From radiation hardened hardware to edge computing in deep space, AI is moving off the ground and into one of the harshest environments imaginable. So what does it really take to run artificial intelligence in orbit? Let's ask the expert. [Music] I'm Maria Varmazis and this is T-minus Deep Space. My guest today has spent decades building computers that don't just survive in space, they think they're. We will explore how AI rated systems are flying today, what they enable for Earth and deep space missions, and why space may be the ultimate proving ground for autonomous computing with Ralph Grundler from AI Tech. So about myself, I've been in the computer industry since the early 90s, so a long time now, and developing computers for a lot of different applications. About maybe 15 years ago, AI started coming back again. AI has these waves of adoption, and the compute power came to the point about 15 years ago where it was viable again. We got involved in that, and we were making big dreams about autonomous cars, and the Jetsons, and everything. We really realized the limitations of AI at that point, but technology developed so fast, and so I have been along with that development. Along the way, of course, the military and other industries got interested in AI, and also in technology and computers that I was working on. So I got involved somehow in developing space-rated computers, and so that led me to this journey here at AI Tech, and working together with customers to get them AI rated computers into space. AI rated computers in space, just that phrase alone. I would love to know what that entails, because I know, as we were saying before we started chatting, I don't think there is a hotter topic right now in the realm of space right now than that entire concept. So you feel like the perfect person to walk me through that. What does that mean to be AI rated? No, wait, space-rated AI system in space. Yeah, exactly. So let me tell you a little bit about AI Tech first, because that's an interesting story in itself. How did we get this name? It's a 40-year-old company to think that when they started a company 40 years ago and called it AI Tech, they must have been geniuses, but that was one of the kind of peak points of AI about 40 years ago. And at that time, the layers that you would use for computation were much less. They're like doing four layers at a time, and it took minutes to do, or less or whatever. But this young man, he graduated from UCLA with a master's in AI, and he thought he was going to change the world. That all farm tractors that are driving on rows should be totally using AI to do that work, because here are these rows, we have the processing power to do that. Nobody should be driving these. And of course, the military said, hey, we drive in rows as well too. That's very interesting technology. You should build computers for us, and that is the story of AI Tech and how we got into the military and also developing space products. Yeah, a lot of people think AI is some new fangled thing. And as you said, indeed, it's been, or it's, it's, it's the latest iterations are certainly newer, but AI has certainly been around for quite some time. Just, yeah, a lot of people, I don't think quite realize that. So I appreciate that. Yeah. Yeah. So about 30 years ago, we got into space, and so we've been flying something like 2 trillion miles without any failure. And we've always wanted to go back to our roots and do more about AI in space. And got contacted about some products we had done for the military by NASA. And we worked together with them and we got some NVIDIA processors flying into space for some of their applications. And so it's been kind of taken off ever since then. So we have about, it's a pretty amazing statistic. I don't know who's doing the math, but we have, we have roughly about 200 objects flying around the earth right now. And so we have around 2 trillion miles without any failure, which is kind of a cool thing to say. And one of my favorite projects is actually is we had some computers that went to Bennu and back and you think, oh, wow, that's, you know, 4 billion miles that should really add up, but it's a thousand, a thousand billions to get to a trillion. So we're still 2 trillion miles plus without any failure. Oh, those numbers. That is truly awesome. And talk about a flight heritage, which is very enviable there. That's fantastic. Yeah. So that's given us tons of experience about how to, how to do things right. And of course, NASA is very experimental and they want to keep things low cost. So we've, we've really kind of become the A.I. radiation experts. If you will, I don't know if I can say that, but we're kind of like launch experts as well too. There's a lot that goes into that. So how do you get, you know, some consumer product into space, right? That's the hard part. And the first real step to that is figuring out what will fly and what won't fly. And that's just doing some basic analysis of the parts that are there and if they can handle any kind of radiation or not. And people kind of always ask me, you know, what is radiation? What does it do? You know, all those kinds of questions. And it's, it's kind of hard to explain because you can't really like see radiation. In fact, we, we joke around that, you know, oh, we could take our products through TSA a thousand times and get, you know, like a single dose