Okay, well let's, let's get started. I'm Tristan Gerra, senior semiconductor analyst at Baird. I would like to introduce Everspin Technologies, a leading supplier of next generation persistent memory technologies. We're pleased to have with us today, Sanjeev Aggarwal, President and Chief Executive Officer. And with that, let's get started. I think you have a presentation for us.
Yeah. Thank you, Tristan, and thank you to Baird for this opportunity. Good afternoon, guys. Try not to put you to sleep after your lunch. Can I have the remote for... Tristan, the remote?
Oh, the remote. We need the slide remote, yeah.
Ah, thank you. All right. So, first of all, I'm gonna make some forward-looking statements, so just a caution about that. So as far as our company overview, Everspin is actually a global leader in developing and commercializing a non-volatile memory, which is magnetic random access memory or MRAM. We're the only company that's actually commercialized or brought to production two different generations of this technology. One is a field-switched Toggle MRAM, and then the newer generation of the current switch spin-transfer torque or STT-MRAM. We have been in production for more than 15+ years. We have an 8-inch manufacturing facility in Chandler, Arizona, and then we also do our 12-inch manufacturing with our partner GlobalFoundries.
The 8-inch facility, we actually rent a clean room space from NXP. It's a Class one clean room, and we've been in production since 2006, actually before our spin-off, which was in 2008. Since then, we've actually shipped more than 150 million-plus products. And very proud to say that actually, a single-digit RMAs from the 150 million-plus products that we've shipped. So very conscious on quality and the technology and the products are really robust, and the customers appreciate that. And that's the reason why we're able to ship into mission-critical applications, industrial automation, IoT, data center. And the sweet thing about MRAM is it's actually radiation immune. So you can actually send it out to outer space, and the memory doesn't get disturbed.
Also, it can hold up to robust temperatures from -55 to 150 degrees Celsius. We have about 2,000+ customers, so a pretty diverse customer base. So not one customer will actually negatively or positively impact our quarterly results. And finally, we are very focused on our IP. We have about 650+, patent and applications worldwide, and we have actually licensed our technology to companies like Seagate, to Toshiba, also the US government, like Honeywell, Frontgrade, and then also to sensor companies like Bosch and Alps. So we have tested our IP, and it has been successfully licensed and indemnified for our customers. So some investment highlights. You know, as I said, we are the leaders in MRAM, and we are the singular domestic supplier of MRAM in the US.
So that actually makes a special interest for the U.S. government in our company and in our technology. And we have had considerable success licensing this technology to the U.S. government and also manufacturing parts in our eight-inch fab in Chandler, which is the only facility that can actually do MRAM in the U.S. So we do both Toggle MRAM, as well as STT-MRAM for defense industrial-based customers in the U.S. We have a pretty diversified customer base, as I just mentioned in the previous slide, and then I'll go through some of the applications in the upcoming slides that will tell you what kind of applications and what kind of customers we've been able to secure over the last 15-plus years. We do have a large market opportunity, and I'll go through that.
Our TAM is actually on the order of $7.5 billion, and if you convert that to a SAM, it actually drops to a little bit less than that, about $5 billion or so. We have barely started scratching the surface of the market opportunity that we have, and I'll get into that in a few, slides as well. A proven management, team, we have people from, Intel, Samsung, and such, with 15+ years' experience. So overall, I think we have a, a pretty diverse team, with experience in both 8-inch and 12-inch, facilities across the U.S. So why MRAM? What is the value proposition that MRAM brings? So bottom line is, it is, combines persistence with performance.
So basically, you take the battery off, and the memory will not forget what you have re-recorded in that MTJ bit. Performance, we can actually switch as fast as an SRAM or a DRAM. Endurance, you can write unlimited times to this memory without actually degrading the technology. And as I said earlier on, reliability, it's a pretty robust technology, operates between -55-150 degrees Celsius, automotive, industrial temperatures, for greater than 10-20 years. So that's the reason why there is so much interest in this technology. And if you take this and try to actually classify the memory, the exciting thing about MRAM is that you can actually, especially STT-MRAM, is you can actually tune this technology to make it look like any memory you want.
So if you wanna make it look like a DRAM, you can tune the technology to make it look like a DRAM. If you wanna make it look like a flash memory, you can actually make it look like a flash memory as well. So on the left over here, you're basically seeing PERSYST memory, and the idea over there is this is persistent, so basically you take the batteries off, and it'll never forget its information. And then it has unlimited read and write cycles, and the PERSYST stands for performance. So it's basically really fast read and writes, and unlimited read and writes, and that's the reason why it's actually designed into many embedded systems. The first technology we brought to production over there was Toggle MRAM.
Then more recently, over the last a year or so, we actually brought out STT-MRAM, and that is a 64 megabit and 16 megabit STT technology that addresses the same markets as Toggle MRAM. And that's basically the non-volatile SRAM market, or if you're familiar with FRAM or ferroelectric memories. So we compete with those memories in this space. We also brought out a part which was DDR compatible or STT-DDR part that's actually shipping into the data center market. Recently, we had an announcement with IBM on the design win that we had for the FlashCore Module. I think we announced it a couple of months ago.
We have been designed in for the last two generations, and this is the new generation that was actually designed in, and it's on the order of one gigabit density. And that's the highest density MRAM today that you can buy in the market, and that was developed by Everspin, and then the technology was transferred to GlobalFoundries, where it's being manufactured since 2017. The next one is UNISYST, and in this case, you can actually unify the data and the code, and that's the reason why we're calling it UNISYST. And what I mean by that is when you turn on your computer or when you turn on a device, there is a configuration for the device that is stored in a memory block, and you can actually use STT-MRAM for that memory block.
Today, I think you're using NOR Flash for it, but you can use STT-MRAM, and you can use it for data execution as well as configuration data, and that's why it's called unified code and data memory. Today, people are using NOR Flash, which actually stops scaling at 40 nanometer CMOS, and there is no solution beyond 40 nanometer. That's the reason why the foundries are actually using embedded MRAM to extend the embedded flash roadmap. Similarly, Everspin is using standalone STT-MRAM to extend the standalone NOR Flash roadmap. There is nothing available today with NOR Flash on the order of 256 megabit or higher, and that's the reason why we are actually targeting that NOR Flash market, and that is a huge TAM. It's on the order of $2.5 billion.
What we plan to do over there is bring out the first part, which is an xSPI family on the order of 1 gigabit or so, because those densities are not available. Then you can actually start using it in system-in-package. And what I mean by that is, let's say you have a 7-nanometer CMOS, and you're looking for memory for that solution. So you can actually use a 22-nanometer STT-MRAM and do a system-in-package solution or an SoC solution to take advantage of the STT-MRAM. Today, folks are using NOR flash, and it's too slow.
So, for example, if you're trying to do an over-the-air update, like Tesla did, like a couple of months ago, all the cars were stationary for 15-30 minutes, waiting for the code to go through, and we could have done it in seconds or minutes if we were using STT-MRAM. So that's the promise STT-MRAM brings to the market, and that's what the automotive guys are excited about using this product. The next one is because it's gonna be automotive qualified, you can actually use it with an LPDDR interface. That is something we'll bring into the market after the first product, which is the xSPI, and finally, chiplets. I mean, once you have this xSPI product, whether it's a 1 gigabit or 256 megabit, you can do a chiplet anytime.
But the reason why we've held back is because the customers need to define the interface. There is no point in Everspin building three or four different interfaces and waiting for the customer to come. So once the customer has defined what interface they would need, we can bring that solution to the market. And finally, a very forward-looking thing, AGILYST. It's basically in development today. It's gonna take at least 2-3 years to actually commercialize it. And the idea over there is that MRAM is non-volatile, it's persistent, it's as fast as SRAM, and therefore, it leads to, and it can be actually securely integrated into the devices that you're using.
What that makes it is low power, so power management, and then it's instant on, and that's very much valued by the FPGA guys, and it's also of value to AI inference on the edge. So if you think about AI inference, what is, what is the major concern that we have over there? There is a lot of power loss when you're taking the data from the server to the edge and back. Then everything that you have on the edge needs to be continuously refreshed, so that the memory doesn't lose its information. So clearly, you can put MRAM on the edge. It's non-volatile, it's as fast as SRAM, and then you don't need to refresh it every time.
So you can actually do effective AI inference on the edge and store the configuration of the information that you're looking for on the edge as well. So that's something that is in development. We're working with some partners, and you should see that in the next two or three years. So this is a slide that actually... There you go. It talks about the persistent UNISYST products that I was talking about. It puts on the y-axis the endurance or the read/write cycles, and on the x-axis it puts the latency or the performance, how fast can the memory perform? So on the very right, on the bottom is storage, right? That is really cheap memory, it's really slow, and the customers really don't care about performance.
