BeSang Inc's 3D Super-NAND costs just 2¢ per gigabyte

[ET3D]

When is this NAND expected again?

This year.

(And what's around the NAND is less than what's around the magnetic hard drives.)

Yes, this is all assuming that the 10x cost saving will materialise, although it will have a big effect even if it's just 5x. As for you getting convinced, it feels to me like you haven't even tried looking into this, and you probably won't, so we'll have to wait and see. I'm not taking this at face value myself, but BeSang has been around for 8 years, sold a license to SK Hynix in 2013, and says it plans to bring product to the market this year, I take it with some seriousness.

By the way, I just noticed that the end of the article says this has 30x density than competitors' processes, and if that's the case (and not 3x as I read elsewhere), that would be even more revolutionary.

As for enterprise (I assume you agree with me that HDD's will disappear from the consumer market), as I said, HDD's will still be there, but as articles like this make clear, SSD's are already getting a big foothold, and price parity will likely accelerate their usage significantly.

[watercooled]

Yes, this is all assuming that the 10x cost saving will materialise, although it will have a big effect even if it's just 5x. As for you getting convinced, it feels to me like you haven't even tried looking into this, and you probably won't, so we'll have to wait and see. I'm not taking this at face value myself, but BeSang has been around for 8 years, sold a license to SK Hynix in 2013, and says it plans to bring product to the market this year, I take it with some seriousness.

Hahaha don't just dismiss my argument as me 'not looking into it'! I have done, probably more so than most, hence my saying it will take more than yet another Powerpoint to convince me it's practical. I never said it was impossible, but I'm not getting excited just yet!

By the way, I just noticed that the end of the article says this has 30x density than competitors' processes, and if that's the case (and not 3x as I read elsewhere), that would be even more revolutionary.

And all the more dubious.

As for enterprise (I assume you agree with me that HDD's will disappear from the consumer market), as I said, HDD's will still be there, but as articles like this make clear, SSD's are already getting a big foothold, and price parity will likely accelerate their usage significantly.

If and only if SSDs (not just plain NAND - I was comparing finished on-the-market HDDs to these NAND prices so the 'stuff' around the HDD is irrelevant) get substantially close to, or surpass, the cost of HDDs will we see them replacing them in off-the-shelf systems. But yeah I agree with that - the downsides of even cheap SSDs aren't such a big deal for consumers. But the enterprise market is huge and price is usually not such a big deal, so even with a price crossover it will only really matter in some segments.

[ET3D]

hence my saying it will take more than yet another Powerpoint

That's a rather bad way to show that you've looked into it, since it implies that you look at the technology as nothing but a powerpoint, instead of something that was demonstrated in 2007 (as some googling found). Coupled with the 'When is this NAND expected again?' comment it made you look rather clueless. You still haven't given me an impression that you've looked into this, although I'll take your word for it.

As for the density, I found some information. Apparently there's 30x density increase per layer. The 3x figure is for a 5 layer BeSang process vs. 48 layer 3D NAND process, which I'm guess is what they consider a sweet spot. I'm guessing that adding more layers could have a detrimental effect on density.

[watercooled]

That's a rather bad way to show that you've looked into it, since it implies that you look at the technology as nothing but a powerpoint, instead of something that was demonstrated in 2007 (as some googling found). Coupled with the 'When is this NAND expected again?' comment it made you look rather clueless.

You're dead wrong then, aside from being basically rude by dismissing what you don't agree with and resulting to borderline ad hominem attacks. Something existing as a prototype does not mean it's suitable for mass production, not even close.

You still haven't given me an impression that you've looked into this, although I'll take your word for it.

Please do. But understand I don't just take a company's claims at face value when they're claiming to beat the likes of Intel and Samsung by 10x. That's an exceptional claim which needs exceptional proof - I see no such thing.

Look at it this way - if it's truly as good as they claim; being cheap to implement and manufacture as well as beating competitors' density by a huge margin, why have SK Hynix not been mass producing it exclusively since they had the chance to license it? And since they're fabless (as far as I know) - who is actually going to license and manufacture it if not SK Hynix?

As for the density, I found some information. Apparently there's 30x density increase per layer. The 3x figure is for a 5 layer BeSang process vs. 48 layer 3D NAND process, which I'm guess is what they consider a sweet spot. I'm guessing that adding more layers could have a detrimental effect on density.

30x density isn't exceptional if you compare it to planar NAND, and you can bet your bottom dollar the big players have considered denser but shorter NAND stacks but all seem to have independently ended up with taller, less-dense stacks for mass production.

I've followed tech news for long enough to have seen tons of exceptional claims in the form of relatively small companies claiming to have come up with a way of beating the market.

[shaithis]

And all the more dubious.

If they have found a way to prevent electron-leakage, then all their claims would make sense....

[watercooled]

If they have found a way to prevent electron-leakage, then all their claims would make sense....

Don't get me wrong I'm all ears if they can demonstrate why they've managed to beat everyone. But like the phrase goes, "extraordinary claims require extraordinary evidence". I'm waiting to see that.

[Tetsuo Ryuu]

As some have said it won't be 10x as cheap, but even so it could be the push needed for SSD to go mainstream, maybe even for future consoles.

[anselhelm]

To qualify that last statement with regards to price, I'm referring to the as-yet unlaunched Toshiba X300 8TB HDD (non-helium-filled, around £280), rather than the more expensive HGST 8TB HDD (helium-filled, around £420).

Seagate 8TB drives have been available for 18months or so for about £180 - prices are crept over £200 recently with the £ falling (this actually started pre-brexit).

I've about 8 - they've been rock solid.

Gosh no. Won't touch Seagate with a bargepoll. Also, those drives are for archiving with terrible read/write speeds.

The drives I'm talking about are actually fit for purpose for general use rather than just large file storage.

[anselhelm]

If this actually works out as cheaply as advised, I'd like to hope that it could at least bring about price parity with HDDs and finally start to bring out their end, though in reality I'm far more sceptical that even that much of the cost-savings would be passed on to consumers.

It will take far more than price parity to end HDDs. Maybe in parts of the consumer market, but lets not forget a firmware bug can destroy all off that data in an instant on an SSD for example. And then there's write endurance which is more than a passing concern in some uses.

LTO tape drives are still heavily used for large-scale backups.

Practically ending the dominance of HDDs will be fine enough for me.

SSDs and HDDs both can suffer from firmware bugs which cripple a drive (*cough* Seagate *cough*) so whilst I understand where you're coming from, I would still opt for a large SSD at an affordable price given the chance.

Sure, SSDs do tend to fail spectacularly when they fail, but then the failure rates overall are lower. Not something I made up. Look up the results from recent tests across the board about this. Means that it's not as simple as one or the other.

As for tape, it's still cost-effective for large backups. The problem is random reads... or reading anything much from it in general

[watercooled]

SSDs and HDDs both can suffer from firmware bugs which cripple a drive (*cough* Seagate *cough*) so whilst I understand where you're coming from, I would still opt for a large SSD at an affordable price given the chance.

Yeah, both can outright fail of course and you probably know what I meant specifically but if not, I mean an SSD can irrevocably wipe an entire drive in an instant, no chance of recovery. A HDD simply isn't capable of failing so spectacularly, so quickly - if the PCB dies it can be replaced, if something mechanical dies it can be replaced (probably needs to be in a clean room ideally), if the head crashes it might wipe out a few tracks but professional recovery can still salvage most of the data, and so on.

If a HDD is failing you'll often get some warning and have the ability to quickly copy important stuff off, even without resorting to the above methods or paying a recovery firm. A firmware bug on an SSD has plenty of methods to choose from to destroy its contents instantly, e.g. corrupting the mapping tables (through unexpected power loss for instance) or just outright flashing the NAND - NAND flashing completes in seconds, it doesn't have to go in and erase every block one at a time at normal write speed like a HDD does. And we've seen it happen on real SSDs because of bugs. It's far less common now they're maturing but it's still going to be a concern.

And then of course there's the issue of data retention, degraded flash can lose charge and therefore data if left alone for long enough, so it's probably unwise to rely on them, especially finer geometry NAND, for archival use which in itself is a far-from-insignificant market. Retention is over-exaggerated for drives in normal use though, the amount of places that misinterpreted a study a while ago was alarming, so much so I actually emailed the source and asked for an explanation, they provided me with this link: http://www.pcworld.com/article/29251...after-all.html

Of course, everything will be backed up anyway, right?

Except professional recovery companies would be out of business were that the case...

Anyway, my point is that the likes of Google, Microsoft, Facebook, etc (and likely organisations such as the NSA, GCHQ and so on) buy HUGE volumes of HDDs for, amongst other things, archival use. That alone drives demand. So sure, if SSDs reach price parity with HDDs (despite claims to the contrary I don't see it happening soon, people assume HDDs themselves are a stationary target which they're not), or even just cheap and big enough for most practical consumer uses while maintaining decent performance, they'd likely displace a lot of HDDs in that segment. But overall it probably wouldn't matter all that much - the market for HDDs would still be vast. Like the comparison I used earlier - tape drives aren't exactly found attached to every desktop any more, but they're not going anywhere for backup use.

[DanceswithUnix]

Anyway, my point is that the likes of Google, Microsoft, Facebook, etc (and likely organisations such as the NSA, GCHQ and so on) buy HUGE volumes of HDDs for, amongst other things, archival use. That alone drives demand.

Companies like Google just cloud store everything, getting redundancy by storing all data in at least three drives spread around the planet, so their live store *is* their archive.

I work for an enterprise storage company. Enterprise drive prices are different, but you just have to hear people talking about the back end storage as either "flash" or "spinning rust" to know in our minds the trays of rust are already as good as dead technology.

[watercooled]

Not necessarily. Because of power/environmental concerns a lot of big companies (Facebook for one like to blog about it) have a lot of nearline storage with older stuff stored on spun-down but still attached drive arrays; even the power supplies etc are spec'd such that only a couple of drives in a rack can be spun up at any one time. They use some software to predict if the contents will be needed in the near future, for example someone scrolling through a photo archive, and spin the disks up ready. It's less and less practical to keep years of rarely-accessed archives permanently online. I even remember reading that they use automated blu-ray and tape drives, not just for backup but for another layer below the HDD nearline arrays. Access time likely goes into minutes, IO speed is very low and yet they're not much cheaper than HDDs per GB.

But the likes of Google basically made a name for themselves by using off-the-shelf hardware and smart software to make it reliable as a whole. For lots of uses the advantages of SSDs aren't relevant, and the drawbacks are problematic, and while they're significantly cheaper per GB (which they will be for some time yet) and offer much higher endurance, there will remain a market for them. If I had a pound for every time I heard an established technology was 'dead', and all that. Say a company decides to use an array of the same batch of SSDs for their storage array - we've seen more than a couple of time-bomb firmware bugs on SSDs. Of course a well-designed array shouldn't do such a thing, but people should always back up important stuff too. Not everyone does it. Unless there's some paradigm shift in SSD architecture, there will remain the looming threat of a bug destroying its whole contents in an instant - there's just some inherent volatility in the medium as it currently exists. Cost is largely immaterial for performance-critical sectors and enterprise SSDs already exist there, and yet the market for SAS HDDs (which can easily cost more than cheap SSDs) still exists. Horses for courses.

Studies like this are interesting too, looking into failure modes of flash drives: https://www.usenix.org/system/files/...-schroeder.pdf
Having read through it I concur with their conclusions - SSDs seem to fail more spontaneously and impacting real user data more frequently than HDDs. It's a grave over-simplification to claim SSDs are more reliable than HDDs.

But all this is largely academic, based on the assumption that we'll see mass-produced NAND reaching price parity with HDDs by the end of the year. Guess we'll have to wait and see!

[DanceswithUnix]

Cost is largely immaterial for performance-critical sectors and enterprise SSDs already exist there, and yet the market for SAS HDDs (which can easily cost more than cheap SSDs) still exists. Horses for courses.

Oh cost is very much a factor. Redundant NAS boxes in front of redundant RAID controllers with redundant fibrechannel switches don't come cheap, but people buy this stuff usually to hit a performance target such as so many thousand virtual machines. Even with a terabyte of ram caching the disks, it takes a *lot* of spindles to get enough io per second to run a big VM farm and that gets expensive fast, or you use flash. The cost of the spindles will dwarf the cost of the controllers etc. You need tiered storage? Not with flash, it is all quick.

Flash was a game changer, it already won.

And yes there is stuff out there about flash failure, storage designers read them too

[ET3D]

who is actually going to license and manufacture it if not SK Hynix?

That's a good question and I'm looking forward to see the answer. As for why this hasn't been implemented, the business reasons are obvious. No company it its right mind wants a high end product which makes a lot of money become a low cost commodity. Also, companies don't want to pay patents. They'd rather wait for them to expire.

Which is why I imagine that BeSang has been having a hard time getting traction, and why even when the technology is released it will take a while to show the full savings. Nobody wants to kill their own market. Much better to artificially keep prices higher and drop them over time, and make a lot more money that way.

[ultrasbm]

it should, but you know how it will be...

It's more likely to be 20% more expensive

[DanceswithUnix]

Nobody wants to kill their own market.

That is the sort of logic that got companies like Zilog to where they are today

If there is a better way, then someone is going to deploy it and gain market share. Best make sure it is your company that gets in there not someone else.

My best guess if this technology has been left on the shelf is if it was seen as too risky. Companies on thin margins may well run away from something that looks like a money sink to get working.