Samsung Aims for 1,000-Layer NAND by 2030, Begins Wafer Bonding at 400 Layers

[Nomad76]

Samsung aims to create 1,000-layer NAND by 2030 relying on its new "multi-BV" NAND design. The Bell reports that this plan involves stacking four wafers to overcome structural limits. Wafer bonding technology plays a crucial role in this progress and Samsung intends to use it to break the 1,000-layer barrier. Samsung Electronics DS division CTO, Song Jae-hyuk, pointed out that wafer bonding allows separate production of peripheral and cell wafers before joining them into one semiconductor. The Bell says this technology will likely appear first in Samsung's 10th-gen NAND (V10), while industry experts think a single wafer can hold about 500 NAND layers when implementing only cell structures. In the past, Samsung has used the COP (Cell on Peripheral) technique, a method that places the peripheral circuit on one wafer, with NAND cells stacked on top. However, as NAND layers grow, the lower peripheral parts face more pressure potentially affecting reliability.

Samsung's plan involves working with China's YMTC, which should offer a hybrid bonding patent for V10 NAND. ZDNet reports that the South Korean tech company will start making its V10 NAND in large quantities in the second half of 2025, with about 420-430 layers. Besides wafer bonding, Samsung adds other technologies to its NAND plan. The Bell points out that cold etching using molybdenum, and other new ideas will start with 400-layer NAND and play a key part in growing to 1,000 layers. Samsung isn't alone in trying to create ultra-high-layer NAND products. Japan's Kioxia also wants to reach this goal through its "multi-stack CBA" (CMOS Bonded to Array) technology. The company's plan is even bolder hoping to sell 1,000-layer 3D NAND by 2027.



View at TechPowerUp Main Site | Source

 

[Bwaze]

We're reading about additional layer breakthroughs for years and years - what does this bring into consumer sector? Maximum capacity is stagnating for 6 years now, that was unthinkable in any other era of home computers... And price wise we're also going up, not down - with many vendors now offering 2 TB drives as largest capacity, to somehow hide the fact how overpriced they are.

 

[azrael]

We're reading about additional layer breakthroughs for years and years - what does this bring into consumer sector? Maximum capacity is stagnating for 6 years now, that was unthinkable in any other era of home computers... And price wise we're also going up, not down - with many vendors now offering 2 TB drives as largest capacity, to somehow hide the fact how overpriced they are.

I guess it's a bit like with GPUs. Why sell "cheap" to us peasants when they can get bags of cash from enterprise. Especially when NAND manufacturers have a habit of artificially limiting production so as not to hurt their bottom line. However, as capacity on a single die grows, prices are bound to go down.

 

[TheLostSwede]

I guess it's a bit like with GPUs. Why sell "cheap" to us peasants when they can get bags of cash from enterprise. Especially when NAND manufacturers have a habit of artificially limiting production so as not to hurt their bottom line. However, as capacity on a single die grows, prices are bound to go down.

What you call artificially managing might also be due to the fact that NAND has a pretty fixed production cost and if the companies that makes NAND are forced to go below the production cost, they would lose money and go bust.
Admittedly we had a taste of what SSDs should cost, but at the same time, we where getting to a point where it might not have been sustainable long term, but I don't know the exact cost of making NAND, but I believe we were getting close to the point where it was no longer profitable for some companies at least and that's why they cut back on production, just like in any other business.
At the same time, what is a reasonable price? I agree that current pricing isn't very attractive compared to what it was a couple of years ago.

 

[Bwaze]

I have very little sympathy for NAND companies that moan and bitch about market oversupply for years now. They are basically offering the same products for more than half a decade - why would anyone upgrade?

Yeah, benchmark figures have gone up, but most real world tests show that they haven't moved the user experience by any noticeable amount. Optane did, and would even more if they offered competitive transfer speeds, but wasn't worth the effort. So what we basically got are power hungry drives that emit a lot of heat for negligible improvements in application load, install times etc.

And if enterprise demand is so great (for all the AI server needs), why moan about consumer space? Are technologies so far apart those are two different product sectors with completely different solutions that don't transfer?

Can we get an affordable 8 TB SSD? No? We're further from them than we were at the release of first consumer 8 TB drive, Samsung 870 QVO 8 TB 5 years ago! How many extra layers have they crammed into chips, making them smaller and cheaper to make since then?