of testing. Right. And so it's hard for people to kind of imagine, but if you, I don't know, do you, do you watch any James Bond movies? I have in the past, not haven't recently, but yes, I have seen James Bond films. For whatever reason, James Bond is always underwater. I don't know if you know that. Yeah, that's true. But like a tuxedo underwater. Yeah, exactly. And there's always people like shooting at him, right? And so you like to see these bullet traces going through the water. And that's what you can kind of think of is radiation going through silicon, you know, through a chip, a computer chip. And as you know, transistors are made out of silicon. So you can think of this radiation going through silicon and leaving like this little trail behind it that eventually will close up or could close up. But those can do some damage or they can do nothing, nothing at all. So it's a, it's a, that is the kind of testing and experimenting that we're doing is to make sure that the parts will survive. So you can think of radiation as being random bullets shooting through silicon. So it's kind of a war zone out there. You know, you think space is so quiet and peaceful, but it's really a war zone. You're being bombarded by radiation from the sun and people say nothing comes out of black holes, but a lot of radiation comes out of black holes and they come and hit earth. And so we have to deal with those. The cool thing about earth is we have this magnetic field that kind of protects us. And it's kind of a plus and a minus because the protection prevents the radiation from a lot of the radiation from hitting earth. But as we know from sunburn, some of it hits, hits earth, right? So, yeah. And I was, I was just wondering about that because I was thinking it, you know, applications in low earth orbit can also benefit from that magnetospheric shielding. But you all have also built systems that have gone way outside of our magnetosphere. So you definitely understand like that, that radiation environment's got to be, I mean, deep space. I can't even imagine how intense that is. What magnitude of intensity difference there is there? Yeah. So low earth orbit orbit is kind of a cool place to go. And that's where really you can, you can get some things to work. I'm trying to think of the correct way to say this. But the way that low earth orbit works with the shielding, you know, and that's where the ISS kind of resides, is that provides that little extra bit of shielding. And so then you can put products up there that you normally couldn't put into deep space. That's the, that's the advantage of our shielding. So we have products in low earth orbit right now that are, you know, NVIDIA based GPGPUs. And that is kind of the way that people really want to go because as I mentioned before the show started that, you know, the software guys really like CUDA cores. And it has to do with their flexibility and their reprogrammability. If you change your algorithm, you're always taking a risk that it won't compile in a new system. And so by using NVIDIA, which is very flexible, we're using an FPGA, which is also very flexible. You have the opportunity to get that to work in the compile much faster than in hardware dedicated applications. And the hardware dedicated applications have their spot as well too, because they're much lower in power. So if your application can work in a dedicated hardware application, then you can take advantage of that lower power. And as you know, power is a premium in space. Absolutely. Yeah, I was going to say, so you have zooming out a little bit about what this all enables. I mean, I'm thinking how often I've heard about, you know, orbital data centers, but also, you know, satellites that are able to actually do a lot of heavy lifting while, you know, on orbit of that data that they're gathering before transmitting it down Earth. I imagine it's going to things like that are enabled by what you all are being able to shield here. Right. Yeah, I know it's exactly exactly what you're talking about is being able to process the data and just provide the relevant information rather than all the data back to Earth. As you know, the pipeline back to Earth is a little bit complicated going through all the atmosphere and lining up with your transmission point on Earth. There's a lot of complicated things that require good data transmission. So the real advantage is edge computing. And when people talk about data centers in space, they're not talking like data centers here on Earth. Those are different. And the reason, the reason why I want to bring that up and it's an important point is because it's going to be more edge computing focused. So for example, I would have a satellite that would go by, let's say this data center in space, transmit my data to the data center. It would do the processing for me and then transmit the result back to Earth. That would be an example of a data center in space because transmitting in space from satellite to satellite is a fairly well solved problem. You know, with Starlink and other companies doing an excellent job of that using optical and things like that to do the transmission. I appreciate you making that distinction because I think sometimes people might have a vision of a server farm in space, which is like adorable. And also, wow, what would that be like? But not exactly, as you said, it is more edge computing and I mean, which comes with its own interesting opportunities there also, as you just mentioned. So the resilience that's going to be needed for edge computing in space for those ODCs is, as I said, an interesting challenge and one that is very existential for the need that is clearly there for this technology to take off. Right. Yes. So like I mentioned, in order to do edge computing in space using commercial based products, it's a war zone out there. What you're really trying to do is keep your commercial devices alive. So you have to build a hardware infrastructure around these systems in order to keep them alive. A simple example is just monitoring temperature and current because what happens when the bullet goes through the water, you know, and again, you know, the ion goes through the silicon. It can short out a transistor and it can either flip the bit or even short it out and create a current short. All of a sudden you have a current spike. So you know, you have to be able to monitor those and shut that down quickly before you destroy your device. And then like I mentioned before, just like in water, it takes a little bit more time. It'll reconstitute itself and close up that gap and be able to function again normally some of the time. A little bit like water and sometimes a little bit not. Exactly. The metaphor works well though visually. It's a very good one. And I'm thinking also for things like accuracy, some radiation flips that bit, you're going to have a lot of higher error rates, which not great for a lot of the applications that I imagine are very important and potentially sensitive if they are being put on orbit. Right. So that's why data protection is so critical in space. And you know what's really cool is I don't know about maybe 20 years ago we started putting these server farms in Denver and in Mexico City. And what's interesting about Denver and Mexico City is that they're very high altitude cities and they had much higher error rates. And at that time we didn't do a lot of ECC protection. And so the server farms were failing and I was actually in the compute field at that time and you know, hey, your servers are failing and no one could really figure it out. And then we realized that it was radiation actually that was the cause. And that's wild. Yeah. And so from that point on servers have become much more careful about handling data and how to make sure that data is correct all the way through the pipeline. So it starts from the collection point all the way through into memory and the memory into processing. You know, how do you protect the data and make sure it has things like ECC protection and stuff like that in order to make sure that you're not, you know, propagating these scolariates. And so that's another really important thing to do when you're providing these edge computers in space is making sure that you're doing your due diligence with error correction. And, you know, monitoring for errors. We'll be right back after this quick break. I wanted to make sure to ask you also about application for what you're seeing for what you all have built. So again, the radiation tolerant AI supercomputers in space. I just love saying that phrase. It's really fun. You know, I can imagine a lot of different applications. I've certainly heard about them, but I'm curious what you all have seen and what kind of interest you're seeing. Yeah. So the easy application is image processing, right? Because as you know, people love to take pictures of Earth and they like to look for things on Earth and then identify those objects and provide that information. It could be anything from whale pods to where your cattle are, to oil rigs, to container ships. There's so many things that businesses want to track in order to not lose their assets, right? Basically. And so that's a huge almost no brainer kind of thing. And in fact, that's one of the one of that. It's very simple applications for AI. And one of the very first ones in space was cloud detection because if you take a picture and it's covered, covered with cloud and you can't see anything, you don't want to transmit that back down to Earth, right? So that was one of the very first simple applications for using image processing space. So that's a that's a easy answer. But there's so many other ways that you can use AI in space. The other easy answer is the deep space applications, right? If you're far away from home, the transmission time takes so long that you want to be able to do a lot of things autonomously. And NASA is great at doing these kinds of experiences and their Mars rovers, perseverance and curiosity. Two amazing pieces of equipment, by the way, at tons of AI technologies included inside them with deep learning computer vision neural networks. And they're using this for autonomous navigation and testing samples and all kinds of things on their Mars rovers. So they were really the first ones to really go deep in AI and put that on space. And another cute fact about Mars is everyone thinks, oh, my God, if you go to Mars, there's going to be tons of radiation. Well, it turns out that Mars has a little bit of a magnetic core as well, too. And so you can think of going to Mars is kind of like going to Leo. It's actually the radiation is not that bad. So if you have your equipment, I did not know that. Yeah. So if you have your equipment off until you land there on