That are familiar with our story, folks that won't be in here. So the intent here is to just give you a taste of our story, enough to get you to see if there's maybe some potential interest, and then we'll invite you to follow up with us to have a dedicated call or even a visit if you'd like. So let's talk a little bit about it. Always the forward-looking statements, they remind me to say, and that's me, maybe a few years ago, a little more hair and a little darker. Just something on Aehr Test. We've kind of had our splash screen for a while. This first one, we're actually now incorporating the new platforms that we acquired through the Incal acquisition. And for those that are following along, that acquisition has gone really, really well for us already. I only have myself a handful.
I've been acquired a few times. I have a solar company. We've gone public. I've acquired a few companies, and I'm sorry, been acquired as well. It isn't always that when you acquire someone, it gets better after you acquire them. That's been my experience here, which has been really nice. The customers, which we knew while we were doing due diligence, continued on in a very positive environment where Aehr Test plus Incal really do make a good story. Both of us have actually been in this burn-in space for over 40 years. We have kind of settled in on what we call the wafer-level burn-in side of things. They have stayed on the package part, particularly on the qual and reliability side.
They've come into the position of seeing a large chunk of the high-performance computing and AI processors, which has made it really interesting because those companies are also asking to go into production and now into wafer-level burn-in, which is a test step that basically screens out weak devices before they are shipped to customers to avoid failures in the field. We have a very large customer install base here. They add to it. We actually haven't added any pictures of their customers now. We need to kind of work through and figure out who we can and can't talk to. We do have some overlapping customers. We definitely have some new customers through the acquisition of Incal and a lot of systems out there. Market drivers, a lot of detail here, but we try and keep this thing updated for those that follow along.
You can see contrast from year to year. But one of the big plays here, and I can't tell you how many times we talked about it, is I'm sure every single CEO up here, except for maybe the last guy, was talking about AI. And it's really a big part of what's driving our time right now, through the acquisition and through inbounds by customers asking us to actually test their AI processors at wafer level. So that's a big one. Electric vehicle, clearly a big one that drove our business over the last couple of years and will continue to do so going forward. And other things that, at a trend-wise, are driving things towards a new type of test and reliability that we're doing at wafer level. First off, you just got to start with semiconductors are growing.
Semiconductors themselves are going to double over the next about six, seven, eight years. And if you look at all the infrastructure and everything that was put in place over the first 50 years, it's going to double in the next seven. That's driving all kinds of capital equipment purchases from test and front end and is definitely driving increased capacity needs for equipment like ours. I get a little bit more specific here. And for those that have heard our story, basically, as semiconductors are growing, they're not becoming more reliable. That's not really intuitive. With smaller geometries, compound semiconductors, and other techniques, the devices themselves actually have a higher failure rate at infant mortality than they did in the past. For us, as a tester company, that's good. That sounds bad.
But through testing and screening, you can actually end up with semiconductors that have ultimately the same reliability long term, but you need to do more testing. The other thing is people are putting semiconductors into more applications where reliability matters, things like autos. And if you listen to me in this room, what, seven, eight years ago, I think you're going to find a slide. And I started talking about how autos were putting things like GPSs in cars and almost as absurd as that sounds now compared to autonomous vehicles and all, but more and more critical requirements for reliable semiconductors going into that. And then the bigger one that's really, really been disruptive is things where semiconductors themselves, instead of being discrete, are being packaged together to combine to meet a functional requirement. And we were seeing that a long time ago.
Just as we see now, you see it from the EV traction inverters that was driving our business last year, co-packaged optics. That's going to be something that's going to, we believe, drive our business a lot going forward. Memory devices, co-packaged in either NAND or DRAM stacks, and then things like the AI processors. I brought an example of that later if somebody wants to actually—you've never seen one up close. These trends, as I described in terms of decreasing reliability, the need for it, and this increasing need for known good die when the devices are put together. Just to slow down and make sure people understand, the devices would be tested independently of each other. If they fail, you throw them away.
But then, when you put them on something that looks like that, which might have eight processors on it and eight stacks of 12-high die, when one of the devices fails when tested at the system level, you throw the whole thing away. So we've been talking about the need to actually test it. And what we do uniquely is able to actually test the entire wafer at a single time and up to 18 wafers at a time. Been in here talking about it for a while. Silicon carbide was one of the first ones that drove it, but we're actually seeing things now in other application markets. If we just look at how we've been growing, our TAMs are growing. This is also something we haven't shared before.
