Tuesday, August 4th 2026

Kioxia and Sandisk Unveil 10th-Gen 332-Layer QLC NAND with 4.8 Gb/s Interface Speeds

Kioxia Corporation, a subsidiary of Kioxia Holdings Corporation (TOKYO: 285A) and Sandisk Corporation (Nasdaq: SNDK) today unveiled their next-generation Quad-Level-Cell (QLC) 3D flash memory technology, which delivers up to 60% increase in bit density compared to their 8th-generation, surpassing 37 Gb/mm² and setting the industry benchmark for bit density, performance and power efficiency.

Leveraging the companies' revolutionary CMOS directly Bonded to Array (CBA) technology which fabricates CMOS logic and the memory array on separate wafers before bonding them together with high-precision wafer-to-wafer alignment, the new generation improves read and write bandwidth compared to the 8th-generation 3D flash memory and becomes the industry's first QLC 3D flash memory technology to reach a 4.8 Gb/s interface.
Additional interface enhancements improve I/O data-out transfer power efficiency. These large power efficiency upgrades directly address the power and cooling challenges of modern AI and cloud infrastructure.

Chief Technology Officer at Kioxia, Hideshi Miyajima, said, "As AI continues to underpin the advancement of society, its applications are expanding from generative AI to agent-based and physical AI, while the use of data is becoming increasingly diverse and sophisticated. QLC technology enables the efficient storage of rapidly expanding data volumes, helping deliver greater performance and scalability for AI systems. Kioxia will accelerate development toward the commercialization of its 10th-generation flash memory products incorporating QLC technology."

Chief Technology Officer at Sandisk, Alper Ilkbahar, said, "As AI training, inference, and hyperscale datacenter workloads drive unprecedented growth in data generation, the storage industry faces increasing pressure to deliver greater capacity, higher performance, and improved energy efficiency. By redefining the performance and efficiency envelope of QLC NAND, our 10th-generation QLC 3D flash memory delivers simultaneous gains in density, bandwidth, and energy efficiency and establishes a new paradigm for high-capacity flash storage to provide a scalable foundation for next-generation infrastructure applications."

Key 10th-genaration QLC 3D flash memory technology highlights include:
  • 332-layer architecture and optimized floorplan design deliver up to 60% increase in bit density, achieving >37 Gb/mm² compared to the 8th-generation.
  • 4.8 Gb/s*3 NAND interface enabled by Toggle DDR6.0 and the Separate Command Address (SCA) protocol to unlock the full potential of fast interface.
  • CMOS directly Bonded to Array (CBA) architecture enhances density scaling, performance, and manufacturing efficiency.
  • Power-Isolated Low-Tapped Termination (PI-LTT) improves I/O data-out transfer power efficiency.
Source: Kioxia
Add your own comment

24 Comments on Kioxia and Sandisk Unveil 10th-Gen 332-Layer QLC NAND with 4.8 Gb/s Interface Speeds

#1
Athena
They don't list endurance of these, IIRC, other QLC NAND was around 1000 P/E, and I wonder if these are more, or less, or the same?
Posted on Reply
#2
Ihnako
I miss MLC drives.
Posted on Reply
#3
lexluthermiester
AthenaThey don't list endurance of these, IIRC, other QLC NAND was around 1000 P/E, and I wonder if these are more, or less, or the same?
QLC 1000P/E cycles?? You're being very optimistic. Try 600 or 650 at most.
Posted on Reply
#4
outlw
1000 P/E cycles seems quite... optimistic for QLC.
I hope it has improved significantly since Gen. 1 QLC came to market, but I seem to recall that QLC was only able to handle 150~300 P/E cycles.
Posted on Reply
#5
lexluthermiester
outlw6669I seem to recall that QLC was only able to handle 150~300 P/E cycles.
It's improved since that time, but let's be fair, even if we accept a P/E rating of 700, that's still terrible durability compared to the P/E of TLC, which is 1800 to 2200 for even the most sub-par samples.
Posted on Reply
#6
Wirko
IhnakoI miss MLC drives.
So build up a stock of TLC drives because otherwise you'll soon miss those too.
outlw66691000 P/E cycles seems quite... optimistic for QLC.
I hope it has improved significantly since Gen. 1 QLC came to market, but I seem to recall that QLC was only able to handle 150~300 P/E cycles.
The SN5100 with Kioxia Bics 8 QLC chips has 600 TB writes per TB capacity declared, so the raw chip endurance must be significantly higher. How much higher? Depends on the optimism level but at least some SSDs must survive that, no?
Posted on Reply
#7
R-T-B
lexluthermiesterQLC 1000P/E cycles?? You're being very optimistic. Try 600 or 650 at most.
Upping the layers helps with durability all else being equal (because they can raise the process node size and fit more electrons in the charge trap).

