Nanotube-infused NRAM promises DRAM speeds with unlimited endurance

Since 2001, a company called Nantero has been working on making non-volatile storage with carbon nanotubes. The technology is called NRAM, and it's designed to deliver DRAM-like speeds with unlimited endurance and low power consumption.

That combination sounds too good to be true, but it's not just a pipe dream. Nantero CEO Greg Schmergel told us the firm has been working with two of the top five semiconductor foundries. Its nanotube tech is already deployed in seven production-grade fabs, he said, and multiple implementations are in different stages of qualification.

Instead of building its own products, Nantero licenses its intellectual property to others. That IP includes the "world's only method for purifying carbon nanotubes" to the standards required by CMOS fabs. This patented approach is compatible with existing fabrication tools and apparently requires no additional equipment investment.

NRAM is created by laying carbon nanotubes on top of a silicon wafer. That layer is then patterned with standard lithography and etching. On a 2x-nm node, each NRAM cell contains hundreds of nanotubes. The tiny strands can be scattered randomly with that geometry, but alignment is required to ensure adequate coverage when the feature size shrinks down to 10-nm territory.

Source: Nantero

Each cell effectively has its own nanotube fabric. The resistance is low when the nanotubes are touching and high when they're not. Applying voltage causes the cell to switch between those states in just picoseconds, according to Schmergel, and writes require a fraction of the power required by flash-based storage.

Unlike NAND, NRAM has "theoretically unlimited" endurance. Chips have already survived 1012 write cycles and 1015 reads, which might as well be infinity. They can purportedly survive just about anything, including extreme heat, cold, magnets, and radiation. A test chip has even been to outer space courtesy of an "open bay" in the Atlantis shuttle. That's some serious geek cred right there.

Although it sounds like the initial implementations will have one bit per cell, NRAM supports MLC configurations via nanotube sub-groups. Test chips have also been created with multiple layers. Interestingly, Schmergel told us that vertical stacking is a higher priority than squeezing extra bits into a single layer. Building up will apparently be required for NRAM to challenge flash memory on cost. It's supposed to be cheaper than DRAM already, though.

Some of the first NRAM products will use DDR4 interfaces. That's mostly for convenience; Schmergel said DDR4 isn't actually fast enough to fully exploit NRAM's potential. Nantero is exploring custom solutions that will.

We won't have the skinny on the first NRAM products until Nantero's partners are ready to reveal them. Schmergel told us to expect the first to be built on 2x-nm tech, so it probably won't be long before we hear something. If everything goes according to plan, the technology could be used in everything from servers to smartphones to embedded applications.

Comments closed
    • GTVic
    • 5 years ago

    Isn’t ” … works with any …” one of the greatest selling points for late night TV infomercials? Hopefully this catches on better than the Car Caddie.

    • travbrad
    • 5 years ago

    Phew. I was hundreds of TB away from running out of endurance on my SSD. Disaster averted.

    • Chrispy_
    • 5 years ago

    Had me worried there;

    I was reading a serious TR news post and got down to the sixth paragraph before I found my [i<]purportedly[/i<].

    • Sam125
    • 5 years ago

    Unless something has changed with CNTs within the last 15 years, good luck trying to scale the production of “CMOS grade” CNTs.

    This sounds really promising but I can’t help but wonder if this news report isn’t just a search for a new round of investors.

    • Peter.Parker
    • 5 years ago

    [quote<]...but it's not just a pipe dream[/quote<] Au contraire, it's a dream pipe!

    • mesyn191
    • 5 years ago

    This all sounds really cool but until it actually shows up in a product I can buy at non stupid prices I don’t care too much.

    There is always some new memory standard that promises to be the end all be all only to never show up.

      • LostCat
      • 5 years ago

      I don’t even care about the prices since there’s still competition, we’ve been shown so many new awesome techs that never show up at all.

        • UberGerbil
        • 5 years ago

        Yeah, if it’s shipping at super-expensive prices for server / HPC environments or exotic military/aerospace applications today, it’ll be at enthusiast-grade pricing tomorrow and in ubiquitous consumer products the day after that (see SSDs, for example). But first it has to be [i<]shipping[/i<].

    • brucethemoose
    • 5 years ago

    [quote<] On a 2x-nm node, each NRAM cell contains hundreds of nanotubes. [/quote<] Correct me if I'm wrong, but that sounds pretty big for a single cell on a ~22nm process. Carbon nanotubes have a diameter over 1nm and can be much longer than that.

      • Hattig
      • 5 years ago

      I think they’re on top of each other too. But as the article says, they’re going vertical, not denser.

      But capacity is probably going to be lower than current DRAM until that happens.

      • Pwnstar
      • 5 years ago

      It’s probably actually 28nm.

    • Duct Tape Dude
    • 5 years ago

    Several questions that time will answer:
    1) How dense is it compared to standard RAM, or ultimately NAND?

    2) Is power consumption sane at scale?

    3) Is capacity too small/bandwidth too good/etc to pursue a PCIe SSD form factor?

    4) How do this tech’s characteristics compare to HP’s memristors (which they refuse to release en masse in standardized formats and are reserving for their Machine)?

    5) Where can I sign up to test new computing technology in space? I’m very well-qualified, having never killed a Kerbal in several dozen hours. At least.

