All right, hi, good morning, everyone. I'm Brian Shin, semiconductor capital equipment and memory analyst here at Stifel. Thank you all for attending day two of the Stifel CSI Conference here in Boston. As I catch my breath here from running around the meeting floors, really glad to have CEO Mike Plisinski of Onto Innovation here with us. Mike might give, like, a very quick overview here of the company, but Onto is one of the leading inspection metrology companies in the semiconductor equipment space. I'll kind of kick it to you, Mike, if you want to talk a little bit briefly about the company, and then we can kind of hop into a fireside discussion.
You want me to talk a long time so you can catch your breath?
Yeah, I'll be there in about two minutes.
What was the question?
Or just, like, if you want a quick overview of the-
Of the semiconductor-
Or no, just of—
Company
Onto, for people unfamiliar, maybe.
So I think for a company of our size in semiconductor capital equipment, one of the unique things about Onto Innovation is the number of different secular trends that we're tied to. So whether it's the advanced nodes or transition to Gate-A ll- Around, or advanced memory, DDR5 and beyond, or packaging, where you see all the activity right now around packaging for AI devices and HBM or CoWoS, panel-level fan-out now being discussed. All of these are markets that we have several tools, well established in for multiple years. So in the case of HBM, you know, our first installations were 5, 6, 7 years ago, and now that's starting to take off, and we have, you know, all three of the HBM manufacturers adopting our inspection and now growing into some of our metrology tools. CoWoS, similar story.
With the lithography, we have several tools and customers installed for both panel-level fan-out and IC substrates, so panel on a different kind of material.
Okay. Yeah, maybe we can. I'll definitely want to kind of touch on some of the things you talked about there, Mike. Maybe before we do that, just to sort of focus for a moment on the near term, any updates relative to the guidance you issued for the quarter and kind of for the year, and even sort of the contour for the year? And maybe just before you answer, I think there was some discussion there that, you know, you saw the second half up, maybe sort of mid-single digits on revenue relative to the first half. But I think, you know, in terms of what that contour looks like, there was a thought maybe that Q3 would be down a little bit, and then Q4 kind of bounces back up.
And so kind of any updates relative to sort of the engagements in the business environment?
So I think, people read in a lot to what I said. So what I said was AI revenue would be down, or digestion, and we still think that. You know, so much capacity was added, we're literally seeing our customers running out of fab space, and they've been reallocating fab space to ramp so quickly, and this is a story that's playing out in both HBM and in, and in the CoWoS or 2.5D packaging. So, so my comments were specific to the AI packaging. The, for the second half, we talked about definitely, growth, I think we said incremental growth. If we look now, you're asking me, you know, what's the update now? What I can say is, if you look at the headlines, everything is actually getting brighter, everything is more positive.
You talk about NVIDIA saying they're still supply constrained. Okay, that's great, that means people need to add new equipment. We see the advanced nodes coming. TSMC has talked about Gate-All-Around ramps and when they expect revenue. That's also positive for us. HBM, the growth in HBM is picking up a lot of the slack in DRAM and memory. That's obviously another good driver for us that we've talked about. And then in the last call, NAND. So we've talked about a surprise on the upside was some NAND growth, driven by the need for solid-state drives to support these AI processors. So, so, you know, from the headlines, things are, you know, positive. We have an updated guidance. We'll probably provide an update in our Earnings Call.
Okay, fair enough. Maybe just one, one more thing kind of related to that, but in terms of the HBM sort of capacity additions, is there an aspect to that that's timing-oriented relative to, maybe, you know, supplier X, Y, or Z's ability to qualify with, you know, NVIDIA, for example, and also even the kind of the volumes and release schedule of some of these more advanced GPUs and AI processors?
Yeah, I definitely think there's a timing aspect. So similar to, you know, the way the OSATs and advanced packagings worked for decades, that we've been in this market. Everyone has to add the capacity if they want to compete for, let's say, the Apple business, or in this case, the NVIDIA business or somebody else. Then they have to qualify and then get the orders. So the trick for equipment suppliers is to make sure we're not over-investing and thinking that everyone who's competing for a single order is going to get that order. So nowadays. It was much harder decades ago. Nowadays, we have this down pretty pat, but in the case of HBM, specific to your question, I think NVIDIA and Hynix are well established. There's strong ties there.