Even though we've had areas of other businesses, realistically, the focus last year and where our revenues were coming from was the silicon carbide put in the EV chargers and the EV infrastructure and inverters and also silicon photonics. But if you look at this fiscal 2025 that ends in May of 2025, we're adding gallium nitride power semiconductors, hard disk drive head components, AI processors at both package and at wafer level. And those that have followed our calls, we've been making investments in the wafer level burn-in for flash memories and DRAMs as well. This is a bathtub curve. If you have never seen this before, basically, in semiconductors, the likelihood that a device fails at day one is higher than over time. And it actually decreases with time, meaning at six months from now, if the device hasn't already failed, the likelihood that it fails decreases.
At some point in time, usually about a year into it, if it hasn't failed, it's going to run at almost perfect until it wears out. Incal, through the acquisition, when we talk about reliability qualification, actually characterizes the wear-out failures. So we're putting devices like all the AI processors through their test systems and actually stressing them to the point of failure to identify how long they will last. What it doesn't do is screen out the devices that are weak or would otherwise fail in production, and the industry right now is scrambling to try and figure out how do we do a production burn-in of the AI processors with all of the chips around them. Silicon carbide, a slide that we shared before, has gone from discrete devices used in the Tesla inverters to everything going to modules now.
So this module, which is a Hyundai-Kia version and Toyota, Honda, Kia, Hyundai, Mercedes, BMW, VW, all of the Chinese suppliers, they're all going to modules. When they do this, this is how you burn it in. And each one of the die on this thing fails at 1% rate during infant mortality. There's 48 die on it. So we would actually have a 48% failure rate in production. This is what drove everybody to go to wafer level burn-in. You burn in the individual die. It's removed before you put it on here. If you actually test this device, it will not fail. So all the OEMs, as the car manufacturers are called, are driving and requiring a certain amount of burn-in time done at wafer level. And so far, we've been qualified on almost every automobile company out there for EVs.
This just talks to the implication of having multiple die on the same package with a 1% failure rate, what the net effect is in terms of the yield loss. These are real data. The AMD processor introduced last week, very interesting example of what these things are and why you would want to do wafer-level burn-in. This is something that I found was really interesting. One of my shareholders actually sent this to me. I guess a week ago, someone referred to this, but Meta had an article in July where they were doing a Llama build or basic one of their training models. During that, they had all these failures. They characterized it and were in a public forum complaining about the reliability of the semiconductors because of the failures. The GPU itself was 30% of them. The memory was 17%.
Half the failures were hardware failures that we can tell you through proper screening would be removed. So they're out pushing the likes of NVIDIA's and saying, "You need to figure out your reliability problem and get a way to test these things because this is really expensive and not funny." We have patents across a wide range of companies or countries, I'm sorry, and a wide range of applications and technologies and processes in wafer-level burn-in. And basically, what we do is we supply an entire turnkey solution that includes a test equipment for electrical testing and stressing, contactors, which is a consumable, if you will, that we call WaferPaks that are bought for each wafer design.
Over the history of the tool, you would come back to us and continue to buy different fixtures, if you will, that make contact with those wafers and an automated aligner for doing this alignment. For those that are close and with more time, we can explain this is completely different than anybody else in the industry, which actually tests one wafer at a time with a dedicated prober, a dedicated tester, and a probe card. In the same footprint of that one wafer system, we're doing 18 wafers. Our cost of test is almost, it can approach an order of magnitude less than the other guys. You say, "Well, it seems like you overshot it." The test times are an order of magnitude longer. That's just the only way people can cost-effectively do this extended reliability and burn-in in production.
So what a system looks like with its door open, you can actually see where we actually insert the 18 different wafers or with a fully automated front end on it for hands-free operation. This becomes really critically important when you get to the memory guys, to all the 300-millimeter AI guys. People do not want to touch these wafers in their fully automated environments. We can actually have, in this case, this one has a 200, 300-mm load port for wafer cassettes for silicon carbide. But on the 300-mm, it uses full overhead handling. So you can actually have the AGVs come and drop off the wafers. You don't have to touch the tool. This is a picture of Sonoma, which is the high-power system for AI. This is an amazing tool for doing characterization of these devices.
