Hi there. Good morning. I'm Brian Chin, the semiconductor capital equipment and memory analyst here at Stifel. Thank you all for joining us here at day two morning session of the Stifel CSI Conference. I'm very glad to have FormFactor with us, and CEO Mike Slessor here on stage. I think he caught a late flight in last night just to kind of be here, so we appreciate that. FormFactor is the leading wafer test probe card supplier, an advanced probe card supplier in the industry, and Mike has been with the company for more than a decade-
Yeah.
at this point.
Yeah.
I think he has a couple, maybe opening remarks, to share about the company, and then we'll kind of hop into a fireside discussion.
Great. Thanks for having us, Brian. This is always a great conference to interact with all of the investors we have here. So thanks again for hosting us. Just to level set everybody before we get into Q&A, FormFactor, as Brian said, a leading supplier of wafer test consumables. People primarily think of us as a supplier of something called a probe card, which is a consumable that acts as the interface between automated test equipment, or ATE, built by companies like Teradyne and Advantest, and the customer wafer. And the interesting thing about this business is, when you're acting as that interface, we call it a device-specific consumable interface, you're needing to essentially build a different probe card for each customer chip design.
And so that drives a very high-frequency demand refresh cycle in our business, and why, you know, it feels a lot more like a consumables business than it does a capital equipment business. Probe cards, and I'm sure we'll talk mostly about probe cards here today, about 75%-80% of our business across foundry and logic, which includes microprocessors, application processors, RF chips, all kinds of different semiconductors. DRAM, which is an area of specific strength right now, driven by high-bandwidth memory, a subclass of DRAM, and flash probe cards, which is really kind of a small business and not a key area of focus for us. The other part of the business that we report as a separate segment called the systems business, these are really engineering probers that allow customers to do very early development.
So, you know, today, where we might be shipping in foundry and logic, volume probe cards for five nanometer and three nanometer devices, we're working in the labs with those same customers on one nanometer process technology and exciting things like optical interconnect, co-packaged optics, brand-new areas of the industry. So nice strategic value in the early engagement and, you know, as we get to talk about gross margins, our highest gross margin segment as well. So maybe with that, we dive off into some specifics.
Okay. Before I zoom out a little bit, I'll zoom in for a moment here.
Yeah.
Just any update in terms of the quarter outlook that you conveyed on the last earnings call, but, you know, a lot of strength in what's kind of been really strong in the past several quarters, probe cards for HBM, which is being driven by AI, then also some other improvement in, in, in logic foundry as well. So any kind of commentary you'd like to share on that?
Yeah, and no real update. We gave our outlook on our earnings call about a month ago for the second quarter.
We actually have that slide up here now.
Thanks, Dan.
Mm-hmm.
And you can see, as Brian noted, some strength sort of across the P&L, stepping up revenue levels from the high 160s to $195 million at the midpoint of the outlook, and a corresponding increase in gross margin and overall EPS. I think a couple of interesting examples here, I know we'll dive into the revenue bridge, but you see the leverage inherent in the model here. We have some fairly specialized technologies that drive quite a bit of vertical integration. We manufacture inside of FormFactor a lot of the technologies, and so when you're not operating at peak utilization, there are some gross margin headwinds. This represents.
Our second quarter outlook represents a nice step up back towards our target model, which delivers 47% gross margin at an $850 million annualized run rate. The bridge associated with the revenue, the top-line piece that then drives this leverage, really a couple of key components, three in particular. One, as Brian noted, high-bandwidth memory. We are a leading supplier of DRAM probe cards, and anybody who follows the DRAM industry now knows that a tremendous amount of energy, expansion, output associated with all three DRAM manufacturers producing HBM, primarily HBM3, some HBM3 in volume. We're a significant supplier in that volume. The other two components come from our foundry and logic segment.
You know, we're seeing and this is true across most of our probe card businesses, but in our microprocessor business, we're seeing significant transition to advanced packaging technologies. This is really pretty profound for a probe card manufacturer because the transition to advanced packaging or chiplets or tiles, depending on what the customer calls it, all the same thing. Essentially breaking apart what used to be a monolithic piece of silicon into individual subcomponents, tiles or chiplets. When they do that, before they reconstitute them back together, that really raises the bar for wafer test and die test, making sure that each of those chiplets is good before it gets committed to the overall chiplet stack. We're seeing exactly the same thing in HBM, right?
Sure, there's a lot of output and interest in HBM, but there's higher test intensity, too, because HBM is a stack of 8, 12, or even 16 individual DRAM die, each of which needs to be tested individually before it gets committed to the stack, which is then also tested. The third component, kind of a seasonal, regular thing that we see, is some strong activity in the mobile application processor space, really associated with our customers refreshing their roadmaps. And a good example of probe cards being a device-specific consumable, even though the handset business, from an end market perspective, not growing all that significantly. All of our customers are refreshing their application processors to go into the phones that are gonna hit the shelves next year, and that is a typical thing we see in the middle part of the year.
So no real updates to this, but I felt like it was worth reiterating both the bridge and revenue, 'cause it'll drive into some of the drill-down questions, as well as demonstrating the real leverage inherent in the earnings model here.
Okay, now, that's helpful, and we will drill down. Maybe we can start with HBM in particular. It's been remarkable in terms of the revenue contribution-
Yeah
..T hat that's driving, you know, 20 now, going on to 30 million per quarter. It basically that level, and you break this out in terms of customer exposures, but you have Hynix now equivalent to Intel revenue. You know, not that it's gonna happen every quarter, but that's fairly remarkable, and pretty uncommon, I think, historically-
Mm-hmm
T o some degree.
Yep, yep.
and so when you think about. And you've described sort of, I know it's a bit of a blanket, but the increasing test intensity from advanced packaging versus sort of, discrete die, I think, like a 20%-30% increase in intensity. But can you describe a little bit what those additional test insertions or probe card opportunities are in something like HBM? Because when I kind of like add them up, it seems like it would be a number that would be north of, you know, 20%-30% more capital intensive-
Yep
F avourable from a revenue perspective. And so, you know, obviously, you have a DRAM wafer. You test those chips once there, but now, can you maybe lay out sort of what those incremental test opportunities are in the-
Yeah
HBM process?
Yeah. And HBM is a good example, and maybe the comparison to draw is, if I were gonna build a 16-gigabit DRAM on a monolithic die versus a 16-gigabit HBM, and try and compare those two, it'll be a bit of a made-up example. But think about the HBM stack being a stack of 16-high DRAM chips. Each one of those 16 chips needs to be tested before it's committed to the stack. 'Cause you can imagine the implications associated with if the ninth die is bad, it's gonna kill all the other ones in the stack. And that's a pretty significant cost problem for our customers. And I think, as most of you know, at the leading edge anyway, the semiconductor industry operates nowhere near close to a 100% yields.
Wafer test is screening out these bad chiplets or bad HBM core die, making sure that the overall stack is going to be good once it gets put together. Now, there's a couple of other test insertions associated with that make believe 16-gigabit, 16-high HBM stack. Once it's all stacked up, it also gets tested, right? So we have a shot at a test insertion for the entire 16-high piece, and in some cases where customers are facing some yield challenges, there's also intermediate test insertions as that 16-high stack gets built, maybe at 8, maybe at 12, to make sure it is good and there hasn't been any defects introduced in the stacking process.
As Brian said, you know, we've kind of guided people to that overall being a 20%-30% increase, and the reason it's not 16 times is, of course, the number of bits on each of the constituent DRAM die is 1/16 the total, right? And so you don't see these, you know, factor of 2, factor of 4, factor of 8 magnifications. But because of the increased number of insertions associated with probing the whole stack, because also, you know, the increased test coverage to ensure that each of those core die is good, we are seeing a step up in test intensity associated with HBM and everything in advanced packaging for very analogous reasons.
Good. There's a lot of just logic in thinking that companies that have had more cycles and years of learning would be further along in sort of, you know, maybe yield maturity, whatever yield maturity looks like-
Yeah
F or this, versus other ones that are, have had less time to really work this out. They're probably ones that will have to even have a lot more test coverage, test intensity in the, in these cases, in, within the grand scheme of people supplying HBM today.
Yep, yep, and I think that's true. I think you also see with suppliers providing metrology and inspection equipment, right? The defect modes associated with stacking die in advanced packages, whether they're chiplets in foundry and logic for microprocessors or HBM and memory, there's a new set of defect modes that the industry and our customers are working their way through resolving, that require new test technologies, like our high-speed probe cards, and new metrology and inspection pieces. So you're seeing a wave of investment, a concentration of investment associated with those applications.
Okay. You know, based on our modeling, and this bogey has shifted upwards, even already this year, but we're modeling your HBM revenue could be up 3x this year. You know, equipment suppliers have talked about two, three, even 6x-fold kind of increases. You know, these multiple effects probably will not persist forever, right?
Right.
But what kind of—you know, I don't know what horizon is, is right, but over the next few several years, what do you think will be sort of that, you know, a CAGR in this business for you when you think about. 'Cause, you know, even more broadly, on memory, you would typically see it behave cyclically, but not grow cyclically over time.
Yeah.
Clearly, you've already taken out your prior, you know, highs on DRAM revenue.
Yeah, yeah.
And you know, so, I mean, within that whole context, like, maybe how would you sort of address that?
Well, I think DRAM for us is really the tale of two sub-markets right now. The DDR5, low-power DDR5, call it regular DRAM or non-HBM DRAM, it continues to operate essentially at cyclical lows. The growth in our DRAM business, as Brian notes, has been, you know, printing a record level in Q1, creating expectations for another record in Q2, really driven by the growth of HBM, which has gone from $10 million a quarter in Q4 to $20 million in Q1. And in round numbers, you know, our guidance implied $30 million in Q2. That obviously, things can't, you know-
Mm-hmm
P roceed at that pace forever. There are some built-in constraints, but I think you can draw a pretty straight line between hyperscalers spending on generative AI. You know, each of these HBM modules ends up, well, multiple of them end up attached to GPUs in these generative AI modules in the data center. I think one of the questions about growth rate is: How much of the investment continues on the generative AI front versus the inference front? How HBM intensive that is? There's a bunch of open questions here, but certainly, over the next, you know, year or two, we do see growth associated with the HBM business, especially when you look at some of the GPU roadmaps and the increased memory content. You know, we're going from 8-high to 12-high to 16-high.
The number of gigabits attached to each of these GPUs is going up quite dramatically, and at least to first order, probe card spending should scale with them.
Okay, and just maybe one last thing here, just in terms of that context. When thinking even beyond Q2, you did seem a little bit reserved about second half. Was that just conservatism on your part, or is there. You know, what I've observed in the DRAM market going back historically, is that sometimes there's customers that kind of come in for a quarter, but then cycle out, another customer kind of comes in.
Yeah.
It's not like there's all suppliers in the quarter at the same time, but is that an aspect to sort of your thought process?
Yeah, I think a couple of comments. One is because probe cards are specific to customer chip designs, they run on very short lead times. Our heartbeat, if you like, is the wafer cycle time in the fab, 'cause customers don't place orders essentially until they start the wafers. So your visibility tends to be much more limited than a capital equipment business might be, where you're running backlog for several quarters. So that's one reason why we don't offer much commentary on the second half. We do have indications, you know, from customers that they're gonna continue to invest, but at various times, those plans change. So that's one of the reasons for it.
You know, I think if you look at how the business, the strength in the business is being driven, you know, that element of HBM, we do think has continued long-term legs. But to your other point, we have seen historical patterns. Quarter customers can go through a digestion phase. They'll order a whole bunch of probe cards for multiple designs, and then run those for a quarter or two, and not need new probe cards from us until they refresh the design or up the volume level. So you can see this in our 10% customers over history, we'll often have very strong quarters, followed by a little bit of a digestion phase, and so I don't think it would be unreasonable to think that we have that in some of these businesses.
Yeah, and so that was my last question, but just slightly connected to that. Are we also almost? I don't know if you agree with this, but the DRAM suppliers have converted a lot of capacity towards HBM.
Yeah.
But they've run out of what they can convert-
Yeah
E ffectively at this point. So I don't know exactly how all those wafer outs trickle out here, but there does seem to be a little bit of a moratorium there before some of the fresh capacity is ordered, installed, and eventually going to output wafer. So there could even be like a, I don't know, a pause in there somewhere in terms-
Yeah
B etween this.
Yeah.
That's just that, I don't know if you agree.
Yeah, well, and that's one of the potential uncertainties or swing factors here. I mean, we've heard about, from some of these customers, constraints on fundamental long lead time things like clean room space now being constrained, fab space being constrained. So I think, you know, there's again, some timing issues. People don't need probe cards, right, until they have the wafers at the end of the line.
Yeah.
That requires equipment capacity in place, qualified and running, and there's some, you know, tremendous shifts in magnitude, as Brian noted, between capacity for regular DRAM and HBM going on right now, that I don't think we've ever seen the scale of in this industry. So might be reasonable that, you know, there are some puts and takes on a quarter-by-quarter basis, but I think the long-term secular trend is pretty clear.
Yeah, and transitioning into to Logic Foundry, you know, part of the reason that why AI, X, Y, and Z looks so good now is because the other things don't look as good-
Right.
- to some degree, right? For the most part. And so even you, you don't have to take HBM out of your revenue stream, but if you did, right, you, you'd be closer to, like, a $165 per quarter.
Yeah.
Right? And that would actually be, yeah, that wouldn't be a superb, like, you know, uptake in the business, right?
Right.
And that's consistent with sort of the bifurcation that, that you're talking about. But, but you are starting to see, it sounds like, from kind of the, the layering of the revenue, the advanced packaging and, and for, for things like PC client-
Mm-hmm.
- some of those System-in-P ackage situations starting to drive your revenue. So you did kinda distinguish that from sort of things that might be more seasonal-
Mm-hmm
I n the revenue stream. So, like, do you expect that to continue to be sort of a, a growth element this year, maybe, maybe a, a stronger tailwind next year relative to more designs committed to, to heterogeneous integration-
Yeah
and die disaggregation, some of these, these words that are kinda strange that people-
Yeah
Talk about a lot? Go ahead.
Yeah. All synonyms for-
Yeah
Advanced packaging, right? The industry's probably gotta figure out one name for this so that we all understand what we're talking about. I do think there's a fundamental tailwind there. If you look in the client PC space, we've seen a pretty remarkable shift. You know, the one of the major client PC microprocessor launches in 2023 and into 2024 is a chiplet-based architecture, and that drives significant volume in the industry. And if you look at that customer, they've now shifted their entire roadmap over to a chiplet tile advanced packaging technology. So put together with the uptick in test intensity we talked about, if we're able to get a client PC refresh cycle out of something like Windows 11 conversion, I do think there's some upside there.
Okay. Interesting. And just talking about the GPU aspect, where they have not necessarily maybe the leader. Like, NVIDIA has not necessarily needed very advanced probe cards historically, but, you know, now as transistor density increases and other criticality emerges, there seems to be some adoption potential there. How is that opportunity unfolding for you this year?
Yeah
R elative to your expectations-
Yeah
and kind of even looking forward?
Well, you're right, and, and I think it ties back to the theme of their adoption of advanced packaging, right? They're now essentially reticle limited on the size of the GPU, and so stitching multiple GPUs together using advanced packaging technologies, things like CoWoS, right? You know, so that offers an opportunity through us, to us through the foundry, primarily. And as those things ramp, we do expect some upside there. But it's a different size of opportunity than something with very high unit wafer volumes like HBM. Micron talked about earlier this year, in one of these generative AI GPU packages, the HBM is 7.5 times the silicon area of the GPU in that. And so for a supplier who's really driven by wafer volumes and unit starts and, and elements like that, HBM's nearly an order of magnitude higher opportunity.
Now, having said that, the GPU piece is important, it's strategic, and an important revenue driver in one of our higher margin businesses, so making some progress there, but I wouldn't expect it, even at peaks, to move the needle the way something high unit volume like memory does.
Okay. Yeah, that's fair. And then you touched on the systems business earlier, and how that could be 20-25% of revenue. You sold FRT, and so that-
Yeah
K ind of changed a little bit the math-
Yeah
On that scenario. But even factoring that in, it did come down a pretty substantial amount, Q4 into Q1, and you kind of talked about it being a little bit suppressed. And so I'm curious, sort of, can you just describe the dynamic that took place in terms of the go-forward part of that business? And also, kind of what visibility you have in terms of that getting back to, say, I don't know, mid- to upper-30 kind of quarterly revenue level?
Yeah
In the systems business.
So our systems segment, in general, again, really enabling customers to do early R&D, applications like quantum computing, which is way out there, right? But we think is a good fit to FormFactor's core competencies of thermal control and electrical test. You know, the business we divested, the FRT business, metrology and inspection for advanced packaging, you know, we made that decision primarily because it was a subscale business where we'd had some nice wins, but were on the brink of needing to make some significant investments.
Mm-hmm.
Got a fair offer from Camtek, and we're able to put that business in their hands, where they had already made those infrastructure investments, so I think a good deal for both. The systems business, being R&D-driven, tends to be a little bit more slower growing, and so to make up for, if you like, the hole left by the divestiture in the revenue stream, that's gonna take a few years, unless we see some really rapid acceleration of something like co-packaged optics and silicon photonics, where we're engaged, but in very early pilot production in our systems business as well.
Okay. All right. And maybe to go back to something that we, you know, talked a lot about earlier, but when we think about the competitive environment a bit, you've talked about how you have a strong position for HBM at the moment. The fact that you have competitors in the memory—in DRAM, that's one competitor, then you have a different competitor in logic foundry, it's a different competitor there. The fact that you do logic, advanced logic, and DRAM, does that give you an advantage in terms of HBM, whereas these other companies are kind of more singularly focused?
Yeah. I think the simple answer is yes, but maybe backing up, you know, this is a fundamental part of our strategy, right? We're, we're trying to be a supplier of scale who can use their technologies, their pretty vertically integrated manufacturing footprint, that are just inherent to these specialized technologies. I can't go to Flextronics and get this stuff built. So, you know, you've got a fixed cost structure to absorb. You've got an infrastructure and supporting the overall industry to support. So, you know, the diversification across multiple markets and multiple customers and trying to be a leading supplier at all those major customers, a really foundational part of FormFactor's strategy.
