Micron and Intel Announce 3D NAND Flash Co-Development To Push SSDs Past 10TB

MojoKid writes Both Micron and Intel noted in a release today that traditional planar NAND flash memory is reaching a dead-end, and as such, have been working together on 3D memory technology that could open the floodgates for high densities and faster speeds. Not all 3D memory is alike, however. This joint development effort resulted in a "floating gate cell" being used, something not uncommon for standard flash, but a first for 3D. Ultimately, this 3D NAND is composed of flash cells stacked 32 high, resulting in 256Gb MLC and 384Gb TLC die that fit inside of a standard package. That gives us 48GB per die, and up to 750GB in a single package. Other benefits include faster performance, reduced cost, and technologies that help extend the life of the memory.

Comments

Re:Storage space isn't the problem.

[Anonymous Coward]

Simple. Don't use SystemD! Get one of the BSD's.

not the problem

[slashmydots]

A monkey could configure a 10TB array right now and power isn't exactly a problem. Putting it in a single drive is neat but the #1 problem with SSDs right now is price. The prices are horribly inconsistent day to day. They can make a 2Tb or 10Tb or 10000TB drive for all I care but what I need for my many, many custom builds at my shop is a low cost 240-256GB SSD.
Once in a while I can get a $90 silicon power S60 240GB SSD. Crucial's MX and BX series hit that low once in a while. All others are perpetually above $100 which is too expensive for a Facebook wonder do-nothing PC with a pentium 4th edition and 4GB of RAM. Some people do reasonably go past 120GB too so I do typically want to use 240GB drives. I blame smartphones' cameras and itunes' automatic backup of ipads and other devices.

Re:not the problem

[LordLimecat]

For a do-nothing pc you should be using any of the hundred sub-$60 128GB ssds.

Re:not the problem

[beelsebob]

You have a faulty SSD if you've got to 12% worn in 1 month. For a non-faulty 840 EVO 1TB that's a physical impossibility. 12% wear would imply that you've written more than 36TB to the drive already, which implies writing continuously at 833MB/s, which is higher than the drive's maximum write speed.

Re:not the problem

[LordLimecat]

What are you doing that's writing ~100-200TB / year?

Re:not the problem

[Christian Smith]

Centralized remote syslog server that also feeds to Elasticsearch which indexes the syslog entries for reporting and searching. After daily aggregate reports are done, a week later the entries are moved to spinning rust and only the canned reports stay on SSD. With the way Lucene works and our indexing is setup, a 100 byte syslog entry can easily turn into 2k of index. The canned reports now only take a tenth of the time they took before we added the five Samsung drives and the historical reports load even faster than that, but the drives are just wearing out too quickly for comfort.

So you're simply using the wrong drives. The 840 EVO is a consumer level drive. You want a more enterprise like drive like Intel 3700, which is warranted for 10 drive writes per day for 5 years. It costs more, sure, but you can surely justify the cost based on the performance and reliability benefits.

Re:not the problem

[guruevi]

You're doing it wrong! Unless you don't care about data loss and calculated out that replacing the EVO's every few months is actually a cheaper option compared to a decent SLC (Intel, STEC, ...) or even RAM SSD over the lifetime of your server, then I would recommend re-examining your setup. SLC's are not only faster, but they're a heck more reliable. If you just care about speed on a single box, get a PCIe based SSD. Desktop drives for this kind of setup is asking for trouble.

I still have a set of 32GB SLC (Intel X-25-E) from 2009 which have digested several PB worth of data in their lifetime as well as a number of OCZ Deneva's with the same workload. Probably the best investment possible.

Re:not the problem

[Anonymous Coward]

Oh please. I've replaced several hundred worn-out consumer SSDs in servers. They wear-out quickly. There's a reason Dell charges $4,998.78 for their write intensive 800 GB SSDs. Poor quality drives like the Samsungs or Intel consumer models often wear-out in less than a year. The Dell drives cost ten times as much as the Samsung, but they're worth it since the Samsungs wear-out so quickly. If there wasn't a problem with the Samsung garbage wearing out so damn quickly, why would anyone ever pay ten times as much to get something that doesn't wear-out so quickly?

Re:not the problem

[Billly Gates]

Nope benchmarks show superior reliability

Likely your drives didn't wear out. No trim in raid with anything under Windows 2012 without intel RST driver screwed up wear leveling as it lost track of virtual rewrites

Re:not the problem

[bloodhawk]

So your doing enterprise level syslog aggregation and archiving on consumer level hardware? really?