That being said, even 1000 is pretty poor if they even reach that. I like good TLC with ram buffer way more. Shame they seem to be killing those in consumer land, slowly.
lexluthermiesterIt's improved since that time, but let's be fair, even if we accept a P/E rating of 700, that's still terrible durability compared to the P/E of TLC, which is 1800 to 2200 for even the most sub-par samples.
That and you must remember all other things being equal, TLC can leverage these same benefits to get even more margin.
Posted on Reply
#8
Athena
lexluthermiesterQLC 1000P/E cycles?? You're being very optimistic. Try 600 or 650 at most.
I thought I was being conservative :laugh:



that is with 144 layer QLC... so for a cool $8K, this can be yours!

this is the graphic that I remembered...


That paints a rather depressing picture, 1K is considered acceptable these days, and I suppose it's true for lots of people that don't do write heavy workloads, so for the average consumer, that still is fine
Posted on Reply
#9
Wirko
R-T-BUpping the layers helps with durability all else being equal (because they can raise the process node size and fit more electrons in the charge trap).
They didn't say they made each layer thicker or a cell's surface area larger. Instead, the taller stacks "deliver up to 60% increase in bit density".
R-T-BThat being said, even 1000 is pretty poor if they even reach that. I like good TLC with ram buffer way more. Shame they seem to be killing those in consumer land, slowly.
It now seems that enterprise datacenter drives are going to be either QLC for dense storage or SLC for caching, with nothing in between. And no one will be developing new TLC technologies just for client PCs, we'll only be getting breadcrumbs from the AI DC table.
AthenaThat paints a rather depressing picture, 1K is considered acceptable these days, and I suppose it's true for lots of people that don't do write heavy workloads, so for the average consumer, that still is fine
You may be confusing the NAND chips endurance ratings (P/E cycles) with SSD endurance ratings (TBW divided by capacity).

Anyways, even depressingly low numbers like 300 or even 240 (for an SSD, not a NAND chip) would be acceptable if the drives worked reliably till the last byte and then switched to a read-only state forever. A lot of anecdotal evidence tells us they often don't, and when they suddenly die, they are dead.
Posted on Reply
#10
R-T-B
WirkoThey didn't say they made each layer thicker or a cell's surface area larger. Instead, the taller stacks "deliver up to 60% increase in bit density".
Yes, adding layers increases density without increasing size or fanout by nature. I'd furthermore be shocked if the nodes have not been relaxed.
Posted on Reply
#11
Wirko
R-T-BYes, adding layers increases density without increasing size or fanout by nature. I'd furthermore be shocked if the nodes have not been relaxed.
Who knows. Analysts are happy to slice and make images and measurements of new processor chips but they almost never think of NAND, and absolutely never think of DRAM.
Posted on Reply
#12
R-T-B
WirkoWho knows. Analysts are happy to slice and make images and measurements of new processor chips but they almost never think of NAND, and absolutely never think of DRAM.
It is indeed a mystery. All I can tell you is P/E rising doesn't come out of thin air. My bet is on the one law of physics that helps them.
Posted on Reply
#13
lexluthermiester
WirkoSo build up a stock of TLC drives because otherwise you'll soon miss those too.
No. Unless there's a break-through in NAND durability, TLC is not going anywhere.
R-T-BUpping the layers helps with durability all else being equal (because they can raise the process node size and fit more electrons in the charge trap).
Not by that much..
R-T-BThat being said, even 1000 is pretty poor if they even reach that.
True.
R-T-BI like good TLC with ram buffer way more. Shame they seem to be killing those in consumer land, slowly.
I've heard/read nothing of this. Very much doubt it. QLC is WAY too slow for premium applications.
R-T-BThat and you must remember all other things being equal, TLC can leverage these same benefits to get even more margin.
Also true.
AthenaI thought I was being conservative :laugh:



that is with 144 layer QLC... so for a cool $8K, this can be yours!
That is marketing nonsense. They have yet to demonstrate that claim in real world applications. TLC however, might be in that range or better.
Athenathis is the graphic that I remembered...
This is closer to accurate. SLC is currently in the 250k range, MLC in the 50k range and TLC in the 3k to 4k range. QLC has yet to be demonstrated to function reliably function above 700P/E cycles, unless I've missed something somewhere.
Posted on Reply
#14
remixedcat
Well those infamous wd green sn350 nvmes have tlc nand w double digit endurance lmao



While seagate firecuda x1070 has qlc w 600tbw for 1tb capacity.

Posted on Reply
#15
Prima.Vera
I though the next big thing in storage expansion are the "new" PLC drives.
That would put SSDs in the same size and durability rank as the HDDs. ;)
Posted on Reply
#16
remixedcat
Prima.VeraI though the next big thing in storage expansion are the "new" PLC drives.
That would put SSDs in the same size and durability rank as the HDDs. ;)
actually HDDs have higher write endurance than SSDs and PLC would have a handful of P/E cycles lol

Posted on Reply
#17
Prima.Vera
remixedcatactually HDDs have higher write endurance than SSDs
Not if you are called Seagate ;)
Posted on Reply
#18
remixedcat
Prima.VeraNot if you are called Seagate ;)
or backblaze and their shitty enclosures that kill drives.
Posted on Reply
#19
Prima.Vera
To be honest, I would love me a 20TB PLC SSD drive for all my backup and storage stuff. For cheap, I mean :D
Posted on Reply
#20
remixedcat
Prima.VeraTo be honest, I would love me a 20TB PLC SSD drive for all my backup and storage stuff. For cheap, I mean :D
couple-three-bucks lol
Posted on Reply
#21
outlw
AthenaI thought I was being conservative :laugh:



that is with 144 layer QLC... so for a cool $8K, this can be yours!
This is honistly quite supprising to me, seeing QLC with a 3000 P/E cycle rating.
In fact, I suspect that this P/E rating is for the complete Enterprise SSD, not for the NAND itself.
That is to say, I suspect that they are actually advertising that the drive itself is warrantied to be written to 3000x its capacity... but the drive, I am sure, has a pile of overprovisioning, and the controller is also working to compress the data transparently, so that less writes overall are needed 🤔

For PLC, I have not looked deeply into it, but I remember reading that its performance is similar to that of spinning rust.
At that point, I really have to question why you would bother with a PLC SSD, over just getting a denser classical storage HDD.
The only real advantage I can see, is that there would be no moving parts with a PLC SSD, which is not that big of an issue for a datacenter anyways.
Posted on Reply
#22
lexluthermiester
Prima.VeraI though the next big thing in storage expansion are the "new" PLC drives.
No. PLC is a not viable. No one has yet to make samples that can survive common use tests. It's durability is pathetic because of all the voltage cycles applied and it's speed is slower than SATA1 because of all the right cycles require to set 5bits in series.

PLC is not a viable tech and is VERY unlikely to be.
outlw6669In fact, I suspect that this P/E rating is for the complete Enterprise SSD, not for the NAND itself.
That's possible. Either way, it's a mis-truth.
Posted on Reply
#23
remixedcat
lexluthermiesterPLC is a not viable.
it is if you are pure evil and make subscription ssds ha ha
Posted on Reply
#24
Prima.Vera
lexluthermiesterNo. PLC is a not viable. No one has yet to make samples that can survive common use tests. It's durability is pathetic because of all the voltage cycles applied and it's speed is slower than SATA1 because of all the right cycles require to set 5bits in series.

PLC is not a viable tech and is VERY unlikely to be.
Well, it's just about something I've read some time ago, about the future of SSDs and their extended capacity for consumers. Personally, I would probably buy one, as I mentioned, just to replace my 8TB spining rust for backup and storage only, but only if it be for VERY cheap, and their time endurance surpasses that of HDD.
Posted on Reply
Add your own comment
Nov 12th, 2026 04:40 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts