 This is really meant to be a birds of a feather session was the the kind of the scope we really want this to be an interactive discussion. This is really about how to get the Linux community more involved and leading Beagle and have Beagle serving the Linux community more right. So this less of us, more of you, but we will give you a little bit of an introduction to to Beagle board if you're not really that familiar with us. We exist exclusively as a nonprofit organization. We seek to advance the the state of embedded systems development utilizing open source right and not just open source software but open source hardware. You know, we want to see things go as open as deep as is reasonable right so as as is possible right that all the way through the the silicon level right. But where we can't have that at least we can have really good documentation and ready supply of processors that we utilize and we open it up as absolutely much as we can because we want this. This is about preserving knowledge and making our jobs easier to make things. This is the leadership of Beagle board.org you're looking at three of the board members. I'm Jason Kreidner. I was a founder of Beagle board.org and yeah, can't even say something quick about yourself. Microphone. Yeah, so I've had background in embedded Linux and sort of wireless router stacks. Sorry, I've been at the booth lost my voice and more recently, you know, microcontrollers and to me those domains have always been kind of separate. So joining the board here with Beagle board is kind of fun I can play in both both areas at one time. And if you haven't checked out MicroBlocks, MicroBlocks.fun that's on the microcontroller side. So I'm very big on the openness of software and hardware to improve education and access to technology and get more people engaged in technology, especially girls and women. Hi, I'm Drew. Work for Bailey Bray and also another board member of the Beagle board.org foundation so two things I really love are electronics and Linux. So when single board computers started, I was really excited about that. Before Beagle, I was working with some proprietary ones and I got frustrated that you could get maybe a schematic, but not a full schematic. You couldn't get the board layout, the processor you couldn't buy. And then I met Jason and a maker for many years ago and he was telling me that the Beagle bone is completely open source. So open source hardware so you can get the schematic, you can get the PCB, you can get the PCB layout. All the parts are available in quantity one from distributors so anyone that wants to can build any of the Beagle designs, modify them. One of the ones I liked was someone made a derivative board that was a logic analyzer, I think called the Beagle logic, which was really fun. So yeah, I was playing around with other boards and I discovered that the Beagle was much more open and that's kind of what hooked me into the thing. And then going to the current day, I'm very excited about the potential of even going beyond the board level being open source. There's a lot of developments now in open source silicon design, so in the future we might even get into that and risk five. I did remember to put a quick slide in just to have us talk about risk five here, but I forgot to put GSOC. So the guy that did the board Beagle logic, Abchec Kumar, now works for Google but he was a Google Summer of Code intern with Beagle board. And I see some other folks in here that participated as interns and Beagle's mentorship in Google Summer of Code. And that's part of one of my favorite times of year, which is now. We're about three weeks into coding now with the Google Summer of Code interns. And to me, that's a great focus. That's where a lot of our energy goes is really trying to mentor students and bring up more people. So we have very much a mixed focus for us. I think education is education of everybody. So continuing education, young education, university, a graduate level education. And it's very much about improving access and enabling people to deal with all the stuff that there is to learn about dealing with embedded systems. So there's a couple of board members on here. I think a lot of people that are around Beagle board and Linux would be very familiar with Robert Nelson. He is the distribution maintainer. He works on the Debian distribution for the Beagle boards and also pulls in. I see a bunch of the TI folks here pulls in all their kernel patches to make sure that they're applied. So we do a kind of a mix of chasing mainline and supporting vendor kernel stuff. So we try to make sure that we're constantly switching back and forth to running both mainline and vendor kernels. And Mark Yoder, right now, I guess he's the only full-time professor on the board. But he has a course on embedded Linux at Rose Holman. And he has written some free online books that are on docs.beagleboard.org. And he's working on another one this summer. And he utilizes those and a bunch of other materials, some up on the Linux wiki and other locations. But he shares all of his teaching materials so that people that want to teach embedded Linux can. Anything else about the leadership? Any questions about the board or CEO or how Beagle board is run? It is a 501c3 organization that supported primarily through royalties but also through donations. Which is a non-profit in the U.S.? Yeah, that may not translate. So it's a tax exempt non-profit. So if you made a donation, if you were in the U.S., you'd actually get to not pay tax on it. That wasn't that money. So what makes Beagle special? We're not just about making cheap computers. We want to make computers that you can make into other things. Ultimately, things that don't look like a computer, they just look like an appliance that just does the job and doesn't make technology get in your way or run a platform that serves some other interest. We want to give as much power to the individual as possible. Which also means not preventing them from going to business. So actually making and selling things. And so our community in general is very entrepreneurial and domain expert. Heavily expert in building embedded systems. And a lot of what's kept Beagle board really exciting for years where we haven't maybe done a lot of updates to the main CPU cores that we're utilizing is that we mix both Linux and microcontrollers and ultralow latency microcontrollers in one platform. We do lots of new amazing things with that like the CNC mills and lighting and audio, like ultralow latency audio, the types of things that you want to do in an embedded system. We're doing real-time control, hard real-time control and all the fun networking and amazing things that come from Linux. We've kind of been everywhere. When a Raspberry Pi goes into space, it's big news. When a Beaglebone goes into space, it's Tuesday. There's actually satellite networks built out of people doing space operating systems and people building their CubeSat frameworks off of Beagle's. So lots of people have written books, most of these on Beaglebone Black. But I think with some of the new stuff that we're coming out with, we're going to see a lot of new interest and new people getting started with BeaglePlay and BeagleConnect Freedom and a lot more materials coming out on those. And of course, there's been just hundreds of add-on board systems that have come out mostly open-source, not all, but a large chunk of them, people creating more open-source hardware. And something I'm especially proud of is that the Linux Foundation has historically used Beagles as their platform for teaching embedded Linux. And I know some other folks here have worked on teaching materials. There's one there, but another book author at the back of the room, and teaches courses with Beaglebones. And even the Linux Foundation themselves do that as well. And of course, there's people create projects all the time and share them. This is, I think, a kind of a fun project that came out of the Jet Propulsion Laboratory. And I just kind of solved this on a National Geographic blurb. And this is something that actually has happened to me fairly often. And when I see a Beaglebone Black, I notice it immediately. But six seconds into a National Geographic special, there's a Beaglebone and their Mars rover prototype. They never talk about it. They never show it again. But they do drive it under the Arctic ice to simulate the environment on Mars. And then people go to school, they learn all about how to build embedded systems. Beagle and they come out and they graduate, they do things like this, like make the world's first cheeseburger robot with 14 Beaglebone Blacks. And so these guys tell me that they would never allow a PLC to be utilized in any of their designs. This is the model that they use as Beaglebones. And so there's just a lot of different... The user base for Beaglebone is really large and a lot of people doing a lot of different things. Even building in products, even ones that require safety certification. So Dremel actually puts Beaglebones into their laser cutters. And it may not seem like such a huge number given some of the other things, but we've sold over 6 million boards. And so all the Beagleboard doesn't get the profits, but there's a royalty based on the utilization of the logo for people that manufacture it with the logo. If you wanted to make it on your own, you can do that and a lot of people do. A huge portion of the designs that come across TI's desk actually start with the Beaglebone as a design and then they modify it. And that's where a lot of people build their custom systems, but if they just want to buy one off the board, just don't blame me, right? You've got the design, you've got all the data sheet materials. Don't blame me if it doesn't work. You've got everything you needed to test it and validate it in your use. But we've done some new things. We've not just stuck on ARM Cortex A8 anymore. That's something that I think a lot of the embedded Linux community doesn't know is that the Beagle has moved past Cortex A8. 64-bit? 64-bit. So this was the first 64-bit Beagleboard, the Beaglebone AI 64, which has an eight-tops neural net accelerator and finally good support for those accelerators in Python. And the accelerators themselves are... There was GCC support for the C6 DSP that's used for part of that. And the C7 does not have GCC support, but there is a tool chain. We actually ship a ARM-hosted tool chain for the C7 processor that's running most of the neural net processing on the board. So the thing about Beagle is we try to ship the boards with all the tools you need to develop for the board on the board. So if you want to program the C7 DSP, all the tools you need to program the C7 DSP are actually shipped on the board. You plug it in, you log in, and you can begin programming the C7 DSP. Maybe a little bit of a learning curve, but it's possible. And most of the source for those are released from TI. I'm not aware if there's any limitations in terms of what's opened up source-wise for the neural net accelerators. My understanding is they provide all of the source for that. Yep, so if you want to dive deep, yes, the technical reference manual is over 10,000 pages, but better to have it and not need it than to need it and not have it. So the information is there. We love Texas Instruments for this because they know how to make the chips available and provide the documentation, including the ARATAS, to all the developers. This is a pretty powerful board. It's a dual-core Cortex-A72 running at 2 GHz. Other numbers in here. It's got PCI Express on here. You've got USB3. And I think that we ended up selling this one for about $180. There's lots of peer-use. There's actually 12 peer-use cores in this. Yeah, explain peer-use. We talked about being able to support hard real time. A lot of what's made Beagle super interesting is the ability to do ultra-latency hard real time. The peer-U is a 32-bit risk core. It is proprietary to Texas Instruments, but there is GCC support for it. And there's LLVM backend work in progress during this year's Google Summer of Code. So it's a 32-bit risk, but it's fairly magical in that there are registers in those cores that are directly mapped to IOPEN. So R30 and R31 in some of those peer-use when you get some of them are staged, but the ones that have IOPEN access are directly register mapped. That means you can sit there and XOR a pin. And there's zero pipeline depth, single-stage pipeline depth. So you can do every cycle. You can toggle the clock. So on the Beagle Bomb Black running at 200 MHz peer-use, you can generate 100 MHz clock in a CPU. And you can sample at those rates. That's how they built a logic analyzer, the Beagle logic. We just used the peer-use to go do the sampling of the IOPENs. And because this is within one system, you also have access to that DDR memory. So what a lot of people do is they do the high throughput stuff with the BFA72s, A8s and A53s and the family. And then they use the peer-use to bit bang that out on the IOPEN. That allows doing things like the... So you can do 3D printers, but you can't spit plastic out as fast as you can control the motors. So where it really gets more interesting is when you start doing CNC mills and things that can actually move much, much faster. Because you can do the complicated path planning for five-axis mills and keep up with the motion path. All those huge amount of pulses and feedback loops that you have to do to drive the motors. So did I explain peer-use enough? Is there any questions about peer-use or anything so far? This really should be more interactive. There are open source libraries for motor control. So there's the machine kit project. So most of the people that leverage the peer-use, it's usually a small handful of people that actually do the programming on the peer-use, but then they release a firmware sets that allow that support to scale out to a lot of other people. So if you look at folks like Bella, they've done this ultra-low latency audio synthesizer and they're using the peer-use to be able to take the synthesis samples and respond really quickly to the codecs. So they have really short buffers. If you look at doing things like the CNC mills, the machine kit has the code in it already and you can configure the geometries in Python. You don't have to spend any time doing the peer-use code because the peer-use code is already written in a way that allows you to hook up numbers of stepper motors and coordinate them however your geometries are. So where you essentially fill up the memory with patterns of pulses and then the peer-use pull those pulse patterns out and drive them out to the motors. Lighting as well. There's actually really popular in Christmas light displays. If you played with the WS-2812s, those are great, but those are at least in order of magnitude more expensive than the type of LEDs that you can use, like they're used for Jumbotrons, right? So you can replace a $3,000 FPGA controller for driving these parallel panels with a $50 Beaglevon Black because of your ability to process the high throughput stuff with the Cortex-A class processor than the peer-use to be able to deal with the bit-banging. And yeah, this has got a lot of peer-use. This is newer. This is our latest board. I think this is the last one was about $180. This is under $100. So this is 64-bit quad-core A53. And you'll notice the size of the heat sink? There isn't one. No heat sink, right? This thing runs nice and cool, quad-core A53. It's really a sweet spot, right? The processors themselves, it's on this side of the board. I know that you can look at the variance of this, that you can look at on TI's website, that are under $5, right? And I'm not sure exactly, which may be a thousand-quantity, right? This is finally the replacement for the AM3 that we've been waiting on for a long time, and it's real, right? So we're no longer stuck in the Cortex-A8 32-bit ARM world, right? We are fully into 64-bit world. So we've got a little bit more memory on here. We've got the two gigabytes of RAM and 16 gigabytes of flash. And we've, you know, the processor is obviously tremendously faster. This one has a two-pair use on it, and it's also got an M4, and so the M4 is accessible for programming as well. So, yeah, there's that. And I think we've done some really interesting things with the connectivity on here. Under here is a CC1352-P7 microcontroller. This also has an M4, so the main processor, SOC, has an M4 in it, but there's an M4 out here in the wireless processor. And it's also got an M0 in it for, it's essentially a software-defined radio. And you can do 2.4 gigahertz stuff, so you can do stuff like Zigbee and Bluetooth Low Energy with that. But you also got sub-gigahertz, right? So you can talk to devices a kilometer away, and you'll get, you know, a reasonable amount of data throughput, right? So some people talk much, much