Microchip Introduces Tiny Cheap Linux Modules

Linux is in everything these days, and that means designers and engineers are crying out for a simple, easy-to-use module that simplifies the design of building a product to do something with Linux. The best example of this product category would probably be the Raspberry Pi Compute Module, followed by the C.H.I.P. Pro and its GR8 module. There are dozens of boards with Allwinner and Mali chips stuffed inside that can be used to build a Linux product, and the ‘BeagleBone on a Chip’ is a fantastic product if you need Linux and want to poke pins really, really fast.

Now Microchip is rolling out with their answer to the Linux System on Module. The SAMA5D2 is a single chip in a BGA package with a small footprint that runs Linux. It’s capable, it’s cheap, and if you want to put Linux in a project, this is your newest option.

The core product in this new Microchip lineup is the SAMA5D2 SIP, a system in package that puts an ARM Cortex-A5 CPU and DDR2 memory in a single BGA package that, with a cursory examination, looks easy enough to design a PCB around and reflow. There are four chips in this lineup, with 128 Mbit, 512 Mbit, and 1 Gbit of DDR2 memory. The 128 Mbit chip is meant for bare metal and RTOS applications, with the higher memory chips capable of running Linux at least as well as a repurposed router.

This chip is at the core of Microchip’s ATSAMA5D2 SOM, a system on module that adds power management (that only requires a single 3.3V supply), an Ethernet PHY, and boot memory into a single package that’s effectively as hand-solderable as a QFN package. It’s Linux on a Chip, or at least as close as we’ve gotten to such a concept.

Adding Linux to a project is hard, and while there are modules and systems that can do it, we’re always welcoming more options given to designers. While these modules and systems aren’t exactly cheap compared to a beefy ARM microcontroller — the SIP starts at around $9, the SOM is available for $39 in 100-unit quantities — this price is quite low compared to other Linux-on-Modules available.

82 thoughts on “Microchip Introduces Tiny Cheap Linux Modules

    1. It is often really hard to explain to management why yocto is not yet a full featured ecosystem like debian/Arch. Often, I feel like I am the only one who experienced yocto build recipes as a heinous mix of legacy patch levels, unstable “head” repo builds, and or kludged dependencies (like the early tar balls in slackware). Sure it supports multiple thin-hardware platforms at once… which is very ambitious, but I just can’t see the multiple concurrent vendor-specific-patched project code-trees as sustainable enough for a reliable production environment yet.

      I like microchip’s 8bit/peripheral stuff, but 1/5th the resources of a Broadcom SoC at 3 times the price is a tough sell in these modern times.
      meh… personally I will pass… and stick with Alpine, Debian (“my precious”), or Arch for now. =)

        1. now that atmel is microchip, microchip wins as microchip is both atmel and microchip
          but in reality they simply had a very wide and diverse range of micros
          personally i liked atmel parts for building things up as i have all the peripherals i need and i can grab a cheap generic part that does everything and does it well
          ii would grab a pic for final production runs where you can shave cents by not including peripherals and memory thats not used

          1. but less capable, a more difficult layout and less traceable
            in reality when picking out a SoM you dont pick up a single SoM, you have to look at all your options and choose whats best for your system
            if you need high speed SPI, LVDS, PCI-E, SATA, GbE, LCD, FPGA cores, exc you pick a module that has that perifierial at the lowest cost

    1. That certainly doesn’t put it in the cheap category mentioned in the article. It might be decent value for money, but it’s not really cheap when there’s something like a Pi Zero to be had for significantly less.

  1. “It’s Linux on a Chip, or at least as close as we’ve gotten to such a concept.”

    That’s relative – we’ve been closer, for example, check out Aria modules (in general, SoMs in this form factor have been around for at least a decade now).

      1. Yes, the Octavo SOMs definitely look like more bang for the buck. Also, easier to do a 4-layer board with since they have a 0.1mm ball pitch. I haven’t designed anything with them yet, but getting closer.. They will become more interesting as they flesh out the no-connect pins on the 400-pin version.

      1. I have heard Orange Pi is subsidized by ??? and sells at a loss, making designing into a product not so attractive. I can’t recall if I got that from a reliable source. Anyone know?

  2. I’ve been playing with an Onion Omega2 which is “Linux” on a wifi-enabled board for under $10 each. I believe there are several other small boards architected like this. These are what I would consider “tiny, cheap Linux modules”, and this new Microchip offering, is a not-so-cheap, not-so-tiny std-Linux-capable PC-on-a-chip. Yes?

    1. The specs on this thing are much better than on e.g. the MT7688 that the Omega2 is based on, or many of the Allwinner or Broadcom SoCs that RPis and clones are based on. One thing that stuck out on the specsheet is the 12-bit ADC — of the SoCs I am familiar with, only the Allwinner A13/R8 sports an ADC and even then it’s only 6-bit, so this Microchip-one has a leg up on that.

        1. Honestly the all-in-one. You need to remember that developing a board comes with an R&D cost, and a module that has everything baked in means someone else has footed part of the bill for your product development.

          1. Yet I still need to design my own PCB to use this part. So replacing this SoM/SoC with something much cheaper and adding two small parts is still better. It might be a better solution, if I’m limited in available board area and/or can’t duplicate other functionalities of this particular part by adding more cheaper parts…

            There is also a matter of the fact this is a new part and probably will end up with rather thick errata soon. And Microchip has tendency to leave their parts, especially 16- and 32-bit ones with thick erratas full of workarounds and suggestions that they might fix the problem somewhere in the distant future, maybe even before part becomes obsolete…

      1. Well, the AM335x has 8 pins of 12-bit general-purpose ADC. Don’t know if these pins are actually exposed on the Octavo OSD335x (BeagleBone-on-a-chip), but I think they’re available on the actual BBB, so I’d assume they are on the SIP too. Aref is limited to 1.8V though.

        It’s really a quite competent ADC machinery too, with a programmable FSM that can make custom scan schemes, has HW averaging of 2/4/8/16 samples, FIFOs, DMA etc. (But, consequently, also a bit of a pain to configure and get it going.)

        Yea, it also has a mode for sampling and monitoring (resistive?) touch-panels, but I’ve only used the general-purpose ADC mode myself.

      2. I thought all the SOCs from Samsung, Allwinner, Amlogics, Rockchip, etc. have 12 bit ADCs. They are phone and set-top box chips. If you don’t see the ADC it is because PCB designers did not bring it out to a pin? But looking I see multiple channels of 12bit ADC only on the Samsung 210 [same ARM core as TI 33XX], 4418, and 6818 (and all the older ones that are EOL like 2440/2451[in most Chinese O’Scopes] and 6410[same core as Raspberry Pi]).

      3. Still, Raspberry Pi + Arduino, or whatever combo with the bare chips, and you’ve got that, plus the Atmel as a slave to do the bit-twiddling stuff it’s good at. Or whatever other mix of big CPU + little micro. Or there’s that board you can already get with the 8 little auxiliary units on it.

        For the much higher price, I suppose this new chip must have something it’s best at.

  3. Skim linxgizmos.com for 10 minutes and you’ll find a dozen more computer-on-module options with every combination of memory, storage, and connectivity you could imagine. Those features won’t make anything special in this market.

    The two things that actually matter for people reading this site are:
    1. Can I actually buy today, and without buying 1000 of them?
    2. Can I run up-to-date Linux and interface to the hardware peripherals without my own team of full-time developers?

    For most of the awesome-looking CoMs on the market, the answer to both of those questions is no. The first one is easier to find a “yes” to, but the second one is very, very hard to satisfy. Most of the hardware vendors provide atrociously out-of-date board support packages (you wanted to run a 3.10 kernel, right?) and spotty peripheral support (we’ll get around to enabling that MIPI CSI peripheral one of these years, we promise!). This is what made Edison unusable, and what keeps Raspberry Pi and Beaglebone hardware atop the hacker market despite their unimpressive hardware specs and power consumption.

    If Hackaday is going to post about every shiny new compute module that shows up, it would be far more useful to evaluate the software situation than to just regurgitate the manufacturer’s feature list.

    1. @Bunsen: IMO your comment is the best so far in this thread.

      You said: “If Hackaday is going to post about every shiny new compute module that shows up, it would be far more useful to evaluate the software situation than to just regurgitate the manufacturer’s feature list.”

      I second this big time! Add value HaD – don’t just parrot a manufacturer’s PR, especially when it comes to Microchip which has a reputation for “hidden costs” at the (often crippled) BSP/development toolchain level, recurring support/licensing fees, proprietary program/debug dongles, on-and-on…

    2. Great questions.

      The SIP and SoMs are available in single quantities at MicrochipDirect now.

      Atmel has been mainlining their drivers since 2002. Microchip continues that tradition. More information on the software story at linux4sam.com

    1. Lazy journalists who don’t investigate anything any more. It’s infested all journalism. They just re-word press releases. Like a student doing his assignment at the last minute, re-phrasing the textbook as he reads it.

      Still at least here it’s just some microchip or other. There’s a world of stuff that actually matters that suffers from this terribly.

  4. Don’t forget that Linux is everywhere.

    My DVD recorder uses it. My two tv sets use it. My blu-ray player uses it. And so does my TomTom One GPS receiver.

    So there is demand for this, cheap and in massive quantities.


    1. I agree, if your app needs the libs and data for GPS or video decoding, etc. and you don’t want to hire the people who can pull that together for a lightweight free RTOS. And as Linux is used for more and more, that option will shrink I suppose. But at the moment? If you can run it on an ESP32 yo can save a lot of costs.

      1. Microchip has lost me forever to the ESP32. I used to be a big fan of Microchip, but the last few chips I used from them had a lot of bugs, the documentation had lots of bugs, no mention in the errata of the bugs, and I wasted a lot of valuable dev time on time-sensitive projects to these bugs. ESP32 is still ‘beta’ but I’m having a lot of fun with it.

    1. But why would I want to run an entire OS on a coin cell? When and where it would make sense to use something like that, when I could use one of thousands of different microcontrollers in 8, 16 and 32 bit versions with extremely low power consumption, flexible sleep modes, etc? My sports tracker runs for almost a year on one button battery, has a 32-bit ARM microcontroller and doesn’t run Linux to do advanced movement tracking and run RTC…

      1. Linux was designed to run on the i386 system on Linus Torvalds desk. It was not designed to run on 64 bit systems or on ARM, these things were hacked in later. It was not designed to run on phones or supercomputers or on IBM mainframes. It was not designed to e a high performance web server. The features necessary to support these sorts of applications was added in later. You assert that things can only be good at the things that they were originally deigned for, and that assertion is patently false.

      2. Seems to do all right in smart phones. But it’s a system-wide issue rather than just a hardware or kernel issue. Having the best low-power and sleep support in the world doesn’t help if you have processes waking up every 100 milliseconds just because. That is an area where lots of software, originally built for desktop or server use, is still lacking, but the desire to run well on smartphones and laptops has been a driver for improvements in that area. For example, when Mozilla were still trying to make Firefox OS, and worked on the MemShrink push, they removed lots of unnecessary wakeup events, and coalesced the remaining events into fewer ones.

  5. Interesting family, the trend towards SiPs is a positive one. The variant caught my attention because it’s targeting RTOS applications (though no indication of which will support it.) At 500 MHz, an M7 is still probably a better bet with more oomph, but having some of that DDR2 on package may be appealing for some use cases.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s