Re:not the problem

[fnj]

$ smartctl -a /dev/sda | grep Wear
177 Wear_Leveling_Count 0x0013 088 088 000 Pre-fail Always - 136

Please post also the bytes written data from smartctl so we can get some idea of what crazy torture you are or are not subjecting it to.

Re:not the problem

[sonicmerlin]

Then buy a Samsung 850 Evo Pro. The 3D NAND is at 40 nm and is extremely long-lasting.

Re:not the problem

[jones_supa]

A good consumer drive could take it just fine.

Re:not the problem

[halltk1983]

Yes, there is. Dual SAS channels, allowing the disc to be controlled on independent control channels from redundant controllers, allowing improved uptime.

Re: not the problem

[halltk1983]

Get more RAM and disable the page file. I did that years ago, and it works just fine. I've had *one* OOM blue screen in 4 years, and that was because I was seeing how much I could open at the same time.

Re:not the problem

[Coren22]

Just to counter this. I have an Intel I bought on 7/23/2009 (looked it up on Newegg) that I don't use much as it is only 32 GB. I also have a 120GB Intel (with a Sandforce controller) from 8/25/2012 that is still working just fine. Intel and Samsung are on the top end of SSDs as they own their own fabs, so don't badmouth without actual evidence.

Re:not the problem

[Coren22]

You want a more enterprise like drive like Intel 3700, which is warranted for 10 drive writes per day for 5 years

Is that 10 a typo?

Re:not the problem

[Coren22]

What is wrong with Newegg? They tend to have the most selection of computer hardware around. Sure you could buy from Ingram Micro, but there is no guarantee that they have any larger selection, and you will pay more.

Re: not the problem

[Coren22]

On top of what Halltk says, if you install Windows 7 and above to a SSD, it auto optimizes for SSD (no page file, turns off defrag, and I am sure other optimizations)

Re:not the problem

[ejasons]

Is that 10 a typo?

Entire drive capacity writes...

Re:not the problem

[LordLimecat]

I might suggest that you need to not be expecting a consumer SSD to hold up to an enterprise workload.

Re:not the problem

[LordLimecat]

He could have gone with the Samsung "Pro" drives, which have held out to 2PB of data writes before croaking; they at least pretend to not be consumer drives.

Re:not the problem

[fnj]

LBAs written are in sectors (512 bytes), so about 37B sectors or 19 TB. The Flash has written (actually erased) about 136TB, so you are running with a wear amplification of about 7:1. If the drive is full, this is not too surprising. The way to get Flash to last longer is to leave free space so that the FTL has room to work in.

Nailed it. This drive is being tortured. 19 TB in one month is 630 GB (almost 2/3 of full drive capacity) written per day, or 7 MBps averaged 24x7, on a budget drive.

Over-provisioning at the outset would have helped a lot, but that is still a hell of a big data load.

Tipping point?

[BenJeremy]

Hopefully this also sees a reduction in the cost of SSDs to bring them closer in line with platter drives, which have only just started dropping into the $30/TB range once more (since the Thai floods gave manufacturers their own Sumitomo excuse to drive up prices).

If the market had progressed more realistically, platter drives would be $15/TB and we'd already have consumer-level 10TB drives, but Seagate and Western Digital took a breathing period to reap profits, allowing SSD technology to start playing catch-up. ...not that SSD makers are off the hook... they've gone to smaller fab processes that shortened the life of NANDs and also have kept prices from falling at a reasonable rate, too.

I think we are two or three breakthroughs from reaching parity on cost per byte for platter and solid state tech, at which point, platter technology will likely become a very small niche market.

Re:Tipping point?

[mlts]

Platter technology will end up being pushed to the NAS/SAN, which is why WD is making their red line of drives.

Perhaps HDDs, now that speed and capacity are secondary, they will start evolving down the path of reliability, perhaps replacing tape as an archival medium.

NAS drives are going to be a big market, especially with devices like Apple's new MacBook with limited expansion capability, so people will use WiFi Direct hard drives as their main backup source, as opposed to USB drives. In this use, capacity is limited on the MacBook, and speed is limited, so drive makers (hopefully) will end up working on leapfrogging each other for reliability and security.

Re:Tipping point?

[Anonymous Coward]

Chip fabrication tech and flash density is increasing at a rate far faster than spinning platter storage. At mass market scales, SSDs are going to win pretty soon because of simple BoM cost.

Hard drives require a lot precision machining and assembly. Chip fabrication and PCB assembly scales much better, and is much cheaper. It's the same pick-and-place followed by oven reflow that /everything/ else uses.

Flash storage is going to win by size, logistics, power usage, and mechanical requirements. Why devote space in your device to a hard drive when a postage stamp sized module soldered down next to the SoC that powers your product is cheaper and smaller?

SSDs are getting tiny already. Have you seen what's inside a modern SSD? A tiny board with 2-3 chips. Most of the inside of that 2.5" drive chassis is empty air. WIth intel talking about 750 gigs on something the size of your thumbnail.. Hard drives are done. Just done.

Reliability is trivial too. Modern controllers with modern wear leveling algorithms let you sacrifice a little storage space to give you as much or as little reliability as you'd need, depending on your target market. Even if your flash cells are less reliable, the massive increase in density lets you devote more of them to wear leveling.

Re:Tipping point?

[swb]

I think major leaps of density will eliminate platters. Why bother with them at all with their ridiculously slow seek times, heat, power consumption? At high capacities they're more of a risk to data integrity due to slow array rebuild times and it takes dozens of them to equal the IOPS of flash. Even now platters are either useful for their high density as Tier 3 in a SAN or in large numbers to get IOPS.

If there was a huge leap in flash densities I think they would get cheap enough that no one would bother, even if they were "unreliable" consumer MLC technology. Vendors could just double the extra flash used for recovery of bad cells and increase the endurance.

Re:Tipping point?

[Anonymous Coward]

A tiny board with 2-3 chips. Most of the inside of that 2.5" drive chassis is empty air.

That's because of the mSATA format. You can already buy a 1 TB mSATA drive. On raw size alone you can fit 7.1 of those into the same space as a 3.5" hard drive.

They just need to get the cost down. Storage solutions in the future will be trays of those chips in a 1U-4U form factor with fans in front and some CPU and networking connections. No more 'hot swap' hard drive 3.5" form factors, just swap an entire blade of SSD chips.

Re:Tipping point?

[Billly Gates]

Where is cheaper and greater ram? We should be seeing 128 gb ram by now and ram drives and caches for faster performance

Re:Tipping point?

[bloodhawk]

64GB and 128GB ram is already available. You just don't see them at the consumer levels as their is no need or demand for them. a consumer benefits more from persistent storage, the advantages of ram drives is really non existent for all but a very small niche of users. Server side we have had this for a long time. The two servers I am currently playing with have 4TB of ram.

Re:Tipping point?

[jones_supa]

Hard drives require a lot precision machining and assembly. Chip fabrication and PCB assembly scales much better, and is much cheaper. It's the same pick-and-place followed by oven reflow that /everything/ else uses.

Good point.

Re:Headline != Story

[edmudama]

750GB per package.

A single SSD may have anywhere from 1 to $alot of packages on the board, hence 10TB SSDs.

Re:Headline != Story

[BronsCon]

This. We already have >1TB SSDs, so clearly they're already using $alot of the current 750GB packages to do that.

Why SSD in a "do-nothing" PC ?

[perpenso]

All others are perpetually above $100 which is too expensive for a Facebook wonder do-nothing PC with a pentium 4th edition and 4GB of RAM.

Why use an SSD in such a do-nothing PC? If you can't go with a regular HD try a hybrid SSD-HD. Last I looked a hybrid with 1 TB HD and 8 GB SSD was under $80.

Re:Why SSD in a "do-nothing" PC ?

[mlts]

Had a similar choice when giving a laptop to a relative. I went SSD instead of SSHD because SSDs are physically more resistant to shock.

However, if given the choice with a desktop... I'd probably still use SSD, just because when I delete a file and fstrim the drive, the file is -gone- for good, since the drive controller will come around, write "1"s to all the pages that file used and call it done. Of course, keeping good backups when using SSDs is wise, just due to this exact thing.

Re:Why SSD in a "do-nothing" PC ?

[RobinH]

I have upgraded 4 computers from HDD to SSD's since Christmas. They were all from 2 to 5 years old, and all of them run like they're a brand new computer. They boot amazingly fast, and they launch programs really fast. Compiling is much, much better too. I wouldn't ever go back to a rotating hard drive for anything other than long term archival storage now. Maybe I'd do a hybrid drive, but really around $100 or so for 240GB is a really nice sweet spot at the moment.

Re:Headline != Story

[BronsCon]

Damnit, that was <750GB.

i can't wait!

[Gravis Zero]

At this rate I'll finally be able to have my entire music collection on an iPod without having to compress my music in that terrible FLAC format. FLAC is a "lossless" format but you can totally hear which bits have been squished into the file for too long! That's why I decompressed all my files and let them sit for a week, so that the bits can breathe.

Re: i can't wait!

[DigiShaman]

You joke, but for some of my favorite music I don't mind leaving it in WAV format. I can afford the space now.

Re:i can't wait!

[smprather]

This truly is the way "audiophiles" think. I'm still emotionally scarred from the crap I read in Audiophile magazine(s) while in high school. https://www.youtube.com/watch?...

Re:i can't wait!

[maestroX]

Be sure to put the disk upright to avoid 1 bits accidentally falling over.

Re:Headline != Story

[beelsebob]

750GB is the amount they can fit one one chip package - i.e. what they could fit on a SD card. A typical gum stick SSD has several of these (usually around 5), hence the 3-4TB per gumstick estimate. A 2.5" drive will typically have more like 12-15 of them, hence the 10TB estimate.

NAND flash is still expensive garbage...

[Anonymous Coward]

More layers of NAND will just multiply the already steep cost. With process shrinks and MLC, flash endurance is falling sharply, and the bits literally leak out over time. In addition to being physically unreliable, the complex software stack required to transform block-erasable flash into useable storage does not inspire confidence, especially when virtually no drives guarantee data integrity with loss of power.

Far more interesting would be solid state storage based on memristors or phase-change memory. It would provide excellent performance and reliability without the complexity and limitations of flash-based SSDs.

Re:NAND flash is still expensive garbage...

[Yunzil]

With process shrinks and MLC, flash endurance is falling sharply, and the bits literally leak out over time.

Uh, that's kinda the point of 3D. Instead of shrinking the cells, just stack them up. It's what Samsung did with their 850 Pros, which have a 10-year warranty. Besides, the TechReport torture test shows that reliability really isn't a problem for consumer-level devices.

Does this mean

[Anonymous Coward]

they are optimizing the design to be 3D printed in my living room? I'm already 3D printing my house, and the car in the garage, it would be nice if we could 3D print a complete house including the laptop in one pass?

Re:Does this mean

[Em Adespoton]

Most microchips are 3D printed... you just wouldn't be able to afford the printer they use, and the chemicals involved would likely make living in your house a thing of the past..

Re:Does this mean

[Em Adespoton]

Whoosh?

Re:licensed from Toshiba?

[Em Adespoton]

Samsung has yet another 3D method they're using in PCIe boards, and those are already shipping in Apple's MacBook Pros.

Competition is good... especially in this field!

Re:Storage space isn't the problem.

[ChrisMaple]

Keep more than one version of Fedora on your computer, in different partitions. I currently have F14, F17, and F20. Use multiple drives, so that worst case there are multiple boot drives available from BIOS. Keep a "Live CD" available on a USB flashdrive.

I'm sticking with tape

[OrangeTide]

Good old reliable tape. None of this fancy random access hard disk garbage that fails all the time, or complicated wear leveling flash nonsense.

Maybe something like core memory or bubble memory if I need some random access behavior.

I hear it's down to a penny per bit, only around 1200 megabucks for 10 gigabytes of Core memory.

Re:I'm sticking with tape

[Anonymous Coward]

Tape is reliable, and core memory (in its modern incarnation as MRAM) provides immensely superior reliability and read/write performance. Unlike NAND flash, it offers true random access at a byte level granularity. NAND flash has its place, but it is the CD-RW of solid state storage technologies.

NAND flash uses a pile of emulation software to fake hard disk semantics, and it degrades reliability and performance. Anything less than a quality enterprise SSD with capacitor has no chance of implementing such a complex piece of software correctly. Even then, those are only useful for transient data storage, and will forget everything if left on a shelf for a few years. It is not wise to use NAND flash without understanding its limitations, and for valuable data, a checksummed filesystem like ZFS and backups are a must.

Personally, I'd rather pay more for a good quality SSD, and stick with (non-SMR) hard drives for bulk storage. Economically competitive SSDs will remain a dream for many years, and I wouldn't trust the commodity ones any more than inexpensive USB hardware. If ignorance and insults are your thing, feel free to learn by experience.

Re:Storage space isn't the problem.

[hairyfeet]

ROFL...needing THREE installs of the same OS along with a LiveCD just to keep a working system thanks to systemd, while the other guy has to use XP just to Google for fixes for his "modern and new" 2015 Linux install? Yep, its soo ready for the desktop ha ha ha ha! Been saying for years, if you don' t demand better, hold asshat devs like Poettering's feet to the fire? Then you deserve the half baked mess that you get. No wonder The Hairyfeet Challenge is celebrating its eighth year unbeaten!

As for TFA....how are the MTBF for 3D NAND? Did they manage to lick the "controller fails and takes out the drive" issue? Because while the speed of SSD is great one thing that royally blows ass is how you get fuck and all for warning before they shit themselves and die. I'm sure some jackhole will pop up with some anecdote (while neglecting to mention he dropped the thing) about "his HDD just died" but since I've done more HDD replaces than many here have had home cooked meals and by and large? You get plenty of warning with HDDs. You get write errors, you get noise, stutters, they will usually give you enough time to get your data off...not SSDs, I thought Intel had the right idea by giving them a finite lifespan but on that big SSD shootout the Intel one did the "no BIOS/UEFI" brick bit just like the rest.

I mean as much money as they are spending increasing size, is it too much to ask to have a little "failover" chip that just leaves the drive in a read only state so you can get your stuff off if it shits the bed?

Re:Storage space isn't the problem.

[Billly Gates]

Actually Hairy this was tested .

Ssds are more reliable than platters these days. My 2012 Samsung pro in raid 0 still functions. There is a free tool out there that gives you health and life of a ssd reported I had til 2025 before it goes kaput. Times are changing and sand force is gone.

Try it ... and system D guy was troll and off topic trying to start a flameware. SystemD and other init replacements were hip since 2005 when Apple did theirs ... until last summer when sys admins who find nothing wrong with dozens of 200 line programs all linked created thousands of threads upon startup felt threatened

Re:Storage space isn't the problem.

[fnj]

I built a FreeBSD 10.1 server with a single root ZFS pool consisting of seven 3 TB drives in RAID-Z3, including a small 7-way mirror of swap space. The process was completely pushbutton using the install UI. Partly I did it just to explore how much difficulty the install might be (no difficulty whatsoever), but the setup has proved very effective in use.

It was pretty cool rerouting some of the SATA connections randomly (even to a different HBA) as a test, and removing two of the drives as a test, and having it still boot to a fully operating state and run fully usably with no intervention or drama at all.

After extensive experimentation and production use, the only real criticism I have that is just head-scratchingly stupid and lame is that there is no sensor capability, and no one in development seems to think there is a glaring problem with its omission. Coretemp works beautifully, but you can't detect fan rotation or access any voltage or secondary temperature sensors.

Re:Storage space isn't the problem.

[fnj]

Simple. Don't use SystemD! Get one of the BSD's.

At least spell ir right, you guys, so you don't look like idiots. It's systemd, not "SystemD". Sheesh! Does anyone write FtpD or HttpD or SmtpD?

Re:Storage space isn't the problem.

[Celarent Darii]

For monitoring motherboard stuff:

http://www.freshports.org/sysutils/healthd/>systutils/healthd

I see your 900 TB

[justthinkit]

I see your 900 TB and raise you to 2 freaking petabytes. Not quite a min raise, I realize...

Re:Storage space isn't the problem.

[fnj]

Thanks for the tip. I'll check it out. The proper URL is http://www.freshports.org/sysutils/healthd/. Unfortunately all the links at that page are now dead and "there is no maintainer for this port". I will try it, with trepidation, though all other sensors related ports for FreeBSD appear to be garbage.

Re:Storage space isn't the problem.

[GCsoftware]

Yeah spell it right - it's 'systemd'. Looks like an OK operating system, shame it doesn't have a decent init.

Re:Storage space isn't the problem.

[hairyfeet]

Sigh...please read TF post before replying...kay? I did NOT say that SSDs were unreliable, hell I use 'em myself as boot drives (because I don't give a shit if I lose the OS, won't be my first Windows install after all) my problem is they tend to fail unexpectedly. And that test? Yeah I'm afraid that means jack and shit, after all if you had run a test on Maxtor and only ran it on 2002 and 2004 it would have shown they were very reliable, 2003 was when they had the bad batch.

Look nothing you are saying changes the fact that its NOT the flash that fails by and large ITS THE CONTROLLER and with current designs if the controller fails? You boot up and no BIOS or UEFI detection. please don't take MY word for it, Google "SSD no BIOS detection" and just replace BIOS with UEFI after the first pass and see for yourself this is NOT an isolated incident and unless you have own EEPROM reader and the skills to rebuild the array? You be fucked son.

At the end of the day the absolute worse case scenario? I spend an hour building a clean box and switch the platters long enough to get the data, i did that not too long ago for a woman who had the only copies of pics of her recently deceased nephew on a drive that fried its electronics, if that would have been SSD? It would have cost her thousands if she could have even found somebody to do it. Its 2015 man, there is no damned reason they couldn't put a backup controller that turns the drive into a WORM medium and alerts the user, no reason at all.