 

[TumbleGeorge]

How many extra layers have they crammed into chips, making them smaller and cheaper to make since then?

Enough for must we have 16TB m.2 SSD if they was decided to make nand chips with double capacity.

 

[Wirko]

I guess it's a bit like with GPUs. Why sell "cheap" to us peasants when they can get bags of cash from enterprise. Especially when NAND manufacturers have a habit of artificially limiting production so as not to hurt their bottom line. However, as capacity on a single die grows, prices are bound to go down.

You're making it sound like production is naturally unlimited like the growth of ivy, and companies actively fight that with pesticides.

 

[TheLostSwede]

I have very little sympathy for NAND companies that moan and bitch about market oversupply for years now. They are basically offering the same products for more than half a decade - why would anyone upgrade?

The NAND makers are far from offering the same product now as they did a decade ago.

Yeah, benchmark figures have gone up, but most real world tests show that they haven't moved the user experience by any noticeable amount. Optane did, and would even more if they offered competitive transfer speeds, but wasn't worth the effort. So what we basically got are power hungry drives that emit a lot of heat for negligible improvements in application load, install times etc.

It's hard to improve the performance with NAND, on an SSD the main way of getting better performance is putting more NAND channels in parallel, but even this has its obvious limits when it comes to random read/write performance. Optande/Xpoint didn't scale well in manufacturing though, which is why Intel/micron called it quits.

And if enterprise demand is so great (for all the AI server needs), why moan about consumer space? Are technologies so far apart those are two different product sectors with completely different solutions that don't transfer?

The NAND should largely be the same.

Can we get an affordable 8 TB SSD? No? We're further from them than we were at the release of first consumer 8 TB drive, Samsung 870 QVO 8 TB 5 years ago! How many extra layers have they crammed into chips, making them smaller and cheaper to make since then?

See, this is where you are getting things mixed up. NAND has barely shrunk at all when it comes to the nodes that NAND is manufactured on, it's even worse than DRAM. Not sure if you saw yesterday's news from micron about their new 1-gamma node, which is at the very least their fifth 10-nm class node for DRAM since 2018. By your logic, that's seven years of no progress.
Stacking the NAND wafers is not making NAND cheaper, as it's a complex and costly process, although TSV (Through Silicon Vias) has made it somewhat more manageable, but due to the difficulty in providing thin enough NAND wafers, there are some serious manufacturing limitations here. Up till now, it's been hard going above 300 layer stacks, which is why we've ended up somewhere around 230-250 tall stacks at the moment. Everyone is trying to make taller stacks, but it doesn't mean they're not working on improving the density per NAND wafer, but it's slow progress on both ends.

 

[azrael]

You're making it sound like production is naturally unlimited like the growth of ivy, and companies actively fight that with pesticides.

Well, I kind of call it artificial limitation when production is scaled back because profits are lower. I get that they cannot operate at or even below cost, but I do think there's room to either lower prices or grace us with higher capacity drives at the same price. Just my opinion.

 

[TheLostSwede]

Well, I kind of call it artificial limitation when production is scaled back because profits are lower. I get that they cannot operate at or even below cost, but I do think there's room to either lower prices or grace us with higher capacity drives at the same price. Just my opinion.

Again, don't mix up the NAND manufacturers with the SSD manufacturers, as they're not always the same company. Obviously, we should expect better prices from the likes of Samsung, micron, sandisk, Kioxia etc. that are vertically integrated, but apparently people are prepared to pay a premium for their products due to the vertical integration.

 

[Bwaze]

There are press releases for basically every major jump in layer number about how that will unlock larger capacity consumer SSDs, last ones from Kioxia and Samsung from early 2024 promised larger capacity drives and lower prices of 8 TB drives. A year later, and all we have is another news article of higher layer count, with price per TB going up, and max capacity actually lowering on many drives.

16TB M.2 SSDs will soon grace the market — Kioxia unveils 2Tb 3D QLC NAND to build bigger SSDs

Samsung's upcoming 280-layer QLC flash could allow for 16TB M.2 SSDs — claims up to 50% higher storage density than the competition

Samsung is prepping 400-layer V-NAND for future 16TB SSD storage and PCIe 5.0 performance

 

[Snoop05]

There are press releases for basically every major jump in layer number about how that will unlock larger capacity consumer SSDs, last ones from Kioxia and Samsung from early 2024 promised larger capacity drives and lower prices of 8 TB drives. A year later, and all we have is another news article of higher layer count, with price per TB going up, and max capacity actually lowering on many drives.

