Make A Compatible Raspberry Pi Clone – But Your Pi Must Die

The world is awash with Raspberry Pi clones that boast fruity names, but those looking for a piece of the real thing will find their compatibility only goes so far. Shaky Linux distros abound and, with a few honourable exceptions, they are not for the faint-hearted. The reason that a market hasn’t emerged for fully-compatible clones is that the Pi people seem to have a monopoly on the world’s supply of the particular Broadcom SoCs that they use, forcing would-be competitors to source the brains of their outfit elsewhere.

It’s easy to buy a Raspberry Pi SoC though, if you don’t mind receiving a Raspberry Pi along with it. So to make a compatible Pi clone for space-constrained applications, the folks at Arducam removed the SoC from a Pi 3 and designed a surface-mount module board for it, making a 40 mm x 25 mm postage-stamp style system-on-module. It’s not a Raspberry Pi, but it runs Raspbian.

Their board is not one that they will be selling, but it does open up interesting possibilities for others with an eye to creating Pi boards in different form factors. It would be fascinating for example were somebody to produce an open-source module board for a Pi SoC. Some of you might be asking why the existing Compute Module was not suitable for them; in the write-up they cite mechanical issues with the SODIMM socket.

This isn’t the first compatible Pi clone we’ve seen. Aside from the intriguing but short-lived Odroid W there was another even smaller Arducam offering that never made it to market.

34 thoughts on “Make A Compatible Raspberry Pi Clone – But Your Pi Must Die

  1. This obviously isn’t a DIY thing; or an answer to the ‘Broadcom doesn’t appear to sell those; at least not to mere mortals’ problem; but element14 does custom spins.

    I’m not 100% clear on why BCM doesn’t sell the SoCs(though given their usual practice it would probably be with minimum order quantities and NDAs); it’s not like the rPi business model has fat margins built in for them; but for any nontrivial quantity I’d probably be rather more comfortable not BGA-scavenging SoCs for my boards, if only for reliability reasons, never mind labor.

    1. Problem with element14 is that they tend to decline small batches. I’ve tried to order 10 units with small BOM modifications and they refused me. So this may be the way to get truly custom board.

    2. “though given their usual practice it would probably be with minimum order quantities and NDAs”

      Pretty standard with almost any ARM SoC on the market, sadly. :(

      1. I wonder why someone hasn’t crowdfunded a campaign and bought enough to fill the minimum order quantity, then send them out proportionally to the one’s paying for the campaign.

        Or maybe the pinout of the chip is under NDA as well?

        If it isn’t, then it would make for a pretty nice way of obtaining a bunch of chips and make a more custom solution able to run a slew of easily integrated software with ease.

          1. I wouldn’t be surprised if that is the case.

            Though, it depends on how one looks at a bunch of people/organizations making a bulk perches together. Is it really reselling? Or just a collective perches as to fulfill the minimum order as a group.

            But I guess such a group perches would end up with a lot of NDA signing and likely have Qualcomm say no due to that alone…

          2. Yup. The NDA is a big part of the issue here.

            Also sometimes the problem is not just the SoC but the companion devices. As an example, the PMIC in the Pi 3 has two “off the shelf” variants, but can be customized using some OTP memory – the units on Pis have been customized, so you can’t replace the PMIC with an off-the-shelf component.

            Qualcomm is nasty about NDAs and they’re sadly one of the best in the mobile industry.

            Mediatek is notorious for forcing Chinese phone manufacturers to violate the GPL. This is why you almost never see MTK-based phones in countries with decent intellecual property laws.

    3. I used to work with Broadcom on set-top boxes, if you aren’t spending millions then you won’t get their attention. And a custom silicon spin required very large volume commitments. Even if you are a customer they have 16 week lead times.
      Interestingly the Roku was a direct cousin of the Pi, they came from the same nursery.

    4. Eben Upton worked for Broadcom for a very longtime and setup this deal with them in the early days of the RPi project. I believe it also gets subsidies from the UK government.

  2. That’s fairly cool n’ all – the desoldering, re-balling and reuse – in a just-because-we-can way at least.
    But not really sure why anyone would do this for a real-world use-case, beyond proof-of-concept. If the “pi foundation” don’t want us to use their precious SoCs, please don’t. Scrapping 90% of a brand-new toy just to get an IC is just… not sustainable.

    Maybe could consider using some board as-is, but other than maybe the Compute modules, I personally probably would not go there either. Not for the kind of stuff we normally do anyway, where any kind of reliability is required.

    For the kind of custom embedded use-cases described here, space- and mech constraints, I wonder why they didn’t go with with easier and IMHO probably better suited solutions, like e.g. the Octavio OSD335x Granted, not exactly small at 27×27 mm, but the total footprint would probably match or beat their 25×40 area. And manufacturable, reliable, off-the-shelf.
    (Also wouldn’t have to sponsor the pi-people to just create even more e-waste, they – and many of us – are doing that quite enough already)

    1. For ‘real world use case’, I can’t imagine there are many people who require the identical cpu/soc compatibility, plus their usage would have similar physical problems with dimm mounting, plus are small enough to not meet the minimum order amount for a custom board.

      I’m not questioning or belittling their needs in this case or anything, I just think it would be rare enough that most everyone else would be able to go with one of the other options easier.

      I’m a bit curious myself about the mechanical issues with the dimm mounting.
      With the removable requirement of a dimm slot removed (har), my first thoughts would involve some heavy duty epoxy and perhaps pcb space for mounting metal reinforcement brackets or something.

  3. Broadcom does sell the SoC. We’ve tried to get them for our product, as it would have been a nice upgrade path from prototyping on a Pi.

    Their MOQ (minimum order quantity) starts off at half a million, so unless you have volumes behind you, it’s pretty much a no-go :)

  4. This is really cool! It would be great if Broadcom were to sell the RPI4 SOC on Digikey.

    Still waiting for that click baity titled article on how the Raspberry Pi foundation botched the USB-C spec on the RPI 4. Even though it really is a minor mistake if not a total nothing burger.

    Please don’t disappoint me hackaday. A title like “Raspberry Pi foundation gets a C on USB-C” or “Raspberry Pi foundation did not see the USB-C”. It would really make my day.

  5. Doesn’t that device still use PoP for RAM? Wouldn’t you have to decouple and reball/remount that as well? That’s a lot of work to make a much more expensive Pi.

  6. Uh huh.

    Where this described as just a hack for the sake of a hack I wouldn’t question it. Great hack!

    But just to overcome mechanical issues with a SODIMM socket?

    My first thought was “could you just solder to the edge connector of a compute module? That would be hard but then so is this.

    Second thought was to just make a bracket that holds the compute module down pressed into the socket. Would that fix it? Was it an environmental/contamination issue? How about running a bead of silicone along the connection after plugging it in? How about doing both things?

  7. Is the compute module not small enough?
    If they are trying to get it small enough to stack boards behind a CMOS sensor I sort of understand, but by the time you put such a thing into a case you would/could have a lot more space to work with.
    Teardown any CCTV camera and you’ll find oodles of space for a couple of several reasons.
    Only really convert stuff actually needs to be that small and it comes with compromise always.
    OK, so I’m talking CCTV – what is their application that requires such smallness ?

    1. Have a look at the Odroid N2 (Amlogic S922X). If I’d absolutely have to look for an alternative It would be my first choice. The Jetson Nano would be my second.

      Having said that my personal experience after buying 50 or so Raspberry Pi alternatives over the years is to always try to stick to the RPi whenever possible.

  8. The foundation should take note that this would be a great product for space constrained and vibrating environments. All the people here plugging all the other sbc’s have no clue what the cost of software maintenance is for real companies. If you’re not shipping thousands of units at ridiculous margins you just can’t afford all the grief of a board without software support.

  9. I don’t quite get the vibration argument. The Compute Model got mounting holes to screw it in place if the retention of the SO-DIMM is not strong enough for you.
    Put some strong loctite on the screws and that thing is never coming off again.

    1. There are additional retainers for less demanding environment.
      https://www.embedded-computing.com/embedded-computing-design/ruggedization-advancements-enhance-memory-module-design

      If you secure it to the main board by screws/standoff, it is not going to move relative to the PCB nor the socket. The cantilever contacts have very low mass so they aren’t very likely to break contact.

      Back in the old days, you could get modules converted by having leads soldered on the contacts. This is done when money is no object. If you really need high G on a DIY scale, you could use flex PCB soldered to the contacts to bridge to your main PCB.

  10. Has Arducam considered asking for some custom raspberry pis? Like if they can crank out some of these smaller modules, at the price of a regular PI, that saves the world on environmental costs…

    This post seems like a good advertisement for other similar chips.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.