I haven't seen—we've heard discussions from Micron and Samsung about wanting to break into that. You know, we haven't seen announcements about that yet. If that doesn't happen, then there's some excess capacity, and that's from, you know, the Micron and the Samsung. But you have AMD, you have Intel, you have several other players that are also trying to now catch up to NVIDIA and get a piece of this AI action. So I don't expect that capacity to last for very long.
Okay. Kind of still within the theme of, you know, advanced packaging, which obviously is sort of a recognized opportunity and something that's driven growth off this down cycle, for the company. Maybe can we take a step back and just remind people sort of what, you know, in your case, the Dragonfly system, but what your systems inspect or measure within this process scheme? And also just what—why the, why it's such a challenging situation for the suppliers in terms from a yield standpoint, as they're trying to get to a certain output, and why they're kind of struggling in that, in that sense, and why that's a good thing for, for us?
Yeah. So the tools do a lot. So the Dragonfly is not just a 2D inspection system. There's several different types of 2D inspection capabilities we do. One that we've talked a lot about is the ClearFind capability we offer, which is unique. It was released maybe five, six years ago, and yet still nobody has been able to replicate that capability. That capability is critical for interconnect reliability.
So ensuring that as you're placing these layers, and drilling TSVs, there's no resist residue in the bottom of the TSV or across a bond pad or any kind of interconnect, such that when you bond this to another die or device, over time, heating, cooling, whatever, that interconnect will break, and now you've just, you know, yield lost a finished device, a final product, with the consumer, and that's a big problem. So ClearFind is able to detect that very, very accurately. So most of our customers do 100% inspection now with the ClearFind, in addition to the 2D defect inspection, and some of the 3D metrology that we added to the system through our some of our unique sensors.
Most recently, customers, as they've been ramping, these ultra-thin wafers that are used for the 2.5D packaging and HBM for the AI compute devices, subsurface defects. So if it's ultra-thin, you're starting to see cracks, and these cracks proliferate, and through subsequent processing, can shatter or break or, you know, cut through several die, depending on the severity. Customers want to detect that. They've used our system in the past to detect it, but on a random sampling basis. So running around because the type of technology available to look through silicon is slow. Well, because the problem is becoming so critical, customers want to do that now 100%, and so we offered, and we announced a tool that can do that, a capability on the Dragonfly that can do that.
Going back to your question, the system has a tremendous amount of capability in one system, and customers are always finding new applications and new challenges for either the capability or additional capability as their process becomes more complex, or they try and get to the next level of yield and output.
Okay, great. And maybe that kind of, you know, is an off-ramp to another question about sort of the process control intensity, where clearly these are critical, high capital at risk applications. And so it, it's sort of the payback is sort of innate relative to spending on process control. Do you see that relaxing over time as some of these companies go through cycles of learning? You know, Hynix would kind of be at the forefront there, and other companies maybe trying to now have to experience some growing pains. Or do you think it's sort of also a moving target or moving bogey in terms of, you know, now you're stacking 12 chips or 16 chips, and now there's other complexity that kind of comes into play.
Do you think it's just kind of shifting anyways, Mike?
I think the general trend is kind of clear. The capital intensity for process control is increasing. So overall, across the... You know, whether it's front-end or packaging, capital intensity generally is increasing for as a percentage of what the fabs are spending. Specific to advanced packaging, I think it's increasing dramatically. So historically, you know, when we were selling into OSATs, it was almost a rubber stamp. Any tool good enough, do a 2D inspection, and then I can show the wafer maps to my customer. Now, we talked about maybe 6-7 years ago, when InFO became so critical to TSMC's roadmaps and how they won the Apple business away from Samsung. That was really the start of the top IDMs investing heavily in packaging to drive those roadmaps for performance advantages, and they demanded a higher level of process control.
So they demanded much better resolution, higher repeatability, matching between tools, all these things that actually are more capable tools we're able to deliver, and that's why we started to see some significant gains in the top IDMs. But the lower end part of the market, you know, that's not a place for us.
Okay. And then in terms of sizing the market, you know, this year, for example-
Sorry.
Oh.
I didn't finish one part of your-
Okay
question before. I just remembered. As far as the you know, you asked if once they stabilize, you know, will it kind of level off? We're not seeing that. So right now we're seeing the pace of these customers innovating, the pace of change increasing quite rapidly, and if anything, we're seeing more applications and more process control intensity. For instance, the example I just gave on the subsurface defects, we're also seeing more pressure on, "Okay, we wanna be able to measure or see smaller defects, smaller variations," so driving more innovation in the optics and the software algorithms that we use to differentiate good from bad. Sorry.
Yeah. No, the good news is that when you look at these GPUs, there's no faster refresh cycle, and it's such a complex situation, it pulls the memory through, and so-
Yeah
something like that.
Yeah.
So when you try to size the market, I know there's not, like, a great third-party estimate, but for this year, maybe I'd say if I think about your addressable market, some of your peers' served addressable market, maybe it's like $600 million-$700 million. Does that ballpark sound right to you? Or I know maybe the- what you include or don't include, maybe-
Right
something like that, but.
Uh, ballpark-
Okay
Sounds about right. I think the- this whole, AI, the ramp of chiplets, the ramp of the, you know, the CoWoS packaging and HBM is driving a huge increase in the packaging SAM or TAM. So the, the amount of investment, the capital intensity versus traditional OSATs or traditional copper pillar or wafer level chip scale packaging, is much, much higher. So we're seeing, really rapid growth in packaging for, for the high end.
Yeah, and if you look at, you know, different equipment companies and, they're talking about 2x, 3x, even 6x growth in their business, that kind of ties into this. You guys were kind of on the front foot and forefront of sort of this pickup, right? But obviously, those sort of multiple growth rates are not gonna sustain-
Right
You know, you know, for many years, right? So when you think about growth modulation, but still growth, I think of it also in terms of, like, how much incremental capacity is being added one year to the next. So it could be how much relative to the HBM capacity added this year, what's it gonna be next year? You know, CoWoS capacity last year, what's it gonna be next year? Those could start to be more similar in terms of the increments, maybe still higher, but do you kind of agree with that philosophically or. I mean, it's not based on a forecast.
That they'll essentially double the current capacity for next year?
Yeah, but if you, if you say you go from, like, 15x, like, just to pick numbers, round numbers, like, if, if TSMC, sorry to throw names out there, but if they had 15K capacity last year, say they went to 30K, that's a 15 increment. Say they and these, these are probably too low, but say they went to 45K next year, that's another 15 increment, and so your revenue is still a strong base each year, but it doesn't necessarily grow. And so it'd need to be, like, a 20 increment next year for actually it to, to register growth. And so philosophically, do you think that's something like that is happening, or is that a little bit too draconian or something?
I think that's a little simple. First, they're growing rapidly. I think maybe even... What we've been told to be prepared for is more growth than that.
Okay.
I won't say what, but more growth than that. But for us, it's not just about their capacity, and that's why WFE is not something I obsess over. It's about their yields, and so it's about the applications we can continue to add, for instance, you know, the acoustic metrology and some of the new metrology sensors we talked about in the last Earnings Call. We're seeing, you know, opportunities there with OCD and even the films. So, for existing capacity, there's opportunities for us to expand our entitlement and expand our revenue.
Yeah. Okay, maybe I'm breaking off to sort of panel lithography. There was a, you know, unverified report out there suggesting that NVIDIA could move one of their Blackwell product lines over to panel packaging. Maybe for a moment, can you just explain sort of what that need is over time? Not to verify that report per se, but what that need is out there and kind of how your positioning would be in terms of your lithography tool, as well as inspection, as that market starts to, you know, be adopted or come into play.
