Intel & Micron 3D XPoint Memory

Non-volatile 3D stacked material that doesn't store charge that's many times denser than flash and changes resistance? Soo...Memristors?

Slower than but higher density than DRAM, faster and higher density than Flash. Still going to be a tough sell aside from servers.

But hey any improvements in non-volatile memory are a good thing! Much love to HP for digging-up this old technology :D
 
Well, so much for nvme, sounds like were back to the data bus being the bottleneck again, LOL.
 
Fantastic news. Shame they could not actually demonstrate it working, but I get the impression that products will come to market next year. I hope for some consumer products to come late next year.
 
So from the sounds of things this will eliminate the need for memory as we know it in a PC. Sounds like all you need is a decent HD with this stuff in it and it will be fast enough to use in place of RAM and also offer permanent storage. Sounds like a win-win to me. Looks like DDR4 may be the last generation of Random Access Memory.
 
Yup, definitely a Memristor. See here, the cell structure is exactly the same (page 6):

http://www.ece.ucsb.edu/~strukov/DmitriWebPage_files/MemristorSymposium2010Strukov.pdf

Also, there's going to be the problem of price. Each of those cells uses Titanium, and the signal wires are Platinum. Yay expensive!

https://en.wikipedia.org/wiki/Memristor#Titanium_dioxide_memristor

This is because Intel thinks everyone can afford thousands of $ for new components, granted even though is using these materials, if it is only using 1/100th of a gram of less then it wont drastically be expensive, but then again this is Intel charging $200+ for a chip that costs them $2 actual cost including margins.

"quote"
Intel and Micron unveil 3D Xpoint™ technology and create the first new memory category in more than 25 years. (huh what happened to HBM I thought that was also new memory, or HMC which is also micron?? if they are also new, then in the first line they lied lol. The biggest concerns for me are of course cost/capacity, power use, form-factor and compatibility as the current high end Intel drives while fast also pretty much REQUIRE specific Intel motherboards just to function which is no bueno for the intended customer base, I would love to see a USB sized ~500gb drive with in the 500+mb/s read AND write sustained speeds for ~$100 I don't see that happening however)
 
It's only expensive to make compared to a flash chip, which uses much cheaper materials.

If they can get it dense enough, they could justify the higher build cost.
 
If you're writing and calling individual memory cells instead of blocks, wouldn't this also require a massive change in programming? Meaning not only do you need to get the hardware out, but everything would have to be programmed for it as well?

Or would it be backwards compatible, with entire blocks being written/read just fine, but to truly take advantage it would have to be programmed specifically?
 
If you're writing and calling individual memory cells instead of blocks, wouldn't this also require a massive change in programming? Meaning not only do you need to get the hardware out, but everything would have to be programmed for it as well?

Or would it be backwards compatible, with entire blocks being written/read just fine, but to truly take advantage it would have to be programmed specifically?

I'm pretty sure this is managed by the controller on the drive.
 
I'm pretty sure this is managed by the controller on the drive.

Yes and no. There's no need for, say, the concept of block size anymore. For stuff that's optimized at a low level (like databases or operating systems), that's major. Also, we don't even have the tools to do much of anything interesting with NVMe's focus on asynchronous data access just yet.

I asked the Intel engineer that was in the Q&A afterwards what was in store on this front, citing the example of what they did with Threading Building Blocks and MKL on the multicore front, and got the "we're just talking about the product today" response :(
 
" from 4.4 zettabytes of digital data created in 2013 to an expected 44 zettabytes by 20204"

I hope to be around in 20204 to see their new drives.
 
A bit confusing, it both sounds like it is, and is not fast enough to replace RAM, the guy in the video uses the example of gaming, specifically in game worlds without loading, so VRAM.

But elsewhere it's hinted that it'll be in between NAND and RAM.

If it's not as fast as RAM, won't it just be a better version of NAND?

