X-NAND promises QLC memory that works at SLC speeds

nanoguy

Posts: 967   +14
Staff member
Something to look forward to: When we look at how SSDs have evolved over the last decade or so, it's hard not to appreciate just how fast and affordable they've become. That process is still ongoing though and with a new technology dubbed "X-NAND," solid state storage could get faster than ever before.

A decade or so ago you could find a 32GB SSD for around $500, and a 64GB drive for $1,100, but today you can find fast 1TB or even larger drives for less than $150. This evolution has taken years of R&D efforts, flash storage manufacturers cramming more bits of data into every memory cell, and fitting as many of these cells as possible onto a NAND chip.

The first consumer SSDs were single-level cell (SLC) drives meaning they could store 1 bit of data per cell, but typical consumer drives today integrate triple-level cells (TLC) and quad-level (QLC) cells, which means they can store 3 bits and 4 bits per cell, respectively. There's even 5-bit PLC NAND in the works, but that won't be here for a while -- 2025 at the earliest.

Most of our readers may already know that SLC NAND offers faster write speeds and higher endurance, but can get quite expensive, while TLC and QLC NAND are a more cost-effective way to build high-capacity drives. On the flip side, TLC and QLC NAND is comparatively slower, which is why manufacturers have had to employ various tricks (DRAM and SLC caches) to achieve good read and write performance, as well as acceptable endurance levels for typical usage in a personal, education, or business environment.

There is a company that claims to have a solution to this problem in the form of X-NAND. The technology was first announced at last year's Flash Memory Summit, but it flew under the radar until this month, when two patents for it were officially approved.

X-NAND is a different approach to NAND memory design and the work of Neo Semiconductor, a company founded in 2012 by Andy Hsu and Ray Tsay. Simply put, the aim of X-NAND is to offer the performance benefits of SLC NAND and the storage density of multi-level cell (MLC) NAND in a single package.

When compared to conventional multi-level cell designs, X-NAND reduces the flash die's buffer size by 94 percent, which would allow manufacturers to increase plane count from 2-4 to anywhere between 16 and 64 planes per die. This allows for greater parallelization of reads and writes on a NAND die and, in turn, could lead to increased performance even for SLC NAND.

Compared to QLC, X-NAND would -- at least in theory -- allow for 27 times faster sequential reads, 15 times faster sequential writes, and 3 times the random read/write speed of the former technology. At the same time, the new technology results in a smaller NAND die with lower power needs, which would keep manufacturing costs the same as with QLC. Endurance is a more complicated story, although the company says TLC and QLC could see an improvement.

It's worth noting these are performance estimates, so we're only looking at potential improvements to conventional NAND designs. Still, with TLC and QLC SSDs set to become the most adopted flash storage technologies across enterprise, desktop, and mobile markets, it's nice to see companies coming up with solutions for TLC and QLC's biggest challenges, which are write performance and endurance.

Neo Semiconductor is currently looking to forge partnerships with NAND manufacturers like Samsung, Intel, Micron, Kioxia, Western Digital, and SK Hynix to implement its IP in their designs, which includes 22 patents as of writing. If you're interested in a deep-dive into X-NAND, you can find one here.

Permalink to story.

 

ypsylon

Posts: 393   +328
When somebody comes up with the solution to QLC hopeless endurance then I'll get excited about that. As thing stands QLC (and all things which will follow it) is just crap for anything else than WORM drive.

I wouldn't trust QLC for anything other than movie storage. Certainly not for important data. It's terrifying seeing QLC drives being used as boot drives by SI.

Like many much touted techs a lot of this stuff never sees the mass production stage. Thing is, manufacturers don't care about longevity. They care about increased yields and profitsssss$$. QLC or PLC with all its hopelessness fit that bill really well. If they did care, they wouldn't have left the MLC, maybe TLC for cheaper drives, stage. No-brainer!
 

