news samsungs 3d v nand stacks multiple flash layers enters mass production

Samsung’s 3D V-NAND stacks multiple flash layers, enters mass production

NAND flash memory typically employs a planar structure with a single cell layer. For years, we’ve heard whispers about three-dimensional memory with multiple cell layers. Now, Samsung has started mass-producing it. The flash giant has started manufacturing what it’s calling the "industry’s first three-dimensional (3D) Vertical NAND." Dubbed V-NAND, these initial chips weigh in at 128Gb (16GB) each—the same density as the flash employed by Samsung’s latest 840 EVO solid-state drive.

Instead of using a floating gate to store the electrical charge associated with data, Samsung’s V-NAND traps electrons in a non-conductive silicon nitride layer. This charge trap architecture is supposed to cut down on the cell-to-cell interference that makes it difficult to scale gate-based flash memory down to finer lithography. V-NAND’s cell layers are connected with a "proprietary vertical interconnect" that can stack up to 24 layers.

Unfortunately, there’s no word on how many layers are used in the first V-NAND chips. We’ve asked Samsung to elaborate—and also to shed light on the memory’s die area and fabrication process, neither of which are mentioned in the press release. The 128Gb planar NAND in the 840 EVO is fabbed on a 19-nm process, and the 64Gb chips in the 840 Series and 840 Pro are made using 21-nm tech. V-NAND is probably in the same ballpark.

Thanks to its unique cell structure, V-NAND promises 2X-10X "higher reliability" than Samsung’s 19-nm NAND. The 3D flash also boasts twice the write performance of its planar peer. Faster, higher-density chips are in the works, as well; Samsung claims its V-NAND tech is capable of cramming a terabit (128GB) onto a single chip.

Samsung doesn’t say how long it will take for V-NAND to make its way into consumer products, but we probably won’t be waiting for too long. Less than five months passed between the introduction of Samsung’s 128Gb TLC NAND and the 840 EVO SSD based on those chips. If Samsung can manage a similar turnaround with V-NAND, we could see the three-dimensional flash pop up in solid-state drives early next year.

Update: Samsung has confirmed that its first V-NAND chips use 24 layers to reach 128Gb. The company wouldn’t tell us the chip’s die size or its exact process geometry. However, according to Senior PR Manager John Lucas, the fabrication process is "roughly the equivalent of a planar process technology in the mid-teens."

0 responses to “Samsung’s 3D V-NAND stacks multiple flash layers, enters mass production

  1. Yeah and like that EETimes article said, that testchip was on 110nm silicon. I don’t care how many layers you have on it; at 110nm it’s not going to be competing with sub-20nm NAND on price.

    Crossbar’s RRAM seems like one of those promosing future solutions that still need to prove that they can be scaled down in size and mass-produced reliably before they can disrupt the incumbent technologies (much like graphene). They have a long way to go.. Maybe another five years

  2. Yuh… And why would they wait to release the tech at a later date if they know it’s going to be obsolete in a couple years? You’d want to milk it for as much as possible before it’s outdated, which is already slated. From the article RRAM has working prototypes out already as well, so maybe sooner then that.

    [quote<]According to the Crossbar press release, a working test chip has already been produced. Company CEO George Minassian says Crossbar RRAM is "easy to manufacture and ready for commercialization," as well. [/quote<]

  3. Well, it is conspicuous that Samsung has this giant Pro-sized hole in their renewed lineup post-EVO release.

  4. [quote<]I would even venture to say that it wouldn't have been 'ready' for prime time if it wasn't for RRAM that just popped out yesterday, which definitely looks to be something that would instantly surpass this type of memory and obsolete it.[/quote<] I doubt the Crossbar RRAM is anywhere near commercial feasibility at this point. Meanwhile, this Samsung chip sounds like something we can buy in SSDs in less than a year. Crossbar marketing is doing well, but I think they are mainly fishing for an acquisition

  5. [quote<]on thursday I bought a new McLaren F1 after I been earnin $8551 this past five weeks and-over, ten/k this past-month.[/quote<] So funny

  6. I’d believe that Samsung would be replacing traditional NAND fab lines with V-NAND according to their regular upgrade schedule if the production costs per wafer are even close. As far as price is concerned, they’d be remiss to not recoup their development costs by capitalizing on the increased margins while the competition catches up!

  7. MRAM is also a resistance based memory, and that’s scheduled for late this year or early next year. Not for finished products, but for chips to be coming off the belt.

