The analysts and, group head of the Energy and Sustainability franchise here at, William Blair. And, so we focus on four different areas: generation, energy efficiency, so once you create an electron, how do you make it go further? Which is where Aehr actually fits in. Energy storage, which my colleague, Mark Shooter, is an expert in, and then, energy services, Tim Mulrooney, maybe you've heard, focuses on water remediation. And so in the efficiency area, we talk a lot about the trend from silicon to, compound semiconductors, such as gallium nitride or silicon carbide. And it just so happens that, simultaneously, the move to modules in semiconductors is causing a quality assurance, or sometimes referred to as burn-in, move from individual devices to a wafer level, which is what Aehr specializes in.
So, I won't tell you the whole story. I'll have Gayn. He's a much better salesman, and we'll talk and brought some toys for show and tell to kind of show the products that they're in. So, so I've got Gayn Erickson, CEO, to my right here, and Chris Siu, the CFO of Aehr Test. Before I turn it to Gayn, I know he's chomping at the bit, please refer to our website for any important disclosures. And with that, I'll turn it to Gayn to share slides for maybe 10 minutes, and then we can do some questions and have a proper breakout up at Jenner A after this.
Okay, thanks, Jed.
All right, well, thank you. I actually, this presentation's available to everybody through, through both our website and, I think through this conference. It has about three times as many slides as I'm gonna cover, but I'm gonna just tap through them, to just do a real quick overview of us, kind of get to some of the key points. And there's some assumption that you folks know who we are, where I'll dwell a little bit more time in terms of what's more notable and, and, and interesting in our story. The green button is supposed to advance. It's not advancing.
The big green button.
Yep, the big green button, or the green light that's lighting.
The big-
The big green button that's lighting.
Yeah, underneath that. No, that's the laser. Maybe you should get closer.
All right. So, a little bit about Aehr Test real quick. So, we're in semiconductor capital equipment, test equipment. I've spent my whole career in that. And we're just part of the supply chain of semiconductor manufacturing, quality assurance, bring up a semiconductor, semiconductor processes. And most of our revenues come from actually what's referred to as production test and burn-in or screening, and that's taking semiconductors that are basically functionally good, but have a higher infant mortality rate than the application would allow. So we'll go through a screening and kind of a stress test on these devices using some interesting formulas and all, to screen out the bad devices without killing the good ones. Okay? And it works now. Forward-looking statements, you've all seen that many, many times, and that is me.
So we've been doing this for a long time, with over 40 years. We have thousands of systems actually in production. Our big claim to fame and what's driving our business right now is all about the transition of semiconductor test towards the wafer level, that you'll hear in a lot of industry conferences that are going on. This has been something we've been embarked on for years and are notably unique in this space. We have a long list of customers. This doesn't even include all of them. Some of them just simply don't let us use their names. But we've been doing this a long time, and people recognize us in our industry for our expertise in this stress and burn-in requirement. Okay? Just a quick update. Our fiscal year just ended.
We had said before, about a couple of months ago, that we were gonna finish with revenues greater than $65 million, profits, GAAP profits of over $11 million. Just to give you a size, you know, we have a great balance sheet. We have no debt. We have great working capital. Technically, actually a lot of inventory, if you follow us right now, just with some of the things that have been going on, of the products that we're out selling. We're really a market share leader in this new segment that candidly, we coined and phrased, wafer level burn-in. And in this case, for the silicon carbide, we have seven customers. We are engaged with all the major suppliers around the world right now, and over a dozen companies actually doing evaluations as we speak.
Candidly, we're still at the very front or early innings of the transition of silicon carbide. You know, basically, we see this as a growing market, somewhere in the 25% range over the next handful of years. A couple other big things. We shipped a high-power system this last quarter that we talked about, that was for silicon photonics and optical I/O for chip-to-chip communications. We have a new lead customer that is doing a very high volume data storage, that's probably gonna be a 10% customer for us this next year, which started June 1st. We put out an 8-K because of these next two bullets, because of the materiality of the engagement with a NAND memory supplier that we have been talking about for several years.
And this is the first call we're making to actually engage with them this year, that we think is gonna drive business in, in 2026. And, a commitment, kind of the biggest one, is- has to do with this AI accelerator. Something most people are shocked at, that you could ever do wafer level burn-in of AI processors at wafer level. We've been able to, do some early prototyping and validation, and now we have a commitment to move to wafer level with intent to go to production. Big drivers for us, semiconductors growing from, you know, $500 billion-$600 billion to $1 trillion, really driven by the AI, green energy, and digital transformation side of things. More and more semiconductors are less reliable.
Most people don't realize that with these smaller geometries and bigger sized devices, there's more need for reliability test and burn-in than there was in the past. There's also more devices like silicon carbide and gallium nitride, and photonics-based devices that have higher infant mortality rates that's driving for it. And lastly, more and more devices are going into multi-chip modules or heterogeneous packages that are driving that burn-in to a wafer level state, which is right where we are particularly strong at. This is kind of an interesting thing around Moore's Law, where even while he was alive, he was talking about probably at the end of Moore's Law, the shrinking semiconductors, in order to continue on with that, you're gonna see more and more different flavors of semiconductors packaged together into what Intel coined a heterogeneous integration.
So this decreasing semiconductor reliability going into applications that really matter has been the big trend that's driving for more and more test, more and more spend, and more and more reliability testing. And then the increased need for known good die related to multi-chip modules is really driving the business towards us. So a lot of discussion has been around the electric vehicles and silicon carbide, and things like this module, which actually has 48 silicon carbide devices in it. This is the main, most people in our industry would recognize this. This is actually the module that Hyundai-Kia is driving as their main platform for all their IONIQ device platform, and the SUVs, and everything. These devices are failing at about 1% each during the life of the car.
So we weed them out during wafer level before they go in here, such that none of them will fail during the life of the car, and it's just a showstopper. And if you didn't do it at wafer level, you'd be doing it at the module level with a 48% yield loss. So it's really simple math, and it's very easy for people to understand, but it's not the only application. Intel did a presentation during OFC just last month, actually showing co-packaged optics, where they're putting optics into chipsets along with memories, and they're communicating for the first time between each other optically. That's a big thing that's driving our business.
NAND has been doing stacks for years, and those stacks are requiring more and more burn-in at the wafer level, and we believe that by the end of the decade, we'll start to see people in mainstream doing DRAM wafer level burn-in as well. And then, notably, because we also introduced this benchmark, it is not NVIDIA. I'll just go out with that, but people like NVIDIA. If you look at what's going on with the HBM memories on the stacked devices, something like this, you can actually see the compute chip alongside of the memory. There's a real desire for both the memory and the compute chip to actually be burnt in before it shows up in these systems.
Okay. Our market drivers, you can follow along with some of them, but the major trends in electrification, electric vehicles, and data centers is driving a lot of our business. I already gave a little bit about the silicon carbide market. For people that don't understand this bathtub curve, basically, all semiconductors have an infant mortality rate that is, fails higher early on than later. The kind of rule of thumb, if it hasn't failed within a year, it won't fail for 20. All we do is we accelerate that year with accelerated features that allow us to take, in, say, 24 hours, make the device look like it's one year old, and every device that's still functioning gets shipped to their end customer, and the bad ones are weeded out.
The trick is to do that without wrecking or decreasing the reliability of the devices that you're testing. There's been a progression of silicon carbide going from the original Tesla Model 3 inverter that oftentimes I carry with me, it's about this large, that was done in discrete devices, and a shift to basically every single electric vehicle company in China, Japan, Korea, Europe, and United States are all building based upon modules, like the one I showed here. That's been driving the wafer level testing to the wafer level for the obvious reasons we talked about with yield.