Mars, you don't have to worry too much about radiation. It's not quite as bad of a war zone. I did not know that. Because I know the moon is good luck. But yeah, I didn't realize. So it's about roughly, like very roughly equivalent to Leo. That's fascinating. That's really neat. I've learned a lot through you, Ralph. I really appreciate that. Is there anything else that you wanted to highlight about what you all are working on? I want to make sure I give you plenty of opportunity. Yeah. So the coolest things that we're working on is the latest NVIDIA Oren processors. I mean, those are incredible that we're working on. It's the NVIDIA AGX Oren 64. And so this is 248 pops in terms of processing power. So this is not the latest and greatest from NVIDIA. But it's definitely something that will provide quite a bit of processing power. There's also 12 ARM cores inside as well, too, providing a lot of processing power. So we have customers that are doing image processing that are so excited about this product flying into space, which we're going to do hopefully this year. We'll be able to get it up and launched. Well, that's exciting. Yeah. It's really cool because it takes a bit of energy and it takes a lot of cooling as well, too. So those are two technologies that are important. And I already mentioned about trying to keep it alive. So we have a lot of electrical techniques to keep it alive and make sure it keeps working. And then the other cool thing is it's in a tank right now driving around in the desert. So it's-- Wait, what? Sorry. I could say it right. What was that? Yeah. So we were also a Millero company as well, too. In fact, that's a bigger part of our business. And so it's really cool because we're able to test out technologies in really harsh applications. And one of the harshest, of course, is in a tank in the desert. And you would think there would be some kind of fan cooling or something like that available to you, but it's not. It's also a cold plate application because it's deep inside the tank. And it's providing image processing for tank operators, which is a very helpful thing as well, too, for them. Wow. So that brings me back to-- so once you have that kind of experience, then you can survive the launch, right? And since we understand space, we understand the off-gassing effects, we understand venting, and we understand that PCBs are critical if they have those little bubbles in them. They expand and explode when they go into space. So you have to make sure that your PCBs are of the highest quality to make sure that-- Yes, printed circuit boards for my friends who don't know what they mean by PCBs. Oh, yeah, sorry. I think people-- No, no, no, it's OK. I knew what you meant. I was a computer scientist for a little bit, so I knew what you meant. Yeah. But yeah, I'm not sure all the audience would know a PCB off-hand, so I figure I would just define it. But yeah, I never thought of that either because, yes, yeah, making sure there's no bubbles in there. Yeah. That is something I've never thought of that. Yeah, so some people think about, oh, you know, why can't I just use a cheap manufacturer in Taiwan to make these PCBs? And the real answer is that they have to be perfect, right? And it's not that air bubbles necessarily would be bad if it was in a place that doesn't hurt anything, but if it's like underneath a pin and it lifts a pin as it opens up-- Oh, forget it, yeah. It's like, you know, there's your one in a hundred chance, but there's so many one in a hundred chances in space in getting to something to work in a space. You have to reduce as many of those as you can. And it's, I mean, the experience of driving something around a tank really sets us up for success in terms of the launch, right? And modern cameras, unfortunately, have kind of taken the stress out of a launch. I don't know if you see, but they look very stable and very still the cameras when they're onboard. Yeah, not that do-do-do-do-do-do-do, yeah. So back to the early videos of launches and the whole thing is like shaking like-- The shaky cam. Yeah, shaking like crazy. And that's really what's happening. And what people don't understand is satellites are mounted in, you know, what something would call like a Christmas tree structure inside the rocket. And so this Christmas tree is like shaking back and forth, you know, it's like your cat trying to climb the Christmas tree. [Laughter] And so being able to make sure that you have that shock and vibration testing all done is really important. And the reason why I mention off-gassing as well too, and it's critical to understand what materials you're using for off-gassing is because you have to vent all these electronic boxes, right? They have to have a hole on them because when you go from one atmosphere to zero, it's expanding, right? So you have to let the air out. And with that as well too is you'll get particles from off-gassing, right? So if you use plastic materials that have a lot of off-gassing, you could potentially coat things like your solar panels, which would be bad. And so understanding all of that from somebody who's been in space for 30 years, you know, with two trillion miles without failures, knock on wood. [Laughter] But no, it's because we do all that testing and all that preparation. That's why we're successful. And that has a cost of time and effort and money that people like to do. We can't