    • Flapdrol
    • 5 years ago

    [quote<]not just a pipe dream[/quote<] badum tish

    • just brew it!
    • 5 years ago

    I wonder how well “theoretically unlimited” endurance will translate into practice once the process matures. Since the underlying mechanism is physical it seems to me there’s still potential for wear. Furthermore, while 10^12 cycle write endurance might as well be infinite compared to flash, it still isn’t enough to make this tech a viable non-volatile replacement for DRAM.

    In the meantime, commodity NAND flash will continue to get cheaper…

    If I had to place bets, I’d guess NRAM will remain too expensive to displace NAND, and won’t be viable as a non-volatile replacement for DRAM due to a combination of cost and write endurance issues. If it makes it to commercialization at all, my guess is that it’ll remain a niche product like [url=http://en.wikipedia.org/wiki/Ferroelectric_RAM<]FRAM[/url<] or [url=http://en.wikipedia.org/wiki/Phase-change_memory<]PCM[/url<], used in embedded applications where its particular combination of characteristics make it especially suitable.

      • ImSpartacus
      • 5 years ago

      Yeah, nand is getting cheap. I can’t wait until respectable tb ssds dip below $200.

      • cmrcmk
      • 5 years ago

      That was my thought, too. 10^15 writes is fine for long term storage like NAND, but if you use it as RAM (assuming a 1GHz speed) you could burn through that life in 2 months, though obviously that would be some pretty intense usage.

      • owmcyehs
      • 5 years ago

      10^12 becomes essentially infinite when you also consider the capacity.

      Imagine a 4GB module connected via a 50GB/s DDR4 interface.
      If that module can withstand 10^12 writes then that is 4 zettabytes of total writes. At 50GB/s it would take over 2500 years to write that much data.

      A good first use for this memory would be cache on NAND based SSDs. Many of the fragmentation problems would go away with a few GBs of non-volatile, unlimited endurance NRAM for cache and lookup tables.

        • just brew it!
        • 5 years ago

        [quote<]Imagine a 4GB module connected via a 50GB/s DDR4 interface. If that module can withstand 10^12 writes then that is 4 zettabytes of total writes. At 50GB/s it would take over 2500 years to write that much data.[/quote<] That's great if you are using it as a flash replacement. Not so great if you're trying to use it as a non-volatile replacement for DRAM (no wear leveling).

          • owmcyehs
          • 5 years ago

          You couldn’t use it as a drop in replacement for DRAM, but you could use it for the same tasks as DRAM. If wear levelling is needed a very course grained approach would suffice – I expect it could be implemented in software.

          I’m still not convinced wear is a problem at all. Consider that tight loops and frequently updated memory locations exist within the CPU caches. Most changes are not written out to RAM. A use case which requires GB/s of writes to main memory is unlikely to be performing those over a tiny address space. (Even if we drop the address space to 4MB – at 50GBps it will still take 2.5 years to reach 10^12 writes per cell).

    • blastdoor
    • 5 years ago

    Cool!

    It’s becoming increasingly clear that memristors are not the only nifty idea in the world of RAM.

      • just brew it!
      • 5 years ago

      See my other post. There are other equally nifty NVRAM ideas which (unlike memristors) are already being widely used in production products. You just don’t hear about them much unless you happen to do embedded development since they are “behind the scenes” devices, used for internal tasks like storing non-volatile fault logs, maintaining device settings, etc.

    • crabjokeman
    • 5 years ago

    Does anyone else see an eye in the black-and-white NRAM Cross Section image?

      • blastdoor
      • 5 years ago

      Now that you mention it, yes….

        • crabjokeman
        • 5 years ago

        Thanks. I like validation when taking the Rorschach test.

        • Neutronbeam
        • 5 years ago

        Eye do too. Totally tubular bro-dude!

      • Liron
      • 5 years ago

      Not only an eye, but three clearly distinguishable short, fat tentacles below it.

      • Tristan
      • 5 years ago

      NSA spy

        • the
        • 5 years ago

        That’s not the NSA style. Rather it is the Illuminati.

          • UberGerbil
          • 5 years ago

          Why not [url=http://illuminatisymbols.info/wp-content/uploads/2013/02/590px-IAO-logo.png<]both[/url<]?

      • Hattig
      • 5 years ago

      Sheesh, that’s Cthulu isn’t it?

      Anyway, if this turns out to be true. Holy hot damn diggety wowza.

      • wierdo
      • 5 years ago

      That’s NSA’s backdoor, you’re now on the no fly list for spotting it. Also bake the FBI a nice cake, they’re coming over this afternoon.

        • crabjokeman
        • 5 years ago

        My doormat reads, “No men in uncomfortable shoes allowed!”
        Also, I haven’t flown in years. I have no place to go…

        EDIT: BTW, I prefer terrorists on my airplane flights as opposed to small children. Terrorists are generally quieter and less annoying.

      • ludi
      • 5 years ago

      I do now, thanks for that.

    • TwoEars
    • 5 years ago

    Since 2001? Holy eff. I want their investors.

      • Liron
      • 5 years ago

      If you’re working with CNTs or graphene, it’s not that hard. Most investors in the early 00s knew that there was no chance of commercializing within a decade.

Pin It on Pinterest

Share This