Show HN: We made a small and cheap network switch | Hacker News Reader

Show HN: We made a small and cheap network switch

docs.murexrobotics.com

666 pointsbyran158 comments

Hello, we're Max and Byran from MUREX Robotics, a high school robotics team from Exeter, New Hampshire. We are super proud to have made this open source piece of technology! It is only 6.9 dollars (actually!) from JLCPCB :) I hope you like it.

You can find us at byran@mrx.ee and max@mrx.ee as well if you have any questions.

We will be putting a small run of these boards for sale somewhere (we have <25 units of stock), probably for $10+shipping. Let us know if you're interested in more!

Board files for everything we make is here: https://github.com/murexrobotics/electrical-2024

Open articleView on HN

Awesome work! It's really cool to see this from a high school team. While designing liquid rocket avionics [1] at Purdue Space Program, we went with a BotBlox switch that cost $80 apiece [2], which I thought was ridiculous. My proposal to in-house the Ethernet switch was vetoed because I was a filthy CS student (joke) and my co-lead (the electronics guy) said it wasn't worth our time designing and validating such a part.

An Ethernet switch for $6.9 directly from JLCPCB is pretty incredible, thank you for making this product sector a tiny bit better :)

[1] https://sagarpatil.me/projects/cms-avi-hw

[2] https://botblox.io/products/micro-gigabit-ethernet-switch

"making this product sector a tiny bit better" is exactly what MUREX is all about :). It's something we honestly believe in and will continue working on as long as we're around. Believe it or not, the Ethernet Switch was the least problematic piece of hardware in our tech stack! If you want to take a look, we have our other boards in the docs as well. Your rocket is so f*ing cool as well! I definitely want to do something similar in college.

Ok this is sick, I love the philosophy of your team. I'll be strongly considering your CM4 carrier and ESC to integrate into future designs.

Also thank you, I've loved working on the PSP rocket! Bi-propellant rocketry is a pretty rare to do as an undergraduate, and you should consider applying to these schools if that's something that motivates you:

- Purdue: https://purdueseds.space/

- Berkeley: https://www.berkeleyse.org/

- UCLA: https://www.rocketproject.seas.ucla.edu/

- Georgia Tech: https://www.gtspaceprogram.com/

- ERAU: https://daytonabeach.erau.edu/about/labs/rocket-laboratory

This is a non-exhaustive list of schools I know that have undergraduate-run liquid rocketry programs.

I'd add

https://www.colorado.edu/studentgroups/cobra/

and Portland State University!

https://www.pdx.edu/psu-space#psas

Undergrads, not catching on fire in space!

Thanks everyone! Noted deeply in the heart. :)

I think you should submit your project page [1] as another show HN, I found it to be interesting

Some thoughts:

- Agreed with not using I2C, I2C has been identified as a root cause in several cubesat mission failures: https://pure.tudelft.nl/ws/portalfiles/portal/10531886/art_3... and https://webapps.unsworks.library.unsw.edu.au/fapi/datastream... (yes clock stretching is evil). I2C should be banned for multi board communication

- Classic CAN have very small 8 byte MTU with frame preemption, which is actually useful for its intended purpose of time critical automotive data transfer. If that 4 byte brake packet is blocked by a 1500 byte packet then your car will crash and explode. But the tradeoff is that this makes it very slow for bulk data transfer

Wtf, I2C in a satellite!? I am by no means an electrical engineer or know anything about doing stuff in space but it seems absolutely obvious to me if you require any sort of communication, that you use differential signaling. Even on earth you can have interference, in space with all its radiation it's guaranteed. I'm surprised anything worked at all with I2C.

Yes, I2C is technically meant for intra-board use. But it works surprisingly well over large distances if you avoid daisy chaining and run one cable to each target. If you use multiple identical target chips you need to route everything individually to a central MUX anyways.

As an example, the Nintendo Switch used I2C over a ~2m cable to communicate between the controller and nunchuck. Worked fine even in noisy household settings with wifi and microwaves and whatnot.

At work we've used sensors for data logging that communicate using I2C over distances more than 20 meters, using plain Cat5 cable.

Thanks for this! I recently designed a sensor board that connects to our main board with I2C, and in chatting with an EE about it she mentioned I2C is not intended for intra-board use. I just put a scope on the signals yesterday and they seem okay. The cable is only 15cm long, and it connects to a multi-use port which would be difficult to make work with differential signals in addition to the other things that port can do. I’ll keep an eye on it but maybe it’s okay.

If you _must_ use I2C, then look at SMBUS if its an option for the parts. I2C's biggest failing is that there is no protocol level timeout, so one stuck device can block your entire bus unless you have a reset line for all the peripherals on it. https://www.analog.com/en/resources/design-notes/guide-to-co...

Classic I2C problem. After your CPU resets, at least clock out a bunch of cycles onto I2C to get interrupted I2C transactions to finish.

Ooh this is great to know thank you!

I2C doesn't really care about cable length all that much. The thing to keep in mind is the interplay between bus capacitance, pullup strength, and drive strength.

A longer cable means more bus capacitance, which means with the same pullup resistor the signal rise time will be higher, which means you need to reduce the bus speed. A stronger pullup will reduce the rise time (allowing a higher bus speed), but each chip's driver has to be able to overpower the pullup too. If the pullup is too strong for the drivers, you end up being unable to send a zero.

In practice your cables can be quite long, you just have to run it at a lower speed. If you really want to push it, there's always transceivers like the PCA9615 which turn it into a differential bus.

Do you mean the Wii?

Yes, of course I meant the Wii, sorry about the mix up. Too late to edit now.

Not an EE here, but I've dragged some circuits together as a hobby and have only used I2C. Why would Nintendo opt for I2C instead of a differential pair? Is there some extra part cost? What part(s) would you use to go from I2C to differential?

I don't know about I2C specifically, but a related device is a serdes (serial-deseria) which converts between a parallel interface and one or more differential pairs.

https://www.latticesemi.com/what-is-serdes

Someone else mentioned the PCA9615 which looks like it'd to the job.

I don't know why Nintendo did it. But it's certainly quite convenient, there are even standard form factor breakouts for the Nunchuck like in the link below. This gets you a controller with accelerometer, 2 buttons and a 2 axis joystick with plenty of libraries available for using it with Arduino, RPi etc.

https://learn.adafruit.com/adafruit-wii-nunchuck-breakout-ad...

You usually don't find I2C in the high end space-rated parts simply due to the added complexity of a simplex protocol but you see tons of single-ended SPI parts. You don't need diff pairs unless it's for high speed or long cable runs. If your controller (FPGA, microcontroller, etc) has the resources, a good idea is to have a single SPI slave per master. Also since SPI is not standardized, vendors may have different signaling requirements which makes it slightly more difficult to put different chips on the same bus. Talking to SPI chips is super easy and can be implemented entirely in 7400-series logic if you want so it's entirely possible to have analog electronics send control signals to SPI devices without a single CPU in sight.

Worth noting though that the switch you linked claims to be a gigabit switch while the switch from murex is 100Mbpbs.

Not sure if that justified the price difference

Congratulations MUREX Robotics team, great job!

There are products at different price points on the market, for example this 55x55mm switch from my company Brainboxes[1] is sub $50. We choose that size so that we could also produce a gigabit option with the exact same footprint. We opted for microMatch[2] style connectors as you can get board to board as well as board to cable options.

Your co-leads decision to buy-in is quite common, as you can reduce time to market and also not have to manage the component lifecycle if you go with an off the shelf option.

[1] https://www.brainboxes.com/product/pure-embedded/pe-505

[2] https://www.te.com/en/products/brands/micro-match.html?tab=p...

That's a sick board! If we had found that before we made it, maybe would have just used this board haha. What is "buy-in"? Is it meaning us using JLCPCB to buy and assemble the chips?

Thanks! Like yourselves we saw a clear niche for a ultra-reliable small embedded board suitable for robotics and other space constrained systems. I'd be very happy to send you one of our products to compare, i'll message your team email.

By "buy-in" I was referring to the parent comment and how the electronics guy chose to buy-in a pre-made module rather than design their own.

Looks so cool! Just one question: IIUC, linear voltage regulators works by wasting voltage delta until target voltage is attained, instead of switching output current as buck converters do, so at 12V input it dissipates heat of up to (12-3.3)[V] * 0.8[A] = 6.96[W] onto the board depending on downstream current draw(of an FE switch, so I imagine would be tiny fractions of 0.8A realistically). Do board feel cool enough to touch as is?(please use back of hand/finger if unsure)

Hello, yes. Unfortunately it is hot to the touch. The board doesn’t draw 800mA, but it does get to like 60C. Max, the main designer for V2, said a heatsink is recommended. Our thermal via solution keeps it within somewhat safe temps. We wanted to keep the cost low, so we opted for the LDO. In our robot we do direct 3.3V with a buck that does 3.3 for our entire system. That’s Max’s power board.

Don’t be afraid to implement a basic buck converter! Something like the MPQ4572 isn’t too hard to get right (right enough for hobby projects at least).

They made an Ethernet board, I don't think a buck converter would be hard for them, but, as they said, it does increase the cost.

Honestly, hooking up a simple Ethernet switch IC is a cool project but isn’t especially difficult - and this is only 100 Mbps also so not very tight tolerances.

Set up your design rules right and I’ve had gigabit RGMII and 1000Base-X work first try.

Yes, if you're an expert at designing PCBs, I'm sure it's not hard. It's objectively hard, though.

For sure, this wasn't that hard. But I think that's great; it means anybody can be an EE. That's a core mission of our team. Our team works hard on PCIe 4.0, USB 3.1 Gen 1, and Gigabit ethernet as well. This is just one board that we made and were proud of. Feel free to take a look at our other boards in the documentation website!

how hard do you find pcie v4? did you look at how big a step is it to v5?

I would think the majority of people would be running this of of 3.7-4.2V lithium batteries or 5v USB so the step down Delta would be minimal, but yeah probably good advice to incorporate a more efficient buck step down for those running traditional 12v systems which usually operate at 14.4v.

Thanks - yeah 60C(140F) sounds somewhat warm, glad you're already planning for a DC-DC upstream. Feeding it 5V could make sense too(3.3V out + 1.2V dropout = 4.5V < 5V), but anyway this project feels like scratching an itch for many. Congratulations to you and your team!

It may be beneficial to have an additional system power rail. Adding 5V produced by an efficient buck converter would allow for more efficient LDOs on the small boards.

LDOs are much smaller and don't produce noise like switch mode power supplies. It's true you will waste some power but that can be minimized by making the dropout voltage as small as possible.

Nice project and good work.

The only thing is it doesn't address a new or existing market as-is because it competes with what already exists. For example, a TrendNet 10/100 compact switch (not a hub) goes for $7.31 including shipping on eBay and it comes with a case and a power supply. Decommodifying a product requires finding niches where there is demand like automotive, aerospace, military, or marine applications. Until roughly 2020, 2.88 MB 3.5" floppy drives were in-use primarily in industrial and turnkey commercial systems long after they disappeared from desktop computers. Dinosaur technologies can live on for a very long time, often in critical systems deemed too expensive to replace.

Keep pushing forward, learning, and getting better.

Btw, if someone made a:

- 48 port 10GBASE-T (802.3an-2006) POE++ (IEEE 802.3bt-2018) 960W-1600W+ (3422W would be the upper limit for type 4)

- L2 (at least) switch

- unmanaged to fully-managed (but no cloud features)

- 4 100GBASE QSFP28 uplinks (unpopulated)

- dual, hot-swappable PSUs

- 2 models: Ports facing either forward or reverse

- 19" 1U half rack depth, and wall mountable

- Most importantly: doesn't sound like a jet engine under full load by leveraging better engineering, such as using some industrial-rated parts, heatpipes, and moving hotter air but less volume

I'd throw down in the $6K price-point neighborhood.

Comparables:

$4800 FS S5860-48XMG-U is close but sounds like a jet engine with dual 1U screaming PSUs and 3 hot swap chassis fans, but only available in conventional top-of-rack forward facing, leading to longer, messier wire management unnecessarily.

For example, a TrendNet 10/100 compact switch (not a hub) goes for $7.31 including shipping on eBay and it comes with a case and a power supply

A 1G switch from Ali with a case, power supply, and the RJ45 connectors is also under $10.

This isn't a desktop Ethernet switch. Lots of robotic parts like industrial cameras and fancy laser sensors use Ethernet for interfacing instead of USB or RS232C. Doing so solves cable length limitations and connection stability issues of those peripheral buses. There would be penalties and overheads of (mis?)using inter-node communication protocols like Ethernet (and TCP/IP), but those tradeoffs are completely acceptable.

What is not acceptable is full sized switches inside of a robot: they're way too bulky. You may be looking into installing a switch inside a humanoid upper arm or inside pelvis. Regular switches and hubs don't fit there.

This product solves that specific robotics packaging problem. Electrically it's a switch/hub, physically it's much smaller than that, that's the point.

I wonder if you could do away with the switch and just daisy chain in many of those applications. It seems to be forgotten that ethernet supports this.

It's not exactly forgotten. There's 2 wire Ethernet/PoE standard (used mostly in automotive) that can be daisy-chained. Designing for it however is non-trivial and few third party devices (again outside automotive market) support it.

Hmm... I thought everything still had collision detection and would work on a bus, like what you'd get using a hub.

There is normally no collision detection beyond 10Mbps. Also PoE would not trivially work daisy chained.

100base-Tx is specified for shared media (hub rather than switch), but I think you need some electronics to make a hub, you can't just do a passive connection, and so you probably end up with a hub IC and may as well use a switch IC instead.

