Toshiba's 768 Gb 3D QLC NAND Flash to Match TLC's Program/Erase Cycles

[Raevenlord]

Not all news coming out of Toshiba is bad or somewhat bad. The Japanese giant still is one of the biggest players in the NAND semiconductor business, no matter the recent woes. Even more recently, though, Toshiba announced they had developed quadruple level cell NAND memory, which should improve density and, therefore, reduce the price/GB ratio on future consumer products, such as SSDs. However, each increase in the number of cell levels bring concerns regarding not only performance, but especially durability, since a higher number of states per cell increases the voltage steps that are applied to it (SLC NAND dealt with two voltage states, MLC with four, TLC with eight voltage states, and QLC will handle 16 of these.) This tends to make errors more common, and the cell's longevity to be compromised due to the amount of variation in its states, which means more powerful error correction techniques must be employed.



According to Toshiba, its 3D QLC NAND targets around ~1000 program/erase cycles, which is close to TLC NAND flash. This is considerably higher than the amount of P/E cycles (100 - 150) expected for QLC by the industry, which means the company has achieved what many thought difficult. Toshiba has begun sampling of its 3D QLC NAND memory devices earlier this month. everything points to mass production on late 2018, early 2019, though, which means we still have a long way to go until we see this technology implemented. This won't be the one to save us from escalating NAND prices; we'll have to look to other, more market and supply-and-demand based factors instead.

View at TechPowerUp Main Site

 

[Prima.Vera]

This won't be the one to save us from escalating NAND prices; we'll have to look to other, more market and supply-and-demand based factors instead.

Time to build more fabs then.

 

[bug]

Time to build more fabs then.

Because those are dirt-cheap...

 

[Prima.Vera]

Because those are dirt-cheap...

NAND profits are huge, especially with current prices.... Samsung and TMSC are already doing it.

 

[bug]

NAND profits are huge, especially with current prices.... Samsung and TMSC are already doing it.

It's not only about the money. Infrastructure and sourcing materials has to be considered. Even the risk of earthquakes or other natural disasters (remember back in 2011 when Vietnam floods upset the HDD market for more than a year?) is taken into account. A fab is not a skyscraper.
There's big money to be made, for sure, but you have to cough up a huge amount before you can cash in.

 

[R0H1T]

I knew Toshiba wouldn't disappoint us

around 1000 p/e cycles

Now waiting for the 200$ 2TB miracle to come true

 

[bug]

I knew Toshiba wouldn't disappoint us

Good for you, because everyone else didn't. And it's still a mystery how did they manage that.

Now waiting for the 200$ 2TB miracle to come true

You and me both, bro.

 

[R0H1T]

Good for you, because everyone else didn't. And it's still a mystery how did they manage that.


You and me both, bro.

Like I said in the other thread, endurance is more about controller than NAND.

The Intel enterprise drives have endurance from a few TB to several PB, the DC 3700 being a prime example. The drives themselves are based on the same NAND AFAIK, though something like a DC 3700 will have better binning however error correction, controller et al have a far bigger role to play in making it more reliable. Toshiba says they have something better than LDPC for EC, seems like that also has a huge role in driving the endurance up.

 

[bug]

Like I said in the other thread, endurance is more about controller than NAND.

The Intel enterprise drives have endurance from a few TB to several PB, the DC 3700 being a prime example. The drives themselves are based on the same NAND AFAIK, though something like a DC 3700 will have better binning however error correction, controller et al have a far bigger role to play in making it more reliable. Toshiba says they have something better than LDPC for EC, seems like that also has a huge role in driving the endurance up.

I very much doubt that. Enterprise drives may fiddle with overprovisioning and employ more efficient error detection techniques, but when a cell can't hold its voltage, the controller is powerless.
Plus, I don't recall anyone ever detecting endurance differences for the same NAND driven by Silicon Motion, Phison or SandForce.

 

[eidairaman1]

Toshiba make SLC

 

[hojnikb]

Like I said in the other thread, endurance is more about controller than NAND.

The Intel enterprise drives have endurance from a few TB to several PB, the DC 3700 being a prime example. The drives themselves are based on the same NAND AFAIK, though something like a DC 3700 will have better binning however error correction, controller et al have a far bigger role to play in making it more reliable. Toshiba says they have something better than LDPC for EC, seems like that also has a huge role in driving the endurance up.

Thats not entirely true. Yes, you can extend NAND endurance by employing better error correction and compensation for degradation, but ultimately it's all down to quality of the actual silicon, how much cycles is going to hold up.

 

[dreamer2908]

It's not only about the money. Infrastructure and sourcing materials has to be considered. Even the risk of earthquakes or other natural disasters (remember back in 2011 when Vietnam floods upset the HDD market for more than a year?) is taken into account. A fab is not a skyscraper.
There's big money to be made, for sure, but you have to cough up a huge amount before you can cash in.

Fact check: it's actually Thailand, not Vietnam.

 

[bug]

Fact check: it's actually Thailand, not Vietnam.

I stand corrected.

 

[R0H1T]

Thats not entirely true. Yes, you can extend NAND endurance by employing better error correction and compensation for degradation, but ultimately it's all down to quality of the actual silicon, how much cycles is going to hold up.

There's also over provisioning & firmware, thanks to bug for reminding me, so basically a whole host of things that make an SSD. For instance Intel's 540s ~ http://www.anandtech.com/show/10432/the-intel-ssd-540s-480gb-review

Third party NAND & third party controller, with tuned (custom?) firmware & 5 years warranty. I'll still maintain that RAW NAND is last on my list of things that make certain SSD's great, or reliable, though binning definitely helps & the type of NAND e.g. 3D vs planar.

 

[Readlight]

If it can not last 1000 years i not interested!

 

[trparky]

It also depends upon the size of the NAND cell. Remember how NAND endurance took a major shit as they shrank past 19nm? Hell, some 19nm planar NAND SSDs even suffered from decreased read performance as the data aged in the NAND cells due to voltage drift which required the SSD controller chip to do more error correction to maintain data integrity. Then all of a sudden they put the brakes on NAND shrinkage and put the whole thing in hard-reverse and went to 40nm with 3D-NAND. What happened? Endurance even with TLC shot way up. I figure that they will go to perhaps 50nm (to store more electrons) which will be capable of doing QLC while keeping endurance the same as TLC NAND.

But that's just my theory.

 

[bug]

If it can not last 1000 years i not interested!

Joking aside, CDs were supposed to last 100 years and we all know how that turned out.