This just shows sort of graphically the math, but the, the nut is, it really is when we see these benchmarks where people are taking devices that have, say, a 10-die module to them and having 10% yield loss, and when they went to 48 die, it's just astronomically impossible to do this without wafer level burn-in. You-- This is like baby pictures for me. You'd have to be a tester, geek, to actually love this, but this is what a test system looks like for us. In the industry, there's nothing that looks even like this because we tell people this tests 18 wafers at a time, and they're like: "What? How do you even do that? No one does that. No one. People can only do 1 wafer at a time.
You're doing 18 wafers at a time." And so it's through novel architectural differences, and IP, and patents that we're able to pull this off, and through the selling process, get more and more customers on our platform. And it really takes the technology of the test system, the consumable contactor, and the alignment mechanism that is a turnkey solution from Aehr Test to get you there. If you followed our story for a while, there was a time when it was, "Are there really gonna be people that are going to do this?" Because you have to be completely all in. You buy all of it from Aehr Test. And some of our lead customers that were notable in our 10-K, because of 10%, were the likes of Intel and Apple. ON Semiconductor, STMicroelectronics.
Some of the biggest names in semiconductors were our first lead customers on this, and now we have a quite long list of customers on this platform. Okay. Big trick around the differentiated in terms of being able to test. One specific thing about what we do that's different is every single device through the burn-in process; we can give 100% certainty that it was burned in. And this ends up being a really important thing, and particularly to the automotive guys, 'cause any device that doesn't get burned in and gets shipped has a 1% chance of failing in the field. 10,000 cars per million would have a catastrophic walkaway event if it didn't get tested. So this is a critical thing and one of the key differentiators that allow us to get the margins that we do.
You really have to understand this. I won't, I won't bother you with what it takes in silicon carbide. We do an example of our cost to test models. We share this all the way up and down the supply chain so that the EV manufacturers understand what it costs to actually get to the quality levels that they're talking about. But you could do a 12-hour burn-in for $0.12 per die, and most people are shocked at that. That's actually a really attractive thing when the device is a $12 device or something. For those that follow us, we're also expanding in gallium nitride, which is being really towards data, solar, and some automotive applications. We think this year will be the first year that we actually start seeing meaningful revenue. It could be in our fiscal year, a 10% customer for us.
Silicon photonics, the high-power system that we ship is pretty exciting 'cause we think it's the leading indicator to optical I/O, and optical I/O is coming. AMD, Intel, and NVIDIA have all announced their plans for chip-to-chip communication. We will all wake up in a few years, and more and more computers will be talking to each other chip-to-chip directly with optical I/O, which is a big deal in our space. This is an example of a co-packaged optics demo that was just done at OFC by Intel. Again, baby pictures for me. This is our fully integrated aligner. This is a really big deal for those that have followed us. We've been talking about this.
We did this through the COVID downturn with the input from the memory guys, which allows us to do fully automated handling of this, which is novel in this space as well, and keep our separation away from yet to be announced competitors. At this point, we still stand alone, really, as the, the primary game in town. We made three more bullets here. For those that followed us, we actually had a big customer that had bought our system right before COVID and then went dormant. They actually just went quiet. They were working through this whole process. They've come back, and it's a hard disk drive application. Think of heads that are billions of units per year that are gonna drive volumes. They started forecast volumes with us this year. Also could be a 10% customer.
People that followed us closely, it's like, "Oh, yeah, I forgot about those guys." They never went away, and this is something that's notable for our forecast. I mentioned the flash memory customer. The first NAND customer committed to an on-wafer benchmark. Probably take us the bulk of the year to actually get through that benchmark. That should tee us up for being able to address their production needs, potentially as early as our fiscal 26, which starts in a year. This is a big deal to me personally in our company. This has been something we really built this platform around, and so this is a notable uptick for us because the NAND flash opportunity is much bigger than the silicon carbide, for example. It's one of the biggest in the world.