BSim500

Posts: 855   +1,941
When somebody comes up with the solution to QLC hopeless endurance then I'll get excited about that. As thing stands QLC (and all things which will follow it) is just crap for anything else than WORM drive. I wouldn't trust QLC for anything other than movie storage. Certainly not for important data. It's terrifying seeing QLC drives being used as boot drives by SI.
We are well passed the "peak" for the SSD speed vs endurance balance and going the wrong way. The almost mindless tunnel vision psychosis "but muh synthetics" or "the death of HDD's" is absurd when the former is the least noticeable metric and the latter isn't going anywhere with the increasingly sh*t endurance of 2021 SSD's in the real world. We are also well passed the peak for tech sites actually doing their own test-based articles vs the collective copy-paste PR infomercials. Imagine if the "We tested an Intel MLC drive back in 2013 and found endurance no problem so that's that, we won't bother ever reviewing it again" etc were applied to CPU reviews. We'd still be waiting for them to review Haswell in 2021...
 

Bawlsdeep

Posts: 141   +153
I look forward to the day when I can get single SSD stick that holds 10TB for less than $1000.

Well SSDs that have low cost per TB always suffers in longevity.

I would not want to buy a 10GB SSD using QLC or even PLC in some years, for 1000 dollars.. It will most likely come with a 2 year warrenty like most QLC drives (if not all) and wear out in a server/NAS environment pretty fast, especially if it's on 24/7 using RAID/ZFS etc.

I can't find proof which kind of SSD PS5 and XSX/XSS use. Probably QLC to keep price down. If thats the case, I expect alot of problems for people some years from now. Especially if they install and uninstall games alot. Alot of write really wears down QLC fast.
 

ET3D

Posts: 1,762   +407
The most exciting thing for me here was the speed rating for SLC. For sequential reads this technology competes with DRAM, and could potentially reach GDDR speeds. This could be revolutionary. The current consoles already use fast SSD streaming for allow for more seamless worlds, but imagine if the data could be read directly from the drive at RAM speed.

It looks like even TLC could compete in speed with system RAM. That's definitely a game changer.

can't find proof which kind of SSD PS5 and XSX/XSS use. Probably QLC to keep price down. If thats the case, I expect alot of problems for people some years from now. Especially if they install and uninstall games alot. Alot of write really wears down QLC fast.

It's unlikely. QLC's slowness is what this article is all about. The PS5's flash is particularly fast, so unlikely to use QLC, but I think that would also apply to the Xbox. I'd also quess that the PS5's flash is a lot over-provisioned, considering its strange size.
 

Bawlsdeep

Posts: 141   +153
The most exciting thing for me here was the speed rating for SLC. For sequential reads this technology competes with DRAM, and could potentially reach GDDR speeds. This could be revolutionary. The current consoles already use fast SSD streaming for allow for more seamless worlds, but imagine if the data could be read directly from the drive at RAM speed.

It looks like even TLC could compete in speed with system RAM. That's definitely a game changer.



It's unlikely. QLC's slowness is what this article is all about. The PS5's flash is particularly fast, so unlikely to use QLC, but I think that would also apply to the Xbox. I'd also quess that the PS5's flash is a lot over-provisioned, considering its strange size.

QLC is not that slow, the biggest issue with QLC is longevity, especially in write heavy scenarios. It's worse than TLC was 5 years ago. Today TLC is pretty decent for regular consumers/gamers tho. MLC is not really used anymore for consumer space. All top M.2 NVME SSDs with PCIe 4.0 uses TLC.

For game loading times, there's not a big difference between high-end NVME drives and mid-end, even low-end often delivers pretty much same result. Even SATA SSD's are not much slower in most cases. Going away from HHDs was a big leap for consoles tho, and it wil affect how next gen games are made (meaning PC gaming too). HDDs held back game dev's.

PS5s weird SSD size is probably because it's a custom ssd. 840GB but only has 667GB of free space to install games on. Highly doubt it's a 1TB drive with 160GB reserved. They would not do this, when Xbox SX has 1TB.

Sony talked ALOT about how crazy and next gen the SSD is, but high-end SSDs like Sasmsung 980 Pro and WD SN850 beats the standard SSD in loading times. It was marketing. Microsoft talked way less about their 1TB SSD and game loading times are the same or even faster than PS5 in some games.

I bet you could put a cheap QLC SSD in PS5 and see identical loading times to the stock drive. It's nothing crazy, going SSD was the only right choice tho.
 