  8. [quote<]So what do you think the chances are Samsung has been sneaking these into their past drives?[/quote<] Zero. [quote<]There is no way for people to test the flash on the drives to see what type it is (or have went out of their way to do so), they have to take the manufacturers word for it. [/quote<] You also have no proof that those drives aren't being powered by NANO-UNICORNS! Given the choices, I'm pretty sure that nano-unicorns are FTW in these Samsung drives.

  9. Read speeds on the Crossbar RRAM is also quite a bit faster. The question is can they get this technology up to the production levels needed to support the consumer market and will they charge a premium over NAND. If they can keep the costs in line (so no more 2x NAND prices) and get production up then I see this doing quite well in the marketplace. It won’t replace NAND unless the prices are similar but it can still do quite well if it performs as they say at a reasonable premium over NAND.

    At least the Samsung development shows that they haven’t stopped improving NAND. It will be interesting to see what other improvements might be coming before RRAM makes it into the marketplace.

  10. I’m sure he’s aware of that. Perhaps you should read the article again? They were talking about 16 GB (128 Gb) and 1 Tb (128 GB) chips. 😉

  11. You can get many microcontrollers in DIL packages too, but it’s not exactly cutting edge technology. ARM chips are a step in the right direction, tho. 😉

  12. [quote<]What I want is speed. It's downright stupid how many phones with lots of memory and fast processors that are crippled by mediocre storage implementations.[/quote<] Which is sometimes brought up as a reason for not having SD card storage. It's probably just an excuse though.

  13. So what do you think the chances are Samsung has been sneaking these into their past drives? There is no way for people to test the flash on the drives to see what type it is (or have went out of their way to do so), they have to take the manufacturers word for it. Samsung would then be able to reap all the benefits from this type of memory while the rest of the industry is ‘stuck’ on normal flash.

    I would even venture to say that it wouldn’t have been ‘ready’ for prime time if it wasn’t for RRAM that just popped out yesterday, which definitely looks to be something that would instantly surpass this type of memory and obsolete it.

    I’m all for new things, the timing of this just seems really fishy. Samsung is the one holding the patent on the memory, the maker, and the user of the stuff… Makes it seem like it wouldn’t be very hard to abuse it.

  14. Keep your skibbys on for SATAe dude. It’s right around the corner, no reason to wish for something that will need drivers and uses precious ports up.

  15. They are 16Gb and 128Gb. A B is Byte and a b is bit. This may help legitimize your ramblings a bit more.

  16. I’m guessing this 16GB chip is 3-layer. Just going from the max of 128GB per chip with 24 layers.

    Quite interesting they’ve already entered mass production. It should require a whole new manufacturing setup and must have taken quite some time. It will probably be expensive at the start so they can create more factories with the same setup for this.

    So, 5 months maybe we see it sold and at a premium. Then, 1 year later it should end up being cheaper and the new standard for flash memory.

    Given the EVO’s 1TB capacity and using the same density chips … maybe these will enable drives up to 8TB.

    Twice the write performance is nice too. It will probably end up doing 12Gbps+ read/write and use the SATA-Express connection.

    Around the same time we should start seeing stacked GDDR memory on graphics cards too (1-2 years).

    Exciting times ahead 😀 .

  17. According to the [url=<]Crossbar RRAM press release[/url<], the write speeds at least should be quite a bit faster. And they've already prepped for the first phase of development, since they've already successfully fabbed a crossbar array at a commercial foundry.

  18. Thanks for contradicting the article without providing any details or references. Are you trying to troll?

  19. This stuff will likely be quite a bit less reliable, quite a bit cheaper, and somewhat slower.

    Beyond those vague general statements there isn’t enough public info. out there to give you more concrete answers at the moment unfortunately.

  20. The RRAM that is likely to be one of those “3-5 years away” types? Wouldn’t hold up much hope.Anyway, they will almost certainly just get bought out by a semiconductor company for their patents. Then we won’t hear anything about it for a while. As is usually the case.

  21. “A DIL package! What a pleasant surprise even if it’s only for a prototype. :)”

    I can see Samsung getting into a patent dispute with McDonalds over this as weren’t they the first to use a Dill pickle on package and that was multi-layer also.
    I think Ronald and Co have moved to 65nm with cheese these days so Wendy’s have them beat as they are still using the Taiwanese Miniscule Cheese Company (TMSC) so are stuck on 130nm. Sometimes thicker is better, hey girls!
    Advanced Micro Dill have announced a hybrid cheese/pickle design which apparently offers a lot of synergy and improves the taste buds ability to process certain ketchups.
    Unfortunately it has been delayed as the engineering team are suffering with a cold so can’t taste anything. On top of that it seems that ketchup manufactures are finding it difficult to develop new flavours that take full advantage of this synergy.
    In response to this AMD have hired Flavor Fav out of Public Enemy to lead their promotional campaign and to develop a new ketchup for in-house use by their engineers. This cocaine based ketchup is said to increase productivity significantly according to AMD’s Roy Taylor’s official blog:

    “Dudes and babes, Yo, how’s it hanging? We are stoked at AMD’s crib as we iz getting a phat amount of work done. This new ketchup is da bomb. Hammer time is officially over as we are now on Flavor Flav time which means we is so kool we often don’t even turn up for work. Not because it’s bad at work because we have loads of sexy and exciting times here, weez just too busy clocking other tings. Laters dudes and dudettes.”

    The International Nazi Taste Engineering League (Intel) have released an official response to this threat to their stranglehold on the throats of consumers via their chief Twit and Pickle Head WILL.I.AM.

    “click click click click click click… click”

    We are waiting for the NSA’s crack team of morse code crackers to decipher his communication before we can report on its meaning. Our Moscow correspondent will get back to us once we have news of developments.

  22. I wonder how this will stack up against the RRAM that is just around the corner as well.

  23. I blew one of those by attaching a printer while the Amiga was on. As a replacement I used a 6526 from a C64. Everyhing worked except for a double click in the Workbench. Probably because the 6526 has slightly different timers as the 8520.

  24. Well, at first I was looking for the bond wires too but then I thought that for a flip chip the connections would be on the other side of the chip. Now that I think of it the 54 pins of the DIL package would never be enough for this chip. So yeah, very likely just a prop, photoshopped or build.

    Edit: Oh god, I just noticed the empty pads for the bond wires. It’s all bull. Why are they doing this? Why do they tease us with a DIL package?

  25. Yeah, I’m aware of the PCIe SSD RAID cards; that’s a workable solution as long as Trim isn’t compromised, but a native PCIe SSD controller like Samsung built for Apple is more along the line of what I’m looking for, whether it plugs into a slot or into a port through a cable.

  26. I could give a rats about capacity- any phone that doesn’t include a MicroSD slot is designed to extract higher margins from larger storage capacities, and I don’t think many people who primarily use their phone as a phone ever come close to actually running out of storage.

    What I want is speed. It’s downright stupid how many phones with lots of memory and fast processors that are crippled by mediocre storage implementations.

  27. That’s what I mean- one of the solutions that uses the PCIe bus, not necessarily the slots themselves, though there’s nothing wrong with that, necessarily, since putting a UEFI-compatible PCIe interface on and SSD controller is fairly easy, a la Samsung’s SSD for Apple.

    Addendum: Given that an SSD controller could be made as wide as needed to access any number of NAND channels, there’s no reason that we couldn’t have an SSD today that could saturate an x16 PCIe 3.0 slot, or any common subdivision.

  28. on 2.5″ SSD they would have to wait until SATA Express gets implemented by Intel/AMD, I am more intrigued on seeing these chips on Phones as the internal NAND. Hopefully with reduced cost we see 32GB/64GB as the default storage and much better I/O performance..

  29. And windowed–for no aparent reason. Strange, but I don’t see any wire bonds to the chip. Photoshoped prop. 🙁

  30. Depends on what you mean. Some “ultrabooks” have PCIe SSDs, but they’re advantageous for thin laptops that can’t fit or power multiple SSDs in RAID. That’s not an issue for desktops.

    So I guess a better question might be: when will PCIe SSDs be advantageous for desktops?

    Probably once there’s a standard native PCIe format. Right now, they use SATA bridge chips and multiple SSD controllers. It’s just RAID on a stick, with potentially increased latency, instead of reduced by the PCIe bus.

  31. At least the 80’s, anyway 🙂 The chips looks almost like one my old Amiga’s 8520 CIA chips…

  32. When consumers can afford them. i.e. unlikely. NGFF is a better solution anyways for the consumer side…

  33. Samsung had no problem spinning a new controller for the SSD used in the Macbook Air- it’d be very hard to believe that they wouldn’t be working simultaneously on V-NAND and a controller to make use of it in a shipping product.

  34. SSDs should take a longer while ,they need new controllers.
    As it is it’s pretty weird that Samsung is starting production now when others are planning for 2015-2016 because they believe only then it will be more cost effective than 2D NAND. So better wait and see what the chip can actually do and at what price before getting too excited.