It might work in so far as you'll get packets through. The performance would be abysmal, though. Modern Ethernet is built around switching.

Would it really be that bad? If the devices were well behaved (ie. not too noisy, no gratuitous ARP etc) and the application could assume that most of the time either zero or no devices will be communicating, would it be that bad?

This is something I might test myself. I have a couple of audio devices that will never be both "active" at the same time. In my current layout I need to run either two cables or just another switch and it just seems a waste. I wonder if I can buy daisy chained cable so I don't have to make one...

Our company is often asked at expos if we supply 2 wire ethernet switches (e.g. Single Pair Ethernet SPE[1]) but the reality is there's little demand (outside of automotive) we see in the Industrial marketplace for this type of product.

The other issue is that there is at least 2 competing industrial connector standards for SPE, the main ones being from Harting[2] and Phoenix Contact[3]. I think this could be a great option for the future and ultimately lead to lower cost cabling and smaller products.

[1] https://www.single-pair-ethernet.com/en

[2] https://images.app.goo.gl/ND9d9x66YNUckS7g6

[3] https://images.app.goo.gl/yWAiUAx4Y5vRmmbZA

Yeah that's the thing: neither of them is really standard, and both are fairly expensive. They are also available in very limited mechanical configurations (angle, mounting). Fortunately SPE does not really require either of them. In our product we're going with our own solution.

> Our company is often asked at expos if we supply 2 wire ethernet switches (e.g. Single Pair Ethernet SPE[1]) but the reality is there's little demand

This reminds me of a funny conversation I had in a small town stationers when looking for an item.

‘No, we don’t stock it. People keep asking, and I keep having to explain that there is no demand.’

Though this wasn’t for an obscure Ethernet variant.

ha, true! However at expos often the person asking is a vendor looking to sell SPE components to us, or someone in the industry trying to determine if a technology is popular. Can be a bit of an echo chamber.

In industrial IO, including robotics, the norm these days is 3 port switches embedded into devices, specifically to support daisy chaining.

Hubs are no-no because of various performance issues, including how more and more gear supports gigabit on the daisy chaining.

I suspect the combination of being able to dissipate 3.4kW of power, and the requested size and noise constraints may put this combination of features firmly beyond reasonable, even before considering cost.

But there is always room to hope. :)

A PoE-supplying switch doesn't need to dissipate that power. Assuming a power supply efficiency of 85%, supplying 3400W to downstream devices means having to dissipate 600W of heat.

Keep in mind that you'd be supplying 71W to every single downstream port. That's an insane amount of power. Something like a Cisco Catalyst 9136I Access Point only comes up to 47W, and that's assuming 16 radios, double 5Gbit uplinks, and a USB device drawing 9W.

A Juniper EX4100 ticks all your boxes.

Adding a sound requirement is stupid and ridiculous. You're wanting to take the power supplies from 48 90W devices and put them all in one dense little box and then complaining that it needs to be cool.

If you have room for 48 PoE devices, you have room for a properly cooled and sound isoloated IDF.

Well done.

For those of us not generally in the hardware world (and thus not 100% familiar with the terminology) could you post more pics?

Especially of the enclosure? I'm not really sure if you are just exposing headers, or if there are regular Ethernet plugs on the board?

Thank you so much! We're so happy you like it. It means a lot to us. We will be adding more pictures right now!

We're exposing 1.25mm pitch Molex Picoblade connectors to make the board as small as possible. The built-in magnetics allow it to be connected to any Ethernet device just from spicing into an RJ45 connector.

The comparable BotBlox SwitchBlox Nano is 25.50 x 25.50 mm, albeit with two fewer ports. This is 44.90 x 42.11 mm. How do you justify the claim of being the world's smallest?

Hi, I'm Max, the lead designer of this V2 revision. You're absolutely correct! My team and I were not aware of BotBlox's 3-port switch, so a more specific description would be the world's smallest 5-port unmanaged switch. The smallest commercial alternative we found was also BotBlox's 5-port unmanaged switch [1], which we beat in both size and cost.

Once again, our mission is to create open-source, cost-effective, and accessible electronics for as many people as possible -- I think our board is much more attractive in that respect and a win in my book :)

[1] https://botblox.io/products/small-ethernet-switch

Since you have thrown ISO/IEC 8877 out the window, which is referenced by 100BASE-TX, you surely can't call it an "ethernet switch."

> It is only 6.9 dollars

> probably for $10+shipping

You could honestly sell it for $30-40 and it would still be a pretty good deal. Meanwhile Blue Robotics be like "that'll be $175 plus $50 shipping and customs fees as a percentage of that $175 fam"

https://bluerobotics.com/store/comm-control-power/tether-int...

God, everything they sell is overpriced to the point of insanity. They could really use some proper competition.

Low volume hardware pricing always look infuriatingly high, but they also start looking hopelessly low once you've dipped toes into it and tried multiplying your spent development man-hours with McDonalds wages or tolerable unit price by expected sales volume.

I've never heard of Blue Robotics, but I doubt that $175 product ships 1k units/year, and even if they did, that's $170k revenue, or 2x entry level engineers salary worth of raw recovered cash before factoring in any expenditure whatsoever, let alone taxes and HR. It probably hardly feeds one, and that's based on an optimistic hypotheticals that they ship a thousand of that product every year.

Large scale multinational corporations ship in orders of million units. That makes it way easier to amortize non-recurring engineering and ship small products virtually at cost.

I mean sure, I understand that. But that doesn't mean I have to like it. BR are really well known to anyone dealing with ROVs or AUVs, but that's not exactly the average person.

Commercial consumer hardware R&D is mostly a fool's errand these days, since if it's something that's worth producing and sells there will be clones that work just as well available almost immediately. I'm not sure how say, Adafruit, sells anything at all to be honest, anything they make gets perfectly cloned and sold on Aliexpress for a tenth of the price. I guess going extremely niche and overpriced is one way to work around it.

I'm super impressed and have a lot of respect for group of teenagers.

I have a question: do you have any idea what your parents did right to get you into electronics and hacking like this? I'm looking into getting my 3 kids into CS and EE, and it's something i do for a living, but i have no clue on how to make this for them. My parents never did anything except buying me all the books i asked for; I was just obsessed with knowing more cool stuff.

Thank you for sharing such a cool project with the world, your future will be very bright, keep it up!

Nice work indeed. However, was there a reason you didn't support gigabit ethernet? I haven't used 100mbit ethernet for more than a decade...

Hi! I'm Altan, another member of Murex. Many of the design decisions behind the switch were driven by the requirements of our underwater robot. In our case, the communication speed was capped by the transfer speed achieved over our tether (we use galvanically isolated OFDM to inject data over our powerlines). Since size and cost were our primary goals, 100mbit was more suitable than gigabit ethernet. While it would have been cooler to have a gigabit switch, it would also increase the size and cost.

GigE needs twice the pin count. IMHO, there's not much room on the board for any more i/o. Certainly gigabit is nice, but there's plenty of applications where 100M is more than plenty.

I’m not sure what your background is, but 100Mb Ethernet is still rather common in embedded devices and applications where the network protocol is primarily intended to facilitate UART serial communication. Just as a general note for context, I will defer to their more specific answer for this particular application.

Neat project!

Agreed. Due to the long lifecycle of manufacturing equipment we still see a lot of 100mb out there, and it’s not even embedded.

I would note that all new products seem to be gbe or better.

If you're doing tethered ROV stuff, the weight of the teather is a big big deal, so adding 4 additional wires is a non starter. For the stuff that goes extremely deep, they use fiber because it's much lighter. It presents significant cost increases of course, which, you'd want to avoid if you can.

You can get neutrally buoyant cables:

https://bluerobotics.com/store/cables-connectors/cables/fath...

It's still mass, drag and cost.

As much as modern Ethernet standards are much nicer (my house is wired for and running 10Gb everywhere, with 40Gb Infiniband to a couple locations too), 100Mbps still has its place. Specifically, anything embedded, slow, and/or cheap. No reason to spend the extra money on 4 more wires and pins and trace routing if your microcontroller only sends a few packets/second.

At that price point, the cost of the RJ-45 port is probably more than the cost of the 802.3 chip and I wonder if the cost of supporting that old chip on a contemporary device doesn't surpass the cost of the components for a nowadays standard 1 GB/s.

Why classic full-fat fast-ethernet (base-tx) instead of single-pair ethernet (base-t1 or t1l/t1s) if you are targeting embedded use?

Thanks for asking! That is a very good question. However, we use bog-standard Ethernet connections, so using Fast Ethernet is super straight forward. Just splice the cable and we're all good.

This is amazing, well done, no notes, would love to buy when available

I think it's been rightly pointed out that you aren't beating commodity parts on price, but you're also not a manufacturing operation with scale and there is a certain niche for which anything with open hardware that's well-documented is a killer feature

Please email us at byran@mrx.ee for now!

One side project I never seem to start is a single board cluster based on Octavo SoMs. The idea is to have 32 cores per board to mimic a Thinking Machines CM-1-like cube. How easy is it to use a PCB to route Ethernet between nodes? What kind of components would go between the SoM and the switch?

That would be super cool! I'm not sure how it'd work together. I would think we need a higher speed fabric-type connection.

> That would be super cool! I'm not sure how it'd work together. I would think we need a higher speed fabric-type connection.

It wouldn't be a thorough emulation of a CM, just something that looks like one. More like a tool to play with clusters and MPI. Not sure there is an Octavo part (and the bundling of all passives in the module is important to simplify the design) with enough RAM to run Kubernetes (although Docker Swarm would probably run fine, but I've never thought about one with 32 nodes).

That said, the Octavo parts have 2 GbE ethernet ports, so we could have two independent networks on the board (which is nice, one for in-cluster management traffic, and other for the outside).

How does one learn about ethernet signalling and the cable interfaces (would we need those if using PCB traces?). What could be the passive components between a switch chip and a SoC on the same PCB? This is where my electronic design skills end.

Very cool. It says fully open source, are the board layout files available?

Yes! It is here: https://github.com/murexrobotics/electrical-2024 under networking/switch Edit: We use KiCAD Nightly 8.99 for the new features. You can view our boards easiest with kicanvas.org and just pasting in the https://github.com/murexrobotics/electrical-2024/tree/main/n...

It is great to see that this is accessible and possible with limited effort/budget!

On the higher speeds, it remains difficult: In the datacenter, 10G ethernet is often standard or even outdated. But for non-mains powered systems, even 10G uplink is hard to come by. I would love to have a switch with 25/100G uplinks in a smaller-than-19"-rack form factor with 12-40V DC power. Building one as a side project might still be too complex - if you would get access to chipsets at all.

We only found ~gigabit and ~fast ethernet chipsets that are easily accessible. However, it's definitely something we will think about in the future! Thanks for the suggestion!

How about the Mikrotik CRS510-8XS-2XQ-IN? 2x 100G uplink, 8x25G/10G/1G downlink, DC power input, and only 320mm (12.5in) wide.

Nice job, I'm always impressed with HS robotics projects. Clever design and nice form factor.

I bought a stack of these:

https://www.amazon.com/Tenda-Gigabit-Unmanaged-Wall-Mount-Pr...

for some projects that need ethernet. They are mid-level quality. Not sure I'd put them on something that would break my heart to get back.

I agree. They're also so much larger than our design! If getting a switch ever interests you, please let us know at byran@mrx.ee. No pressure of course!

Your group is incredibly cool, keep it up. It's exciting that kids in high school have access to this type of activity. I love living in the future!

Thank you! Attempt the impossible :)

Seriously, my high school was mostly boring and pointless. I’m in awe of the opportunities some kids are able to find/create today.

Why do they say "a high school robotics team from Exeter, New Hampshire", instead of "a robotics team from Phillips Exeter Academy"?

FYI: Phillips Exeter Academy is an elite, private high school in the United States. Mark Zuckerberg went there, too.

This is truely awesome! Hat's off. If you're experience in college is anything like mine you will find that 90% of the incoming freshman class is looking at "electronics" for the first time (assuming you're going EE, but similar for CE or CS), you will be way ahead right from the start.

That this switch is small and light might make for some interesting UAV applications as well!

Thank you so much! Super excited for college. I think it should fit right into an UAV. I see a lot of use from Ardupilot forums (which we use). It’s the reason we have the LDO as well. We wanted to keep the application possibility open.

Congratulations! Impressive work for any age.

Is creating bespoke parts a requirement for your robotics competition or just a part of your team's ethos?

Our mission is to create open source electronics that democratize technology for as much people as possible! This is a full list of stuff we've done this year: https://www.murexrobotics.com/mrxEE. It isn't a requirement for the MATE ROV competition (I wish it was though; it would be super cool!).

Hey can I just say I appreciate y'all's mission and energy in the comments? Very inspiring, very welcoming—you're the best!

Thank you so much! This energy has waked us all up!! We’re all living together right now preparing for the world championships next week in TN! This is soooo exciting.

I wish we have this 2 years ago, this will be extremely useful for our cubesat. 115k2 baud UART is too slow for ground development

Hello! I'm Osbert, another elec member on MUREX. For sure, we use 115k2 UART on our "custom" ESC serial communication protocol, MASCP (murex async serial communication protocol). We stream 2 IP cameras, our CM4 board (open source as well!), and our mrxPLC (DC powerline data injection technology). Thanks for the encouragement. It means a lot to our team!

Nice job guys. For your next challenge, see if you can find a cost-effective way to add VLAN management, or do something cute with the form factor, like make it fit inside a standard electrical socket wall cavity with AC adaptor, or inside another ethernet port.

PS. Try to reference vendor application notes or datasheets instead of stackexchange where possible.

Thank you! This goes in our underwater robot, and it has all the features that let the system work without anything else. However, we will definitely keep your suggestions in mind for our future revisions. We're seeing how to make this board even smaller, while keeping the cost down. 7 dollars (technically 6.9) is the mark to beat.

> full Bob-Smith style termination for all center taps

Man, that's some serious inside baseball right there.

Thanks! Do it right, do it MUREX!

hmm ... not the cheapest. $7 is just for the PCB (assembly) right? you need an enclosure and power supply as well.

monoprice #41710 is $10 all in with a price break starting at qty 2.

maybe you want to qualify your description with 'embeddable'.

> built-in magnetics

interesting b/c the website says `external magnetics`

Given this was made by a high school robotics team and it accepts up to 12V, it's likely that the intended use case is on a robot that already has a 12V battery

Hello! I just want to clarify some of these points. Saying it is the world's smallest switch is a bold claim, and there are definitely alternative options around. However, our board scales really well. The main reason we say it's the cheapest is because the real BOM is only around 4-5 dollars. Our small run of ~25 boards came out to ~$7 each. If you agree with my interpretation, perhaps the words "smallest and cheapest" can be together?

Magnetics: We have built-in "external" magnetics, so the physical ports don't have the magnetic inductors required in the Ethernet standard. This makes the ports super small, allowing us to use 1.25mm pitch Molex Picoblade connectors. Again, thanks for the suggestions. We're still learning and the team is taking all the comments very seriously.

The "real" BOM is what you can actually get. If you don't have the funding to make 10,000 boards in order to push to cost down to $4, then the BOM is $7.

oh wait, so the BOM is $7? that's not the complete cost, you have to assemble it as well.

Fast Ethernet +/- pins go through a special 1-turn transformer for electrical isolation installed between the PHY and RJ45 connectors, commonly referred to as "the magnetics" in Ethernet world. It's physical ferrite and copper object molded like a chip, therefore it can't be trivially integrated into a silicon or in MCM. Hence most Ethernet chip require one external to the chip, and in this case it is indeed external to the chip, and integrated into the board at the same time.

(yes a cap in-line works too if you know what you're doing)

Congrats.

A few months ago I wanted to make a small lab out of a few SBCs. Looking for a cheap 10/100 switch, I was surprised to see prices are this low; got a tplink 8 port / LS1008 for 10$ from Amazon.

Thank you! We also have those switches. They are fantastic cheap tech as well.

“We are super proud to have made this open source piece of technology” - of course it’s open source - you live in the state of “Live free or die!” My favourite state motto. Thanks!

Not a hardware guy,how does one crimp and secure those custom ends?

You can buy them right of Digikey by searching "4 position Molex Picoblade cable assemblies" or similar!

If you are interested in such things, check botblox. They have a similar sized switch, a smaller one with fewer ports, a stacked one for Gigabit... it's very expensive though.

Ah you're right –– the team and I were not aware of BotBlox's 3-port switch! I guess a more specific description would be the world's smallest 5-port unmanaged switch. We do however beat BotBlox's similar 5-port unmanaged switch [1] slightly in size and miles in cost. I'm Max by the way, the lead designer of this revision.

[1] https://botblox.io/products/small-ethernet-switch

Are you telling me one does not need an EE degree to make hardware like this? How do high school folks have access to resources / mentorship to make such a polished product?

Absolutely not!! This is what MUREX is all about. We are on a mission to show anybody can do this. Change the world one small bit at a time. This was all self learned from the internet, trial and error, lots of passion, and this community. Our mentor doesn’t have a background in engineering so he just gives us lots of emotional support and some physics help. Love you guys!! This is amazing.

I hope your mentor see this thread, they are doing a great job if your working anything to go by.

Most major chipsets have their design specs in PDF format online with all the info there for interfacing on a hardware level. Many drivers these days are open source. Kicad is free to use for designing PCBs.

And finally https://www.youtube.com/@PhilsLab has some good tutorials.

Finally, goo into things with an attention to detail to polish a board. A hobby board takes far less time than a product board, but all it is is time and thinking about each attribute more. You don't need third level education to have that work ethic.

Best of luck!

This team is at the second best high school in the USA. they have the resources and the money.

Hello! We got 5k for the entire robot, and our school does not provide much engineering support. MUREX is entirely student run! Of course, we are incredibly thankful to be able to attend such a great school. we are first this year :)

This extremely cool, when I was in high school a few years ago we made some power adapter boards, but we couldn't have hoped to build a functional network switch, great job!

Hi, I'm Crane and I am also part of MUREX robotics and this is very inspiring to me as an electrical engineer who is starting out.

> We will be putting a small run of these boards for sale somewhere

Somewhere like http://tindie.com?

Maybe! If you’d like one, please email me at byran@mrx.ee!

I was PEA ‘10 :) This is amazing. You’ve come a long way from the robotics class I took senior spring…

j/c but why are we still maxing out at 1gb ethernet connections? why have the speeds essentially not progressed in 20 years? you can get a 40gb connection by just using usb-c on modern machines. what's going on? (curious about why this is the case industry-wide, i think this project is really cool)

The intended use-case for this board is some robotics applications. I don't think parts of robots need to communicate via gigabit. 10/100 Base-T is presumably easier to implement and thus cheaper and smaller (which is the point of this project)

You can get 40+ Gb/s over USB-C or connections like DisplayPort and HDMI by using thick expensive cables that top out around 10ft of length. Beyond that, price starts shooting up as you get either optical transceivers or active retimers/redrivers built into the cable assembly.

Ethernet over copper is designed for cable runs of over 300ft and has to be much more forgiving of poor quality cables and connectors. That means for the same level of complexity and power consumption in the transceivers, you're just not going to be able to get as much bandwidth.

Ethernet equipment suitable for 2.5Gb/s and 5Gb/s in consumer equipment (cheap, low power) is now readily available, but there's not enough demand to drive pricing down to parity with 1GbE and completely supplant it. 1GbE is good enough for most consumer use cases, especially given the dearth of multi-Gig WAN connections in the consumer market, the lack of popular use cases that would benefit from slightly faster LAN connections, and the continuing improvements to WiFi.

2.5gbit is quite common and cheap.

Combination of power and distance.

Copper Ethernet standards tend to be specified for over 30 metres, and go (preferably) up to 100m. That’s pretty tricky to do, you need quite thick cable with individually foil shielded pairs to achieve those kind of long distances at 10Gbps. 40 Gbps USB-C on the other hand is recommended to travel over a maximum of a metre or so of cable, with the recommended being 0.8m (2.6ft) of cable or less! Thunderbolt cables that go longer need active driver chips inside the cable in each connector to make the whole thing work.

Then there is the power issue, 10Gbps Ethernet uses significantly more than 1Gbps, so a 40GBase-T switch would be even more power hungry.

The combination of these has meant that basically most people just use fibre if they need more bandwidth.

Some details I would like to see:

- Does the backplane support full 1.25Gbps throughput?

- Does is switch at line rate for all packet size?

- The switch chip supports LACP, port mirroring, vlan tagging, QOS, these would be amazing to integrate into a product even for slightly higher cost in a 5 port switch

- It's generally a hard requirement for me the 801.3az be off by default or can be turned off. I've had far too many issue with it enabled on network

Hello! It supports 100Mb throughput, nothing more, nothing less. It's been tested with iperf. I believe it switches at line rate, but don't quote me on that haha. The chip's extra features can be enabled with an EEPROM, which we removed to cut costs and size further. It's on V1 though.

Sorry if I missed it on the page but are all the ports Auto-MDIX?

We didn't try, but I think so. We always ensured the crossovers were correct. Sorry I can't give a better answer!

I would love to have one of these, if still is available!

Please send us an email! We will try our best to get you geared up. byran@mrx.ee

Can you produce a 10Gb version of the same?

Cool.

Is this designed for personal or industrial use?

Hello, I'm Crane and I'm also on the MUREX Robotics team. The price point we are aiming towards is $7-$10. Our network switch is currently used for personal purposes, in our underwater robot, and we look forward to expanding it for industrial purposes.

Given the caseless PCB and non-RJ45 connectors, it looks like it's designed to be used on robots, and I can certainly see it being used that way in my robotics team

This is awesome, you all rule

Thank you it truly means a lot to the team!!

Can you make a 2.5g version

Unfortunately, it's not really useful for our use-case. Our robot only uses ~kB amount bandwidth.

Excellent work, stay bold.

Thank you! It means a lot to our team!!

Typo: “commerically”

Thank you! Noted.

Great work!

Thank you!

impossible is now possible

unleashing innovation :o

Now do POE

That’s one area we are exploring!

It looks great! Have you considered running it through https://www.quilter.ai/ before sending it out? It's free.

You can hear more about it in this podcast: https://wandb.ai/site/resources/podcast/episodes/ai-in-elect...

I actually met with the Sergiy as well a few months back! Super cool guy. I also highly recommend quilter - thought it was super cool (not paid haha).

Quilter doesn't have the best support for impedance matched traces. We do 90Ω on a 4 layer board and Bob Smith termination which has some pretty goofy design requirements.

Reply on news.ycombinator.com