Also, poor HP, beaten to the punch, there must be some serious cursing going on there.
 
The arrangement looks a lot like RRAM/memristors. Memristors don't necessarily need platinum. HP's memristors have cells built with hafnium oxide, for example. Intel and Micro are being very secretive about the devices, so they may have found another material and don't want to disclose it yet.

(huh what happened to HBM I thought that was also new memory, or HMC which is also micron??
HBM is just DRAM connected with TSV and organized as a wide memory device. TSV have been used for a while for stacking memory on SoC and other chips, used in that way for over 15 years, and actually dates back to the 1960s.

edit: TheRegister mentioned this:
An Intel spokesperson categorically denied that it was a phase-change memory process or a memristor technology. Spin-transfer torque was also dismissed. Whatever it is, Intel and Micron have been developing it for about ten years.
XPoint might be what was previously referred to as "New Memory B" coming in 2017, but is still a RRAM design. "New Memory A" is probably the crossbar based RRAM Micron is currently sampling.
 
Last edited:
Well, I guess the next build is going to wait until something is settled so I know I'll be able to use devices which run using this tech.

This is what's going to lead to multi TB SSD's, yes?
 
2TB NAND based SSDs are out already. :p

If these type of 3D XPoint devices can be scaled as expected and stacked, it should help drive down SSD prices based on those. But we haven't seen its performance yet or release time frame...
 
The arrangement looks a lot like RRAM/memristors. Memristors don't necessarily need platinum. HP's memristors have cells built with hafnium oxide, for example. Intel and Micro are being very secretive about the devices, so they may have found another material and don't want to disclose it yet.

Fair enough. They hardly even mentioned enough details to pin it down as Memristors, so I let my mind run wild.

This was less of a product announcement, and more of a marketing field day :D
 
Fair enough. They hardly even mentioned enough details to pin it down as Memristors, so I let my mind run wild.

This was less of a product announcement, and more of a marketing field day :D

It was a press conference focused on the technology- definitely wasn't presented by them as a product announcement. The flash memory space is hypercompetitive and they're a long way off from shipping products with this tech, so I don't really fault them for not disclosing the secret sauce- it would allow others to catch up.

Public companies do funny things around the timing of releases of market-moving news like this, so I won't try too hard to dissect their motives for tooting their own horn at this particular moment.
 
Well, I guess the next build is going to wait until something is settled so I know I'll be able to use devices which run using this tech.

This is what's going to lead to multi TB SSD's, yes?

Might be a while, they said pricing somewhere inbetween NAND and DRAM.
 
If it isnt too much slower then dram and the 'blocks' are much smaller I wonder how much speed can be made up with a very wide bus and a lot of interleaving.

Having long term non volatile high speed storage is a nice though... there is an advantage to being able to just plain turn off/unplug your computer and knowing dram is now clear.... this could make virus/malware removal a lot more difficult.

Still being able to stick orders of magnitude more data in my GPU could make for some nice eye candy.
 
If it isnt too much slower then dram and the 'blocks' are much smaller I wonder how much speed can be made up with a very wide bus and a lot of interleaving.

Having long term non volatile high speed storage is a nice though... there is an advantage to being able to just plain turn off/unplug your computer and knowing dram is now clear.... this could make virus/malware removal a lot more difficult.

Still being able to stick orders of magnitude more data in my GPU could make for some nice eye candy.

How do you figure, virii/malware hides on your drive and loads into ram, it never just stays in ram...
 
So this is something totally new and not their version of Samsung's 3D V-NAND they are/were working on?
 
A good summarize about 3D XPoint from Cnet :

http://www.cnet.com/news/intel-and-micron-debut-3d-xpoint-storage-technology-thats-1000-times-faster-than-existing-drives/

"Intel and Micron today unveiled their all-new memory technology called 3D XPoint (pronounced "cross-point"). This is a new class of memory that can be used both as system memory as well as nonvolatile storage. In other words, 3D XPoint can be used to replace both a computer's RAM and its solid-state drive (SSD).

