Operator: We’ll take our next question from David Duley with Steelhead Securities.
David Duley: As far as high bandwidth memory goes, could you just talk about how much more inspection and intensive it is versus just a standard DDR 5 product? As my second question, as far as your advanced nodes business goes for calendar 2023, how much would you expect the memory business to be down and how much would you expect the foundry logic to be down?
Michael Plisinski: Again, I’ll start with the last one. For 2023, I don’t have the numbers directly in front of me, but I would expect memory to be down probably double what logic is down. Doesn’t help you too much, but I don’t know if logic might be down 20%. DRAM memory might be down double that, 40%. As for HBM and the attach rates, the process control intensity versus normal DRAM, it’s a good question. The number of applications, so I would say it’s two to three times more, process control intensive, at least for our applications, on our Dragonfly, than normal DRAM and DRAM packaging. And that’s because of the layers and the amount of metrology and inspection involved in preparing these die to be stacked and to be connected to either die in the stack.
So there’s a lot more metrology coinciding with our inspection in order to make those packages yield. And the metrology is on our inspection tool, so that drives the utilization of our inspection tool, it’s all integrated, just not to add any confusion.
David Duley: As far as high bandwidth memory goes and your outlook there, I’m assuming that’s just what the one lead DRAM manufacturer. I don’t really think the other guys have a lot of product in the marketplace at this point. Is that the way to look at it? Or are you working with more than one guy in high bandwidth memory?
Michael Plisinski: We’re working with more than one, but one certainly dominates more than the other.
Operator: We will take our next question from Vedvati Shrotre with Jefferies.
Vedvati Shrotre: I guess I wanted to ask about the packaging a little bit more. So you talked about the $90 million opportunity over three quarters, is that more towards HBM versus like a CoWoS capacity? Can you characterize kind of what the split is between heterogeneous integration versus HBM?
Michael Plisinski: Yeah, I would say 60/40. So a lot of it is on the logic side. So CoWoS and, say – and then the rest on the HBM.
Vedvati Shrotre: For my follow up there, you have TSMC talking about doubling their CoWoS capacity, Amkor is sort of tripling their 2.5D capacity. So how should we think about – like, how does this translate into a revenue opportunity for you? Is it like a direct link of doubling capacity means doubling revenues for you? Or how does that equation work?
Michael Plisinski: So far, it looks that way. There’s a very high attach rate of what we’re seeing. Right now, that would be how I’d model it. If yields improve and they start to drive, let’s say, a normalization and an optimization, they may reduce the process control, but if anything, we’re seeing an increase in process control as they’re trying to improve yield, and that’s why I mentioned the increase in process control intensity. So we think maybe even – since the beginning of the year till now, maybe 20% or more increase in process control intensity on the logic side, so heterogeneous packaging on the logic side [indiscernible].
Vedvati Shrotre: If I think about what your install base was in 2021, 2022 for like a Dragonfly inspection tool, most of that was for kind of smartphone markets where they do [indiscernible]. How is the fungibility of those tools? Like, as the utilization is low on maybe smartphones, are they fungible? As in, can they be used from in an info process to a CoWoS process? Is that a possibility? Or do you need a different set of tools or different concepts?