16TB M.2 SSDs will soon grace the market — Kioxia unveils 2Tb 3D QLC NAND to build bigger SSDs

Samsung's upcoming 280-layer QLC flash could allow for 16TB M.2 SSDs — claims up to 50% higher storage density than the competition

Samsung is prepping 400-layer V-NAND for future 16TB SSD storage and PCIe 5.0 performance

View attachment 386781

Yep, every two months there is press release grom each vendor about increasing NAND layer count, followed by announcement of their flagship consumer drive using single NAND package on 75% empty PCB and top SKU as 2TB
Glad to see the recent Samsung announcemet not being the case but for the most part this applies more often than not

 

[Frank_100]

Isn't it time someone makes a serious effort to bring re-ram or phase change memory to the market?

 

[RUSerious]

The NAND makers are far from offering the same product now as they did a decade ago.

It's hard to improve the performance with NAND, on an SSD the main way of getting better performance is putting more NAND channels in parallel, but even this has its obvious limits when it comes to random read/write performance. Optande/Xpoint didn't scale well in manufacturing though, which is why Intel/micron called it quits.

The NAND should largely be the same.

See, this is where you are getting things mixed up. NAND has barely shrunk at all when it comes to the nodes that NAND is manufactured on, it's even worse than DRAM. Not sure if you saw yesterday's news from micron about their new 1-gamma node, which is at the very least their fifth 10-nm class node for DRAM since 2018. By your logic, that's seven years of no progress.
Stacking the NAND wafers is not making NAND cheaper, as it's a complex and costly process, although TSV (Through Silicon Vias) has made it somewhat more manageable, but due to the difficulty in providing thin enough NAND wafers, there are some serious manufacturing limitations here. Up till now, it's been hard going above 300 layer stacks, which is why we've ended up somewhere around 230-250 tall stacks at the moment. Everyone is trying to make taller stacks, but it doesn't mean they're not working on improving the density per NAND wafer, but it's slow progress on both ends.

Well said. The majority of gains in density, capacity and performance are going to enterprise compute and large data centers which are a bit less price sensitive and need that density and performance to remain competitive. That said, prices are coming down slowly for 4TB M.2 drives - I'm guessing the high inflationary period in the US has had a lot to do with the lagging price/TB curve for consumer SSD.

 

[Pizdarenkowitch]

3D trench cell for next-gen NAND Flash is way more exciting than this crap.
Check out : https://www.imec-int.com/en/articles/imec-improves-memory-window-3d-trench-cell-next-gen-nand-flash

 

[Bwaze]

That said, prices are coming down slowly for 4TB M.2 drives - I'm guessing the high inflationary period in the US has had a lot to do with the lagging price/TB curve for consumer SSD.

There have been 4 TB SSD drives for 150 -180 EUR in mid-2023. This one is an older PCIe 3.0 drive, younger models were a bit more expensive. Now most of the drives are at least 70 EUR more expensive, and all the models that are coming out now are either even more expensive, or end at 2 TB. So how are the prices coming down? In which direction?

 

[TheLostSwede]

Isn't it time someone makes a serious effort to bring re-ram or phase change memory to the market?

No one has managed to produce it in large enough bit capacity, it's all stuck at something that's barely useful for storing a router OS in today.
This company is talking about high capacity ReRAM at 12 Mbit or 1.5 MByte...

Reasons to Choose ReRAM｜RAMXEED

www.ramxeed.com

 

[RUSerious]

There have been 4 TB SSD drives for 150 -180 EUR in mid-2023. This one is an older PCIe 3.0 drive, younger models were a bit more expensive. Now most of the drives are at least 70 EUR more expensive, and all the models that are coming out now are either even more expensive, or end at 2 TB. So how are the prices coming down? In which direction?

View attachment 386791

I stand corrected. I guess I should have said 'were'. The time frame when I looking to buy, they were going down. Decided on a 2TB drive because I already had 2 1TB Drives. So, the last year has sucked