all be like Elon Musk and just, you know... There's only one of them. I mean, it's only one of Elon. So yeah, the rest of us cannot be him. That's true. Yeah, so Elon Musk has the huge advantage of his. He can really just shoot anything into space because he's the richest person in the world by far, right? Not even close. I don't think seconds, not even close. A runaway first, this is true. This is true. The rest of us can't do iterative testing and go, "Oh, well, that's a few billion dollars gone." Exactly. You know, for us, you know, especially small startup companies and even larger companies, and NASA especially, is any kind of failure in space is negatively looked on by the company. And so we're doing everything we can to make sure we remove all those one in a hundred chances to make sure that our customers are successful. And that's what we really ask to do with going back and testing and testing, analyzing and testing, and then creating a product that can fly into space. So we have NVIDIA processors flying in space right now, so we're excited to send more up there. And then we're also excited to use other technologies. For example, we're also looking at doing things like using hardware accelerators. There's a lot of them out there that are very good that can really reduce the power. That's kind of the magic trick, right, is how do you put something into space that can do a lot of processing with the least amount of power. And that's going to be one of the keys for autonomous spacecraft for sure, which I know is another much longer term thing that a lot of people are really excited about. It's not as immediate as orbital data centers, but certainly that is going to be the future for a lot of space exploration. It's going to require autonomous spacecraft, so it's going to be fascinating to watch that evolve. And it is fascinating, honestly, even right now. Yeah, no, it's really fascinating. And the thing about AI is that it can process tons of data and look for anomalies or specific data points. That's what AI is great at, and that's the perfect application for space, whether it's, you know, like we said, for deep space missions or analyzing images around from Earth or even images from deep space, sending them back to Earth, right? You want to, maybe you're looking for a particular satellite or object in space and AI can detect and track that kind of stuff. And then also it'll be extremely important for other kind of missions to enhance safety or optimize mission operations. I'll give you another real simple example. Right now, there's so many things flying in space that you have to do evasive maneuvers in order not to collide, right? I think you've had shows talking about that. And one of the things that AI can help with is not only predict when you're going to have a collision, but also predict a movement such that you won't have a collision for a long time in the future. And the reason why that's important is because fuel is a premium also in space. Remember, energy and fuel are premium in space, and AI can really help optimize that and make sure you save on fuel. And then also you can't forget about AI in landing and launching and now landing, right? Yes, yes. Space X is a great example of using AI. And that's one of the reasons why Elon Musk can say, "Oh yeah, it crashed and blew up. We have tons of data," and he's using that data as training data for the next round to make sure he doesn't exceed those kinds of limits. But when their rocket boosters land, they're using AI to direct those and land those. So AI is already all over in space, and it's just going to get more and more exciting. And I'm just really looking forward to seeing what people do with it. It's just absolutely amazing. It truly is. Ralph, I've really enjoyed speaking with you too. Like genuinely, this was a lot of fun, and I've learned a lot from you also, which is the best part of my job. So thank you so much for speaking with me today. I really appreciate it. Thank you for being on the show. Yeah, thank you so much, and thanks for giving myself and AI Tech a moment to explain all the cool things about AI and space, and looking forward to talking to a lot of people in the future about it. It's just going to be really exciting. That's T-Minus Deep Space, brought to you by N2K Cyberwire. We'd love to know what you think of our podcast. Your feedback ensures we deliver the insights that keep you a step ahead in the rapidly changing space industry. If you like the show, please share a rating and review in your podcast app. Or you can send an email to space@n2k.com. We are proud that N2K Cyberwire is part of the daily routine of the most influential leaders and operators in the public and private sector, from the Fortune 500 to many of the world's preeminent intelligence and law enforcement agencies. N2K helps space and cybersecurity professionals grow, learn, and stay informed. As the nexus for discovery and connection, we bring you the people, the technology, and the ideas shaping the future of secure innovation. Learn how at N2K.com. N2K Senior Producer is Alice Carruth. Our producer is Liz Stokes. We are mixed by Elliott Peltzman and Trey Hester, with original music by Elliott Peltzman. Our executive producer is Jennifer Iben. Peter Kilpe is our publisher and I am T-minus host Maria Varmazis. Thank you for listening. We'll see you next time. [Music] [Music] [MUSIC]