So we don't really play a part over there in the storage industry. So we have UNISYST, which is actually going after the NOR flash market. Like I just said, that the over-the-air updates take forever. So you can actually do it two orders of magnitude faster, and then you also bring about one and a half orders of magnitude more endurance, and that is something that the customers are asking for and is of value. So that's the UNISYST that's actually in design today, and we hope to bring it to product in the next year or so. On the top up there is the PERSYST, and that's where you're looking for unlimited read/write cycles, and you're looking for fast memory. For example, if you're on the manufacturing floor, and your robots are actually controlling what's going on on the manufacturing floor with PLCs.
If you lose power today, and if you're using SRAM, everything on the floor is scrap. But if you're using our technology, our MRAM or STT-MRAM technology, the robots will quickly write the data back to the PLC, telling them what they were doing when the power was lost. So when you turn on the power, there is no scrap, and you can actually pick up exactly where the robots were before the power was lost. So that's where the data logging comes in. Another example would be casino gaming on the floor, right? Every time there is an action on the casino machine, that data needs to be recorded as fast as possible to the server, and that's where our technology comes in. So IGT, Novomatic are some examples that are using our product, and I think we are basically...
We have a pretty good capture of the market, almost 80%-90% of the market. Data logging is also important, for example, for medical instruments. So if you have a pacemaker, you want to collect the data as soon as possible, and that's where PERSYST comes in. Think about EVs, your battery management system or your terrain management systems. You want to collect the information on the battery health or the terrain health as fast as possible and as often as possible. So that's where the PERSYST memory comes in. Some examples of what PERSYST are and UNISYST are. So this talks a little bit about the market scope. So you can see, in the maroon color, it's the $700 billion or $0.7 billion market space.
That's where Toggle has been playing, since our inception in 2008. That's where you have, NVSRAM, ferroelectric memories, as well as emerging non-volatile memories like, RRAM and, MRAM over there. And we believe that, MRAM is the most superior solution, at least as far as standalone is concerned. Actually, I should mention that this only talks about the standalone market. This doesn't take into consideration the embedded market, which is actually much larger, than this one. And then the UNISYST, which is basically the, dark orange, I guess, that goes from $2.9 billion or so-$4.9 billion. That's the UNISYST product that we have, in design today, and we will, bring out a product next year.
AGILYST is not shown on this because AI inference markets are so difficult to predict right now, so we haven't even given a shot at that, but obviously, it's a pretty large market. So here are some of the examples that I talked about earlier for PERSYST. You can see data logging is a primary feature with read/write cycles that is of interest over here. That's there in the aerospace and transport. It's there in the, you know, medical recording devices, the EVs that I mentioned for the battery management system or the terrain management system. I mentioned the casino gaming over there. Battery charging units, PLC automation is big for us, and then also space because of the radiation immunity that our technology naturally has.
So most electron memories like SRAM and DRAM, because of the electron excitation, they actually lose the data when they go into space. But MRAM, because it's magnetic-based, there are no electron... Even if the electrons get excited, it's not gonna lose its information, and that's the reason why it's radiation immune all the way up to 1 million rad, which is sufficient for deep space. So, some UNISYST applications. So if you're looking at on the image on the left, you basically have a processing module. It has NOR Flash, which is where the code memory is stored that I was talking about. Then you also want to do some data logging, and then you also have some data storage.
So like I was saying earlier, data storage, we would not by itself go after, but you can replace all those three with one memory, a UNISYST STT-MRAM. So now you can do NOR flash, you can do data logging, you can do code configuration, as well as storage, all in one memory, and so it basically simplifies the architecture of the system that you're using. Another example is the FPGA, and I was talking about over-the-air updates. Here it gives an example, I believe, for 128 MB, the program time is on the order of 2 seconds versus several minutes with NOR flash. So really fast memory, a strong value to the automotive customers that we have. Okay, so here's another example of where UNISYST can go, whether it's embedded or standalone.
So if it's a closed architecture, it's probably an embedded solution. If it's an open architecture, then there is some embedded, some standalone, and in the hybrid architecture, obviously, it is some UNISYST and some memory over there. And here, again, our value is performance, reliability, and power and speed. So over-the-air updates that I was talking about or data logging that I was talking about. So if you're looking at the closed architecture, that's where the foundries are making a killing. They're replacing embedded NOR with embedded MRAM. And then the open architecture and the hybrid architecture is where you have a system and package solution or an SoC solution, where STT-MRAM plays a part, and that's where we believe that we're gonna make some good inroads when we bring this part out in the next year or so.