Last edited:

Wereweeb

Posts: 63   +134
The most exciting thing for me here was the speed rating for SLC. For sequential reads this technology competes with DRAM, and could potentially reach GDDR speeds. This could be revolutionary.

Bandwidth, yes, latency, no. Even SLC NAND has high latency when compared to DRAM. You'll still need to buffer all the relevant data in RAM/L3 cache, and then the speed will be subject to their latencies, so no, no revolution there.

You could add an in-between SLC buffer to which you copy colder data to reduce the effects of TLC/QLC latency in cache misses, but that's independent of X-NAND and could have been done already (See Fuzedrive reviews for a taste of that).

At most it could help with startup and shutdown times, when you're copying a lot of data from the SSD to DRAM. SSD's are otherwise already "fast enough" that they're not the bottleneck. The bottleneck in computing speed is mostly in ferrying data from DRAM/caches to compute.

The only revolution here is making QLC viable as a technology.
 

Wereweeb

Posts: 63   +134
QLC is not that slow, the biggest issue with QLC is longevity, especially in write heavy scenarios.
But that slowdown is precisely what happens when you write straight to QLC. Of course it's quick when it's writing to SLC cache, it's still writing to SLC.

Thankfully, X-NAND tries to solve this so that you're always writing to SLC.
When somebody comes up with the solution to QLC hopeless endurance then I'll get excited about that. As thing stands QLC (and all things which will follow it) is just crap for anything else than WORM drive.
QLC has no endurance problem. If you're just a normal consumer, you don't need more than .3 DWPD. No, you're not downloading 333GB of files every single day.

And yes, let's ignore the .1 DWPD QLC SSD's, early QLC drives (Like early planar TLC drives) were very bad and just a way for companies to stress test QLC. Welcome to capitalism.

If you weren't just a normal consumer, you would understand the endurance requirements of different workloads and the class of SSD that you need, that they aren't typical for consumers, and you wouldn't be complaining about QLC in general or as a concept.

The main problem with QLC SSD's currently is that they're not much cheaper than TLC drives, so there's not much reason to downgrade. And that's because there's a class of consumer that will just buy the cheapest SSD's at a certain capacity, so they can get away with higher margins. That'll eventually solve itself.

And then you should complain about PLC/HLC up until companies stop releasing versions of PLC/HLC drives that are substantially worse than QLC. Again, welcome to capitalism.
 

BSim500

Posts: 855   +1,941
QLC has no endurance problem. If you're just a normal consumer, you don't need more than .3 DWPD. No, you're not downloading 333GB of files every single day.
The underlying issue that causes lower endurance also causes lower unpowered data retention, which very much does affect consumers who are told these will "replace" HDD's only to learn the hard way that using them as a backup drive for archiving irreplaceable data, then putting them in a cupboard for several months on end can have very different unadvertised outcomes vs the marketing pitch.

I wouldn't touch QLC with a barge pole for any backup drive that isn't plugged in every week, as even popular TLC drives start to show significant read speed slowdowns down to just 15MB/s because some cells in the drive have lost so much charge that the controller has to rely on advanced ECC to (hopefully) reconstruct the data when reading it back after months of being powered off, well before the "1 year guarantee" is up, and QLC has significantly less overhead vs cell charge loss than that.

The real problem is, no one is bothering to do serious grown up testing on this stuff that could easily lead to data loss if people rely on that now woefully out of date 8 years old presentation slide that was done on MLC drives, and really doesn't mean a thing for "less endurance than some optical RW discs" mediocre QLC SSD's.
 

BadThad

Posts: 635   +681
Thumbs up to any tech that makes things faster. I just hope the licences fees aren't prohibitive for the manufacturers. If any of the majors signs, then the rest almost have to in order to be competitive. It will be interesting to see how it plays out. Let us not forget 3D Xpoint which has now been "removed" from the consumer space.
 

ZedRM

Posts: 631   +395
Unless this new NAND they're promising is as DURABLE as SLC also, it's a waste. Durability is much more important than speed. 3DNAND TLC is on a good level of durability, QLC is not something for use as a boot drive and PLC is a joke.
 
