About time too
1TB in a 2mm mobile package
10TB in an SSD
A 2.5in SSD enclosure is a lot more than 10 times the volume of a 2mm mobile package.
Something does not seem right here?
IMTF, Intel Micron Flash Technologies, a partnership between Intel and Micron, has a 3D MLC NAND technology, which will be used to build 10TB SSDs in two years. With 3D flash a die is made up from layers of ordinary or 2D planar cells stacked (as it were) one above the other. The news came in a webcast for Intel investors …
The enclosure is likely to contain control electronics and perhaps more redundancy than a mobile device. Plus there's less need to go with full-on bleeding-edge chips in a larger enclosure. IOW. consider the 1TB measure generous and the 10TB one conservative.
That said, it seems 3D NAND is about to emerge, and if the price points drop as advertised, this may be at last be the sunset of spinning rust as a mainstream storage tech. If the tech evolves to use more aggressive cells (shorter-lived but can hold more), it may even find a use as a bulk storage medium provided its caveats are kept in mind.
suggesting tablet and mobile phone storage capacity could increase
Will be interesting to see how this pans out with certain manufacturers and suppliers who would rather you paid for rental of cloud storage instead, and don't give that much local storage even now and make that non-expandable...
Even though it can be done it'll be interesting to see if market and user pressure means that it will be done, and we go back more to a local storage solution rather than a cloud/streaming one.
Mobile networks that offer unlimited data will love it. Those that offer metered data will hate it :)
Mobile networks could massively increase capacity within a year if they wanted to without additional spectrum. Increasing cell tower density, MIMO, simultaneous transmission and reception on the same frequencies (assuming neighboring frequencies wouldn't be affected), increase the number of sectors per site. It's in their interest to ensure they keep supply and demand finely balanced so as to ensure they can offer tiered products and also to keep their capex as low as possible which keeps the shareholders happy.
"Mobile networks could massively increase capacity within a year if they wanted to without additional spectrum. Increasing cell tower density, MIMO, simultaneous transmission and reception on the same frequencies (assuming neighboring frequencies wouldn't be affected), increase the number of sectors per site. It's in their interest to ensure they keep supply and demand finely balanced so as to ensure they can offer tiered products and also to keep their capex as low as possible which keeps the shareholders happy."
Trouble is, just about all those things you describe will require infrastructure investments: costly infrastructure investments (particularly more towers, which require permits, land/space acquisition, maybe regulatory clearances, etc.). Others have to wait for new phones to come onto the market capable of using the new tech, which means a lag time of at least a year. As for metered data, customers are already touchy about those since the "unlimited" genie left the bottle years ago. It won't be long before they hold the mobile companies to the promise. And any attempt to rescind the unlimited promise will be met with resistance: likely from competitors eager to cut in. So it's an even more delicate balancing act between keeping customers sated and raising enough capital to plunk down for those soon to be needed infrastructure investments.
So, in light of this and so presumably the ever more real possibility of the all-flash data center, what do we then think of this story from a couple of weeks ago right here in your daily El Reg?
Does this mean that although magnetic storage is really reaching some hard limits of physics, we have a very good alternative in this type of technology?
By the way - this is intended as a genuine question. I'm not sure myself about how this will turn out in terms of price / GB.
Kryder's Law is still on the way out due to the molecular limit, but if this tech pans out, a transition tech will soon be in place that will allow storage to continue growing; at worst, we'll experience a brief hiccup as there is a brief gap between spinning rust running out of steam and 3D NAND hitting the mainstream.
OK, I should have asked my question a bit more accurately. Kryder's law is something only relevant to magnetic storage, but a purchaser of a hard disk doesn't care what's in the box - iron oxide or silicon. It's a way to store bits with a cost, power requirement, reliability and performance characteristics. In fact, generally they prefer silicon considering seek times, noise and power - although the reliability part is moot.
It's possible to considers a revised Kryder's law that is about bits per cm3 - that would be a better way to phrase the question probably.
The article I was referring to suggested that business models based on the concept of ever-cheaper storage were soon doomed. If we believe in a 3D flash future, then maybe these business models aren't so bad as the article suggested - at least for the next decade anyway.
"There has been no iron oxide in hard disks for over a decade at a minimum. "
It's a metaphor. Language is based on the use of metaphors to represent new things by overloading the meaning of old words. The word "computer" was originally used for people who did calculations.
A "screen" was a relatively 2D construction used to hide things - possibly with decorative painting. It would seem that "picture", "painting", or "canvas" would have been more apposite words to borrow as names for our electronic devices. The inheritance was probably via the usage for the larger cinema screen.
I can foresee this replacing Bluray as a long term storage medium, write once variants of 3D-NAND without the ability to rewrite would be stable for decades.
Also having movies on a microSD sized card would be very handy, save on storage space AND make Apple mad for not installing microSD on their products :-) :-)
They've been selling them for years, did they start calling it "3D" once they exceeded 8 dies in a stack? Stupid marketing wins again?
When Samsung announced this I was hoping they were using a large number of process steps to actually create an individual die with that many layers. It would be something crazy like a 100 metal layer chip, but surely that's more efficient than making 32 dies and stacking them? (At least if you work out the wafer handling/processing to allow that many metal layers)
This 3D NAND process isn't stacking individual lumps of silicon, it appears to involve layering insulating material and new semiconductor substrate on top of a previous layer, etching and metallising that substrate, providing interconnecting vias through down into the previous layer and then doing it again as many times as they can get away with for decent device yields.
I asked in another forum (populated by semi guys) and got the answer. They deposit alternating thin films of two materials, position some hole sites using a single exposure, then etch boreholes down through all the layers, then fill the holes with metal. That gives you a stack of gate all around transistors in series. Pretty cool, and nothing like either stacking dies or processing 100 metal layers.
The reason they're using somewhat older processes than this is because the holes can't be perfectly cylindrical, but are conical to some extent, so if they start too small at the top they'll dwindle down to nothing before reaching the bottom.
Well 100 metal layers would be a *lot* more compact than stacked die.
Last time I checked even with thinned wafers the best that could be done (in 2013) was 17 layers in a microSD sized package and this was with optimum packing with all wafers in use.
The newer 64 and 128GB microSDs are just more dense (18nm instead of 22)
So what is the erase cycle I see that this is never talked about lately but it is the prime ware on a SSD. Yes I know about leveling but really number of erase cycles is still an important point. Also a lot of small writes is a lot harder on the erase cycle then lots of big writes. If you change a single bit the block has to be swapped out and erased. So clams of 20 gig per day for 5 years life may not mean much if you simply change 1 bit on one sector as fast as you can. I suspect early failure
Flash controllers these days are smart enough to handle lots of small writes. There are ways to mitigate the wear such as caching small writes briefly and then flushing them in bulk so that a cell only needs to be rewritten once. These days, it's not the flash chips which fail first (and even so, they tend to simply become read-only) but the controllers, which can happen regardless of the flash chips.
Higher density doesn't always mean lower prices
Write wear and speed aren't a major issue.
Firstly: large disks have far lower percentages of writes, which means the sparing load is much reduced.
Secondly: Samsung stepped back to ~40nm architecture when they went to 3d, specifically because it's a lot faster and more robust (and cheaper to fabricate)
Spinning rust drives _are_ less reliable than flash ones, the failure rate at work is testament to that (thousands of drives, flash fails at a rate of less than 1% of rust)
Now that Sammy are offering 3-10year warranties and almost all the spinners come with 12 months, the math favours flash even though rust is much cheaper upfront.
I like it. 64GB uSD is proving too small. My MP3 and docs collections currently are pushing 150GB and I would like to fit them all into a phablet.
Also - what if this stuff becomes small and cheap enough that it is more convenient to add more storage to servers or PCs than to actually erase anything? That would be interesting?
They will milk the life out of this technology on release, selling to enterprise only. I doubt we will see it at a good price point for consumers anytime soon. Think about the start of SSD's and price accordingly, because this is a replacement technology again now for HD's.
About the only thing available now which can handle these huge amounts of storage is the USB stick. It needs a very specific extension to the SD card interface to get past 32GB. An external SSD could use something such as Ethernet to connect, and I have a small box-thing which can connect an SD card via wifi or Ethernet.
I can remember USB expansion cards which used a PCI slot, had the usual row of sockets accessible from the back of the computer, and had a single purely internal USB socket. It may have been meant for one of those connector boxes that occuply the space used by a floppy drive, though these now use a different internal physical connector. not compatible with an unadorned USB stick,
If I had a couple of 64GB USB sticks in a RAID configuration, already possible, that aren't dangling in open air, that might be interesting. Move that approach to the TB scale. But I have seen similar multi-card RAID using SDHC, and I wonder if they can get the reliability.
Biting the hand that feeds IT © 1998–2021