Sign in / up

back to article All the flash world needs is more TLC, suggests report

Triple-level cell (TLC) flash is heading towards being nearly half of all flash shipments. A report from DRAMeXchange says TLC NAND flash will account for nearly half of the total NAND flash output in the fourth 2015 quarter. DRAMeXchange_TLC_chart The report author, Sean Yang, says TLC NAND is shifting from memory card and …

COMMENTS

Post your comment
House rules Send corrections
Add to 'My topics' unstar *
  1. Anonymous Coward
    Anonymous Coward

    Let's hope they fix the data degradation problem

    Problem with TLC is data degradation. Well, all flash do degrade over time (stored charge wear out a bit in each cell), but TLC, having 8 levels instead of 4 levels per cell, is much more prone to this issue.

    Tell that to the buyers of products like Samsung 840 TCL-flash based producs, and iPhone if I recall correctly also, where after writing data to flash storage, see dramatic performance hit after a month because of the flash controller having to read back this "stale" data.

    For people who doesn't know, a lot of TLC-based SSD were (and are still in many cases) plagued with issue where you write a file, leave it there for a month, and when read back, it read back 10 times slower. Yup, that 500MB/s SSD now perform sub-50MB/s (even 15~30MB/s read time in some cases). Yup, slower than a bad performing hard disk. Just Google 'TLC SSD problem'.

    The way for now the problem is avoided is to actually make sure every sectors is re-written at least once a month, and preferably more often than that, possibly even in background. I guess the next controllers/firmware could do that. Since write endurance is at least 10K cycles per sector, writing each sectors once a week still make only 520 writes per sectors in 10 years; nowhere near enough to wear the flash, but essential to keep read speed par specs.

    1 0 Reply
    1. phil dude
      Reply Icon
      Thumb Up

      Re: Let's hope they fix the data degradation problem

      I have been interested in the "bit rot" solutions provided by ZFS and BTRFS, checksumming all data.

      A question if I may. Is this what TRIM is supposed to address?

      P.

      0 0 Reply
      1. Anonymous Coward
        Reply Icon
        Anonymous Coward

        Re: Let's hope they fix the data degradation problem

        No, this is not related to TRIM. TRIM basically tell the SSD that a sector is no longer in use, and can be erased and put back to the free sector pool. Erasing a sector is the most time consuming of flash memory. So, each time the host wish to write to a sector, the sector must first be erased (and this take time). So, it's much faster to write to an already erased sector. TRIM tell the SSD that the sector is no longer needed, and the actual erasing of this sector can be done in background by the SSD controller itself, without affecting the host (without slowing the OS).

        SSD have also become smarter and writing to a sector already containing data will actually write to a new sector from the free sector pool, the the old sector automatically erased and put back to the free sector pool. This is part of the controller wear leveling and is transparent to the PC. The PC will always see the same sector, even though the data can be on any physical sector of the flash. The controller do it's magic and work hard translating "logical" block address (LBA) to physical flash location. However, when you just delete a file on the host OS, no data is actually written back to the now obsolete sectors. So those sectors aren't actually erased. And with time, those leftover sectors can accumulate and then the SSD won't have the choice but to erase sectors on-demand (i.e. when the OS ask to erase it). TRIM is just a way of saying that the sector is no longer in use, and erase it in advance.

        Problem with TLC "bit rot" is that TLC is more prone than MLC flash, and hence show much more the symptoms. From what I read, no-one ever "lost" data, but recovering the data is much more time consuming. Not sure exactly how it's done, probably using ECC and other techniques.

        0 0 Reply
        1. Bronek Kozicki
          Reply Icon

          Re: Let's hope they fix the data degradation problem

          Also, checksums in filesystem will not speed up reads of degraded memory cell. Either the read will fail completely, possibly after long wait (in which case checksum + redundancy will aid the filesystem in transparently recovering the data), or it will take a long time because underlying hardware will need this much time to read all cells reliably.

          1 0 Reply
    2. I. Aproveofitspendingonspecificprojects
      Reply Icon

      Re: Let's hope they fix the data degradation problem

      > Demand will start to pick up by the start of the third quarter because the cartels will have made their money back (and some) by then

      FTFY.

      >> after writing data to flash storage, see dramatic performance hit after a month because of the flash controller having to read back this "stale" data.

      You are assuming that the data you thought you had backed up did actually make it to you large, brand new USB 3 storage medium. I consider my new toy vapour ware until I have the time to check out 50-odd GB of data.

      The problem is I am too frightened too look.

      0 0 Reply
  2. Anonymous Coward
    Anonymous Coward

    OLC

    octa-level cell when

    256 succulent states

    0 0 Reply
  3. Alan Denman

    There's nought as cheap as chips.

    So Trim all the Fat, boys

    0 0 Reply

POST COMMENT House rules

Not a member of The Register? Create a new account here.

Forgotten password ?

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon