Intel and Micron's new '3D XPoint' memory is 1000x faster, more durable than NAND

Posted:
in General Discussion edited August 2015
Semiconductor technology giants Intel and Micron on Tuesday announced 3D XPoint memory, the fruits of a joint endeavor into non-volatile memory technology the companies claim is the first major breakthrough in the space since the introduction of NAND flash in 1989.




On paper, 3D XPoint sounds like a cure-all for modern computer memory bottlenecks. The technology is said to be up to 1,000 times faster than NAND with up to 1,000 times greater endurance, while packing in 10 times the density of conventional DRAM components.

The culmination of more than a decade of research and development, 3D XPoint boasts what is described as a stackable, transistor-less "three-dimensional checkerboard" architecture in which memory cells can be addressed individually as they sit at the intersection of word and bit lines. This design allows the system to write and read data in smaller batches, which in turn facilitates faster and more efficient operation. And with low latency overhead, a single 3D XPoint module can serve both system and storage needs.

Perhaps most impressive is that 3D XPoint is not a one-off proof of concept, but is already in early stages of production.

"One of the most significant hurdles in modern computing is the time it takes the processor to reach data on long-term storage," said Mark Adams, president of Micron. "This new class of non-volatile memory is a revolutionary technology that allows for quick access to enormous data sets and enables entirely new applications."

Any number of computing applications can benefit from such fast and efficient technology. For example, 3D XPoint could accelerate machine learning systems, or enable medical researchers to study genetics and track diseases in realtime, the companies said.



Intel and Micron plan to release sample units to select hardware manufacturers later this year and plan to roll out independent product lines through 2016.
«13

Comments

  • Reply 1 of 55
    solipsismysolipsismy Posts: 5,099member
    The technology is said to be up to 1,000 times faster than NAND with up to 1,000 times greater endurance, while packing in 10 times the density of conventional memory.

    Any chance this could place the system NAND on Apple's A-series package for an even smaller logic board?
    Perhaps most surprising is that 3D XPoint is not a one-off proof of concept, but is already in production.

    That is both welcome and surprising.
    Intel and Micron plan to release sample units to select hardware manufacturers later this year.

    Great, but what kind of cost per gibibyte are we talking here? What kind of write count are we talking?
  • Reply 2 of 55
    It's a cure-all for modern computer memory bottlenecks?

    I think IGZO and Liquid Metal have competition in the unicorn department! The next iPad is doomed without it!
  • Reply 3 of 55
    MacProMacPro Posts: 19,728member
    Wow!
  • Reply 4 of 55
    Never underestimate Intel.
  • Reply 5 of 55
    doggonedoggone Posts: 380member
    Sounds very cool. Once the production costs go down this will spell the end of disk drives.

    My main question is how long until Samsung tries to steal the technology and try and sell it themselves.
  • Reply 6 of 55
    MacProMacPro Posts: 19,728member
    The Beeb have a good write up here... http://www.bbc.com/news/technology-33675734
  • Reply 7 of 55
    polymniapolymnia Posts: 1,080member
    Yes, please!
  • Reply 8 of 55
    flaneurflaneur Posts: 4,526member
    The Beeb have a good write up here... http://www.bbc.com/news/technology-33675734

    Thanks for the link.
    It's a cure-all for modern computer memory bottlenecks?

    I think IGZO and Liquid Metal have competition in the unicorn department! The next iPad is doomed without it!

    Snort. 3D is a gimmick.

    /s
  • Reply 9 of 55
    misamisa Posts: 827member

    Intel and Micron plan to release sample units to select hardware manufacturers later this year and plan to roll out independent product lines through 2016.

    Sounds a bit vapor-ware'ish to me. If Intel rolls out SSD's themselves to put the proof in the pudding, then maybe we will see something cost-competitive with NAND flash and finally see an end to mechanical hard drives.

    The reason that mechanical drives are still being produced at all is because NAND wears out, quickly. TLC has 10,000 writes, MLC has 100,000 writes, and SLC has 1,000,000 writes. NAND also loses it's charge over time, so without being refreshed every so often, eventually you end up with a "blank brick", not a "read-only" one as some people have thought. So for archival use, NAND isn't good at all.

    Xpoint might be able to replace NAND on the M2 PCIe storage, but it seems a bit silly to make SATA3 SSD's with it if it is indeed 1000 times faster.
  • Reply 10 of 55
    solipsismysolipsismy Posts: 5,099member
    misa wrote: »
    Sounds a bit vapor-ware'ish to me.

    Really? This sounds like the least vaporware-ish announcement about a breakthrough technology I've ever heard about.
  • Reply 11 of 55
    prolineproline Posts: 222member
    Quote:

    Originally Posted by EricTheHalfBee View Post



    Never underestimate Intel.



    Why not? Intel is far from invincible and some vaporware doesn't change that. 8 years after the iPhone's ARM debut, Intel hasn't been able to achieve any kind of dominance in mobile at all. They're human. They have their hits and misses.

  • Reply 12 of 55
    proline wrote: »

    Why not? Intel is far from invincible and some vaporware doesn't change that. 8 years after the iPhone's ARM debut, Intel hasn't been able to achieve any kind of dominance in mobile at all. They're human. They have their hits and misses.

    Intel got their butts kicked by AMD in the early 2000's.

    Have you seen AMD lately?
  • Reply 13 of 55
    Quote:

    Originally Posted by Misa View Post



    The reason that mechanical drives are still being produced at all is because NAND wears out, quickly. TLC has 10,000 writes, MLC has 100,000 writes, and SLC has 1,000,000 writes. NAND also loses it's charge over time, so without being refreshed every so often, eventually you end up with a "blank brick", not a "read-only" one as some people have thought. So for archival use, NAND isn't good at all.

     


     


    I think that the reason that mechanical drives are still being produced is actually, cost and capacity.  While there are now SSDs with 1TB+, HDDs are at 6TB+, with some Enterprise Drives reaching 10TB+.  The cost per GB is also vastly in favor towards HDDs, I don't have the exact cent/GB in front of me, but there is a major difference.

  • Reply 14 of 55
    frank777frank777 Posts: 5,839member

    Apple's contribution to this new technology should have been to give it a proper name.

  • Reply 15 of 55
    mdriftmeyermdriftmeyer Posts: 7,503member
    Nice to see the lack of EE and Physics holders posting on how awesome this breakthrough is when it's a rip off of a memristor and not that impressive.
  • Reply 16 of 55
    Never underestimate Intel.

    It's amazing what they can do when their CPU business is gone to shit...
  • Reply 17 of 55
    flaneur wrote: »

    Snort. 3D is a gimmick.

    /s

    Noooo! It can't be. I read it on the internets!
  • Reply 18 of 55
    thepixeldocthepixeldoc Posts: 2,257member
    misa wrote: »
    Sounds a bit vapor-ware'ish to me. If Intel rolls out SSD's themselves to put the proof in the pudding, then maybe we will see something cost-competitive with NAND flash and finally see an end to mechanical hard drives.

    The reason that mechanical drives are still being produced at all is because NAND wears out, quickly. TLC has 10,000 writes, MLC has 100,000 writes, and SLC has 1,000,000 writes. NAND also loses it's charge over time, so without being refreshed every so often, eventually you end up with a "blank brick", not a "read-only" one as some people have thought. So for archival use, NAND isn't good at all.

    Xpoint might be able to replace NAND on the M2 PCIe storage, but it seems a bit silly to make SATA3 SSD's with it if it is indeed 1000 times faster.

    Your "Death Star' argument is mute.

    From the BBC article. Read It. Get Informed. Be Amazed.™

    The advantage is that each memory cell can be addressed individually, radically speeding things up.

    An added benefit is that it should last hundreds of times longer than Nand before becoming unreliable.
  • Reply 19 of 55
    ksecksec Posts: 1,569member

    On BBC they site it will still be significantly cheaper then DRAM. Let just say it will be half the price..... ( I think significantly means even lower, but i assume 2x cheaper for now. )

     

    DRAM is trading to $2.5 per 4 Gb (512MB), going as low as $2.

    3D Xpoint could be $2.5 per 8 Gb ( 1GB ), a 30GB would cost $75, and retailing for $150 or $200. For something as revolutionary as this, it isn't expensive at all.

     

    But before we get ahead of ourselves, this essentially mean the whole stack of storage, RAM needs to be rethink and retested.

     

    Should Xpoint lives as a buffer on SSD? Or better yet lives with DRAM, with DRAM moving closer to CPU. ( Stacked DRAM much like HBM in GPU )

     

    Yes this is exciting, but it would properly be years before we get the full potential.  

     

    P.S - And it might turns out to be a perfect fit for Mobile Phones, When you dont want the constant refresh of memory dragging battery,

  • Reply 20 of 55
    palegolaspalegolas Posts: 1,361member
    Wow, great stuff! The fact that they're using medical research as an example makes it sound expensive though ???? But that's to be expected when launching new technology. I hope it's straight forward to implement in computers and phones etc. That there are versions that communicate over common busses etc.
    Is 1000x faster than NAND enough to be faster than current RAM memory?
Sign In or Register to comment.