News YMTC: Our 3D QLC NAND matches endurance of 3D TLC NAND

[Admin]

YMTC's X3-6070 3D QLC NAND can sustain 4,000 P/E cycles, according to the company.

YMTC: Our 3D QLC NAND matches endurance of 3D TLC NAND : Read more

 

[usertests]

1. We don't believe you.
2. How about we improve the endurance of TLC NAND instead?

 

[ThomasKinsley]

My MLC SSD has lasted 12 years and is still kicking. If QLC can last long it would be a net positive.

 

[_Shatta_AD_]

1. We don't believe you.
2. How about we improve the endurance of TLC NAND instead?

How about you show us how to run a memory company and revolutionize the ssd industry? Actually no one cares about your paltry opinion.
Improving the endurance of TLC NAND doesn’t move us towards higher capacity. You really wanna stick with 2TB mainstream NVMe drives for another 4 years? Somebody gotta make a move to improve QLC NAND. You can only stack them so high before other limitations come into play. Also, improving the endurance of QLC will in turn benefit TLC as well. So keep those irrational comments to yourself.

 

[usertests]

Also, improving the endurance of QLC will in turn benefit TLC as well.

Yes, that was my point.

QLC NAND will only ever improve capacity from TLC by 33%. So you get ~5 TB instead of 4 TB, wow big deal. Adding more layers will continue to be the primary way to improve capacity.

 

[cyrusfox]

These numbers appear to be 6-10x what consumer drives warranty for on QLC, and 4x what the best enterprise QLC has achieved. I believe the P41 Solidigm drive has the best endurance at 400 cycles (e.g.1tb drive warranty is 400TBW). For enterprise QLC best I know of is also Solidigm, the D5-P5430 which does 1000+ cycles.

China claiming this base nand doing 4k cycles... skeptical, I do wonder how much the base Solidigm chip is capable of as there is definitely some write amplification occurring alongside wear leveling, so either Solidigm is including spare capacity to account for that, or a lot of controller magic, and I expect the true NAND R/W cycles to be somewhat close to what the drive is warrantied for in terms of cycles (TBW/drive size ≈ NAND cycle spec). If YMTC has the goods, lets see it on product! But an endurance warranty from YMTC is a whole different can of worms as I would not expect the same warranty support as Solidigm or any of the other manufacturers here. Love to see the tech pushed and still hopeful for a PLC release someday.

 

[Li Ken-un]

QLC NAND will only ever improve capacity from TLC by 33%. So you get ~5 TB instead of 4 TB, wow big deal. Adding more layers will continue to be the primary way to improve capacity.

Giving 5 TB from 4 TB as an example is a puzzling choice (even with the squiggly line) since that is 25% (the same as a density increase from QLC to PLC).

Comparing 4 TB to 3 TB is the accurate portrayal of a 33% increase. Both of those capacities exist as real products. Where has anyone even seen a 5 TB SSD?

 

[usertests]

Giving 5 TB from 4 TB as an example is a puzzling choice (even with the squiggly line) since that is 25% (the same as a density increase from QLC to PLC).

Comparing 4 TB to 3 TB is the accurate portrayal of a 33% increase. Both of those capacities exist as real products. Where has anyone even seen a 5 TB SSD?

It's called rounding down. 3 TB is uncommon, most consumer lineups skip from 2 to 4.

Any more China shills want to weigh in? Get your shots in.

 

[Li Ken-un]

It's called rounding down. 3 TB is uncommon, most consumer lineups skip from 2 to 4.

Wait what? There are zero 5 TB SSDs. That is about as uncommon as it can get. Among the numerical values you could have used, you chose one which had to be rounded (and with only one significant digit) rather than two clean integers. Seems a bit deceitful.

Any more China shills want to weigh in? Get your shots in.

Okay. I’ll bite.

Are you implying that any posts in this thread have been shilling? And if so, does the speaker of the facts detract from the veracity of the words? (see: the shill gambit)

 

[wbfox]

Running press releases from anyone is annoying enough. Running press releases from the country that has lied in most if not all of their press releases (using old TSMC stock, scraping off the intel logos, etc..) for the past few years is what a group with an agenda does.

 

[bit_user]

a data retention period of one year at 30 degrees Celsius, which is on par with TLC solid-state drives.

Thanks for this detail. I'd point out that hard drives traditionally offer a retention period of 5 years or longer.

 

[bit_user]

2. How about we improve the endurance of TLC NAND instead?

They probably have a pseudo-TLC mode, I'd guess. I think NAND makers tend to support lower-density operation both for more critical applications and for dynamically-sized write buffers.

 

[bit_user]

Improving the endurance of TLC NAND doesn’t move us towards higher capacity.

QLC only adds 33% more capacity over TLC. So, it's hardly revolutionary.

Also, having better TLC endurance could be important for shrinking cell sizes or increasing layer-count. So, it's not an irrelevant concern. Might also be relevant for extending the data retention period, which is important for some.

You really wanna stick with 2TB mainstream NVMe drives for another 4 years?

There are plenty of QLC NVMe drives. As I mentioned above, going from TLC -> QLC equates to going from a 2 TB drive to 2.66 TB. If you went all the way to PLC, you'd get another 25% more capacity, but with even greater compromises.

The main drivers of density improvement are using smaller nodes and more layers. There isn't a lot of mileage in further increasing bits per cell, especially when comparing with 3 bits/cell as the baseline.

BTW, the article doesn't mention how many Gb these dies are, nor their area or process node. I have no doubt that you can deliver a higher-performing QLC implementation on an older process node, but that runs counter to the goal of trying to maximize density. Isn't YMTC one of those NAND-makers that has been restricted from using higher-density fab equipment? In that case, they could be partly spinning a weakness into a perceived strength, here!

Also, improving the endurance of QLC will in turn benefit TLC as well. So keep those irrational comments to yourself.

So, you're acknowledging TLC endurance is a relevant concern? How is it then also irrational?

 

[bit_user]

These numbers appear to be 6-10x what consumer drives warranty for on QLC, and 4x what the best enterprise QLC has achieved. I believe the P41 Solidigm drive has the best endurance at 400 cycles (e.g.1tb drive warranty is 400TBW).

I think a big part of that discrepancy is that you're looking at user writes. Due to write-amplification, the amount of device writes is much higher. Also, the device is continually internally generating internal writes as it patrols for bit decay and rewrites data when it finds blocks with recoverable errors.

there is definitely some write amplification occurring alongside wear leveling, so either Solidigm is including spare capacity to account for that,

No, they specify endurance in terms of host writes, since a user doesn't have much control over the other stuff.

still hopeful for a PLC release someday.

No thanks. Big regressions for such a small capacity improvement.

I'm sure it'll happen, though. Too many unscrupulous device makers out there, trying to save a buck by any means necessary.

 

[cyrusfox]

I think a big part of that discrepancy is that you're looking at user writes. Due to write-amplification, the amount of device writes is much higher. Also, the device is continually internally generating internal writes as it patrols for bit decay and rewrites data when it finds blocks with recoverable errors.


No, they specify endurance in terms of host writes, since a user doesn't have much control over the other stuff.

As the NAND makers don't publicly announce their NAND P/E cycle capability, I have always estimated it on what the drives can achieve(warranty for). And for that, these stated 4k QLC P/E numbers are unreasonable/ unrealistic based on what is on the market today ( top QLC drive on the warranty side only boast ≈1060 total write cycles)

And on the topic of PLC

No thanks. Big regressions for such a small capacity improvement.

I'm sure it'll happen, though. Too many unscrupulous device makers out there, trying to save a buck by any means necessary.

Same was said about TLC when we moved from MLC... And look how the P/E chart hasn't held at all, albeit the switch to 3D stacked cells really changed the game for NAND(so not exactly fair). Controller improvements deserve a lot of the credit as well. I love the technological progression! And perhaps as we move to PLC other innovations will follow.

My ancient 1tb 660p drive that is only good for 200TBW is still at 80% drive health. Most consumer use is write once read many. PLC should fit the bill fine for the majority of uses.

 

[bit_user]

And on the topic of PLC

Same was said about TLC when we moved from MLC... And look how the P/E chart hasn't held at all, albeit the switch to 3D stacked cells really changed the game for NAND(so not exactly fair). Controller improvements deserve a lot of the credit as well. I love the technological progression! And perhaps as we move to PLC other innovations will follow.

You realize the more ECC bits they need, the less the gains you get from packing more bits per cell, right?

My ancient 1tb 660p drive that is only good for 200TBW is still at 80% drive health. Most consumer use is write once read many. PLC should fit the bill fine for the majority of uses.

If you think I'm concerned about write endurance, you're wrong. I'm always thinking about what happens when a machine stays switched off for a couple months, or a thumb drive goes that amount of time without being plugged in. There's only so far they can push density before NAND flash almost needs to be treated like volatile memory.