further than that, but that's kind of how we've tested it as, you know, a kilometer in open air. What about the GPU? Oh, my. That is a good one. So thank you, Nishant. So it's still an imagination PowerVR GPU, but something's different. TI has made an agreement with Imagination to release an open-source GPU driver, so both the kernel and user space, no user space binary blobs. It's still, yeah, finally. So this is leading the way with OpenVR's open-source GPU stack and Linux. So it's on the least candidate three right now. They're still working on a lot of the compliance, but that work is moving on and I think in a lot of ways, it's usable and functional now, right? But I think for getting Mesa and all of the Vulcan compliance stuff, there's still a fair bit of, there's a lot of work left to do, but it's real now. So, yay. Yeah, so I'm so used to just pretending we don't have a GPU, because the last thing I want to do is talk about one where I have to use closed-source binary blobs in Linux and no more. Since we're talking multimedia, I see you have a CSI2 connector for I think the first time on the Beagle board? No, actually, there's two CSI connections on BeagleBund AI64, so an AI64 has some reasonably powerful image signal processors or ISPs on it. So it's definitely something to note that BeaglePlay and the AM62 do not have integrated ISPs. So I think utilizing something like Lib Camera and using maybe either GPU, to me it makes probably the most sense, like actually doing GPU acceleration to perform the image signal processing functions. would make a ton of sense, right? So it's got a four-lane CSI, but no ISP. Okay, and the CSI2 connector in the play, is that compatible with Raspberry Pi cameras or is that not? It's the 22 pin that you'll have unlike the Pi Zero. So it's the four-lane version, right? So it's the same connector footprint, right? I think they seem to be focusing more on the lower pin count two-lane version. Two-lane one. Right, but we didn't want to do that. And you can get... There are people that make the ribbon cable so you can get to the two-lane version if you want to use the Pi cameras. Okay. But software work to be done, right? So let me also ask what is the ISP support current status for the bigger board? The AI64, there are ISP loads for a number of fixed cameras, but they're blobs. They're binary blobs that you run on the M4s that work with the accelerators in that subsystem. They're not... Oh, you're going to tell me they're going to change that? Please tell me you're going to change that. I see TI folks raising their hands. Yeah, so we have an intern actively working on lip camera support and an open V4L2 driver for the ISP, so it will work for bigger board AI64 as well. So using the accelerator hardware, so like the... Nice. Yeah. Woo-hoo! Chris? Hi, I got a couple of cameras. No, a couple of questions from the streaming chat. So the first one, the question is... Hang on, I'll put my glasses on. Yeah, can you say something about the Beagle 5, the RISC-5 version? We do have a... Guy says, when will it be available? I've been on the list for over a year. Yeah, so we did cancel the Beagle 5 Starlight project. It was very unfortunate. It was... I think it was a lot of personal pain. There was a lot of blood, sweat and tears that went into that project and a lot of passion to get that. But the passion for RISC-5 is not gone from Beagle at all. So I would say you might not have to wait that long for something new from Beagle around RISC-5. That's probably as much detail as I can give. What is Chris saying? Yeah, yeah. The thing is, we do care deeply about the RISC-5 community and we will be doing things to help advance development of the RISC-5 ecosystem. Okay, just one more quick question again from the chat. So Alex points out that there are still two binding blobs on the AM62XX and it goes someplace. So the question is, is there a plan to get rid of those binding blobs as well? The R5, there's a set of code that is involved in the clock trees. And I think the statement from TI is that we won't be able to get rid of that binary blob, but it's not running on the A53s, right? We are still able to load the boot loader. So you can boot and run some things without it? I don't know, Shant, do you want to speak any better about what's on there? It's not 100% dependent to have that blob? Yeah, so there are two binding blobs. One is the security enclave binding blob, which is signed. And the R5 binding blob that does the device management function. If I recollect right, the source is actually available in one of the SDK versions, but it's a pain in the ass to build it up. Sorry, my friends. The system is abstracted away for the device management function to work. Expanding that functionality with additional software should be possible, but you want to be careful. It's not a general-purpose processor. It's one of the core systems that's running here. So I don't know if that helps. I was trying to give you the connectivity, but I'm okay with all the questions. I think I've seen... Sorry, I've seen maybe a single pair of Ethernet in the Bigel Play. Could you give more details about it, especially if it's a Macsec capable of loading or a PTP capable? There's no... The PHY itself doesn't have any clock recovery, I think, but you can hook it up to the peer use so that you can try to do your own kind of timing. But you'd probably have to look... You'd have to have a little bit of external circuitry in order to do timing recovery. I think that's a super interesting thing to do on that, but it is a 10-base T1L, right, so the long range. We do provide a non-standards-based power over data line of 5 volts at 250 milliamps without the PD discovery. So you just have a software mechanism to essentially turn that power on the lines, mostly so you could hook up microcontrollers to it over fairly short distances and drive communication to a string of microcontrollers. That's the vision there. And more to experiment with power over data line. If you want to run the real long distances, you're going to want 24 volts and you're going to want the power to discovery. While Jason speaks about the technology, I just want to provide some ideas. People have phone lines, phone cabling in their houses that isn't being used anymore. You could use it to wire up microcontrollers or some other sensors in your home. And there was a group here from Test Automation. They have these adapters that you could use to use that power over data line to run test information in and out of the board and set up big test board farm automations. So there's a lot of interesting examples we've heard about. The cabling is extremely forgiving. Better to be twisted, better to be shielded, but if it's not, it's still going to work. It's a very nice, robust technology that I recommend people get some time playing with. It's a lot of fun. And then there's traditional Gigabit Ethernet. But this sub gigahertz wireless is one that we're doing a lot more around immediately. I think we have more plans to do things around the single pair Ethernet as in an accessory capacity. But right now we have Beagle Connect Freedom for kind of wireless peripheral expansion. So it's a way we've gotten really, at least me personally, he's gotten really hooked on this micro bus thing. It's a nice little standard connector that's freely licenseable, like the logos if you just follow the standard. And it's got SPI, iSquared C, UR, PWM, analog and power and a nice regular connection. And we've worked with Microelectronica to have an identifier. And I think Vaishnav's in here. He's actually submitted two RFCs for the micro bus to turn into a bus in the Linux kernel. So an actual probable bus so that on the here you've got the micro bus directly on and it probes the one wire EEPROM that reads a manifest so that you can automatically load the drivers without a device tree overlay. So there's just one device tree entry to describe the bus and put the bus in there, but it's because it's a probable bus you don't need to have a device tree entry in order to discover the board. We have manifest files for 150 boards so far out of the roughly 1500 that Microelectronica makes and they're in the process of putting IDs on all of their boards. So I think that's a very promising thing to help move along the IAO subsystem and all the sensor stuff inside the kernel. But we also have this concept of remoting that interface. We've actually taken gray bus from Project ARA and enabled that to work across and that's part of what my talk following this will be about so that you can actually get that micro bus and all the discovery and everything to work remotely over there for making sensor networks. So this is the board as it ships right in the enclosure. There's a slide to remember to talk about risk five, check. Talked about risk micro bus. We didn't spend a whole lot of time talking about it. We did mention that we have an eight tops accelerator in the AI64 supported by the Python libraries. That's NeoDLR, Onyx and TensorFlow Lite in an open stack. Briefly touched on gray bus. I think we kind of hit stuff. We get a lot of interactive discussion, but we'd like to carry that discussion forward and actually get some guidance from the Linux community. We've started a discord recently. Maybe that's a little more accessible if people don't like IRC. IRC is still a great way to get in touch with a number of us to just chat online. But we've started a podcast. We're happy to have other people come and talk on that. We want to do blogs. We want to talk about what you're doing. And we want to try to find ways to make ourselves more accessible so that you can get inputs on what we're doing in the future. I think we have two minutes. So there's quick questions. Is there any questions? Yeah. What do they want? You can tell us now what you want. You said that the subgear's radio on the Beagle Play is sort of like a software-defined radio. So is there support to build our own protocols with that radio? Support. So the TI SimpleLink SDK does have the ability to kind of, you can mix frequencies and modulation standards and change the way that that radio is performing. I'm not, I'm speaking a little out of school, but it does and it supports, they've got software loads for Amazon Sidewalk and Wyson and a number of different of these things. But I don't necessarily like using that code too much. Whatever gets integrated into Zephyr and exposed through Zephyr is something I'm far more comfortable with. But I think those are the more knowledgeable might work on expanding those features out. And you can, there are some ways to do some patches on the M0 that's actually kind of, you know, driving the radio modulation. But I'm, you know, right now as far as I know they just publish entry points into the ROM code. Not a full stack. Think of time for one more quick question. Or suggestion. Or suggestion. Any beat replacement for the Pocket Beagle to do 64-bit? That sounds like a really great idea. Awesome. You can, yeah, feel free to come talk to us at the booth. We're booth number one right at the coffee break as you go towards where the paid bar is, right? So, yeah, go that way. Out here into the left, all the way to the left. Thank you very much. Really appreciate your time.