Intel and Toshiba announce 3D NAND

Both Flash Forward and IMFT have announced that they are sampling 3D NAND, with shipping products expected within a year.

The most interesting part to me is the fact that Intel-Micron is producing 256 Gb MLC and 384 Gb TLC dies, compared to 128 Gb that everyone else is making. Intel has also previously claimed "disruptive cost" for their new NAND, so I'm thinking we'll be seeing 3TB SSDs in the $500-600 range within a year.

All in all an exciting time for SSDs.

 

until someone releases a 2 TB SSD, nothing would excite me, I had an 840 PRO, then got the 850 PRO, performance wise, they're the same maybe a few digits higher in benchmarks for the 850 but nothing earth shattering. 3D NAND, another gimmick. Just my 2 cents worth

 

Not looking forward to any of the new TLC stuff. If this type of NAND really becomes a predominant one I'm not going to be happy camper. But that's just me.

 

The most interesting part to me is how Intel is shunning the latest charge trap technology (which may be part of the reason for Samsung's TLC nand issues?).

Not only do we get the benefit of larger nodes (and hence, more electrons per cell) and 3D vertically stacked nand, but we also get it with the technology that has proven itself over the last half decade too. Something Samsung should have considered too - instead of being 'new' and 'innovative' and now sitting on a ticking time bomb of a problem.

See:
http://www.overclock.net/t/1512915/...es-benchmarks-needed-to-confirm-affected-ssds

See:
http://www.overclock.net/t/1507897/samsung-840-evo-read-speed-drops-on-old-written-data-in-the-drive


See:
http://www.overclock.net/t/1512915/...ed-to-confirm-affected-ssds/430#post_23719258


With even the 850 EVO possibly showing similar problems (last link above) for Samsung (keep reading the posts after the linked one too).


With the foresight and the wisdom to change/progress, but change wisely and methodically (I'm sure they are testing charge trap technology as we speak), this is another reason why Intel is #1 in my books and my systems.

Yes, I'm sure the cost savings are there to use the newest tech - but obviously, Intel doesn't think the pennies saved are worth the possible risk to themselves and consumers from the unproven process, yet.

This TLC (Intel's) may be worth a closer look. But I myself won't have it on my 'real' workstations. Even if offered in 10TB capacities today. In 2018+ though, I may be drawn to it if the real world data I gather from the web and my own tests show it to be as trustworthy as the drives we're using today.

And even then, a SATA3 interface would be the last thing I would support in any new drive going forward (from today). PCIe 4x+ lanes is where the performance future is for storage subsystems...

 

What? The 840 and 840 Evo both used floating gate technology, so there is no way charge trap is responsible for their issues. Samsung only introduced charge trap tech with their VNAND. And all your last link shows is a single data point that doesn't show slow downs with old data at all. The overall performance might be lower than expected, but that could be an issue with that person's rig or that particular SSD, or it might be the 850 Evo doesn't quite live up to its promises. But that data doesn't show any indication of the crippling slowdowns in old data that the 840 Evo experienced. We need more data before we can say for sure that the 850 Evo is reliable, but right now, I have not seen anything showing that is *isn't*.

If anything, if the 840 Evo issues aren't just an issue of Samsung not knowing what they are doing, and are actually a fundamental issue with the current implementation of TLC NAND, moving to charge trap would probably be safer. If there is a fundamental issue with planar floating gate TLC NAND, it is more likely to carry over to 3D floating gate TLC. But moving to charge trap means a whole different method of storing charge, so going to 3d charge trap TLC is more likely to eliminate the current issue.

 

See:
http://www.anandtech.com/show/7237/samsungs-vnand-hitting-the-reset-button-on-nand-scaling

Note the date on the above article: August 21, 2013 2:50 AM EST, by Anand Lal Shimpi himself.


Also note that the V nand in the 850 series is simply version two of Samsung's sad implementation of that 'tech'.

 

Yeah, but that is still only in reference to VNAND, which is Samsung's name for 3D NAND. The 840 series still used planar NAND, with VNAND appearing first in enterprise drives and then finally making a consumer appearance in the 850 series. That article was from 2013, but that just means that VNAND was in development for a long time before it made it into a consumer application.

Also, check the date of this article:
http://www.anandtech.com/show/7173/...iew-120gb-250gb-500gb-750gb-1tb-models-tested
by Anand Lal Shimpi on July 25, 2013 1:53 PM EST

So reviews for the 840 Evo started popping up a month before that article about VNAND, which was still in development at the time, was written.

 

Hmmm, I guess I stand corrected (for now).

But I'm sure I've read that Charge Trap Flash (CTF) designs were used by Samsung before V nand. And not just in their CF cards either. Btw, this is not new tech, AMD and Fujitsu had a version of this since early 2000 (if I remember correctly) but they called it something else (nor?).

 

It looks like AMD and Fujitsu called it "MirrorBit Flash Memory" but that was NOR Flash, not NAND, and I can't find what they used it for.

Samsung and a few others apparently researched Charge Trap Flash for planar NAND SSDs as a method of achieving smaller processes. But that was back around 2006, when it was believed that floating gate flash could not be shrunk past 50nm. Obviously, manufacturers found a way to get around that limit, as we have since seen floating gate flash get shrunk all the way to 16nm, and it appears Samsung never implemented planar CTF in SSDs.

I couldn't find anything that explicitly said the 840 or 840 EVO used floating gate technology, but that is possibly due to the lack of alternatives. Just like 20 years ago, you never heard people say "I have a CRT computer monitor", it was just "computer monitor" because that was the only technology there was. Once LCD monitors became a thing, you heard people say "I got a new LCD monitor" or "I still have a CRT monitor". Likewise, two years ago, you never heard "this SSD uses planar NAND" because there was no other kind of NAND. Now that 3D NAND is in use, you see "this SSD uses new 3D NAND, while that one uses planar NAND". So if CTF and Floating Gate were in use, you would expect to see "this SSD uses Charge Trap, while that one uses Floating Gate", but that doesn't seem to be the case.

And if Samsung had already been using CTF in their planar flash, when they moved to VNAND, then CTF wouldn't have been a very big deal. Instead, with the 850 series, they emphasized their use of CTF technology right alongside the switch to VNAND.


However, even if CTF tech is not responsible for Samsung's issues with the 840/840 EVO, it will be interesting to see what, if any, issues pop up with the new technology. Considering Samsung's recent history, it certainly wouldn't be surprising if some sort of flaw was discovered in the next year or so. I'm definitely interested to see what Intel and Micron/Crucial come up with, but I'm afraid Intel's offerings will cost an arm and a leg, and Crucial really needs to offer something more compelling than their latest models (the BX100 costs almost as much as the MX100 but performs worse, while the MX200 costs more but doesn't perform any better than the MX100).

 

I was hoping this wouldn't turn into a Samsung flamefest, but maybe that was asking too much. In any case:

1. TLC is here to stay, and it will keep growing. It's more than a bit premature to judge a technology harshly just because one company messed up their first gen product. If you understand why this bug occurred, then you'll also understand why this is less likely to occur on the 850 EVO, which no one's confirmed has the same issue yet.

Also keep in mind even the best companies screw up sometimes. Time will tell whether this is a fluke or Samsung will turn into the next OCZ.

Sandisk has also been shipping TLC flash in their Ultra IIs, and no one's complained about slowdowns yet so it's unlikely to be a fundamental problem with TLC. BTW they're also using CTF for their 3D NAND, so expect CTF flash in the next gen Extreme Pros (Extremely Pro?).


2. 3D NAND is also here to stay and not a gimmick. Planar scaling has basically reached it's limits at 15nm/16nm. Maybe there'll be one more node, but it's unlikely. The only cost effective way to scale now is vertically.

If you really want to see your cheap multi-terabyte SSDs, then 3D NAND is the only way that'll happen.


3. The cost and capacity implications of this are huge. IMFT's 256 and 384 Gb dies basically mean they're planning to make 2TB and 3TB SSDs, or larger.

Anandtech did an estimate of the die size, and the conclusion is that this a pretty big step up in bit density. Those cheap SSDs that are now $0.35/GB are using 15/16nm NAND, and the IMFT 3D TLC NAND easily doubles the bit density. If they can keep yields up I think we'll see prices drop to $0.20/GB in a year.

To put this in context, 2.5" hard drives right now are $0.05/GB. This means the cost will be within a factor of 4. When I bought my first SSD 5 years ago, the cost difference was more like a factor of 20.

PMR scaling is done and there's nothing really slated to replace it yet so magnetic storage costs will only drop very slowly. Three of the four major NAND manufacturers are now on board with 3D NAND and have indicated they plan to continue scaling aggressively to 128 layers or beyond. I wouldn't be surprise if within five years SSDs will reach price parity with notebook HDDs. At that point magnetic storage is done (at least for laptops). This is in my opinion the first nail in the coffin for laptop HDDs.

 

Where is the flamefest? So far, the discussion has been on point.

1) Samsung has messed up more than it's first gen product (the original 840), it has made the gen two product considerably worse (the 840 takes a long time to slow down when reading old/unchanged data - the 840 EVO only needs as little as a few weeks).

Sure, time will tell how Samsung is going to respond in the end, but so far the response is not encouraging.

• 840 TLC drive issue is basically ignored from Samsung... and continues to be ignored still.
• 840 EVO issue fix does not work (since October 2014). Promised fix 'deadline' was passed at least two weeks ago, if we're being generous.
• Latest rumbling is that if new firmware isn't issued, Magician will have the 'refresh' routine built in. What? I'd rather use Puran DiskFresh instead of intentionally installing Magician to touch my O/S files.

In truth, I don't care that they missed the deadline. I personally don't care about the original 840 issue as I only bought 1TB EVO's with this issue. What I do care is that the Samsung drives are obviously flawed in comparison to every other SSD available with regards to this issue and should be recalled with full refund offered.

If that happens, then I'll say Samsung squeaked past another OCZ comparison. Right now? The same pile with a different name smells as bad.

SanDisk's Ultra II is not based on Samsung TLC design. That is why nothing has been reported yet. I still don't recommend it to anyone though. Not when other proven designs are just a few dollars more.

2) No argument about 3D nand. It's about time. But hopefully, if any issues arise, I won't be the one beta testing at that time.

3) 5 years ago, the cost factor was in that 20x range or more. Yet, real life performance was below HDD speeds (compared to my multi VRaptor setups back then). Needless to say, the 'new' SSD's at that time cost me nothing but time (they were returned promptly).

PMR may be almost done, but it has been updated and pushed even further (HGST is the choice drive right now).

See:
http://forwardthinking.pcmag.com/none/328899-hard-drive-makers-eye-increased-density-20tb-drive

HDD's will not be replaced by SSD's any time soon. Not for anyone with big data, anyway.

Notebook HDD's have been 'dead' to me for a few years now. The SSD's finally reached the minimum capacity necessary, performance (in all aspects) finally surpassed HDD's and reliability and power consumption were finally tamed and balanced vs. the performance offered.

Notebook HDD's only exist today to make the initial cost of a new notebook seem reasonable. What I tell clients is that if they don't spend the $$$ to install an SSD in their new notebooks, any troubleshooting needed in the future of that system is essentially costing them 2 to 3x more, minimum. In either time (if they do it themselves), $$$ if they pay me or anyone else, or both.

How I justify the additional cost? I tell them to ignore any additional warranty on the new system and consider the SSD as part of that cost. Sure, hardware fails occasionally. But the more guaranteed issues are software related, not physical. And my rates are not going lower any time soon.

 

I seriously doubt we'll see price drops from these money gouging corporations anytime soon... SSD's have become cheaper over the last 4 years however, they could easily be at half the prices they currently are at...

 

That's because pretty much any new drive (from Samsung, Sandisk, Crucial, Intel doesn't matter) is going to be saturating the SATA3 bandwidth limit. There's simply no way to make a drive faster using SATA3.

The 840 series came very close, the 850 pretty much saturates it. The only way to make the drive faster is to go to PCI-E, in which case the drives are essentially as fast as the NAND chips on them. Most PCI-E SSDs easily do over 1.5gb/s.

I also find it funny that you say "until someone releases a 2TB SSD" and then claim "3D NAND is another gimmick". You do realize things like 3D NAND are going to be what makes 2TB+ SSD drives a reality right? That's why it's a such a good technology and I'm glad that others have started making their own versions of it. Competition breeds innovation.