Toshiba produces the world's first 4-bit QLC NAND Flash Memory

[Coaxialgamer]

Quote

Toshiba produces the world's first 4-bit QLC NAND flash memory, which offers increased levels of storage density and cost/storage ratios. 

 

Unlike the typical MLC (Multi-level call) 2-bit and TLC (Triple level cell) 3-bit NAND that we see in modern SSDs Toshiba's QLC (quad-level cell) delivers 4 bits of information per memory cell, greatly increasing the amount of storage that can be fitted into a single memory die. 

 

Toshiba's new 64-layer QLC flash offers capacities of 768Gb (96GB) of storage per die, which is a huge increase over Toshiba's 3rd generation 512Gb dies which uses Toshiba's 3-bit TLC NAND. This new NAND will allow Toshiba to create a 1.5 TB SSD package with a 16-die stacked architecture, which is simply insane by today's standards. 

 

Traditionally when creating NAND that fits more data, bits, into a single cell the memory loses some performance and longevity, especially when it comes to workloads with heavy writes.  This is why not all modern SSDs use TLC, 3-bit, NAND today. It is fair to assume that this QLC NAND suffers from similar problems. 

 

 

Quote

Samples of Toshiba's upcoming QLC powered devices will be shipping to SSD and SSD Controller vendors in June, with the company planning a demonstration at the 2017 Flash Memory Summit in August (7-10). At this time it is expected that this new type of memory will have some performance shortcomings, but its sheer cost/GB will certainly be of interest to both consumers and device makers. 

Toshiba says that their QLC NAND requires 2x the accuracy of TLC technology, requiring the company to combine advanced circuit designs with their 3D Flash technology to overcome these technical challenges.   

At this time it is unknown when we will first be seeing this new type of flash in consumer products, especially given Toshiba's current financial situation, though it has huge potential to offer a lot of value for money, especially when combined with a fast caching technology.

They've done it again! Hopefully reliability doesn't suffer too much. The dream of an inexpensive multi-terabyte SSD is getting closer by the... Day? 

 

https://www.overclock3d.net/news/storage/toshiba_produces_the_world_s_first_4-bit_qlc_nand_flash_memory/1

[Billy_Mays]

Yay cheaper shit for more budget people

[sazrocks]

Doesn't that mean it's write durability will be terrible?

[Coaxialgamer]

Just now, sazrocks said:

Doesn't that mean it's write durability will be terrible?

Worse than TLC. It could potentially be mitigated by using caching etc, wear distribution etc. Most workloads don't require tons of writing though tbh. 

[sazrocks]

1 minute ago, Coaxialgamer said:

Worse than TLC. It could potentially be mitigated by using caching etc, wear distribution etc. Most workloads don't require tons of writing though tbh. 

Really? I have quite a few terabytes written to my 120GB 850 evo after using it for only 1 year.

[Coaxialgamer]

Just now, sazrocks said:

Really? I have quite a few terabytes written to my 120GB 850 evo after using it for only 1 year.

That's actually not that much. 

The 850 evo is rated for around 2000 P/E cycles, meaning an endurance of around 240TB on the 120gb 850 evo. 

[yian88]

Does this mean i can get 1TB ssd for ~100$ yes or no? if no throw it to the garbage bin.

[DocSwag]

I really don't see QLC being reasonable for consumers in the near future. Compared to tlc performance an endurance go down A LOT and you get what, 1/3 more capacity?

 

Until 3D NAND really goes off and allows us to go back to much larger process nodes or something else happens, I don't see QLC being a great option.

2 hours ago, sazrocks said:

Really? I have quite a few terabytes written to my 120GB 850 evo after using it for only 1 year.

Part of that reason is the 3D NAND Samsung uses is a bit special. They decided to go back to 40nm in the 3D NAND stuff rather than sticking to the smaller process nodes other companies are using when making 3D NAND, which considerably increased performance and endurance at the cost of die size.

[leadeater]

19 minutes ago, DocSwag said:

Until 3D NAND really goes off and allows us to go back to much larger process nodes or something else happens, I don't see QLC being a great option.

This actually is 3D NAND from what I can tell, 64 Layers using 4 bit technology.

Quote

Toshiba says that their QLC NAND requires 2x the accuracy of TLC technology, requiring the company to combine advanced circuit designs with their 3D Flash technology to overcome these technical challenges.   

Twice the accuracy requirement worries me, likely means much higher latencies which is what makes SSDs fell so much faster.

[DocSwag]

10 hours ago, leadeater said:

This actually is 3D NAND from what I can tell, 64 Layers using 4 bit technology.

Twice the accuracy requirement worries me, likely means much higher latencies which is what makes SSDs fell so much faster.

What I meant to say is that 3D NAND allows us to increase process node by enough to alleviate a lot of the performance and endurance issues. I was just too lazy to write it out  

 

I totally agree, mlc was reasonable since you doubled capacity and TLC was still kinda reasonable since multiplying by 3/2 is kinda worth it but only getting 1/3 extra capacity with QLC over TLC while performance and endurance, especially write speeds, suffer tremendously has me worried.

[LucidMew]

iirc QLC has been around for a while now (and a quick google search turned up http://www.tomshardware.com/news/qlc-nand-ssd-toshiba-facebook,32451.html ), and what is new is the 3D stacking of them.

[JoostinOnline]

3 hours ago, sazrocks said:

Doesn't that mean it's write durability will be terrible?

The write limit is incredibly exaggerated.  No average user is ever going to reach it.  The drive is far more likely to fail from other causes before that, or just be replaced for something bigger.

 

It does mean the drive will be much slower though.

[Juanitology]

I've had a 250GB 850EVO for a couple years now and I'm a pretty heavy user. According to the Samsung utility I've written 39TB on it, so I'm about half-way through its rated endurance. and I've used it for everythign from an adobe premiere scratch disk to installing 150GB of steam games several times over and many reformats. vI doubt the average consumer would hit the endurance rating before 10 years, and by then they'd be obsolete anyways. 

[Coaxialgamer]

4 hours ago, LucidMew said:

iirc QLC has been around for a while now (and a quick google search turned up http://www.tomshardware.com/news/qlc-nand-ssd-toshiba-facebook,32451.html ), and what is new is the 3D stacking of them.

Those were just plans on a roadmap. No chips had been made. 

[Coaxialgamer]

5 hours ago, DocSwag said:

What I meant to say is that 3D NAND allows us to increase process node by enough to alleviate a lot of the performance and endurance issues. I was just too lazy to write it out  

 

I totally agree, mlc was reasonable since you doubled capacity and TLC was still kinda reasonable since multiplying by 3/2 is kinda worth it but only getting 1/3 extra capacity with QLC over TLC while performance and endurance, especially write speeds, suffer tremendously has be worried.

It's worth noting that the decreased endurance isn't actually a direct cause of being a TLC. What causes limited endurance is the increased voltage needed at the cell level. That issue could potentially be mitigated. 

[Agosto]

NEW QLC NAND!

AMAZING ENDURANCE! It lasts for a full MONTH of light use!

[DocSwag]

4 hours ago, Coaxialgamer said:

It's worth noting that the decreased endurance isn't actually a direct cause of being a TLC. What causes limited endurance is the increased voltage needed at the cell level. That issue could potentially be mitigated. 

Well, kind of but not exactly.

 

Due to the fact that you need to differentiate between a lot more voltage levels as you increase the number of bpc the boundaries between the voltage levels start becoming more fuzzy which in turn requires using more voltage to figure out what the hell is in the cell which requires more time which increases latency eventually making the cell completely unusable. Whew, that was quite a run on.

 

So yes, technically it isn't direct. But QLC does significantly decrease endurance. Whether it's direct or indirect doesn't matter that much, imo.

 

I mentioned a way that it could be mitigated, which is by going back to a larger process node.

 

For the most part though I'm not as worried by endurance as I am by performance. Doubling the number of voltage states is probably gonna make write speeds go down A LOT.

 

Right now though QLC only seems like it's gonna be used for very specific enterprise applications where it's "write once read many." I just don't see QLC becoming viable for consumers in the near future.

[Zodiark1593]

1 minute ago, DocSwag said:

Well, kind of but not exactly.

 

Due to the fact that you need to differentiate between a lot more voltage levels as you increase the number of bpc the boundaries between the voltage levels start becoming more fuzzy which in turn requires using more voltage to figure out what the hell is in the cell which requires more time which increases latency eventually making the cell completely unusable. Whew, that was quite a run on.

 

So yes, technically it isn't direct. But QLC does significantly decrease endurance. Whether it's direct or indirect doesn't matter that much, imo.

 

I mentioned a way that it could be mitigated, which is by going back to a larger process node.

 

For the most part though I'm not as worried by endurance as I am by performance. Doubling the number of voltage states is probably gonna make write speeds go down A LOT.

 

Right now though QLC only seems like it's gonna be used for very specific enterprise applications where it's "write once read many." I just don't see QLC becoming viable for consumers in the near future.

Cheaper flash media to store console games and movies? 

[DocSwag]

5 minutes ago, Zodiark1593 said:

Cheaper flash media to store console games and movies? 

For console games that actually would make a lot of sense.

 

I'm not sure when in this day and age you're actually use a drive only for storing movies and never writing or erasing to it. Unless you mean a special type of drive only for storing movies and stuff in, say, a plex server.

[The Benjamins]

13 hours ago, sazrocks said:

Doesn't that mean it's write durability will be terrible?

 

13 hours ago, Coaxialgamer said:

Worse than TLC. It could potentially be mitigated by using caching etc, wear distribution etc. Most workloads don't require tons of writing though tbh. 

 

13 hours ago, sazrocks said:

Really? I have quite a few terabytes written to my 120GB 850 evo after using it for only 1 year.

 

Ya and this is for increase capacity so if the durability goes down by half but now you get 2x more space the endurance of the drive would stay the same.

 

and remember their was a endurance test done were even the worst 250GB drives lasted more then 400 TB written to it. so I think a 1-2TB QLC drive will be fine.

[ChineseChef]

This kind of tech may be good for larger "bulk" storage.  Great for low write high read stuff.  Assuming it does the lower price for more storage that the MLC and TLC did compared to their predecessors.

[mariushm]

It could make sense if they make something like 960 GB QLC (1024 GB with 64 GB hidden as spare to extend life)  PLUS 16-32 GB SLC or Intel's new memory thing (whatever is called) as a write buffer/cache.

 

We have now SSDs that use MLC/TLC memory chips where a portion is locked in "SLC" mode, and those portions are used to accelerate writes. SanDisk uses this trick on a few of their drives.

 

It wouldn't surprise me though to see value/budget controllers with maximum 4 channels and no extra memory used with 4 of these chips to save costs. They'd have lousy write endurace (if you get 40 TB for a 64GB drive, maybe you'd have 20-25 TB with QLC) or about 400-500 erase cycles vs 700-1500 erase cycles (with tlc) and 2500-5000 with MLC

 

[Doobeedoo]

Yeah, endurance will be very low though. It's great for WORM I guess for large capacity these will have. 

[DeScruff]

Well according to that article its a 50% boost to capacity. So the real question is how much endurance does each individual bit lose?
With proper planning of wear leveling, the high volume may mitigate that.

Regardless, I wouldn't mind a "Storage SSD" that isn't meant for your OS, but just a place to hold all your media files, or whatever. I'm tired of having to wait for my HDD to spin up, and how loud it is, just to start listening to some music.

[Sakkura]

On 30/6/2017 at 6:21 AM, Terodius said:

I've had a 250GB 850EVO for a couple years now and I'm a pretty heavy user. According to the Samsung utility I've written 39TB on it, so I'm about half-way through its rated endurance. and I've used it for everythign from an adobe premiere scratch disk to installing 150GB of steam games several times over and many reformats. vI doubt the average consumer would hit the endurance rating before 10 years, and by then they'd be obsolete anyways. 

And the rated endurance is very conservative, the underlying NAND is rated for far more writes.

 

For a comparison point to more regular usage, I've had a 120GB SSD since 2012, and it has 28TB of writes on it. That's 4% of its lifespan according to the SMART statistics, meaning it's internally rated for about 700TBW. Which would last me about another 120 years at this rate.

 

It's MLC though, but then again if it was QLC the capacity would be doubled. People don't seem to realize that although newer tech (more bits per cell, smaller process node) often reduces the write endurance, the increasing capacity partially counteracts that trend.

 

I wouldn't really want a 120GB SSD based on QLC, but at more realistic capacities it's fine for consumer use.