Toshiba QLC 3D NAND squeezes a fourth bit into flash cells

Data density is critically important to manufacturers of flash storage chips. Increasing the number of bits in a given amount of die space allows for reduced manufacturing costs, potentially lower power requirements, and gets large amounts of information into ever-shrinking packages. Flash memory comes in flavors with a single bit per cell (SLC) to three bits per cell (TLC), and now Toshiba has announced that it has started manufacturing quad-level cell (QLC) 3D NAND chips. Like it says on the tin, these chips are capable of holding four bits per cell.

Toshiba's QLC chips are built on the company's BiCS 3D manufacturing technology, and the first dies out of the gate store 768 Gb (96 GB) each. The company says a package with a 16-die stack can store 1.5 TB. The previous-generation TLC technology could "only" pack 1 TB into the same package. In its press release, Toshiba touts the potential applications of these higher-density flash chips in datacenters, though it's not hard to imagine the technology trickling down into larger-capacity storage devices for mobile devices and PCs.

Toshiba says it started delivering sample QLC flash chips to SSD and SSD controller vendors earlier this month, and that some of the first silicon will be displayed at the 2017 Flash Memory Summit in August. There's currently no word on longevity and performance figures for the new chips, but the achieved density is impresssive on its own.

Comments closed
    • Fonbu
    • 2 years ago

    QLC will definitely drive capacity up, of SSD’s. And this QLC could drive a new SSD market segment usage scenario for mass storage class SSD’s. But as others mentioned the fourth bit requires many more voltage changes which reduces the endurance. But in a mass storage purpose built drive this would probably be suitable. I imagine more provisioned space will be needed with this QLC and the power consumption will go up. Power will increase because of the nature of QLC and the necessary error checking on the controller that supports flash.

    • Kraaketaer
    • 2 years ago

    I for one am really looking forward to QLC NAND. Why? Well, the endurance “issues” of TLC have been massively overblown, and while QLC does step into dangerous territory (~150 write cycles) I’d be more than happy with that for a general storage drive. Along with more layers than today, we ought to see 2TB SATA SSDs in the ~$200-250 space. That would be amazing. I would probably stop buying NAS HDDs then.

    Not to mention the room for creativity with partitioning and running the NAND in different modes: using QLC as MLC should be as simple as the controller treating every group of 4 voltage states as the same (with some light ECC to even out the transitions). I’d gladly take a drive with 2TB of QLC flash where 1/4 of it was set aside for a 256GB MLC partition (whether static or dynamic), or something similar. Or, of course, a 128GB SLC partition. Yes please. And make this all user tunable/configurable through software.

    • ermo
    • 2 years ago

    Which drive(s) are currently considered the best bang-for-the-buck V-NAND/3D TLC drives?

    Am I correct in assuming that, for consumer purposes (more reads than writes), 3D TLC NAND is *plenty* fast and durable?

    • mikewinddale
    • 2 years ago

    I wonder, is it technologically feasible for a manufacturer to let the end-user choose between SLC, MLC, TLC, and QLC? I mean, is it technologically possible for a TLC chip to be treated as MLC? If so, could some SSD manufacturer add a feature to their software toolbox to let the end-user choose? That would allow one product to cater to both casual users and enthusiasts. You could market an SSD as a 1 TB TLC, and then someone could convert it into a 666 GB MLC. Is that feasible or possible?

    Unless I am mistaken, pseudo-SLC caching already involves treating MLC or TLC as SLC. So why don’t we see SSD software toolboxes allowing users to convert an entire TLC SSD into MLC?

      • willmore
      • 2 years ago

      If the controller on the chip can do it, there is no reason that what you propose would not be possible.

      • mikewinddale
      • 2 years ago

      I suppose one problem might be, what if you’ve already written data to the drive? If you’ve already written data in a TLC fashion, what happens when you switch it to MLC?

      I’m sure engineers could figure a solution better than I can, but one possibility that springs to my mind is, the toolbox only lets you switch TLC to MLC or MLC to SLC if the data you’ve already written to the drive, is less than what the drive’s new capacity will be. E.g., if you have a 1 TB TLC and try to convert it to a 666 GB MLC, but you have 700 GB of data on the 1 TB TLC, then the toolbox simply won’t let you make the conversion. But if you have 500 GB of data on the TLC, and try to convert it to 666 GB MLC, then the toolbox will let you.

      Once you convert the drive, then the drive will, in the background, slowly migrate your data. So if you’ve got 500 GB of data written in a TLC fashion, it will slowly rewrite your drive in an MLC fashion, in the background. Every time you make a new write to the drive, which requires erasing a page anyway, it will just erase a TLC page and rewrite it as several MLC pages.

      Or maybe, it will do it all at once, sort of like when you encrypt your drive, and it tells you that your drive will take a few hours, and it gives you a progress bar.

        • tacitust
        • 2 years ago

        I think a more likely scenario would involve the manufacturer taking advantage of being able to use a single manufacturing line to create a range of products, the mix of which they can change simply by flicking a switch in the firmware and slapping different labels on.

        But if you wanted to leave it up to the customer, the simplest approach would be to only allow a switch in capacity when “formatting” the drive — i.e. if you want to keep your data, back it up first.

      • just brew it!
      • 2 years ago

      I think the only way this would make sense would be to tell the user to back up all their data and blow away existing drive contents. Then repartition and reformat. Aside from that I don’t think there are any major technical hurdles.

    • Chrispy_
    • 2 years ago

    At some point, controller complexity is going to drive costs up more than the density increase can compensate for.

    SLC = 2 voltages and 1 bit
    MLC = 4 voltages and 2 bits
    TLC = 8 voltages and 3 bits
    QLC = 16 voltages and 4 bits.

    If you extrapolate the pattern out, a 6-layer-cell NAND chip has only double the data density of TLC, but it has 16x more ECC and controller complexity. The poor thing has to read, write and error correct for 64 different voltage levels and yet we were running into voltage drift issues and huge performance penalties from controller overhead with only 8 voltage levels in the early TLC days.

    • meerkt
    • 2 years ago

    Uh oh.

    Reliability/performance concerns aside, I’m also starting to wonder if SSDs will even compete with HDDs on price/GB.

      • DancinJack
      • 2 years ago

      Hard drives have been in commercial use for literally decades. Give it a little time…

    • ronch
    • 2 years ago

    And the bottom just wasn’t deep enough and they had to dig a little bit deeper still.

    • chuckula
    • 2 years ago

    [url=http://bit.ly/1PXONqM<] Screw everything, we're doing five [s<]blades[/s<] bits![/url<]

      • davidbowser
      • 2 years ago

      Thank you!

      I sometimes forget just how awesome the onion is. I need to put it in my daily feed.

      • morphine
      • 2 years ago

      I must say, I’m severely disappointed with my writer for not using the obvious Gillette joke. He shall be flogged.

        • CuttinHobo
        • 2 years ago

        I can see it now, the big motivational poster at TR HQ:

        “The floggings will continue until writing improves.”

          • morphine
          • 2 years ago

          Actually, at TR HQ we regularly distribute [url=http://i1.kym-cdn.com/photos/images/newsfeed/000/238/219/71f.png<]paddlins'[/url<], pinches, award the Typo of the Week prize (sometimes multiple times), and overall dispense threats of bodily harm.

            • CuttinHobo
            • 2 years ago

            Paddlin’? [url=https://www.getyarn.io/yarn-clip/5e3340af-4f4f-43ba-95e8-6b04f99dc4ca<]I believe it's more of a kickin' situation.[/url<]

            • Generic
            • 2 years ago

            Expected O’ Brother; was not disappointed.

            Thank you. 🙂

        • Peter.Parker
        • 2 years ago

        …Or at least some Indian burn.

      • Chrispy_
      • 2 years ago

      Damn, you beat me to it!

      • tipoo
      • 2 years ago

      My favorite part is that the article came out the year before the Fusion 5 blade or something.

      Now we’re at 7! (it really sucks, don’t try this one)

      [url<]http://www.dorcousa.com/pace-7-sva1000/[/url<]

        • just brew it!
        • 2 years ago

        I refuse to pay ridiculous prices for blades. After trying various options, I’ve settled on Meijer twin-blade disposables. Relatively cheap, and they don’t suck like most of the other disposables.

          • warriorpoet
          • 2 years ago

          My current favy is Harry’s.

          Yes. 5 blades.

          Huzzah irony.

        • willmore
        • 2 years ago

        Dorco rocks. I switched to them a few years back and it’s saved me a lot of money. I find their three blade cartridge works best for me. I really like that all of their handles and cartridges are interoperable.

    • xeridea
    • 2 years ago

    TLC wasn’t really cheaper than MLC, so it’s unlikely QLC cheaper than TLC. Also likely to be large speed tradeoff.

      • morphine
      • 2 years ago

      [citation needed]

      • willmore
      • 2 years ago

      In defense of the ‘TLC will make flash cheaper’ arguement, it did come out just in time for mobile device production to go insane and suck up all the available flash memory production. That pushed the price of all flash devices up past where it had been.

      It likely would have been much worse if TLC hadn’t come out when it did.

      I, too am worried about the speed and endurance tradeoff QLC will likely bring.

Pin It on Pinterest

Share This