Hybrid SSD design combines NAND, memristor-based ReRAM

Hybrid storage solutions are nothing new, but they typically combine solid-state and mechanical storage. A team of Japanese researchers has now developed a hybrid SSD design that relies on two flavors of solid-state storage: NAND and ReRAM. Unlike DRAM, memristor-based ReRAM is non-volatile, allowing it to serve as cache in a drive otherwise made up of traditional NAND.

TechOn has more details on the “SSD architecture,” which has been modeled on a 256GB drive with 1GB of ReRAM cache. The ReRAM would be used several ways. It would cache incoming writes and push them to the NAND only when there was a full page of data to write. Frequently accessed data would be stored in ReRAM, which could also serve as spare area for the controller.

The group anticipates its hypothetical hybrid would offer better performance, lower power consumption, and a longer lifespan than typical SSDs. Actual drives based on the approach might not be far off, either. HP, which is producing ReRAM in cooperation with Hynix, expects the first chips to be available next year. Samsung is working on memristor tech, too.

If ReRAM chips offer better performance and longevity than NAND, I can see hybrid implementations becoming popular. ReRAM is likely to be expensive when it comes out, but you wouldn’t need much of it to serve as a cache. The storage folks also have time to hone their caching algorithms with implementations that combine solid-state and mechanical storage, though those algorithms would certainly need tweaking for purely solid-state hybrids.

Comments closed
    • ronch
    • 7 years ago

    Although there’s considerable concern regarding the future of SSDs, I’d like to think that our engineers and architects will be able to come up with something that will ensure its continued success. Those saying that SSDs are doomed are a little too pessimistic. I look forward to the day when I can buy multi-TB SSDs for cheap.

    • Arclight
    • 7 years ago

    Oh dang,new reasons to get BSODs (or frowny face of death for those willing to install W8).

    • NeelyCam
    • 7 years ago

    I’ll give it 6mo until AMD copies this (like they copy everything from Intel). Maybe they’ll call it “FreRAM” ‘coz ‘free’ is as much as it’ll be worth.

      • Airmantharp
      • 7 years ago

      It seems like you read the title of the article, at least, but forgot to read enough to see that Intel isn’t mentioned here. And why would AMD ‘steal’ this? AMD doesn’t fabricate memory (or anything, actually), and they don’t make SSDs or their controllers.

      So what exactly are you talking about, aside from a failed attempt to troll?

        • NeelyCam
        • 7 years ago

        [quote<]So what exactly are you talking about, aside from a failed attempt to troll?[/quote<] I'm at -12 already... I'd call this pretty successful, no?

          • Airmantharp
          • 7 years ago

          Yup :).

          • rrr
          • 7 years ago

          Would be better if it was less obvious and people would be mistakenly upvoting you because you’d be taken at face value and it would sound like you praised AMD.

      • pogsnet1
      • 7 years ago

      This is SSD, not RAM. AMD does not make SSD. It only makes RAM because most RAM are optimized for Intel not AMD.

        • Airmantharp
        • 7 years ago

        Technically, they don’t ‘make’ RAM either- they re-brand Patriot modules, and Patriot doesn’t make the memory that’s on those modules, and may not make the modules themselves either.

      • Kurkotain
      • 7 years ago

      let the people bite, and the downvotes commence.

      • UberGerbil
      • 7 years ago

      Wow, you’re really off the deep end on this one.

      • ronch
      • 7 years ago

      Well, if they can make something better, you’d buy it, wouldn’t you? And if not, then at least AMD would provide some competition to help keep the players in this field a bit more honest.

    • Bensam123
    • 7 years ago

    Soo…. why wouldn’t this be used on a mechanical instead of on a drive with MLC? Why wouldn’t ReRam just outright replace MLC as soon as it drops in cost?

      • Airmantharp
      • 7 years ago

      It probably will, over a multi-year period. I would think that demand will keep it more expensive until literally everyone is making it instead of MLC NAND.

    • Theolendras
    • 7 years ago

    Finally a potent write cache technology.

    • Shinare
    • 7 years ago

    I am definitely looking forward to the day where memristors change the face of the computing world. We certainly need something to stir the pot.

      • Arclight
      • 7 years ago

      Right now i think that the main bottleneck is the RAM…..after we get passed the memory wall we will certainly see a big change even in low end computers.

        • Airmantharp
        • 7 years ago

        I hope you’re not talking about memory bandwidth or price, and if you’re not, are you referring to the problem of DRAM being non-static?

      • rrr
      • 7 years ago

      [quote<]We certainly need something to stir the pot.[/quote<] I think this will do nicely: [url<]http://www.smoothharold.com/wp-content/uploads/2008/03/spoon.jpg[/url<]

    • rrr
    • 7 years ago

    Sure enough, it will be interesting to see, what in practice ReRAM technology brings to the table. Will it alleviate longevity concerns arising from die shrinks of NAND memory? We shall see when retail products hit the market.

      • bcronce
      • 7 years ago

      “Will it alleviate longevity concerns arising from die shrinks of NAND memory?”

      It is currently slated for actual System memory ReRAM to be sold in 2014. If ReRAM can handle being 14GB/s DDR3 , it can handle the little bit of writing HDs have compared to system memory.

      • Geistbar
      • 7 years ago

      The lifetime concerns of NAND are technology specific — because of the way flash works, it has a cycle limit. Die shrinks add additional physical constraints that lower the number of cycles, all else being equal. Other memory technologies, including memorsistor based ones, do not have this same limitation; they might have other limitations, and some others might even have cycle limitations, but not for the same reasons as NAND. It’s not really an issue with this tech, once/if it takes over.

    • UberGerbil
    • 7 years ago

    But if Memristors follow the path their proponents assume (production on existing fabs and rapid adoption without any major hitches) the price should drop rather quickly. That would make hybrid devices such as this merely transitional products that mark an interesting changing of the guard.

    Hypothetical products are useful for planing purposes, but only to the extent they’re modeling something that actually happens. I’m still waiting for a significant commercial implementation.

      • bcronce
      • 7 years ago

      It is going to only be transitional, but like you pointed out, how fast will the prices come down?

      ReRAM is supposed to be ~6x denser than FLASH. That alone will drop the prices. FLASH also has a negative scaling for reliability and access time, this alone limits its usefulness to a few more generations.

    • jjj
    • 7 years ago

    What they fail to say is why use ReRAM instead of a few other non-volatile solutions (MRAM,PCM,FRAM). After all the ReRAM is used just as a fast cache so i’m not so sure that the use of ReRAM is all that relevant- it is true ,however, that (3D) ReRAM is what a number of companies expect will win this race.
    The numbers quoted are also for very specific tasks and mean little for general use perf.

      • Theolendras
      • 7 years ago

      From what I see it seems probable that it could boost write speed considerably and shine in various I/O loads.

        • OneArmedScissor
        • 7 years ago

        I think he meant in comparison to those others, which have similar advantages vs. volatile DRAM, like not requiring refresh power.

        However, density still sets some of them apart, and ReRAM even beats flash.

        Scalability also goes along with that. I don’t know about all of them, but FRAM isn’t able to be made at modern nodes, and it may never be. MRAM still appears to be experimental at just 65nm.

        This is why we aren’t going to just wake up one day and find that 28nm SoCs use it, even if it wouldn’t take up much space.

          • Theolendras
          • 7 years ago

          Sorry I think I didn’t saw the obvious on jjj comment.

      • bcronce
      • 7 years ago

      MRAM: Still in design phase, yet alone prototypes. Huge issue is scaling smaller than 90nm

      PCM: Still in prototype phase. They haven’t even gotten to the “engineering mass-production” phase

      FRAM from Wiki: “thus the cost per bit of flash memory is orders of magnitude lower than that of FeRAM. It sounds expensive

      ReRAM: Scheduled to be ready for retail in 2013 and should be a fraction the cost of current SSDs once mature.

      I think you under-estimated how soon ReRAM will be out. It’s just around the corner.

    • DavidC1
    • 7 years ago

    Now all we need is some manufacturer that pairs this with traditional platter HDD so it becomes a Hybrid Hybrid HDD.

      • Theolendras
      • 7 years ago

      This would be somehow weird. You could also forget NAND altogether in that scenario.

      • Airmantharp
      • 7 years ago

      Honestly, I think Hybrid HDDs will go the way of the dodo- primarily because the two main manufacturers (WD and Seagate) don’t seem terribly interested in consumer SSDs.

      With various software and firmware solutions that provide SSD caching, as well as the relative ease of putting an SSD alongside an HDD in a typical Windows system, the extra cost of a ‘hybrid’ HDD just doesn’t seem worthwhile.

      HDD manufacturers are already seeing the crunch, and with rapid SSD price drops, will be forced to put all of their cards onto the table. Here’s looking to $99 5TB drives!

Pin It on Pinterest

Share This