And if you followed us over the last decade or so, you can see our M&A dollars have been spent to do that, our organic R&D dollars have been spent to do that, and that'll continue to be a fundamental piece. I think for HBM, in particular, you know, the acquisition of Cascade Microtech in 2016, we acquired that because we wanted to be in RF probe cards, but it's lent a pretty good DNA of high-speed electrical engineering in the company, right? And I can't draw you a direct line between HBM's success and our position in RF, but RF test has become one of the core competencies of the company.
I think it's really important in not just HBM, but across the industry as test speeds go up because of, you know, the different memory interface buses, but also because of the increased test complexity associated with these chiplets. They've got to be tested at speed before they get committed to the stack, and so that RF piece is an interesting technical synergy.
Yeah, I guess. And with DRAM, and even if it's, you know, mostly HBM-centric, it's putting more upward pressure right on the manufacturing and operations-
Yeah
O f the company. Maybe, you know, foundry logic has not hit its full stride-
Mm-hmm.
but eventually, you know, things typically improve. Some of these trends kinda come through on packaging.
Yeah.
You may have this sort of, I guess, a good problem of having a very hefty DRAM and logic foundry business to support within your infrastructure. So how are you sort of handling preemptively that, especially if lead times are not, you know, super long-
Yeah
F or customers? How are you kinda handling that, you know, dilemma?
Yeah. Well, I think a couple of things. You know, we'd made some pretty significant capacity investments, capacity footprint investments, from 2020 on through 2023. We ratcheted CapEx back this year some, mostly because these were, you know, clean rooms, long lead time equipment, things that we needed to have in place to really serve upside demand, like we're seeing here in the second quarter. But, you know, we've made a bunch of changes in the company, both organizationally and to people, to set us up to scale.
I think you can see that advanced packaging is becoming a real innovation driver with our customers in the industry, and we feel like we're in a strong position to serve that, as a lot of the value from the supply chain shifts a little bit from the front end, you know, transistor fabrication, to the back end, advanced packaging and integration. And so we've positioned the company both with the CapEx investments, but also some of the organization and people changes we've made over the past couple of quarters to set us up to scale, not just to the target model, but beyond the target model.
Great. Great. Maybe just at least addressing the financial model here a little bit. You know, if you take, if you take the second quarter, and you analyze it, you're getting close to $800 million. Your target model, which you haven't put an exact timeframe on, but it's, it's to be an $850 million dollar company, and so you're not too far off the mark. You still need a bit, a bit of a push and a shove to kinda-
Yeah
T o get up there. But when you think about sort of the model beyond that, you know, say, you know, a $1 billion is kind of a nice round number, so $150 million incremental on top of that. What are kind of like loose parameters in terms of what you think will drive that growth to that next station in terms of revenue, and also sort of what the incremental fall through from a margin perspective could be?
Yeah
O n this? Because that's one area where I think some of the steps you're taking are maybe to impart some resilience into the gross margins.
Yeah. Yeah.
Maybe if you could just comment on that-
Yeah
A t some level.
Well, I think, you know, as somebody with the market share positions we do have, I think getting to beyond the model, well, to the model and beyond, from a revenue perspective, will require some recovery in the high-volume markets like mobile handsets and PCs. You know, but if you look at then, okay, crossing that, what does the P&L look like once we're able to achieve the target model, maybe on a quarterly run rate basis, as we did last time? We'll put a new model for it. What will that look like? I, I think, you know, gross margin expansion is the key thing that we've both demonstrated historically, but also is one of the two fundamental metrics we drive inside the company, that and market share gains. So I think you can expect to see some gross margin expansion.
Shai, our CFO, leads a gross margin task force across the company. That's a very important initiative, right? You see us making some nice progress, but in the second quarter, basically based on volume and absorption. You know, if you then look at drop through, we've also demonstrated a pretty good discipline historically on OpEx, and so that increase in revenue with an increased gross margin, you'll see expansion on the gross profit line, but we'd also expect to be getting more leverage out of our OpEx structure. We'll continue to increase R&D spending to keep pace with our customers and drive a competitive advantage, but SG&A is an area where we really want to get some strong leverage out, and you'll see that.
When we're able to hit our model and then, share a new model, those will be some of the attributes associated with it.
Great. Thank you, Mike. That's all the time we've got-
All right.
but I appreciate your being here.
Yep
And participating.
Thanks for having me.