Using SIMMs To Add Some Extra RAM On Your Arduino UNO

A Single In-line Memory Module (SIMM) is a type of memory module containing Random Access Memory (RAM) which was used in computers from the early 1980s to the late 1990s (think 386, 486, Macintoshs, Atari STE…). [Rafael] just made a little library that allows you to interface these modules to the Atmega328p-based Arduino UNO in order to gain some memory space. His work was actually based on the great Linux on the 8bit ATMEGA168 hack from [Dmitry Grinberg] but some tweaks were required to make it work with [Rapfael]’s SIMM but also to port it to the Arduino platform. The 30-pin SIMM shown above is capable of storing up to (hold on to your chairs…) 16MB but due to limited amount of available IOs on the Atmega328p only 256KB can be used. Our guess it that an SPI / I2C IO extender could lift this limitation. A quick (shaky) video is embedded after the break.

70 thoughts on “Using SIMMs To Add Some Extra RAM On Your Arduino UNO

  1. I have often though of ideas like this but never got around to working out how to make it happen. Really cool. When I saw the photo and read the first few lines I was all like “WHAAAT?!!! YEEEAAAH!!!” And about wet myself with “storing up to 16MB..” (that is a LOAD for a little AVR or PIC) but then cried a little when I saw “only 256KB can be used.”

  2. Can we get an Arduino like thing with lots of IO AND a fast processor AND megs of ram for $30 or so? Is that too much to ask? RasPi *kind* of does this but not reallllly. Needs more IO. Being able to run linux while also doing reasonably real time IO would be nice too. Maybe some kind of LCD screen so you can actually get feedback as to what is going on. Anybody?

    1. Yes, that is too much to ask. The Arduino gives you a taste of the power of embedded system. If you need something with the specs you’ve listed but don’t need embedded Linux just build your own.

      If you have the skills to actually use the features you spec’d it should be trivial for you to design and populate a board.

    2. Look at some of the smaller ARM chips like the STM32F10x or STM32F30x series. They cost peanuts, run up to 150MHz, have tons of I/O, decent amount of on-board RAM and some have external memory interfaces, so adding more is not a problem.

      1. They also have a lot of lovely features like DMA and an FPU that works with pretty much all of the internal peripherals.

        I’ve used the TiVA (same renamed chip as the stellaris launchpad) and its great because you can setup an ADC sequence that automatically samples periodically and copies the results to memory with DMA. So you just get an interrupt every time you have a new value in memory ready to use.

    3. This is why people tend to pair a Linux device with an MCU (like a Pi and an Arduino).

      Running a complex OS lets you easily take advantage of a complex and powerful processor. And gives you ready-made drivers to access consumer peripherals that can do something useful with all the data it will be processing. But pretty much precludes generic real-time I/O.

      Combine the two and get the best of both worlds. I’ve been pairing older used Pogoplugs with a bare PIC. About $25 for the combo, and I get all the CPU and RAM, plus high and low level connectivity, I need.

      Of course there is no requirement to run Linux on a Pi or other device which would normally run it. You could always code bare metal, and then you can do whatever you want with all its capabilities. Is that too much to ask? ;)

    4. BeagleBone Black @ $45.

      Tons of IO, some IO routable to one of the two DMA capable 200MHz real time coprocessors (PRU), 512MB RAM, 2GB onboard flash plus microSD, 1GHz Cortex-A8, HDMI out, Ethernet, USB HS host and device, onboard power controller with soft power off support and LiPo/LiIon charging.

      Debian, Android, Ubuntu, and many more Linux distros supported by a large community. Almost all code open source (GPU lib and power management coprocessor blobs, both optional), all design files open source, and all parts can be sourced in onesies.

    5. Well, there’s the latest STM F4 Discovery board, it comes with 2GB of flash, 256K of onboard and 8MB of external RAM, a 2.4″ LCD screen with touch encoder and an ARM M4F (with hardware floating point) @180MHz for less than £25

    6. Not at $30 USD but you can for $60 to $100 depending on how much support you want.

      PhonenixA20 $60 <- Very poor support
      CubieTruck $100 <- A20 based alternative with what appears to be good support
      RadxaRock $100 <- Very green board with a growing community

      Then there are any of the Arduino "compatible" boards with an ARM on board.

  3. When all of the college student rental leases expire in August you can walk down the street and find old PCs waiting for the garbage man. I usually pop the case off the side and see if I can get some memory, hard drives, optical drives, etc. for my junk bin. The memory interface makes for a fun weekend project, and I don’t feel bad soldering to the DIMM (yes, you can do this with somewhat more modern memory too) just for kicks since it was going to be ewaste anyway.

  4. My God, people WHY?!? Why to add RAM in this way? You get a bit more RAM, but basically make the controller useless, because it is wasting all time refreshing the DRAM and all its pins are taken up by the RAM interface …

    This sort of thing has a value only as a neat hack demonstrating that it can be done, but its practical value is zero. The first thing to do if you need more RAM is to look whether you can can get static RAM – they cost only a few bucks these days in megabyte capacities. Of course, a sensible person would use a serial RAM on this type of micro, not a parallel one.

    An actual engineer would choose a controller with enough RAM or at least something that has an external memory interface option (some larger ATMegas or XMegas have it, then many of the small ARM micros do as well).

    Mind-boggling, honestly …

    1. “This sort of thing has a value only as a neat hack demonstrating that it can be done”

      There you have it in a nutshell, Jan. Some folks enjoy doing proof of concept stuff, myself included.

      1. Like seeing if you can screw your feet down to the floor using sheet rock screws and a gold ol’ manual screwdriver… Completely pointless and hours of pain and frustration all so you can say “I new I could do it!” and hopefully post it to youtube to impress your hipster friends.Just imagine if these people dedicated all that effort and brain power into something useful!!! We might have a cure for Cancer!? If only grandma would have let you come up from the basement! But then we’d have fewer pictures of cats with bread mushed over their cute little heads and less arduinos running TRS-80 emulations. Super important ground breaking and fun stuff!

        1. With some hardware modifications you can free up a bunch of pins as well as cycles. You always need a proof of concept before you can advance or at a minimum see if it is viable to advance

          1. Or, I dunno, you could save all that effort and get some modern chips that save 50 or 60 pins, have a nice compatible SPI or I2C interface – built in no less, is static RAM so you don’t have to worry about losing data and refreshing DRAM constantly and you can use the full memory without resorting to all the gotchas of above. Engineers somewhere have already solved all of these problems, packaged them in a nice little IC at a dollar or two a pop. But forget all that! Proof of concept? The fact we were using these in 486’s in 1980 something is your proof that they once worked, and man moved on. Or is it just stabbing in the dark like a mad man at something that, yes is possible, but makes you look like a raving lunatic for trying when there is a better way and has been for years? Don’t let me stop you from trying to turn the earth backwards, But here’s a chisel and a big lump of rock, now go make a better wheel or something…

        2. The point is learning and experience. There’s no need to reinvent the wheel, but you can’t design the next big sports car if you don’t understand wheels.

          1. They’d be pretty hard to find, only made for a couple of years, and they weren’t really practical at the time. For 256 or 512K of SRAM there’s better options now.

    2. What an incredibly misinformed comment!

      It only takes 2% of the cpu time to refresh the RAM in ASM version (i wrote it, i know)
      and there isnt another easy way to get 16MB of ram attached to an AVR easily
      so for 2% loss of performance you get access to 16MB of ram. sounds pretty useful to me. like for example the linked linux project used it :)

        1. i refresh all of ram, and do so in a very tight asm loop. look at the linked linux project! my project is linked in the original story. go look at it. it properly refreshes all of ram and uses it all, provably!

          ugh…reading..

    3. my god why the fuck does clothing store sell clothing?
      BECAUSE ITS A CLOTHING STORE

      this site is devoted to DIY andor “because we can”

      this site, BTW, is called HackADay, get it?
      hack, as in poorly executed modification FFS
      nobody is forcing you to read this site EVERY DAY.

      you keep coming back to post, every day about how this site is not good enough for you. so stop coming back and seek mental help.

      this site is devoted to it and you will never change that.
      we come here to share non practacle projects because we all like reading about non practacle things.

      YOU are the weird one here! YOU are the misfit here!
      WE already know its a hack,
      WE figured this because this site’s name is HACKADAY!

      WARNING THIS CONTENT MIGHT NOT STAND UP TO YOUR STANDARDS!
      ENTER AT YOUR OWN (risk? lol)

      remember, we do not owe you anything.
      don’t act like you already paid us to come up with a solution.

      and of course your right, its all just a huge conspiricy
      to get you to do shoddy work on the job. (facepalm)

  5. 23LCV1024? SPI Serial SRAM chips with 128KB of capacity that cost about £3 each in lots of 1. No worries about the number of available IOs or keeping DRAM refreshed, and they are available in DIP packages for easy use in hobby projects.

      1. Just to prove a point, you can get 1MByte for about 2 bucks at some of the usual sites. Don’t think they make 16MBytes in that format, but still, it’s in the same sortof ballpark. And 1MB of RAM on a serial connection is probably more use for a microcontroller than 16MB that uses up all the I/O pins.

        Then again there probably aren’t many uses for 16MB of data on a microcontroller like that, otherwise they’d make them that way. It’s a proof of concept, and I suppose managing the refresh in software is quite clever. But just about nobody connects RAM chips up to MCUs, it’s the opposite way of thinking. For the money and the hassle you’d just buy the right chip in the first place.

        1. Getting linux to run on a micro like a CortexM3 takes megs of RAM. The docs for uClinux say a 2MB min, other places say 4MB or even 16MB is required. Have a read of [link]http://electronicdesign.com/embedded/practical-advice-running-uclinux-cortex-m3m4[/link] for the why and how. Short answer: cost, power consumption, I/O flexibility, etc

          1. Probably about half or a third of the price. So yes, if you wanted 16MB RAM on an 8-bit MCU with only a couple of pins for actual I/O, then yes, this would be cheaper. Assuming you have a source of obsolete memory from 15 years ago. And you add something to provide the extra pins so you can have more than 256K.

    1. Are you also amazed that ford has not lately introduced an update to the model T?

      ATmega is pretty much deprecated for new designs. Sure people will continue to hack on it but anyone starting from scratch will be much better served by more modern architectures like ARM.

      By the way you can get some very powerful ARM and MIPS processors that will blow away any atmega chip and will fit into the same form factor. See Teensy 3, mikroe-1367, fubarino for examples. These all make atmega look sick in comparison.

    2. There likely isn’t much money in it because when they use DIP versions, people can easily just build their own boards or breadboard it. Notice that the ‘duinos have been migrating mostly to SMT packages only. With the exception of the original Uno, of course.

      So, why go through all the effort of officially designing something and coding in official support when most people wouldn’t buy the board? Especially when you already have the Mega?

      As you already linked, several people have already unofficially supported it. I was actually one of the people that helped put it together (though I think maniacbug was working independently on his own, IIRC.)

      The QFP version of the 1284 gives you some more IO. You can find dev boards with all the plugins for the Arduino IDE.

    1. Aaah, I remember those! Remember RAM compression software, too? For Windows 3.1. Use compression to have your 4MB RAM act like a very slow and diseased 6MB RAM! SOMEbody bought it! Back at the height of RAM pricing, just a little while before VLSI chips adopted the pricing model of jellybeans. Actual sugar ones. The RAM famine of the early 90s, it was terrible!

  6. It’s even easier with SDRAM. It has self-refresh mode, burst mode, so very little job is required to make a use of it. Also timings are not very critical (e.g. clock signal doesn’t need to be stable, and memory can survive more than datasheet says without refreshing). I wonder why people are afraid using those. Too dense pins maybe?

    1. Needing 130-odd pins wiring up is probably a hindrance to the hobbyist market, and the borderline-magickal routing requirements. Also the baroque clock-multiplied data signals and whatever else. Can they even run at speeds below 600MHz or so?

      1. > Can they even run at speeds below 600MHz or so?
        You are thinking of DDR with terminations. They are both clock edges, so actual clock rate is 1/2. I think you can get DDR 400 i.e. 200MHz parts (or slower parts?)

        SDRAM (without DDR prefix) come with 3.3V TTL signals, so they don’t go above 200MHz. You can probably find them in embedded system or really old hardware running at 66MHz.

        There is probably some lower limits on how slow the CLK can run at or otherwise stuff like auto refresh won’t refresh fast enough. If you want to play tricks with CKE can be used to effectively slow it down, but keep the CLK running fast. This is not for bit-banging in software.

    2. 1. 3.3V logic level
      2. Larger RAM array, so more rows to refresh. Higher overhead. ~15.6us per row
      3. Commands are synchronous, so you’ll need extra cycles to toggle the clock. No you can’t just hook up the clk to PWM etc because commands and latency are *synchronous* to clock.
      4. Auto refresh cycles still needs the clock.
      5. Larger memory array means you are burning up lots of I/O and they need to be fast – can’t be from SPI or I2C. (to satisfy #2.) Can’t simply tie them to fix level as the address pins are also used for sending commands.

      Between 2 & 3, you are spending a lot of time talking to the chip.

      it is easy to do it in a CPLD/FPGA, but a waste of time if you have to bit-bang in software. Not sure why anyone want to access large memory array in a 8-bit processors. Use a 32-bit processor with external memory interface and get compiler support.

  7. I did it in 2003 with a 20 pin AT90S2313 to log RS232 communications to two 1Meg 30 pin SIMMs. The trick is to connect the data out lines to the address lines with 47 ohm resistors to reduce pin count and avoid shorting the RAM outputs to the MCU port when bashing through an address. No need to hold address lines once the address strobe is low. Reading – output row address, RAS low, output column address CAS low, switch your AVR port to input, read the data in, RAS and CAS high. When writing use late write / !OE controlled write. Output row address, RAS low, Out column address CAS low, Output Data, !WR low, release CAS then RAS. Put a timer interrupt to toggle CAS every 15 microseconds. Needless to say the application was written in assembler to fit in 2 kB.

Leave a Reply to dmitry grinbergCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

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