Intel's First 10 nm Chips to the Market are 64-layer 3D NAND

[btarunr]

Non-volatile memory often has the first pick of a new silicon fabrication process, as it is a low-risk development. A NAND flash chip is essentially a sea of transistors, with a fraction of R&D cost of something as specialized as a CPU die. It should come as no surprise, then, that the first chips to be built on Intel's swanky new 10 nanometer fabs will be a 64-layer 3D NAND flash memory, the first of its kind for data center applications.

With its 10 nm process, Intel is introducing FinFET Hyper Scaling, Intel will increase transistor densities by 2.7 times over the kind of densities one would traditionally expect from 10 nm. This lets the company scale up NAND flash storage densities by just that much more. The first 10 nm 64-layer 3D NAND flash chips will have high data densities, while at the same time, Intel will be able to push low volumes, characteristic of a new process. This explains why the first SSDs built with these chips are targeted at data-centers, so fairly expensive, high-capacity SSDs can be pushed to customers that can afford them.



View at TechPowerUp Main Site

 

[evernessince]

Samsung already has a few 64 layer NAND product for high end enterprise customers and they have been ramping up volume since june for mass market products. Right now their 960 pro only uses 48 layers.

I would personally buy samsung before Intel as Samsung have proven time and time again that it's SSDs are the best performance and endurance wise. Intel offers middle of the road products at much too high prices. Pretty much Intel's calling card.

 

[Prima.Vera]

But they are still going to sell 500GB NVMe drives for premium. Greedy bsterds.

 

[Boosnie]

[...] Intel is introducing FinFET Hyper Scaling, Intel will increase transistor densities by 2.7 times [..]

Does this mean we'll have 1TB SSD for the today price of a 256GB? [lol]

 

[Wark0]

On StorageReview i didn't read that the NAND was 10nm ?

 

[Valantar]

Does this mean we'll have 1TB SSD for the today price of a 256GB? [lol]

If production cost per wafer was the same across nodes, then yes, in an ideal world. However, as smaller nodes are more expensive to manufacture, price per wafer rises, which means bit-cost doesn't drop quite as much. Combine that with an ongoing shortage (and actors who seem intent on racking up all the tell-tale signs of price manipulation, non-competition and other illegalities) and intense demand and you have quite the different scenario.

I guess in a decade or so, we'll see Intel, Samsung and a few others receive "huge" (but economically insignificant) fines from the EU or the FTC for this whole mess. In the meantime, users and consumers are stuck with the bill, lining the pockets of callous investors.


More to the point of the article: I thought one of the key advantages of 3D NAND was the ability to produce higher densities at larger nodes, and expanding capacity by added layers rather than shrinking litography. Have they adjusted the NAND design to be more innately 3D, and thus hold more electrons even when manufactured on smaller process nodes? Or are we once again nearing the "TLC needs eight voltage states, and our NAND cells only fit roughly that many electrons" limit?

 

[Vya Domus]

It's not that these SSDs are expensive because they are for datacenters , Intel SSDs have been grossly overpriced even for the consumer products.

 

[ZeDestructor]

Samsung already has a few 64 layer NAND product for high end enterprise customers and they have been ramping up volume since june for mass market products. Right now their 960 pro only uses 48 layers.

I would personally buy samsung before Intel as Samsung have proven time and time again that it's SSDs are the best performance and endurance wise. Intel offers middle of the road products at much too high prices. Pretty much Intel's calling card.

Most of that is the controller, not the NAND. Took Samsung close to 2 years and mass-market 3D NAND before they matched Intel's S3xx0/730 series of SSDs, and a solid year before they had anything to show against the P3xxx/750 line at all. Intel, it seems, likes to come in, disrupt the market with a massive change in storage performance, then let it all languish until their next major breakthrough. It's not necessarily a bad thing, but it does leave plenty of room for competition to encroach.

 

[Prima.Vera]

Most of that is the controller, not the NAND. Took Samsung close to 2 years and mass-market 3D NAND before they matched Intel's S3xx0/730 series of SSDs, and a solid year before they had anything to show against the P3xxx/750 line at all. Intel, it seems, likes to come in, disrupt the market with a massive change in storage performance, then let it all languish until their next major breakthrough. It's not necessarily a bad thing, but it does leave plenty of room for competition to encroach.

After Samsung released the first 830 series, Intel was left behind so much is hilarious. Currently I cannot recommend any of their SSD products to to extreme crappiness of their price/performance ratio.

 

[R-T-B]

If production cost per wafer was the same across nodes, then yes, in an ideal world. However, as smaller nodes are more expensive to manufacture, price per wafer rises, which means bit-cost doesn't drop quite as much. Combine that with an ongoing shortage (and actors who seem intent on racking up all the tell-tale signs of price manipulation, non-competition and other illegalities) and intense demand and you have quite the different scenario.

I guess in a decade or so, we'll see Intel, Samsung and a few others receive "huge" (but economically insignificant) fines from the EU or the FTC for this whole mess. In the meantime, users and consumers are stuck with the bill, lining the pockets of callous investors.


More to the point of the article: I thought one of the key advantages of 3D NAND was the ability to produce higher densities at larger nodes, and expanding capacity by added layers rather than shrinking litography. Have they adjusted the NAND design to be more innately 3D, and thus hold more electrons even when manufactured on smaller process nodes? Or are we once again nearing the "TLC needs eight voltage states, and our NAND cells only fit roughly that many electrons" limit?

It's not so much node density cost increases because as you note, 3D-NAND is actually manufactured on a smaller node. It's more recoup of R&D costs. That shit is not and never will be free, even in an "ideal world"

Oh, and then of course there is profit. Nom nom.

 

[ZeDestructor]

After Samsung released the first 830 series, Intel was left behind so much is hilarious. Currently I cannot recommend any of their SSD products to to extreme crappiness of their price/performance ratio.

Erm.. what? While it is true that the 830 held the top spot in it's time, the S3700 (and it's lower-endurance, lower-sustained performance brethren including the 730) ruled supreme from 2012 to 2014, which is where the 850 Pro comes in while the 850 EVO was firmly in the trading blows region.

Of course, all that didn't matter anyways, cause by 2014 Intel had the P3700 out ruling the roost with it's NVMe-ness up until the Samsung 960 twins came along in 2016.

Then again, Optane once again took the crown a bit later in 2017 (originally planned for a 2016 release), though this time it's firmly "blood of the firstborn" pricing, rather than just expensive

Looking at that history, Samsung looks to be about 2 years behind what Intel can do, and that Intel doesn't really care much for iterative chart-topping, preferring instead of release much, much more impactful performance increases. Prices be damned.

PS: if you ask around the datacenter peeps, you'll hear a common theme of Intel not always being the absolute best choice, but always being a very good choice, especially when it comes to firmware niggles. This sort of consistency is hard to get, and Intel has it, while Samsung hasn't quite reached there.

 

[Prima.Vera]

Ze, read again, I was talking about the PRICE/PERFORMANCE rapport of Intel's SSD products.

 

[ZeDestructor]

Ze, read again, I was talking about the PRICE/PERFORMANCE rapport of Intel's SSD products.

At the high-end, they've been pretty competitive. I mean, you wouldn't call an 850 Pro or a 960 Pro good value for money now, would you?

EDIT: Also, there is a non-zero value for other stuff like support and firmware quality, that those of us buying Intel value and pay for.