all sorts of other issues that comes into.

We left out lithoscaling, right? High-NA changes and which is the ability to collect life in the optical system.

Heading in the right direction.

You definitely conveyed the intensity.

There is intensivity.

So I guess the competitive arms race that you've seen at the leading edge logic/foundry space, all else equal, that tends to be a good thing, not just for you guys, but for the overall industry. There's one logic company that's going through quite a bit right now, and I'm sure you guys are getting questions as well. They've reduced CapEx publicly. Is this sort of a 0 sum where if one of your big customers is sort of declining in share, there's another one that picks it up. So net-net, you're all good? Or should we sort of consider this as a net albeit mild negative?

Well, we're certainly running the company that way. We're not going to talk about specific customers and you can get different customers all. We work with customers in different ways and how they spend on process control could be different given their strategies given the mix of their products and so on. In general, our belief is that there's supply to support the demand environment and that there isn't really, in the long run, any inefficient spend where people are going to add incremental supply beyond what's required long-term demand -- for long-term demand. It's key for us to understand some of the moving parts. You don't want to plan for both, right? So -- and I think our customer engagement is such that we have a pretty good indicator in terms of what our customers are planning for from a supply point of view, and it aligns with the demand expectations. So look, our view has always been that the industry is going to -- there's a growth rate for semiconductors that capital intensity continues to rise, probably drives a slightly faster growth rate in WFE because there's not any real drivers to drive intensity down. Mixed WFE logic/foundry WFE will grow faster over time than memory, as I said earlier. And so that, given relative process control intensities across the segment will enable process control to grow faster. KLA is positioned in certain markets, which are inflecting and growing faster. We talked about packaging, we talked about optical inspection, should enable KLA to grow a little bit faster than that. That's how we think about the company. That's how we size it, which is important to set expectations with our own teams and how they think about what kinds of opportunities they're chasing, how much we can spend. If you think you're chasing a 15% growth rate opportunity, you get a very different program budget than if you think you're chasing something less than that. So it forces our teams to make tough choices to really have an intimacy with customers to make sure that we're designing the systems and collaborating with them in a way to design the systems to meet their needs. And look at the industry ends up spending more, we've proven that we know how to ramp our business to support it. So that's generally how we think about in the long term. Obviously, episodically, you have different strategies from customers or regionalization or other things that can drive short-term I'll call it, inefficiency. But in the long run, we're planning around a model that's consistent the way I described it.

Great. That's helpful. High-NA EUV, any thoughts on timing? Does it matter for KLA? And also maybe talk a little bit about the [indiscernible] inspection tool that you guys have been working on.

Sure. Why don't I let Ahmad start.

Yes. So in general, we like EUV. And the reason for that is because if you change the pitch at the reticle, that reticle when goes into a scanner it will print smaller defects and smaller features. I mean the idea is not to print smaller defects. The idea is to print smaller features, but because it has a shorter wavelength of light. And in case of high-NA, you're really not changing the wavelength of light. The light stays at 13.5 nanometer, but you changed the numerical aperture. So that you can go from 0.33 NA and 0.55 NA. So essentially, you're improving the resolution on the system, and you will end up printing smaller features, but at the same time, we will print smaller defects. So it drives process control intensity. Of course, it drives scaling for our customers. So we like high NA EUV. Of course, there's many challenges with high NA EUV, and I don't want to go into those because the companies that are responsible for it should cover those well. And I think our customers are waiting for those problems to be resolved before they essentially go into volume production. These high-level issues are related to overlay stitching, the throughput of the machine and the cost and the cost of ownership and the existing fleet that they have, which is current 0.33 NA can they do double patterning with it? And what is the scaling? What is the timing? So KLA is very judicious about when we release our systems to match the customer's time line. If you release the systems far too early, then you essentially have done the R&D, but you're not able to ramp the system. If you are late, then you have the opposite problem. So the idea is to be able to target the market at the right time. Today, 4.33 NA, I would say most of the reticles are going through KLA, 193 system and is able to find most of the defects and we have Gen 5 print check system that is able to find the defects in the wafer fab. And we have a recall system also reticle recall system that does that. Now as the pitches get smaller and smaller, either by using 0.33 NA or 0.55 NA optical systems will run out of gas, and then you need an actinic inspection system. So we have two views of that market. First is that we are releasing an electron beam multicolumn reticle inspection system for pattern inspection. When -- as you know, reticles are actually written by a mask writer, which is an e-beam writer, that writes the patterns. So we wanted to make an actinic pattern inspector, which is more or less aligned to the wavelength of the pattern writer. So we are releasing a 8XX system that does pattern inspection and qualifies the reticle first. And then we're developing a second system, which is 7XX which a EUV reticle inspection system. We think the timing is aligned with when high NA goes into production. We also have a close collaboration with ZEISS, a partnership. We already do that with them on our brightfield systems, our Gen 4 and Gen 5. And now we have solidified that partnership and extended it to actinic mask inspection, will work together to ensure that we are able to bring actinic to the market together. One of the great advantages of that is they are common to the scanner. And each time the scanner will change the wavelength and/or change the numerical aperture KLA will have the ability to import those optics inside our system. So there should be no significant delay between when a scanner comes out and when an inspector comes out. That's why we think it is a very valuable partnership, and we have done that and recently talked about it. So that's our plan. Again, today, we do mostly everything with our 6x6 systems, and it works quite well. Most of the reticles go through it. The pitches are still not that aggressive in our optical inspection system, along with massive amount of machine learning, we're able to find all the defects that the customer cares about. And then it moves to 8XX, which is a pattern inspector and actinic inspection with a collaboration with [ ZEISS ] in line with when the market needs it.

High-NA timing seems to be shifting out a bit from a production point of view. Just lately, just some of the information you're hearing, I think it becomes economic questions about utilizing the existing fleet of 0.33 systems and how that plays in and multi-patterning and so on. So we'll see how that plays out. But until you get to high NA, you don't really get the feature size shrinks, we talk about pitches that means the size of the feature sizes until they shrink on the reticle. So we're able to meet all the production requirements with optical systems today. There are use cases where customers are using higher-end systems for pathfinding. But from a production point of view, the market requirements are being met by the systems we sell today. Of course, those requirements change as you move into high NA. But for now, we feel like we're pretty well positioned. And having systems earlier than you need them, doesn't necessarily what customers are going to balance their technical requirements with their economic motives. I don't think -- even if I had the system today, I don't think that you'd see much difference in terms of how customers are deploying and planning for their production, given the current specifications from a technology point of view. So we think we're in a good place. Ahmad talked about our recent partnership that we've established that ensures we have a road map that's synergistic with the litho roadmap over time to be able to have a product line to meet some of these requirements at the high end, but you're also in this part of the market, you're serving reticles. I mean the reticle market is broad with multiple inspection points between what's happening in the [indiscernible] and also what's happening in the fab. Good thing about KLA is we're a portfolio company. And so we have products that address all those parts of the market. And so as technology shifts and changes, Ahmad said it earlier, and I think it's great. We get invited to every party, and that's important for KLA to make sure that we're not a point product competitor that we can leverage the portfolio to meet the needs of our customers as they're evolving and working through their processes and the maturity of that from very advanced requirements, driving high technical requirements to more economic and efficiency drivers in high-volume production.

Okay. Great. Very insightful. I didn't want to squeeze in a question on China. As a percentage of total sales, you've sort of been tracking in the low to mid-40s. You're guiding absolute revenue levels to stay relatively stable, I believe, but as a percentage of revenue to come down, are there any puts and takes by application type, that's part one. And then part two is you probably heard this as well, but investors are concerned that a lot of the spying is pull in and it's actually not going into use. Do you have the ability to see whether or not they're just being pulled in or there is real demand there.

Okay. So in terms of the overall -- yes, in the half to half, and I've always talked about China in terms of half to half, first half or second half, generally because you have a lot of greenfield activity. So any time you're building a new fab, there's always a variety of factors, the factors that can impact timing. There's also potentially rev rec implications in terms of timing of rev rec when you're dealing with new customers. So as I -- half to half, the absolute value of the business is roughly the same. We're seeing growth from our non-China customers, which is driving, I think, the half to half percentages, will start to come down and come down more meaningfully in 2025. It looks like Q3 is higher and then we'll start to see that trajectory down in terms of Q4. And so as you look at what we expect into next year, I think it stays stable. And I'll use that turn, you could grow a little bit, it could shrink a little bit, I just think that there's some stability in terms of expectations there. There are memory customers, which memory is down year-to-year compared to last year, given where some of the technology thresholds are in terms of the WFE that's being spent that we have exposure to. We are restricted in terms of market access from some of that. logic/foundry is growing. Infrastructure, which was a big part of '23 is down in '24. That's wafers and reticles where a number of our peer companies aren't exposed to the manufacturing process for those things. So that influences KLA as well. So there's some puts and takes across the full exposure. But I think as we look at what we expect into next year is I expect general stability. Utilization rates, particularly as it relates to a new fab, tend to be lower, but they are there. I mean, well, we're installing all the equipment that we're shipping, when you're a new customer and you're trying to qualify processes and you're trying to ramp limited production and you're trying to meet technical milestones in those environments, utilization rates tend to be lower and our more established customers that are more capacity phases utilization rates are higher. So it's a bit of a mixed bag, but we're installing the tools we're shipping and customers are in different states of maturity in terms of what they're trying to accomplish from a strategic and technology point of view.

That's great. We have time for maybe one question from the audience, if anyone has one. If there is [indiscernible] thank you.

Maybe just an observation you might want to comment on, but you have some longer-term financial targets yet consensus is not near them. What do you think the community is missing versus your longer-term financial targets? And are they still relevant even though you've had this transition here in the middle?

Well, look, we have -- I think we have a 2026 plan that we put out there and defined our last Investor Day back in 2022 of $14 billion in revenue and $38 of earnings as you back up and look at the puts and takes in that, I mean, obviously, it's predicated on a certain WFE level, which was in the low to mid 120 range when we laid that out. I'm not going to comment on WFE expectations for 2026. But I think as we start to move forward, I think we see a pretty good expanding environment over the next couple of years, given what we qualified earlier about N2 and investments in memory and so on. Our share of the market, we're ahead of that. We talked about a 7.25% share of WFE were in the mid- to high 7s today and expect that to continue given some of the things we talked about earlier. So I think we're ahead of the game there, about 0.5 point or so, 50 basis points further along than we thought. Our service business, which we haven't talked about, is growing at the high end of its range, 12% to 14% and would expect that to continue moving forward. So there's an upper there. We've seen some pressure in our more consumer-centric markets that were strong during the COVID period, but there was some pull in as it relates to the mobile market. and how it affects PCB. Obviously, we exited the flat panel business. So it's a little bit of a drag on revenue, not so much on the profitability line. So we feel pretty good about the trajectory of the business. If the WFE levels there, we feel pretty good about the line of sight to the targets. And that what's included and embedded in the plans in terms of operating leverage and expectations around growth, expectations around capital allocation are all completely consistent with the way we thought about it, and we believe has a trajectory beyond the time frame we're talking about.

I'd love to go on. Unfortunately, we're out of time. Bren, Ahmad, thank you so much for the insightful conversation. Appreciate it.

Thank you for having us again.

Thank you.