Mike Slessor: So Charles, I think as we’ve talked about in the past, seasonality has become a little bit elusive in this business, and I think that’s true here in 2023 as well. If you look at the way Shai parsed the bridge from Q3 to Q4 for you, FRT at the midpoint of our guide, the absence of FRT is about half of that $7 million delta. We’re then seeing weaker foundry and logic demand, which we believe is short term, really, given the outsized contribution from foundry and logic in Q3 from a set of major designs there’s a bit of digestion going on of our customers as they use those probe cards for these major designs and major products. Offsetting that some is some incremental strength in DRAM which, as we highlighted, really is driven by HBM and some DDR5.
But those are the different pieces to it. Again, calendar seasonality, although there are components to it, something like the strength in HBM is superimposing what might be the — superimposed on top of what might be the regular calendar seasonality and cyclicality in DRAM. And as I said, I think we’re seeing some degree of digestion of the strong third quarter results in foundry and logic.
Charles Shi: I was hoping you can provide a little bit more clarity on how much HBM revenue you got in Q3 and how much you were expecting in Q4? Because you always said Q3, Q4 is like more than half of DRAM revenue of your largest DRAM customer stuff, but we actually don’t know, right, your largest DRAM customer, how much revenue is it in there within that $28 million to $30 million revenue for — I mean 2, 3-ish in Q4. So can you kind of help us understand what has been the progression of HBM revenue over the last two quarters and going to the next quarter? Because in June, you did say it’s $10 million-ish in June. We just want to understand how lumpy is it, is it showing any sequential growth over the last two, three quarters?
Mike Slessor: Yes, there is going to be volatility quarter-to-quarter. I ballpark you that the absolute magnitude of it in the third quarter of HBM inside DRAM was about the same in round numbers, $10 million, we do expect that contribution to strengthen in the fourth quarter. But again, with probe cards being specific to each individual chip design and things like DDR5 also showing some significant activity and different customers ramping different designs at different times, we want to look at this on a multi-quarter basis, similar to the comments I made around market share. So on an absolute level, about stable. It was heavily concentrated in one customer. As I said, surprising perhaps that half of our DRAM revenue at this customer was associated with HBM.
But we’re looking at this as a longer term trend where we continue to want to ride this tailwind, but also build differentiation and share not just the one customer who’s leading in HBM DRAM right now but each of the three major DRAM customers, and we’ve got activity with all of them.
Operator: And our next question comes from the line of Vedvati Shrotre from Jefferies.
Vedvati Shrotre: I guess can you please help — now that you have HBM and your chiplet die based microprocessors ramping, can you help me understand if there’s a difference in test intensity for HBM devices versus chiplet MPU?
Mike Slessor: No, I think across the board, we see similar increases in test intensity and we’ve quantified that for you in the past somewhere between 20% to 30%,and I think that’s a good rule of thumb for both HBM and some of the logic and foundry designs for that matter. Some of them are quite a bit higher than that, some of them are less than that, because it’s a function of what the customer yields are for the individual chiplets, as well as the degree of complexity that they need to test each individual chiplet. HBM is interesting because you get another — you get multiple test insertions. Each of the chiplets as in the MPU case is tested. But one of the interesting things we’re seeing in HBM is as the stack is being assembled, there’s some intermediate test insertions as well and some of them at very high speeds to make sure the die is going to work, that all kind of rolls up to a 20% to 30% test increase but there’s a pretty good variance around that number, too.
Vedvati Shrotre: And then for my second question, so it goes on to the earlier answers you provided. So in foundry logic that just you’re competing with another competitor for each and every design. Do you — when do you know if you were the market share leader on a particular design, is that — like do you understand this before the design starts to ramp or is it kind of an aftermath?
Mike Slessor: No, you understand it before the design starts to ramp, because our customers don’t really want to surprise us. Our lead times are relatively short. So you do have visibility into what designs you won and what designs that you have gone to your competitor. But that is not entirely predictive of the revenue, because as we saw in the third quarter, customers can accelerate or decelerate their wafer start plans on that particular design essentially in real time based on their customer demand. And so you can have won a majority of designs. But to put it simply, if you don’t get lucky with the ones that are ramping, that might not result in a majority of the revenue based market share but that swings both ways. Our teams are very focused on working with our customers to make sure we’ve got the maximum number of design wins.
But as I said, picking the winners is something that’s extremely difficult to do and probably impossible to do, to be honest, given the dynamics in the industry.
Operator: Our next question comes from the line of Tom Diffely from D.A. Davidson.
Tom Diffely: Mike, just another question on the photonics business. So obviously, it’s been good for your systems business over the last several quarters. But what does that look like in a couple of years when it goes from the lab to the fab, what kind of opportunity is that for you?