So we've been big believers that as the markets continue to drive towards heterogeneous packaging or chiplets or large die, for instance, the GPUs from NVIDIA, that the economies of scale on a 300 millimeter wafer just aren't there, and they're gonna have to move to panel. Whether it's, you know, panel level fan-out or another type of panel packaging, doesn't matter too much to us. We have very strong positions in panel for lithography. It's strongest in panel level fan-out. There, I think we have probably over 80% market share for that. And in addition, we have some unique technology. So panel-level fan-out is tricky because the die shift and rotate, they move around during the processing. So you can imagine a stepper that's trying to drive overlay, it, that's a challenge.
We solve that by using some unique algorithms and the Firefly inspection, which can do additional metrology while it's inspecting, leveraging those algorithms, feeding forward into the stepper so that we can get about a 2x improvement in throughput over what they would have had to do, and this is proven by our customers, and increased yield. So there's some real deep moats around the panel-level fan-out for our inspection and lithography. So any growth there would be pretty exciting for us. I do think it takes time to qualify, so, you know, I wouldn't expect, "Oh, that report came out, and boom, something's gonna happen." Even if it's true, there's gonna be some qualification time.
The good news is there are several lines out there, you know, a lot of lines out there that are using steppers that are qualified for a process, and it's just a matter of qualifying that particular product. So could be an option. I don't, I don't know.
Okay, fair enough. We spent a lot of time talking about packaging, but maybe let's focus one more on the panel litho-
Packaging. Sorry, very important point. So I talked about panel-level fan-out. The other type of packaging, the IC substrates, the key there is being able to with wide field, high-resolution optics, being able to print the interconnects very fast, with very good overlay on a very unstable substrate. So it's cheap, but it's the way that all of the servers high-performance servers have been packaging their technologies. That is our X500, and there's very interesting trends happening there. Obviously, the X500 is we've talked about one or two generations ahead of the market. If the market's printing 5-8 μm RDL, the X500 is proven down to 3, but we don't think the substrates can support a 3 μm process.
Glass is a big transition to watch for, and you hear a lot about that from Intel and some others about the move to glass, to be able to go 3 μm and below. We have capabilities that'll be running in our PACE Lab that show already 1 μm wide field optics in being able to support glass. There's a lot of activity in the panel area that we're already positioned pretty well for. Sorry, I had to throw that in.
No worries. In terms of the more wafer fab equipment sensitive business, for the company, it, you know, it was 40% of the company's revenue, you know, like the Atlas product, for example, for OCD metrology, but it was 40% of revenue, 2022, when the company was at $1 billion. You know, recently, it's 10% of revenue, even less in some quarters. And so certainly, you can anticipate some positive mean reversion there with sort of cyclical improvement. But so, you know, this is. It's just inflecting now, and it's hard to maybe talk about, you know, altitude, but when you think about WFE, that's probably gonna be higher than it was in the last cycle in terms of the peak, and you're expanding addressable market, and maybe even some share pickup here or there.
Would you not expect revenue in advanced nodes, which is WFE-centric, to get back to the level it was in 2022 or even be higher?
I think 2022 was impacted by massive spending through COVID, right? So you had massive spending on DRAM, on servers, everybody working from home. How does IT support it? Blah, blah, blah. So WFE encompasses all spending, not just the high-end advanced nodes where we're aligned, and our share has or our opportunities have expanded in those advanced nodes. So we are seeing new opportunities, and we're continuing to add new opportunities in planar films, in some acoustic metrology, and of course, the Atlas OCD, and even integrated metrology to, for that matter. So there's some good movement there. Will it mean we get back to 2022 levels next year? I don't know. If the spending is there, we should exceed it. So if the spending on DRAM, NAND, and logic on advanced nodes exceeds 2022, we should exceed our revenue-
Yeah
'cause we are, we're keenly positioned there. But when you look at just WFE, you have the 32 nm, 20 nm You have a lot of legacy nodes, 12 nm and 11 nm, that we're not – that's not where our products have the dominant position.