But it tests 88 processors, which doesn't sound like a lot, and it isn't for high-volume production, but it's the highest on the market. This thing has higher power than anything else and is actually cost-effectively and effectively testing these devices. And customers are scrambling right now to get these. We have a backlog out already, five, almost six months of tools, and it's increasing as customers and we're bringing on more capacity to catch up to bring that down as people are scrambling to get this capacity for the AI processors that are out there. We have a motto, if you will, or a kind of a tagline of testing without compromise.
Burn-in historically has been wrought with all kinds of trade-offs of people putting them. They even oftentimes refer to them as ovens, where you put the devices in, you bake them, and you turn them on, and you assume that that is testing. That's nowhere near good enough. We have full automated test equipment, precision reference sources, digital signals, receivers to be able to tell every single device is tested and prove to the automotive guys with traceability that every single device was tested, and that's one of the key ways that we differentiate against other companies. This AI processor opportunity is actually something that's consuming us right now. Most of my day right now is spent on both wafer level and the processors on the AI side of thing.
A company approached us and said, "Listen, is there any way you can use that new high-power system that you introduced last year for Silicon Photonics?" We work with them. And what we've said publicly is we are currently on-wafer at our facility testing multiple wafers in parallel to prove the feasibility of correlation to remove all of the failures at wafer-level so that they do not need to do it at package or worse at system-level. The key is to get away from system-level burn-in of these things, test that device before it's put on here, which will drastically increase the yield while also increasing the quality. And then lastly, we've told people that we have been engaged. The FOX system was actually originally designed for flash memory.
What we have said is we captured one of the interests of one of the big flash manufacturers who's asked us to build them a custom WaferPak for their next-generation device that will allow us to be able to test their high-volume flash memory wafers in production. The flash memory opportunity is significantly higher than silicon carbide, that as folks watch this, and so this is something we're really excited about, but to put it in perspective, it probably takes us most of the year through May to actually prove out the feasibility and get us to the next step. Our goal would be able to then capture enough interest to capture an order or commitment for our system that we could deliver the following year.
It's got a little bit of some time ahead of us, but unless you talk to the R&D guys, and that feels like tomorrow. So I'm really excited about this. This has been a passion of mine since coming to the company over 12 years ago. And we're always getting into memory. The other subtle thing is what we call a WaferPak, which is the probe card contactor, is being designed to be able to handle DRAM at the same time. That's all I'll leave you with. Then lastly, optical I/O, an area that we've clearly differentiated. We've been a market leader in there. There's more and more evidence now to show that the AI processors and high-performance computing segments are actually going to start having chip-to-chip communication with optical interfaces, which is something that we excel at in terms of a test and burn-in.
With that, I'll just summarize. It looks like we have about eight minutes to take some questions. So I'll go ahead and do that. Any questions today? Yes.
You may have said this, but I'm curious. With one of your units, how many things can you test in a day?
Oh, that's a good question. Yeah. No, I didn't quite say that. So in silicon carbide, for example, we're testing 18 wafers at a time. The piece that's missing is how long does it take to test them? And so we've been intentionally vague or broad range, but in public forums at the technical conferences, we always refer to typically between 6 and 24 hours. And one thing I'll say is please do not ship a car with an inverter with less than 12 hours of burn-in. So people know that. The OEMs listen to what we say.
We've tested more silicon carbide wafers across the industry than anyone on the planet, for sure. And they look to us for opinions. And that is a strong opinion. By the way, we have a vested interest in long test times, obviously. But we see the data. You can actually see when you're done what is the likelihood of failing. And if you're not down into that range, it should never be shipped into a car. So in 12 hours, right, I'm doing 18 at a time, you get 36 wafers a day. A wafer might have something like five to 800 devices on it. You divide that into cars, and you get yourself something like 10 - 12 cars per wafer. Another way of thinking about it. And is every one getting tested? Every single one of them.
If you miss one die, every single die has a 1% chance of failing. A Tesla Model S, the one I have, has 96 of these die in it. It is guaranteed to fail in the life of the car. So the OEMs are frantic and paranoid to ensure that it is 100% traceable that it was tested.
Maybe one more question.
Maybe. Go ahead.
Is everyone sending these to you?
No. No, we sell them the.
So you're selling them, or are you training them?
We sell them the equipment. They buy the contactors from us as well and everything. We train it and turnkey it. And then we prove it before we go. Now, having said that, we are at any given time, we're probably testing eight to 10 different customers' wafers in our facility through all the qualifications that we're going through worldwide. Yes.
