Atym’s Founders Weigh in On Progress to Date

Atym's co-founders Jason Shepherd and Stephen Berard recently caught up on the company’s progress to date, including our recent expansion into addressing Linux capable devices that traditional container technologies like Docker can be too heavy for. The following is the transcript from the conversation, and you can also catch a recording of the entire discussion here.   

Jason Shepherd So we're coming up on a year since closing the seed raise, and it's about three years since we really started incubating Atym [before] formally launching last year. So, a little bit has happened... and not just our logo evolution! 

 Stephen Berard Yep! 

 Jason Shepherd Just a little bit about what we've seen over the past year. I mean, we obviously launched the core product, we put Ocre out into Linux Foundation and that's growing. We have a number of really good customer names on board. What are some of the surprising things that you've seen over the past year? 

Stephen Berard That's a great question. I think one of the things that I've been really happy to see is this acceptance of containers on smaller and smaller things, right? The idea that hey, hey, containers can be small and efficient and can fit on embedded devices. When we started this three years ago, people thought we were crazy, like that'll never work or yeah, good luck with that. And you demo it to folks and they're like, that's great... but [the containers] are gonna be big or not run well. And I think now you're seeing - hey, actually that's a thing and there's a lot of value there, right? And I think it mirrors what people saw with Docker. There's a lot of like, why would you do this? You and I saw the same thing with VMware, right? It was like, oh, VMware, that's nice. Maybe for a test environment, but nobody will run a VM in production. It's not fast enough. Fast forward to where we are now. If you're not running it in a VM, like, why, right? 

 Jason Shepherd Yeah. 

Stephen Berard So I think we're seeing a lot of parallels on the embedded side, which is good. 

Jason Shepherd Yeah, it's been surprising too. I mean obviously you shaped the core technology back then, but the number of folks that we've come across that have also landed on WebAssembly as as a solution for this [is encouraging]. Great technology, horrible name! 

 Stephen Berard Right. 

 Jason Shepherd Seeing that bet being validated by customers, you know very large names that have been doing a bunch of research and you kind of see the value in what we're doing, you know the commercialization.  Of course, Ocre - that open [source] component which makes it more palatable in terms of not being this proprietary technology that you make your entire infrastructure investment on. Obviously we open cored it for a reason. So, that's been really cool and yeah, just seeing the evolution and the acceleration of interest over even the past number of months has been great to see. 

 Stephen Berard Yeah, [the joke] I always say is ”WebAssembly, not the Web, nor Assembly”. It's the greatest technology you've never heard of. And there's a whole Alphabet soup of things out there. There's WASI, there's the component model, there's, you know, WAMR and there's all this other fun stuff out there. But I think one thing that's really kind of come up to the top is the need to be able to build on this technology today. There's a lot of new stuff coming that's very focused on the cloud and distributed computing. But there's a lot of interest in being able to do this in the embedded space and we're starting to see that in projects like the Embedded SIG which is part of the WASI subgroup. And also, things like the Long Term Support (LTS) for WASI P1 that we‘re instrumental in moving forward. So, I think that's something that just wasn't a thing a year ago. It’s good to see that there's a lot of progress, and what's really amazing about that is the amount of interest we've been having. Across everybody that we've talked to, I don't think I've talked to a single person where like, yeah, I'm not really interested. Everybody's been like, hey - we should have had this two years ago. 

Jason Shepherd   Yeah. And on top of you know the Ocre project within the Linux Foundation, all the [standardization] work that you've been doing is super validating. Seeing you working with Siemens, Schneider, Emerson, Bosch, and Amazon - a bunch of different folks in the embedded SIG, the LTS (long term support) version of the core [technology]. So yeah, a lot of really good stuff happening that I think we're just scratching the surface on. The types of deployments we'll see, I mean obviously we've been mostly industrial-focused to date, but we’re starting to explore a bunch of different markets, getting pulled into consumer and automotive. You know obviously there's a lot of opportunity in healthcare, and smart spaces in general. A big thing that's been surprising for me that we never really thought about. I mean, obviously with microcontrollers, it's about changing the way you do embedded today - stop the ”source code soup” and the insanity of brute force monolithic firmware. “Cloudify it”, all that stuff.  It's been kind of a nice to have for a long time to streamline embedded and firmware development and we make it more modular and feel more like the cloud.  Definitely a pattern that we've seen is anything associated with AI on device, we're seeing a lot of opportunity where it's like the person that knows AI is not really the person that knows embedded firmware typically.  

Stephen Berard  Absolutely. 

Jason Shepherd   Like we have the one industrial customer where it's like literally - hey, I want to do embedded A I on device and there's the embedded group and then there's the guy that knows machine learning. That's a completely separate group and [with Atym], both sides don't really have to change their workflows and tools to be able to work on it. So, that's been an interesting pattern to see develop. 

Stephen Berard Yeah, and [when] you look at it, part of it is an organizational issue and part of it is technology, right? Like how do organizations work? Like what's the skill set of somebody who's working in the machine learning, computer vision, you know, smart signals kind of space and the tool sets and how they operate. And then there's the embedded team or the edge device team, right? And they don't work the same. They don't work in similar cadences. And so, having the ability to decouple those and not have to synchronize from an organizational perspective, especially large organizations like industrials we've been talking to, that's a game changer, right? That's pretty amazing. On the technology side, I know this has been a problem for a while. There's been a lot of machine learning out there and there's a whole sector of Machine Learning Operations or ML OPS that you hear, right? And surprise, surprise, the way they solve this in the cloud is through containers, right? They containerize the builds so they get that repeatability. But their focus has been on those large cloud-based servers. So, when you want to bring that down to the edge and particularly to a microcontroller, well, those generally until recently haven't been containerizable targets, right? And so, we're able to change that now. We can bring that code and run that on a constrained device. Whether t's a Linux device on the edge or or a microcontroller, and we can do that in a in a footprint that's size appropriate, right? Yeah. 

Jason Shepherd It's funny, [I've come from] more of the cloud and IT space, while working a lot in the industrial and operations world over the past number of years. The past 10 years in the IoT and edge space. I've been caught several times with typos on our slides where I put 256 megabytes of memory [as the runtime footprint] instead of kilobytes because it's just it's just a whole different, you know, in terms of the MCU side. But so you know, ”tiny” is an overused word in edge because it's very relative. But when we say tiny, we're talking pretty small, you know, in kilobytes of footprint. 

Stephen Berard Yeah. We're talking pretty small. 

Jason Shepherd But the other big thing that that I think has been interesting over the past year. So we're talking, you know, microcontrollers - clear opportunity there, revolutionize how you do embedded.  Anything associated with Edge AI. Cybersecurity, we're seeing a lot around Cyber Resilience Act. And of course, just reduce engineering friction in general, make it easier to develop, decouple different life cycles of apps on a device versus having all the firmware jumbled together. But it's been also very clear - a little bit of a surprise over the past year - this opportunity for a Linux based version of the runtime. Once you get above a gig of memory or so on a box [it’s a] solved problem - there are a lot of great solutions, and of course you've got Docker and then you get up further and you're in the Kubernetes world. There's good stuff happening there, lots of commercial offers, but it's become clear there's this ”tweener" class of device - I can run Linux, but if I use a traditional container runtime like Docker, I've just eaten up a couple hundred extra megs of my memory that I could otherwise use for something else. That "tweener” class is like set-top boxes, routers, switches, different types of appliances and digital signage sticks. Imagine a little HDMI stick you can stick in. It's WebAssembly. It's basically a web server. I can show graphics and do all that. 

Stephen Berard Right. 

Jason Shepherd There's this in between class of a Linux capable device that [Atym] would become an alternative to Docker with our Linux runtime that we just created. I don't know if you can talk a little bit about that. 

Stephen Berard Yeah, I think it's really exciting. There's a tremendous amount of these embedded devices or devices people traditionally call embedded that actually run a Cortex A processor. We have them in our phones, they're like supercomputers, but there's a whole set of those, as you say, that are much smaller size, limited memory, limited CPU power. They are Linux capable and so being able to run our runtime and run containers across MCUs and [CPUs running] Linux is, in my opinion, a game changer  I think the other thing though is when you start looking at where you deploy these, there's also a slight difference like when the things are in a data center, even if they're small devices, it's control. It's a kernel environment, it [needs to be] IT managed, or managed in some way. But when these things are at the edge, that's a different problem set. And how do you solve that from a manageability, deployment, monitoring and software update perspective? You have a million of these. Digital signage sticks spread across malls all over the world, right? That's no longer in a controlled network environment. And so being able to manage that through the same control plane we have now, I think is gonna be very, very useful. And then how you manage that as a fleet, how you monitor that as a fleet is obviously something that's really core to what we do. 

Jason Shepherd Yeah, even there's some folks we've chatted with. They're like, oh, well, yeah, Linux, we've got Docker. Well, hey, wouldn't it be nice to have 200 megs [free] to run a more complex AI model or other applications? Or I don't know... pick a lower cost BOM, lower hard cost hardware, save some power... you know, anything like that. But then also you have to remember that Docker doesn't come with a console. So, you're still on your own to deploy devices and manage both the hardware and applications and whatnot. 

Stephen Berard Right. 

Jason Shepherd This notion of us evolving to where we have the Atym Hub, our core offer and then different flavors of runtimes - both Zephyr, the current RTOS reference and Linux, which could be any flavor - certainly, I think we're just scratching the surface on where it could go.  

Stephen Berard Yeah, yeah. It's really interesting. I think you and I have both seen this too. We're like, how many times have you seen something that's in an embedded box? So it's a little metal box somewhere on a factory floor plugged into a machine that's effectively just a small, full PC, like an Intel NUC or something like that. And all it's doing is talking about a zero port and putting that thing on the network to report some data. But you're using a full PC for that and the BOM cost on that's probably like two, three hundred or more. But the reason why they choose that is that it's much easier to manage from an IT perspective, right? They want to be able to manage that via containers or whatnot. Well, imagine now if we could cut that BOM cost [by] a tenth, but offer the same manageability, if not better, right? And so now you can replace that PC with something that is a lightweight Cortex A processor, potentially 8 or 16 megs of memory. So BOM costs in $20 perspective, suddenly that makes that solution much more affordable, right?  

Jason Shepherd Or sub $10. I mean, it raises a question- what is embedded, as a space. I mean  historically it's I'm embedding functionality into some sort of device. It tends to be fixed in function. This is what it does, and it does it really well. But those lines are blurring. We're starting to see these types of devices becoming more software defined and the blur between we call ”device up” and ”cloud down”, or you can look at it like OT blurring with IT, all that kind of stuff. 

 Stephen Berard  Right. 

Jason Shepherd   Just yeah, these things are changing. I think that's the most exciting thing about what we're doing at Atym - just seeing these worlds blurring and the fact that we're only scratching the surface on the opportunity. [With] this new Linux expansion, it'll be really cool to see what new types of use cases we see. But obviously in parallel there's all kinds of stuff on the MCU side with the existing, initial runtime. 

Stephen Berard Yeah. And, and if you look, I mean, we're both in a better world and I think a common theme that we talked about when we were there in Nuremberg [at Embedded World] early in the year was the change that AI is driving in the microcontroller space. You're starting to see microcontrollers now that have large amounts of memory from a microcontroller perspective, maybe 4-8 megs of memory and AI acceleration, right? And so that's for me is also really exciting. And then to be able to then say, well, I actually want to run on this device or I need more CPU power, so I'm going to run on a Cortex A. Not have to choose, right? You don't have to make restrictions. You can choose where you need to run and you can choose from a performance and a cost perspective. That's for me really exciting, right? And so seeing that, I didn't really expect this two years ago - that we'd be seeing microcontrollers with this much memory, with this much acceleration to support these use cases. And to do it in these low power envelopes is fantastic. 

Jason Shepherd  Yeah, I mean, obviously we've been seeing co-processing for a while, but this notion of a hybrid architecture with an M class and A class on the same device. I'm going to spin up like super fast actions on things on the M class. Very, very low power. I'm going to bring up the A class if I want to do more advanced processing. With this architecture, I can run the exact same binary on both sides, or even further up [the edge continuum]. I mean technically run in the cloud. That's not the idea here, but you technically could run Atym containers anywhere. 

Stephen Berard  Yep.  

Jason Shepherd It's just this mash up of different architectures and device types and skill sets and all that sort of coming together, and the lines blurring I think is the most exciting thing. 

Stephen Berard And I think if you look at containers, CI/CD, agile and other stuff, those are all like things that are really good at enhancing innovation that makes us deliver software faster, better, also other stuff. But what really enables that is a set of tools behind that, right? It's things like GitHub and Jenkins and build servers that support those things, right? And the tools that the cloud folks have had, the desktop and regular software developer folks have had for a while has been lacking on the embedded side. Som you couldn't do containerization on these microcontrollers. You couldn't very easily do CI/CD and deploy that very quickly because the updates were big and potentially risky. So having a proper set of tools that matches the way people are already working in the cloud and not have to compromise because you're on an embedded device or a small low end, you know, PC style device. I think that's really exciting, right? I think that's going to open up innovation. 

Jason Shepherd Yes. 

Stephen Berard    You know, quite frankly, look at the amount of devices in that space, right? It's pretty exciting to see what people are gonna do with this technology. 

Jason Shepherd    Yeah. We should wrap up. I love the story of when we were working with that customer that we got the runtime going on the device for. I think it had like a meg of memory or something like that. You've now abstracted all the hardware and it's ready for you to deploy containers when plugged into our hub. We [then] get on the phone with the machine learning engineer and he's kind of like, oh, what do I have to change? You know, he’s used to Docker and Linux and CI/CD. What do I have to change to work in this new environment? And we're like... the target directory. 

Stephen Berard   Yeah, you [just] have to pass dash dash sysroot to a different different CMake file. 

Jason Shepherd    I think the very key is this - it's enabling entirely different camps of developers to work together on new classes of devices without having to like fundamentally change what they do. So anyway, well here's to what's gonna happen in the coming year. It has been super exciting, and it's been great working with you on this and with the team. 

Stephen Berard   Yeah. I mean, you look at where we were a year ago and look where we are now. It's fantastic. So, it's been an exciting ride and I'm sure the next year is gonna be even more exciting. Looking forward to it!