So this gives you an idea of the customers that we have today. On the very left is the enterprise market. So we have, the IBMs, the Broadcoms, the Microchips over there. And so that's actually a huge volume. It's a high-volume, low-margin business. Industrial automation is a high-volume, high-margin business, and if you go to the very right, the mil-aero and transport, that's a low-volume, very high-margin business. But you're looking at industrial automation, you have Siemens, Schneider, Mitsubishi, Omron over there, and those are the ones that are actually using it for the programmable, logic controllers. Medical, you have Nikkiso and Canon. On the casino gaming, like I was saying, Grixis, IGT, Novomatic are some of our larger customers. And in mil-aero and transport, I'll actually hold that for the next slide.
You'll see that over here, our technology is actually being used on the Mars rover, the Perseverance. We are actively collecting data from Mars today. So it's, you know, a testament to the radiation immunity and the robust temperature that we're talking about. We are also on the Lucy mission to NASA today. And then also, like I mentioned, we are in the EVs for battery management systems as well as the terrain management systems. An example over here is Lucid Air. We are also designing with the Bugattis, and we are also designing with the BMW race engines as well. So a pretty wide application and customer base over there. In terms of capabilities, we do everything in-house. We have our own design team.
We have our own manufacturing facility, at least for eight-inch, but for twelve-inch, we obviously go to a foundry, which is GlobalFoundries in this case. In terms of design services, we have done all the designs from for our parts, which includes parallel interface, a serial interface, a DDR interface, and then also any custom interface that the government, the U.S. government has acquired off our our design team. So we do all of that in-house. We have a team that's based out of Austin. Our manufacturing facility is actually in Chandler, Arizona, which is where we were incorporate where we were formed. We've been in production for more than 15 years today, and the beautiful thing is that we actually do all our R&D in that eight-inch facility.
I say that because of two things: one, because we are doing it on an 8-inch facility, we actually are able to control our R&D costs. The second thing is, because it's all in-house, our IP is well protected. What we did, for example, with GlobalFoundries was, we developed all our technology in our 8-inch facility, and then we transferred it to GlobalFoundries for 12-inch, developed their embedded MRAM product for replacing embedded flash, and in return, brought our standalone product to foundry over there. So they're doing embedded MRAM for themselves, and they're doing standalone MRAM for Everspin since 2017. So we've been in production with 12-inch since 2017 now, so almost seven years.
Three different products, 256 megabit DDR3, 1 gigabit DDR4 for the storage market, and then the most recent products that we brought to market, the 64 megabit and 16 megabit xSPI family of products over there. And now we are actually working on the 22 nm FDSOI, and that's the 1 gigabit product that's in design, and we hope to bring to product next year. So what's in the future? We talked about PERSYST and UNISYST persistent line of sight, but what comes in the future? And that's where the AI inference is where we believe that this technology is gonna play a role. So the first one on the left is the MRAM for FPGA, and we talked about it as part of UNISYST, right?
So you have the configuration memory that is stored in the NOR Flash today, and I think storing it in STT-MRAM brings much value because you can get rid of the redundancy, and you can execute in place, you can upgrade the configuration without any loss of data or time. The second one is, we actually built a solution for the U.S. government building a strategic rad-hard instant-on FPGA, and we are calling it distributed MRAM. And the idea over there is, when you turn on the FPGA, all the configuration from the NOR is downloaded into the SRAM, which then executes the lookup tables for the FPGA. Our solution eliminates the NOR completely and replaces the SRAM with our MRAM IP. So it's not the standard IP that you can actually buy from the foundries.
It's our IP that we have developed, and it's actually as fast, 2-5 nanoseconds speeds as compared to SRAM. And you can actually do a distribution on whatever density you need. 1 kilobit, 1 megabit, 256 bits, as opposed to if you go to a foundry, you're stuck with a 16 megabit macro or a 4 megabit macro or a 2 megabit macro. So that's why we call it distributed MRAM, and it can go all over the chip, and it's instant on, and it's reprogrammable. Now, this same solution, as you can imagine, has application in AI on the edge as well. So you can actually update your code for the configuration on the edge, and you can also, given the speeds, 2-5 nanoseconds, you can do fast inference on the edge.
And that's where we have some programs in development, and we hope to bring that product to market over the next 2-3 years. This literally talks a little bit about our roadmap. The first row over there is persistent memory. That's in production today. The Toggle MRAM and the two blue ones that are basically our xSPI family, 16-megabit and 64-megabit. One, the one in design is a UNISYST, or the green blocks in the middle, that talks about our 256-megabit to 2-gigabit xSPI family that can be used for system-in-package, as well as chiplets.