The stock's been on a bit of a roller coaster over the last 12 months.
I'm sorry?
The stock has been on a bit of a roller coaster over the last.
Mostly just down, it felt like. Yeah. No, I mean, so one year ago at this time, the world was thinking that all of us would be driving EVs and there'd be no gas cars in six months, which was always ridiculous. We never repeated that. We've always held that we thought that the car penetration rate of battery electric vehicles would be 30% of 100 million cars in 2030, so 30 million cars. In reality, through all of the thick and thin of it, that's back to where people think it was. But there was a lot of people thinking more than that last year.
We weren't repeating it because I'm looking at the fabs and saying, "They're not shipping anything out of that dirt fab next year." They weren't there yet. Go ahead. No, why don't we just do a couple? Sorry. Go ahead.
So it seems like you had a lot of success with EVs, and now there's a pivot, almost a worldwide pivot towards AI chips and AI in general. Do you see that disrupting anything? Because you said it's consuming most of your day. Do you see that disrupting?
Yeah. So I've actually been pretty clear that it's not a pivot in that in reality, we still love all those silicon carbide guys. We have done the benchmarks. We've been qualified. We have the test systems. We have small incremental enhancements to our roadmap. Customers are asking us for quotes on systems that we can build today in a year.
So it's just not in consumer manufacturing where I'm spending more time on the newer products and markets and customers. Yes.
Can you say anything about the TAM for these markets? I mean, let's say that you get the conversion.
Yeah. Yeah, yeah, yeah. Let me try it this way because I did it a little earlier, so it'll be faster for me to do it. Starting at lowest to highest. Even though there's billions of heads built for disk drives, that's probably the smallest market for us, and we didn't spend much time about that, but that's one that'll come in this year. Maybe it's $10 million a year or something along those lines. Keep in mind we're doing about $70 million, okay? The next one might be GaN, which is a wild card, depending on where it goes.
It has the potential of being bigger than even silicon carbide, but I would still take the under on it. Silicon carbide would probably be the next largest one. This is one we've talked about before when it hits its stride when most of the cars are kicking in, 2026, 2027, 2028. This could be maybe a couple few hundred million dollars a year of business. And then flash is definitely bigger than silicon carbide, and DRAM is the biggest in terms of potential. If you look at a flash fab, it would take somewhere between 60-100 of our $4 million tools for one floor. And DRAM is.
Service?
We get a service revenue, a couple few percent, maybe a little bit more. And then the consumable element, which isn't really consumed, is the contactors needed with each new design.
So maybe the device lasts for three or four years, and then it's replaced. They would throw the other ones away and get new ones. So it's an ongoing revenue. And for example, last quarter, we actually shipped no systems, which is like, "Whoa." We did all of our revenue, met and exceeded expectations all on WaferPaks , just the way the timing worked. So in general, we'd expect about 50% of our business is coming from that consumable, if you will, WaferPak , and half of it in systems business.
Real quick, I mean, you guys made that recent acquisition.
Yeah.
Over recent M&As. Any other M&As that you guys had? Do you see any growth in your future? Is it going to be more?
You know what? It was actually the first company Aehr Test ever bought, and it went really well.
We put some infrastructure and things in place to help us do that well. I won't get into all of it. It would be easier to do another one. There's some interesting things that we could look at. It's not hinting at anything, but there's probably a balance. There's a lot right now organically on our plate. Anyone? Yep.
Competition? What's the landscape right now?
The biggest competition.
Competition.
The biggest competition is package part. About three quarters of all semiconductors by volume of wafers, total wafer starts, are actually burnt in today. Not 1% of them are done at wafer level yet. So the biggest competition is package going to wafer. With modules and things, it becomes very obvious to do it. We've had a couple of small competitors try and do it on a prober as a different alternative, and they've crashed and burned.
One of the next biggest customers right now, we have over 1,000 wafers of capacity. An 18-wafer system would be 18. We have over 1,000 of them out there. The next biggest competitor, I think, has 12. So we're currently dominating that small space that we think is going to grow. We're paranoid, and we'll continue to be too, particularly around China, where even though we have patents in China, they are more likely to try and step on it. And if they do, we actually do plan to take steps against it. Not so much that China itself is so good because anybody who might use those products from it could be threatened by the fact that they might end up having that source taken away. Okay? Thanks. Time's up. I'll go outside, okay? Thank you, everybody.