Kopin (KOPN) CEO Michael Murray: Breakthrough MicroLED-Based Datacenter Optical Interconnect Tech

This is very exciting technology that kind of meshes with Moore's law, if you will. So, Moore's law, we all know, isn't a law anymore. GPUs are running at 1.6 terabits and 3.6 terabits. The problem is the middle GPU to GPU. Copper wires are the same copper wires and sets that I used when I was in high school engineering class. And fiber optic lasers are fiber optics. They're power hungry, they're big, etc. Micro LEDs are right in between the middle. This is a new type of communication technology that I think is going to break a couple of the walls down in AI adoption. So that's one. Two, this is very much something that Kopin already can prove that we can build. So I'll walk that through. And then three, the company is in a very unique place in this transformation plan that I've put in place. Our US defense business is flying. NATO is now growing and this third kind of leg of the stool if you will is certainly going to be a strong revenue driver this year and next year and for the rest of the company.

Well, thank you very much, Martin. Thank you for having us. I'm excited to share a lot more information today.

Sure. So, Kopin just received a $15 million purchase order to develop what we call our neural IO solution. And essentially what that is, Martin, is the convergence between a microLED and a bidirectional transceiver. And Kopin has already had these in production actually for multiple years now. Kopin actually invented neural display which was the world's first bidirectional micro display. We're coupling that with our microLED technology which is ultra bright and efficient and low power. And we put those two things together to create neural IO which is breaking down the copper wall. And what the copper wall is is the ability for GPUs to connect through copper sets as an example. But now the problem that we have with that technology is that this technology is actually gating AI adoption rates because it's just simply not fast enough and the bandwidth isn't there and it's too power hungry. So microLEDs sit above copper technology but right below fiber optic technology which is obviously very power hungry but can solve some of the bandwidth issues. So very exciting technology for Kopin, new market penetration for us and a great partner that we're going to go focus on this market with.

In terms of bandwidth obviously we're able to run multiple channels using a microLED at various bandwidth speeds. So as the bandwidth increases with multiple channels you can provide more data through that pipe. So obviously from a copper standpoint it's tapping out at around 800 Gbits with multiple channels we're able to achieve about 1.6 terabits which we want to demonstrate this year. Next year we think we can get to 3.6 which is going to enable multiple new AI GPU transceivers. Furthermore when we think about fiber optics as an example certainly the bandwidth is there but unfortunately they have to remain on, they create a lot of heat, they use a lot of energy and physically they're bigger than a microLED so those multiple channels that are required will gate the adoption rate for fiber optics. Now we're not great at everything, microLEDs are fantastic for GPU to GPU technologies, board to board technologies as well as rack to rack. So, we're good to about 15 to 25 ft using a microLED. It's much more economical. It certainly is less power and it has a tremendous amount of capability to provide the bandwidth that is needed in today's data centers and tomorrow's.

In some cases there may not be a cable at all. If it's short enough distances, we can potentially, and this is one of the things that we're testing currently, is actually using no cable whatsoever and actually transmitting the data using light over the air. This will allow the ability to cool a data center from a board-to-board perspective much more easily since you don't have copper connects, you don't have fiber optic connections everywhere. So board to board we think we can actually accomplish a lot of that using just strictly light, an over-the-air transmission or using a wave guide which is basically a light pipe if you will for the layman's terms, and or we can also use a fiber optic cable for longer ranges but ultimately I think the point here is a microLED acts like a laser for only certain distances but it uses a fraction of the power and it's also a fraction of the size and cost.

Yeah, great question. So this purchase order is actually for the development of the prototype. We have many of the pieces already working. As an example, we have a microLED in production. We're very unique in that we're the only US company that builds microLEDs in production. It's currently flying in an aircraft. Secondly, we've already invented and patented neural display. That's demonstrable. People have seen it. They've worn it. They understand that we've built that technology and it has an AI enhanced back plane already in the device. So those two pieces are already in production. We know that we can do these things. The next 15 million that, or the 15 million we received recently is to productize that into a demonstrable form. We'll announce that later this year potentially at CES is what the goal is. And then from there, we're focused on actually making that technology production ready, meaning through the fabrication plant that we have here within Westboro, Massachusetts, which we received a $15.4 million IBAS award for. This is public knowledge. The Secretary of War announced a $15.4 million investment into Kopin to build the microLED production line here in the United States, which also makes us very unique in the fact that we can walk upstairs and work on this technology and not have to go to multiple different fabrication plants. So that first 15 million is to develop the proof of concept and actually make the device demonstrable. We'll get that in the hands of folks that are building GPUs and also work with the interfaces that we need to work with. We've signed multiple confidential NDAs with some of the largest companies in the world to work with them on making sure that we can communicate with them and then transmit data with them either GPU to GPU or board to board or rack to rack. The next order that we're expecting in 2027 will be about the same level if not a bit bigger to productize the device and get it through to production.

So for Kopin we always try to stick to our knitting with defense first. Obviously getting AI to the battlefield and as close to the battlefield without a lot of power consumption behind these data centers will be beneficial. So Kopin's working with many government agencies to bring this technology forward. But for the greater AI infrastructure there are significant headwinds around growth meaning firstly the bandwidth issue which we talked about Martin earlier which is 1.6 terabits or 3.6 terabits having to drive across a lane that can only support 800 gigabits per second as an example. It's just not working. So that's an issue that we're working to solve. And obviously the power consumption and cooling which we talked about but overall when you look at the brownfield AI infrastructure things that are already existing today we can help with that level of technology with these types of interconnects that will lower power increase speeds and bandwidth without really having to do much else. But for the greenfield applications where processing speeds are reaching well beyond Moore's law like 1.6 terabits or 3.2 at 2 terabits per second. That's where I think there's going to be a lot of new technology like our neural IO capability that's going to allow these data centers to communicate more crisply. The bandwidth will be there, the cooling will be there, there'll be lower cost. Shrinking the footprint of these data centers is critically important for water usage as you correctly stated. So, there is an impact that we're going to be making to society with this new type of technology. But more importantly than anything, this is a new adoption rate of a new way of electrically connecting devices. Using optics, whether it be quantum, whether it be GPUs or CPUs, being able to connect over light, is going to be the most efficient, cost effective, and lowest power solution. And this really needs to happen now. This is the first, I'd say, chapter in a new novel of electronics. And I'm very pleased and proud to be part of it with Fabric AI.

I think there has to be Martin because what happens is as these new chips come out, whether it be the H200 as a perfect example, as they come out with new chips, there's going to be the legacy data center that you also have to connect to, right? So, I think there's going to be interface boards. A lot of the architectures do support field upgrades. We won't be able to upgrade everything obviously because there'll be some hardcoded and soldered boards and what have you, but we do see the connectivity between the current existing brownfield and the new technology is a great place for Kopin. I'm not going to go into exactly why. There's some technology that we built into neural IO that really helps us do that. And it's something that we've been working on through software but I think there's an opportunity for Kopin to take a leadership position merging the brownfield and the greenfield and allowing that co-processing to happen across various types of technologies CPUs GPUs and memory with existing architectures into the new architectures and that's critically important for the adoption rate and cost structure.

For some of your readers and listeners, they may not be aware that Kopin is a 40-year-old company that I took over just 3 years ago and we're in our second stage of my transformation plan here at Kopin, which is really focusing on US defense first. We've now brought up our NATO and European allies using a relationship with Theon International, a fantastic company out of Greece and this technology is actually leveraging a lot of the know-how around our micro display technology and capability. I've taken a bit of a different view of it. We're adding more software capability, much more synthesis in terms of being able to manufacture this technology. And this is now the second part of our transformation plan here at Kopin. And it very much leverages current technology that we have, but adding in some software components that normally companies and investors wouldn't necessarily think about with Kopin. And it's really working well for us based on this new focus that we have, which is defense first. And we have a medical business as well that's really important to us. And this is kind of the third leg of the stool, if you will. This AI infrastructure leg of our business is really going to solidify the company moving forward. And it's a great growth driver for us. I think it's going to be highly profitable and an exciting market. We're trending right now to be entering a $69 billion market in AI, infrastructure, communications by 2030 is the date and data that we have. So very exciting part of our new look and feel here at Kopin. And we're super excited to be part of it.

I would say as somewhat of a summary, Kopin is very unique in that one, we already have a microLED that is capable of 1.8 million foot Lamberts of brightness which is literally blinding. So it already feels and acts a lot like a laser. Two, we invented neural display several years ago and it's fully demonstrable. Many of our investors have seen it and actually worn it and made it work. And then thirdly, the ability to connect these two technologies and build them right here in the United States in North America and support North American CPU, GPU, and memory companies is very unique for Kopin. But the point is, we have this technology today. Companies that are trying to adopt or create this type of technology, they've never done it before, let alone have their own manufacturing line capability. So, our competition, of which we I'm sure will have some, they have to figure out how to build a microLED first and then build it somewhere where they can control the infrastructure. Kopin already has these pieces. We're just putting them together this year so that we can deliver them to the market next year.

You, too, Martin. Thanks for having us.