Ultimately, DRAM will even be bigger, and this is a leading indicator to it. The last thing is an AI accelerator. Actually, we teased people at the last earnings call. If you pay attention to every word I say in this, I had introduced this high-power system that can do 3,500 W per wafer. The next best is, like, 300. I mean, people are shocked that we can do that, and were wondering what we were up to, why we were doing that in the optical I/O space. Well, one of the things that we're also working on in the background was the need for a really high power at wafer level for AI processors. Again, the processor itself, the advantage would be to do it before you put it into a large system or a large with HBM.
But otherwise, you still have to burn these in. All of these are being burned in today. So they're all going through a multiple-hour test beyond the normal functional test to weed out the infant mortality, and when they do it, they throw away the whole thing. There's also people that do it at system-level test, something all the way like this. So we offer a much more cost-effective way of doing it. Just one sort of technical, geeky thing. If you talk to someone in the space and say, and repeat, "Gayn said that they're doing wafer level burn-in of a processor," you'll hear sort of a cough, something with the baloney. The reason is, there's no way you can do that.
If you understand all of the the processes and the equipment that's been out there today, no one would be able to project the ability to do wafer level burn-in based on everything they know. And the reason is, a device like this takes hundreds of amps of current, and if you're doing full wafer, you'll be doing thousands of amps of current. Well, nobody can put 1,000 amps of current onto a wafer, and we do. So it's like, well, how do you do that? I brought this fun little toy here. This is a battery cable. It's rated at 125 amps at room temperature. At burn-in temperature, it'll do less than 100 amps. So to do 2,000 amps, you need 20 of these. How are you gonna get 20 of these onto a delicate wafer and burn it in?
The answer is, you can't. There's no way you can do that. In fact, you don't wanna do that. If you try to put a 100-amp power supply on a wafer, when one of the devices fails, you'll completely burn up the probe card and destroy the wafer every single insertion. So people say, "Well, you can't do that." Well, we do it differently. We do it instead with some novel ideas of thousands of power supplies, each one with its own capability, that allows us to protect the device and the probe card, and nobody does that. So we're doing things differently, and it's helping us to create new markets.
We also just say at the bottom, this is all about wafer level burn-in, but a lot of the information we've been learning about is around the package part burn-in, and so we'll just leave you with a teaser, stay tuned, related to some package part burn-in opportunities that we'll be announcing in the coming months. That's it.
Is this on? Oh, great. So Gayn, thank you for the presentation. It sounds like you got a lot going on, so maybe if I could just try and simplify with silicon carbide to start. It might be helpful if you talk about silicon carbide with the inherent defects in the material.
Mm-hmm.
That seems like a great place for to kind of prove out your technology. And I thought you did a good job talking about the module, but maybe you could just walk people through that yield-
Mm-hmm
of having two die in a, you know, in a package versus having 48, and, and kind of just go through, 'cause that cements the economic-
Right
value proposition a bit more.
All right. So the devices that are being used in silicon carbide for these, for the inverters, are simply a switch. They're a MOSFET, okay? The oxide is used as a gate to turn on and off the current. You apply a voltage, when you do that, current flows. When you turn it off, current doesn't flow anymore. If you do that in a fancy way, you can actually get a DC battery to push current up, then turn around and push it down again, create an AC waveform, and that's actually what drives the three-phase electric engine. The trick is that when you turn the switch off, it needs to turn off, and if you turn it on, it needs to go on quickly.
The material defect is actually in this oxide layer that is the protective surface that it turns it on and off, and it's extremely difficult to make that perfect. In fact, it's impossible to make it perfect today, because the more robust you make it, the slower it goes and the more resistive, so it actually affects the performance. So people are riding the line to make it as thin as possible so that it has the highest performance, but it has the ability that a few percent or 1% will fail in its early life. So the optimum configuration is the ones that have about 1% failure rate. So they actually design them in such a way that they're right on the edge, and they use testing to actually weed out the infant mortality. For clarity, what does a failed device look like?
It actually doesn't work anymore, and unfortunately, it works in the worst way possible, and it actually no longer stops current. So when you go to turn it off, it doesn't turn off, and so it dumps the entire battery current through one leg of this thing, and it'll take the module, and it'll completely molten the whole thing. It'll melt all of the metal and everything else, 'cause you'll have thousands of amps that go through here. And that we call a walkaway event, okay? And that is because when this happens to your EV, it doesn't catch on fire. They've made really good provisions, 'cause that is the failure mode, but you get out of the car, and you walk home, and the entire inverter engine has to be replaced. It is a really, really nasty failure.
Now, that individual device is actually done in parallel in order to be able to have more current. So for a 400-amp module, you might actually have 4 or 8 of these devices in parallel with each other, all switching on and off. There's some electrical goodness to it, but any one of those failing will actually create that failure. So there's about a 1% failure rate, and so if you were to put these in here, the likelihood that this fails during the life is basically every other one would fail. Now, you'd say, "Well, that's ridiculous. Why did they do that?" Well, because the silicon carbide is so much more efficient and so much better at electric vehicles, that it is won out in basically every one of the designs that are going into every EV going forward, and they're all putting them in modules.
But the by-product is, you would have to do wafer level burn-in. Our test with our lead customer proved that it was feasible and it created this new market, and we're actually directly talking to the OEMs, which are the EV manufacturers, about that quality to ensure that it'll be there, so that these don't create millions of walk-home events.
Maybe just to put some numbers around it, so tell me if I'm wrong here: So if, if that device that you're holding up costs about $600, and it has 48 die in it for somewhere between $5-$10 for each die-
Mm-hmm
so let's use 10 to make the math easy here. That'd be 480, die,
Dollars, yep
where you have $480 out of the $600. So if you have one die that fails, that would be a yield that would be unacceptable to-
Sure
to sell that module. Original designs, which I think are like this right here-
Yep
where you would have two die per package.
Yeah, that's what Tesla's original design was. It had two die in it.
If I have one that fails in this, I'm only losing $20 versus losing the $400.
That's right. And technically, the one was already dead anyhow, so you're losing 10 more. So the perfect is you'd only lose 10, they were losing 20, but it was cost-effective to do it in that package form, and so they did it that way. But as you go to the i f you go to the module, no way.
So for people in the audience, the chip is actually what's failing or what's passing that's inside this package, but you need to have it in a package to get current into it.
Correct, to be able to make contact and to get the heat out of it. That's right.
So the move to module is kind of you need both in parallel. You need the issues with silicon carbide or gallium nitride-
Mm-hmm.
that increase the failure mode of the material, the silicon oxide-
Mm-hmm.
as you talk about, and the move to module that triggers this wafer level burn.
For sure. That is correct. It is, it is consistent with multiple trends of semiconductors that are less reliable because of size, process, compound semiconductor defects, that are being put together with other semiconductors into a package with more of the same devices or different flavors of devices. That is the discontinuity that's driving wafer level burn-in that 2, 3, 4 years ago, nobody had even discussed.
You've proven this with one customer in silicon carbide.
Seven customers.
Well, you've got seven-
Only one that's really ramped.
but one that's ramped to, you know-
That's right.
a high volume. So if the trend to modules is to continue in silicon carbide , presumably those other six will need to scale to some level. Can you talk about and make the bridge now to gallium nitride into some of these other markets that you talked about? Because is it the same value proposition that you're having to go to a module in memory, or is it something else that's driving in these-
It's mostly similar with some subtleties. So in NAND, for example, people are putting multiple devices in the same stack or in the same package to achieve the density requirement and the high speed for, like, solid-state disk drives. And if they don't burn them in or cycle them, and they have a failure in that, they could take out the entire package. So in that sense, it's very similar. Why they're in parallel is for greater memory density, you know, more, you know, more data, whereas this is for higher current, but it's a similar analogy. DRAM, similarly, is, it really with HBM and the AI processors, the new architectures are taking these devices, and they have 6, 8, or 12 devices stacked, and then four of them around here.
This is a very new problem that is what I think is gonna start driving for wafer level burn-in over the next few years, because this is a very expensive problem. But there's some things that have to be done to the DRAM to be able to enable wafer level burn-in, and we've shared that with the DRAM suppliers to be able to do that. So we kind of hold that to ourselves, but there's an opportunity, we think, as we build the wafer level for NAND, it can apply to DRAM. GaN, by the way, gallium nitride, is a little different, because there's not that much evidence that GaN is gonna be put into modules. So it's like, well, why are people talking about wafer level burn-in? And I don't have it.
Maybe in the breakout session, I have a little GaN package, but if you look at the GaN RF and GaN power devices, they're these very novel, high-speed, very expensive packages. And so you go buy a GaN device, go look on DigiKey or something. Instead of it being a $10 chip, it's an $80 chip or a $400 chip. You know, it only has one chip inside of it. The complexity of the thermals and the high speed and the packaging are what drive the cost. So simply, all they wanna do is they don't want to package up that cheap die and have it fail for one package. So it's not as obvious, but it's for the same reason of trying to avoid the cost of the packaging to have it fail in package form.
Do you replace test, or is this in addition to test?
It's clearly in addition to test or shifting the burn-in from a package or module level back to wafer level. So if you're a tester company, which I consider, like, Teradyne and Advantest, Cohu, we are a process step in line with them, and we don't really compete with them.
I guess, you know, as you look across your, your businesses, do you still see silicon carbide as kind of the near-term opportunity for you from a scaling, or have some of the recent discussions from, you know, GaN, silicon photonics?
Yeah.
or memory, have those pulled in such that it's changed any of your expectations?
Yeah, so-
in terms of ramps?
The way I might say it differently is, if you go and look at what we've been saying over the last couple of years, we've been talking about the newer applications. Candidly, the AI we've been holding really close to our chest, but like memory, for example, the GaN application, the silicon photonics application, they have been coming. Last year's euphoria related to silicon carbide and EV kind of dwarfed it all, but we were still working very aggressively on those programs. They haven't really pulled in; they're just coming on time. And it isn't that silicon carbide is going away, it just doesn't seem to be as shiny as it was last year. You know, when we were building models, a couple of years ago, the whole premise was that by 2030, 30% of vehicles would be electric.
And at that time, it was like, "Come on, that's a shocker. That would be amazing if it was." And we said: "I think it's going to be, here's why, and they're all gonna be using modules. That's gonna drive $1 billion-$1.2 billion of our equipment cumulative over that period of time." And that was our value proposition for that one segment. Last year, the numbers were going crazy. There were people throwing around ridiculously higher numbers. Everybody's gonna drive an EV tomorrow, and it, it wasn't that we were that smart, it's just we weren't repeating that. It just didn't seem consistent with what we were hearing with the customers. The customers were all focusing on those large ramps from the likes of VW Group and Mercedes and BMW and, and domestic, that were in 2026 and 2027, consistent with the 30%.
This year, they're just still doing the same. And we saw something that I guess the chairman of Toyota, who's always been really negative on electric vehicles, made a comment a couple of weeks ago, and he said: "Listen, this whole EV thing, it's not what it's all cracked up to be," which I think got him fired, but originally, he said, "It's gonna be about hybrids." And I agree. I think that what happened in the last year has really emphasized why the importance of hybrids, and hybrids, no doubt, are gonna be bigger, which also use electrification infrastructure, than they were going to be. But he still says b ut EVs are still gonna be 30%.
So I don't think it's just we've kind of gone through this euphoria cycle and back up again, but Aehr Test is really busy, and silicon carbide is still gonna be at least half of our business this year. So we are excited to have multiple segments in addition to silicon carbide as we continue to grow.
All right, so anyone who is inspired to do some more math and to get into some material science, we can take it to the breakout in Jenner A , and let's thank Gayn and Chris for the talk. Thanks.