Sure. Next year, it might be a little bit steep and aggressive, but, yeah, even thinking of it, like, as a two-year kinda horizon or something like that.
Yeah. Then I would. Yeah, I mean, we've, we've said we're, we're expanding, and the planar films is an incredibly—it's a much larger market opportunity for us. So that alone, even incremental improvements there, is, is significantly additive to what we had.
Even through, you know, as sort of a segue to that, even through the end of this year, if there was to be upside that emerged in one of, you know, whether it's—do you think if before year-end, maybe in Q4 this year, not guidance, but do you think it would be more likely to be advanced foundry, DRAM or, or memory?
I think the Gate-All-Around node has been more well-documented as to their timing of expansion. So by inference, I would say that would be-
Okay.
But we are seeing NANDs, you know, some NAND increases, and that again comes to less about their capacity going up, but there's moving more towards high-stack 3D NAND, the higher 200+, and that's driving needs for some of our new metrology, the acoustic metrology. So that drives incremental revenue. So it's for us, it's not just factory expansions. Like I said before, it's yield, and it's what's expanding. If it's the advanced nodes, it's very good for us.
Yeah.
Leading edge, good for us.
Okay. I didn't really ask about competition, but, you know, one observation is that all the very largest companies in capital equipment are now quantifying their exposure to advanced packaging, whereas, you know, only one of them had before. And so you can kind of register what that per capita exposure is. It's not as high as yours, but it's, you know, usually it's less than 10% for the bigger companies, and then for yourselves, it's, it's a, a good double digit.
Yep.
But what are you seeing or expecting from companies like KLA in process control? You compete with them, obviously, in the front end. When it comes to packaging, what do you think of competition? Even now, Applied Materials is partnering to enter that panel lithography business. So it validates the market opportunity, but it's not a non-trivial competitor, but also, so.
And so with the AMAT tool, it's not a new tool. So that's a tool we've seen in the market for quite some time. We've seen it in display, and then they tried in packaging, so wafer packaging. So I don't think there's any big shift for us from that partnership. If anything, it's great. It might provide some opportunity for, you know, price uplift, et cetera, et cetera. And with KLA, I don't think it's KLA's first attempt. I mean, they've been talking about packaging for a long time, since trying to acquire August, when I was at August, and eventually, we were trying to merge. Well, it's a long history.
But after that, they acquired ICOS, they acquired Vistec, and they were trying to build out the package. So we've been competing, or KLA has been trying to break into this market a long time. I think we know how to—I, I think we know where they'll approach. They're definitely a powerful competitor. They have a lot of technology, but strictly around 2D inspection. All the other capabilities and challenges of packaging, they're coming up to speed on the, the high warp, the film frames that, you know, there's a lot of different algorithms and changes that customers expect and demand from a single tool, and that's gonna be non-trivial at a certain price. So the battle will be on, it'll make us stronger, and we'll see where it goes.
Okay. I definitely want to get at least one question on the financial model in here. And, you know, this just our estimate, not your guidance, but we, we modeled you by Q4 of this year, annualizing to about $1 billion. So kind of consistent with the prior peak in 2022, annualizing the quarter. Do you expect to be back within sort of that target model in terms of gross margins, operating margins? I think it's like 56%-57% gross, 31%-32% operating margins by that timeframe. Or is that a little aggressive, relative to NICS, relative to some of the cost synergies, other synergies that you're trying to capture?
No, I think we've been pretty consistent. We said we would exit 2024, meaning Q4, within our long-term operating model. So having the effect of, yes, advanced nodes returning somewhat, but also all of the supply chain initiatives we've been implementing across the organization, and then in particular for lithography, you know, to bring those up, those margins into the model. And all that has been tracking, and I believe Mark has been guiding, you know, just some incremental improvements, quarter to quarter.
Okay. So, on track?
On track.
Okay, perfect. I think that's all the time we've got, Mike, but really appreciate you joining us at the conference.
Thank you very much, everybody.
Yeah.