And then the last one up there is the AGILYST, which is a distributed MRAM that I was talking about, that has relevance for reprogrammable FPGAs, as well as for AI inference on the edge. So that is something that's coming over the next couple of years. An introduction to the executive team. We have Anuj Aggarwal, he's our CFO, with experience from Intel and Thermo Fisher. David Schrenk brings about 15 years of experience from Intel. Amit Shah is our VP of Backend Operations, and he has experience with Marvell and Semtech. Yong Kim is our VP of Product Development, with his strong history with Cypress, which is now Infineon. Kerry Nagel is our VP of Technology for R&D.
He's been the father of our technology and has developed the Toggle MRAM and STT-MRAM all the way from our Motorola days. So almost 25 years with Everspin, as we went from Motorola to Freescale to Everspin. And then Khaled Barakat brings a wealth of experience in operations and fab manufacturing and quality from Intel and Samsung. I believe that is revenue. So, you know, since Anuj and this management and I took over the company or became part of this company, you can see that we have consistently gone to profitability. Our cash flow has gone from -$16.6 to $11.7. We are now debt-free with almost $35 million in the bank. And our mantra is to make sure that we are profitable quarter-over-quarter.
We basically went profitable for the first time in 2021 since our inception in 2008. So, it's been a good performance, and it's a great team, and we are dedicated to staying profitable and developing new technology. Thank you.
Thank you. Quick question in terms of the CAGR that you see, I think high single digit to 10%. When do you think you return to this, you know, given the catalysts that you've mentioned during the presentation? And also, if you could talk about pricing. I know it's not necessarily topical given the performance advantage that you offer. But at the same time, I think you've mentioned in the past that from the 2x premium in pricing that you have currently, you can get to a lower level than that, and presumably, that can also help adoption in addition to the value proposition you're already offering today.
Yeah, good question. So the first one is on the single-digit to 10% growth year-over-year. I think once we are through this downturn with the industrial market, I think we can return to that next year or so. And we believe that with the UNISYST coming into production next year, that's actually going to be an inflection point for Everspin, given the huge TAM and given that there is no other memory that actually replaces flash for the standalone. For embedded, obviously, DRAM plays a role, but for standalone, resistive memory or DRAM does not play a role. So I think we have the whole opportunity to actually capture a large market over there of the $2.5 billion that exists out there.
So that all leads me to the question of pricing versus performance. So we have from day one always been a value-based solution. We don't really compete on cost per bit, but we're obviously aware of what the competition is. So today, like Tristan mentioned, I think we are 2x the cost of NOR Flash, for example, at higher densities. But given the fact that the customers are actually doing redundancy 2-3x for the same density, and we can replace 3 NOR Flash with one STT-MRAM, I think it's gonna be a good economics for the customer to actually replace the NOR Flash with STT-MRAM.
The other thing is, with this Embedded MRAM being offered by the foundries as a solution, I think the cost of ownership of the tools and this technology will go down, and that will automatically bring the price down, even for Everspin, as we go to these foundries for manufacturing our parts. So I think there is promise over there.
Right. Time for one more question.
Just on the embedded MRAM within the foundries, what's the risk and opportunity in that space? You sort of touched on that, but might they be more price aggressive, and you touched on the fact that it's quite lucrative for them at the moment. Just explain that right now.
Yeah, sure. So there are at least three foundries that are actually offering embedded MRAM to replace embedded flash. GlobalFoundries, that actually took a license from us, so every time they make a sale, we get a royalty payment from GlobalFoundries. The other one is Samsung, and then there is TSMC. So both Samsung and TSMC have talked about their next generation nodes. So TSMC is already looking at 7-nm, sorry, 16-nm and 7-nm, and they already have 22-nm in production, similar to GlobalFoundries. Samsung has 28-nm in production, and they're talking about jumping straight to 7-nm and 5-nm to offer embedded MRAM on those nodes. I think, as the volumes increase, I think these foundries will try to match the pricing of the embedded RRAM or embedded flash.
So the pricing will go down, but I think it's a, it's a good problem to have, because once you have more volume, and if their pricing goes down, then the pricing that they offer Everspin to standalone also goes down. So I think, we won't feel necessarily a price pressure. I think we will take advantage of the volumes, and the pricing that's going down for the embedded pricing, for the embedded wafers.
Great. Sanjeev, thank you very much for presenting to us today.