Zephyr Marks 10 Years of Embedded RTOS Innovation

Hey everybody. Welcome back to Minneapolis. We're at the Open Source Summit, and we're having a little chat about Zephyr, which is celebrating its 10th birthday.

Congratulations. Thank you very much. All right.

We're excited to finally... The last 10 years have gone by so fast. This is true.

A decade. It's a decade in the making. It's a decade in the making, indeed.

For folks who don't know what Zephyr is, kind of explain where it fits in the lexicon of things, but we're really talking about embedded systems, right? Yep. It's very much an embedded system operating system.

It's a real-time operating system, and it's been used where Linux doesn't fit. Okay? Linux doesn't get smaller than three meg.

Right. Zephyr starts at about 8K and goes up. So we're bringing a lot of modern software development practices into the deep embedded space.

And so if you're familiar with Linux already and things like Device Tree and Kconfig and so forth, those mechanisms are in Zephyr. And so, you're able to put the pieces you only need into your image, and there's a lot of things, so you can have been pre-integrated into the code base, and you just build in what you need. So you don't have to go buy another library over here and integrate another library with this or that thing with that and do your integration.

A lot of this stuff has all been pre-integrated. " Sure. And so that's kind of how we work with Zephyr.

Sure. So what are some of the use cases that we're seeing with Zephyr these days? " Yeah.

There's a lot of new stuff that's coming in. You're seeing it a lot in wearables. Mm-hmm.

It's showing up in a lot of the wearables where you want to keep your battery life going for a long time. Everything from hearing aids to biomonitors, things like that. It's heading into medical, I'm seeing.

And tags for tracking things across the planet. So we're seeing those types of applications where it's being used in wind turbines. It's being used in, familiar with Keychron's keyboards?

Mm-hmm. Yeah. They're running Zephyr now, and then we're up in satellites too.

So do you think, I feel like we went through an era where compute resources were considered infinite. Now, we're doing these use cases that are really close to the edge and really personal devices and stuff in space and is there a new appreciation for the art of coding? I think the art of coding was always there.

I think we're making it possible for people who did one subs to improve the security posture and to collaborate on the common denominators. Mm-hmm. And they have not been used to doing that up till now.

I think that's it, because these devices, to a large extent, have always been with us for the last 10 years plus, right? Mm-hmm. It isn't like Zephyr's created new ones.

And people have ways of doing it, but developers can now do it much more efficiently. " Zephyr's got over 1,000 boards in its repo right now. Right.

Okay. We've got over 300 sensors sitting in the repo. You Kconfig and draw the pieces you want to create your platform, and that's tremendously fast.

We've got a lot of cases where people have gone in and, "I'm just going to prototype and kick the tires," and all of a sudden, their management's impressed that they're done within a week. Yeah. And that suddenly makes a condition for them to start joining the project or start getting active upstream.

So where is that line between what Zephyr does that we could consider important but non-differentiated value and what developers are going to do on top of Zephyr and kind of say, "This is where we add value"? Yeah. So things like the talking to the hardware is where Zephyr's doing it.

Zephyr's focusing on security. Zephyr's also going to be focusing on a certain amount of the safety analysis. Okay?

And that's things that everyone will be able to share. The differentiated values are how you are designing your interfaces, how you are branding things, what capabilities you're making available. The devices of looking at power over time.

These are the things that people will put into their products, and they will have solid capabilities for their products. Whereas the Zephyr will be the, and quite frankly, is to a large extent right now, a lot of the underpinning. Mm-hmm.

So where does the project go from here? " Well, that's why we did the survey. So there's a nice little survey out, which hopefully you guys will include a link to with your project.

And we reached out to RTOS community. Now, obviously, we're reaching out, so we get more Zephyr responses than others. But we got a statistically significant number of other RTOS users who were telling us what they cared about, where they weren't, so forth.

And by contrasting what Zephyr's using and what they're doing, that gave us some spots to clearly aim at to try to improve our story. And so there's-So, some of the robotics areas is an area I think we're going to be trying to get more interaction and engagement in, because I think it's a natural fit for Zephyr to a certain extent. And so there's some stuff there now, but improving the story for those in these market segments, to get them closer to being able to use Zephyr.

We've been doing a lot of things we can to be CRA ready. The EU Cyber Resilience Act, it calls for certain things, and most of those things Zephyr can do now. But how we can make it easier for manufacturers to adopt us in that space and to do more homework, so we're starting to look at those types of issues.

Finally getting our 61508 safety certification is going to be a big one for us. Mm-hmm. And right after that, we've already got the contract signed to work on 26262.

Yeah. So the safety certification would allow people to deploy Zephyr in what kinds of use cases? Anything that has some degree of safety criticality with it.

Mm-hmm. So things like in cars, in planes, in the transportation sector, which is another one, and in some of the home automation, in the appliances, in medical devices. All of these areas have aspects of safety because it involves people.

Mm-hmm. So getting it to the stage where we can actually get this information there and available and used, is going to basically prove we can do it, and so that others can see the pattern and then use it to do their assessment too. Because safety is always a system assessment, it's not just a component assessment.

And so we're trying to make our component be a very easy one to use to basically certify a system. Mm-hmm. And then put evidence to make it possible.

Do developers need to think differently when they're building these types of applications? And I'm asking this question because historically, a lot of developers built things that were batch oriented, and the processes were sequential. I look at this and it's real time, and many more things are happening in parallel.

Is that a different way of thinking about how to build an application? I think embedded's always had an awareness of the hardware underneath it. Mm-hmm.

At some of the higher server-level stuff, they don't have to need to know quite so what's going on. But embedded's always had a certain amount of awareness, and the people doing bare metal definitely know the details, the gory details. Whereas when we up level up to the, Zephyr lets you do a certain amount of abstraction.

So those who are comfortable with embedded Linux are usually pretty comfortable with Zephyr. Those who have been working in other RTOSs, there's a bit of a migration curve to deliver some of the specials, like the device tree and things like that. And the same for the people coming in from bare metal.

But the hardware awareness and that mindset has sort of been in embedded, I think, so the details are different, but that's about it. Yeah. A lot of folks have an interest in all things AI these days.

Oh, yeah. What will it take to kind of take AI all the way out to the edge in some of these use cases? Oh, it's already there.

All right. Well, what does that look like? Well, there's a product I learned about five years ago, so it's been there for a while with us, that is a little device with a camera in it that sits in the top of a garbage can and uses AI to scan the contents of a garbage can to determine whether it should be emptied or not.

Mm-hmm. That involves a certain amount of pattern recognition, what are the levels. Yeah.

And you get to save resources by, oh, hey, this garbage can is empty still, so there's no point sending a truck to it, right? And so saving resources by monitoring that saves fuel and saves other things. So it's at the cost of a battery lasting for a certain period of time, and so you're going to want to maximize your battery life of it in the device, and you want to make it sure it's rugged.

But yeah, waste monitoring. Another case for what it looks like is, Atheros put out a satellite last year up in orbit, and they used Zephyr to do a certain amount of pattern recognition with simple AI at the edge to recognize the conditions, to bring up the more expensive FPGA to do the detailed analysis. So the same way you're looking for a set of conditions.

So using Zephyr for filtering for certain conditions in order to do something more power expensive is already in use, and it's that sort of mixed system that's used. You'll we see things that are AI agents based on LLMs kind of running out at the edge as well, because I think most- Yeah ... of what you just described, I would call machine learning kind of, or applications.

Yeah. But remember, agents mostly, so yeah. So those machine learning, I think some of the agents will be deployed potentially to do some of the analysis, and you might see some of these things at the very detailed edge communicating up to clouds where these things are deployed rather than being down there.

Like TensorFlow Lite's working with Zephyr and has for a while. So has Edge Impulse, and then I think there's a third actually that I know that's out there. So we've got various, the code out there.

So things like Magic Wands and training those- Mm ... those are using Zephyr today. Ah.

I think Disney's Magic Wand, I think Leaps is using Zephyr. So, these sorts of applications, the AI is there for recognizing gestures, recognizing the video, and so forth, and that will, I think, happen at the edge. I think when you build up a Zephyr image, you don't usually have a command line.

It was added as a capability, but for that, it's not the detailed, and most of the agents right now are working through a command line type of interface. And so whether or not those are actually going to get there, it might. Nothing to really stop it, but I don't know of a case for that one right now.

What do you need from the community? You've been working on this for a while, but there's plenty of folks out there who want to contribute to something, but they don't know how to get started. But what do you guys need from them, and what are you looking for?

Oh. Well, what we're looking for is people with expertise in safetySo they can help us get our requirements set crystallized in with the code base, and so we can get the safety certification. That's my big one, personally.

And we need to get people helping us do reviews, all the janitorial things that the projects tend to need. The bulk of the code base at this point in time is drivers and sensors. And so people who are willing to help us with restructuring the code so that we can be more efficient at scale for analysis, not for performance, but for analysis.

Those are the things that, if you ask me the detailed ones that I really need, those are what I need. But there's a couple of stacks we'd love to get in the code base. We'd love to get a native Matter stack in there.

Mm-hmm. So if there's any Matter developers, would welcome you guys. Various vendors have their own versions of it, but getting some collaboration going there would be a really good thing for the project, as well as hopefully make things a little bit more, how should I say, flow through the ecosystem better?

Yeah. Well, is there a crossover between Zephyr and all the other open source projects that are out there, and are contributors working across multiple projects? Yeah.

There is some. Some of the experts tend to work in multiple RTOSes. We are license compliant with NuttX, for instance.

We are not license compliant with the Linux kernel. On the other hand, sometimes the same developers will contribute code under one. If they own the copyright, they can choose which license it goes out under to which project.

All right, cool. So there's collaboration possible. Well, folks, you heard it here.

Zephyr turned 10. There was a birthday party. Sorry you missed it if you were watching this, but join the project, because next year we'll have another party here.

Thank you very much. All right. And we'll be back in a minute.