Last edited:

ET3D

Posts: 1,762   +407
Bandwidth, yes, latency, no. Even SLC NAND has high latency when compared to DRAM. You'll still need to buffer all the relevant data in RAM/L3 cache, and then the speed will be subject to their latencies, so no, no revolution there.
Latency may be in microseconds instead of nanoseconds, but the increased bandwidth will still open up completely different paradigms.

Just for games, compare a 6GB/s drive that can supply 100MB of textures and other 3D data per frame at 60 FPS to a drive that can supply 2GB per frame. That's a whole different ballgame.

You want to analyse a big database? Reading at RAM bandwidth will speed that up significantly.

This will all require redesigning how computers work with drive data, but when disk read bandwidth is no longer a bottleneck, this opens up big opportunities.
 

Bp968

Posts: 236   +167
The underlying issue that causes lower endurance also causes lower unpowered data retention, which very much does affect consumers who are told these will "replace" HDD's only to learn the hard way that using them as a backup drive for archiving irreplaceable data, then putting them in a cupboard for several months on end can have very different unadvertised outcomes vs the marketing pitch.

I wouldn't touch QLC with a barge pole for any backup drive that isn't plugged in every week, as even popular TLC drives start to show significant read speed slowdowns down to just 15MB/s because some cells in the drive have lost so much charge that the controller has to rely on advanced ECC to (hopefully) reconstruct the data when reading it back after months of being powered off, well before the "1 year guarantee" is up, and QLC has significantly less overhead vs cell charge loss than that.

The real problem is, no one is bothering to do serious grown up testing on this stuff that could easily lead to data loss if people rely on that now woefully out of date 8 years old presentation slide that was done on MLC drives, and really doesn't mean a thing for "less endurance than some optical RW discs" mediocre QLC SSD's.

This is like complaining that your civic wont tow your yacht. Its not designed for that job. Using a qlc nvme drive as a removable backup is a bad idea. But using any ssd (or even many usb sticks) is a bad idea for offline storage.

Hdds work ok for offline storage but honestly anything that's important to you should go on multiple media's in multiple locations. Cloud storage (amazon prime is unlimited and free for photos), LTO tapes (their cheap now as long as you stay at lto4-5 or less), and local online backup like a nas. And if its vital to you then you should use a checksum utility on all of them so the files can be checked and repaired and the files should be checked on a rotation yearly.

Sorry, but there really isn't any truely reliable "put it on a device and forget about it" backup solutions anyone should trust. Certainly not single point backup solutions.

I always tell people a single hdd as a "backup" is no backup at all. And most ignore me. Its why I push most users to cloud storage. Its far and away the safest option for 99% of users.

As for QLC, its endurance is just fine. I have a intel (with a 5yr warranty) thats still doing fine after 4 years. I expect another 2-4 from it before I relegate it to backline duty as a game storage drive. If I dont have a new mb that supports more than 2 nvme before that point then ill grab a pcie add in card to hold some more ssds. My PC is hardrive-less now and it will stay that way. Ill be getting a qnap or similar nas at some point and stuff it full of drives and use its 10gb connection and treat it as local storage.
 

ZedRM

Posts: 631   +395
This is like complaining that your civic wont tow your yacht.
Very poor analogy.
Using a qlc nvme drive as a removable backup is a bad idea.
Their point, which you missed, is that it shouldn't be a bad idea. It should be a brilliant idea. Quality NAND should store data long term. SLC and MLC do this well. But with the chemistry formulations being used for TLC and QLC, long term powered-off data storage becomes a problem and it shouldn't be such a problem.
I always tell people a single hdd as a "backup" is no backup at all.
Nonsense statement. A hard drive used for backup is still a backup.
As for QLC, its endurance is just fine.
With P/E cycles being counted in the mid-hundreds, no it isn't. Not close to fine at all. QLC is GARBAGE. Full stop.
I have a intel (with a 5yr warranty) thats still doing fine after 4 years.
I'd bet real money your drive SMART data says otherwise..
And most ignore me.
Yeah, after reading your comment above, it's easy to understand why.
 
Last edited: