Micron 3D NAND Status Update

by Billy Tallis on 2/12/2016 10:00 AM EST
Comments Locked

19 Comments

Back to Article

  • jjj - Friday, February 12, 2016 - link

    Nice that it has 4 planes, with 48GB TLC dies i was wondering about perf per SSD capacity.
    The 768Gb , capacity wise, seems in line with what others will have. Do wonder if we see any 4 bit per cell in 2017 or that's too soon. 16GB xpoint + 64GB 4 bit NAND with just 2 dies would be fun in a phone.
  • extide - Friday, February 12, 2016 - link

    How does 768Gb, capacity wise, seem in like with what others will have? I don't think anyone has announced anything bigger than 256Gb ? Maybe I missed something? That seems like a WAY bigger die than pretty much anyone else has said anything about.
  • jjj - Friday, February 12, 2016 - link

    Others do have 768Gb on their roadmaps in 2017 so even if this will be gen 2 for Micron (it's not clear yet that it is), it is in line with (some) others., capacity wise. The precise timing fort each remains to be seen.
    And don't forget that die size is a factor too. Micron has a big die, Sandisk/Toshiba for example have smaller capacity now but in a smaller die that can fit in a microSD.
  • willis936 - Friday, February 12, 2016 - link

    I can't imagine xpoint controller's being power efficient enough to have a noticeable increase in performance over nvme/half amplitude pcie (as is done in the 6s). I'd say that second die would be put to better use doubling storage :D
  • jjj - Friday, February 12, 2016 - link

    Can't say i have a clue about the xpoint controller and what perf and power it can get.
    In phones one of the main goals would be to get rid of some or all RAM for a big gain in battery life- no idea how they plan to implement it and before we have more details it is futile to speculate.
    As for 16GB, xpoint gen 1 die is 16GB so the option of using less is not there.
  • name99 - Friday, February 12, 2016 - link

    XPoint replacing RAM (as opposed to replacing flash) requires
    - appropriate ordering instructions in the instruction set (ARM or x86)
    - OS modifications to use these instructions

    ARMv8.2A includes the appropriate instructions, but there are no implementations yet, and it seems unlikely that these will arrive before at earliest H2 2017. (A72 isn't yet out there, and as far as I know Kyro is ARMv8.0A. Apple MAY ship the A11 as 8.2A, but for various reasons I think the A10 is just a slightly faster, slightly improved A9 --- like the A7 to A8 transition.)
    Intel seems unlikely to get those instructions into Atoms or even mobile before they get them into Xeon, and likewise to do their usual clusterfsck of supporting them so randomly across processors that it will be ten years before anyone can rely on them.

    So no CPUs in the near future, which means that even if Android ARE working on this support in the OS, they can't push it out (or even test it) for a while --- and when it comes to storage you want to test a LOT on REAL hardware before you ship it into the field.

    So it's certainly coming, but I'd be amazed if I saw it in Android before 2019 or so. Apple COULD have it as early as 2017 because their tight integration means they could even have test silicon in house right now. It would be worth looking at the storage talks at WWDC this year and next year to see if they provide any hint as to ways of doing things that most make sense on an NVM design.

    [It's also possible, and even makes a kind of grand sense, that it's at the transition to an NVM design --- NVM hardware, NVM OS, NVM APIs --- that it makes the most sense to perform the Mac switch to ARM. If this gives Macs a substantial boost in "user-level" performance, even if CPU performance temporarily goes backwards a little, the transitions would be a little smoother; and it's the sort of thing that would be vastly easier to do with an Apple-designed CPU that while waiting on Intel's timetable.]
  • jjj - Saturday, February 13, 2016 - link

    When i mentioned xpoint it was a 2017 matter, and that at best.
    Anyway, xpoint in anything is a 2017 matter , it won't ship much , if any this year. In phones it is expected that it will take a bit longer than in server and PC SSD, because of the extra complications. It might even be unlikely for gen 1 to get in phones because of the price. They might actually be aiming for SoCs built to work with it, maybe advanced packaging but i would rather not speculate on this before we have proper specs. Oh, and i did say some or all RAM. Some phones this year will have 6GB of RAM, one option would be to cut just some of that and add xpoint and maybe hide it from the OS - kinda like Marvell's FLC i guess, except they do it on NAND.
  • phoenix_rizzen - Monday, February 15, 2016 - link

    There's a couple of phones out already (or have been reviewed with shipments coming very soon) with Cortex-A72 cores inside. Mostly MediaTek SoCs, I believe.
  • fishjunk - Friday, February 12, 2016 - link

    What does this mean for the future price/capacity of SSDs?
  • hojnikb - Friday, February 12, 2016 - link

    this means cheaper and more durable ssds
  • jjj - Friday, February 12, 2016 - link

    Price is more a factor of demand vs supply than costs. Sure lower costs facilitate the price declines and more capacity per wafer increases the supply. This year the supply will likely increase by less than 35% while demand should be somewhat in line with that, unless the global economy stumbles.
    With this first gen 3D NAND Micron claimed a 25% reduction in costs over their 16nm planar.
    For the second gen they claim a 30% cost reduction over the first gen.
    NAND prices do tend to decline some 30% or even more per year. In SSDs there is the controller and all else that don't have the same kind of price declines.
    But the prices we see depend on capacity vs demand. If the 4 players ( Toshiba+ Sandisk , Samsung, Micron+Intel and Hynix),overestimate demand and increase capacity by too much, the price declines are high. If they get it right, the price declines are mild. Last year they got it a bit wrong so this year they slowed down the capacity expansion.
  • easp - Friday, February 12, 2016 - link

    Wake me up when they start talking about putting more capable logic on the dies.
  • justaviking - Saturday, February 13, 2016 - link

    Sorry to say this, but...

    Very first sentence = Grammar error.

    "We've got some more information..."

    "We have got," is not correct. It should be "We got some..."

    I'm far from perfect, but then again I'm neither a journalist nor an editor. People who are should be held to a higher standard. It's not just here. I can hardly read a single news article (CNN, Huffington Post, Fox News... anywhere) without encountering at least one major error. Bad grammar, missing words, all sorts of problems.

    Sorry to rant, but when the second word on the page (not counting "Update") is wrong, I just had to say something.

    The rant is over. I love AnandTech. I'm going to finish reading the article now.
  • jjj - Saturday, February 13, 2016 - link

    Not so much in English as opposed to American English and since American English is just a dialect, one can argue that it's not the best choice.
  • stephenbrooks - Saturday, February 13, 2016 - link

    Huh? They're both correct. One is simple past tense "we got some information [in the past]", one is present perfect "we have got some information" (replace "got" by "obtained" to make this clearer).
  • justaviking - Sunday, February 14, 2016 - link

    It is probably no surprise I regret posting this.
    But as long as I did, I have a question and a note to Billy.

    Wouldn't the equivalent of "we have obtained" be "we have gotten," not "we have got"?

    I'm not an English teacher. But I see so many horrible transgressions in "professional" writings it irritates me, and this *minor* thing caused me to vent my frustration earlier today.

    @Billy Tallis - I am sorry. Correct or not, my comment probably sounded much harsher than I intended. It was a good article, which is why I visit AnandTech.com daily.
  • mikato - Monday, February 15, 2016 - link

    Yes.
    "We've got some more information..."
    should be "we've gotten", or "we got", or how about "we have". Seriously, just say "we have" if you have it.

    I admit I find myself saying things with "got" too often when I should just say "have". But it is natural to want to reference the act of getting something with "got" as it provides slightly more info (usually implies somewhat recent acquiring of the object and through an action taken and not just passively receiving it somehow).

    Here's a different kind of simple mistake... missing a "be"-
    "The first 3D NAND Micron has ready for the market will produced to the endurance standards for client drives"

    I do often see much worse in traditional news outlet online articles, and it bugs me too. Missing words all over the place sometimes.
  • Nexing - Monday, February 15, 2016 - link

    What about nms?
    The article just says "Micron's first generation 3D NAND takes the form of a 256Gb MLC die and a 384Gb TLC die (compare with their 128Gb 16nm MLC and TLC)" and two other comparative mentions to nm measures. No single reference to the litography actual size of this 3D nand.
    Samsung 3D nand, so succesfully represented by 850 pro line, has had a high level of success -at its base- because it returned to 40nms, hence avoiding all the endurance and rather short term instability problems brother by smaller die size (iex 840 gen) and then taking advantage of stacking, lower power consumption,etc. that 3D nand provides. Therefore, what is the actual Micron data in this quite relevant regard ?
  • Kristian Vättö - Monday, February 15, 2016 - link

    The cell size is estimated to be similar to Samsung's V-NAND, so we should be looking at something between 30 and 40nm.

    http://www.3dincites.com/2016/01/2015-retrospectiv...

Log in

Don't have an account? Sign up now