The companies claim that 3D XPoint is a major breakthrough in memory process technology, the first new memory category since the introduction of NAND flash in 1989. It's said to be extremely fast and durable, up to a thousand times faster (both in read and write speeds), and it will have higher endurance than existing NAND Flash memory currently being used in SSDs. What's more, it also has as much as 10 times greater density, leading to much more storage capacity in the same physical space, while remaining as energy efficient and affordable as existing NAND flash memory.

Mark Durcan, Micron's CEO, says the new technology is not to be confused with the 3D Flash memory used in Micron's latest SSDs, since 3D XPoint is a completely new class of memory.

According to Intel Vice President Rob Crooke, initially 3D XPoint will use PCI Express (PCIe) as its means of connecting to an existing computer, as PCIe currently has the the fastest bus speed of any peripheral interface. However, since PCIe is still not capable of handling the potential speed of 3D XPoint, in the future new ways will be developed to mount it, which will likely require entirely new motherboard architecture"
 
So this is something totally new and not their version of Samsung's 3D V-NAND they are/were working on?
XPoint is not NAND. NAND memory is basically an electron storage cell/trap. ReRAM has material that changes resistance when written, retaining state without power applied.

XPoint is apparently a second generation ReRAM, using cross-point topology to allow stacking. The only similarity to 3D V-NAND would be that both are stacked memory devices (3D).

I'm skimming through some papers and the potential things Intel and Micron could have improved on are the cell materials and isolation device between stacking layers.
 
As long as the latency isnt too many orders of magnitude different from SSDs and the price makes it accessible, it should get a foothold.
Each chip is 16GBytes, its bandwidth will scale proportionally with size of the memory device, to a point.
Hopefully this allows them to get a comparable bandwidth to todays PCI-E devices.
(It would be nice to have the b/w figures)
Next gen (more layers - a major point of this tech) could scale seriously.
 
A good summarize about 3D XPoint from Cnet :

http://www.cnet.com/news/intel-and-micron-debut-3d-xpoint-storage-technology-thats-1000-times-faster-than-existing-drives/

"Intel and Micron today unveiled their all-new memory technology called 3D XPoint (pronounced "cross-point"). This is a new class of memory that can be used both as system memory as well as nonvolatile storage. In other words, 3D XPoint can be used to replace both a computer's RAM and its solid-state drive (SSD).

The companies claim that 3D XPoint is a major breakthrough in memory process technology, the first new memory category since the introduction of NAND flash in 1989. It's said to be extremely fast and durable, up to a thousand times faster (both in read and write speeds), and it will have higher endurance than existing NAND Flash memory currently being used in SSDs. What's more, it also has as much as 10 times greater density, leading to much more storage capacity in the same physical space, while remaining as energy efficient and affordable as existing NAND flash memory.

Mark Durcan, Micron's CEO, says the new technology is not to be confused with the 3D Flash memory used in Micron's latest SSDs, since 3D XPoint is a completely new class of memory.

According to Intel Vice President Rob Crooke, initially 3D XPoint will use PCI Express (PCIe) as its means of connecting to an existing computer, as PCIe currently has the the fastest bus speed of any peripheral interface. However, since PCIe is still not capable of handling the potential speed of 3D XPoint, in the future new ways will be developed to mount it, which will likely require entirely new motherboard architecture"


Sounds good, I just skimmed the article and 3D v-nand popped into my head.

I know they are working on that too to compete with Samsung.
 
Pcie4 at +30GBs should be here in time for this tech. If they can saturate a full 16 lanes that would be impressive. Give me that and a cpu with 8+ gigs of HBM.
 
I honestly didn't expect to see secondary storage go out the door until the first quantum/graphite-based chips in desktops.
 
I have not read the entire article, but I don't see this in the consumer space immediately. It will likely be released / marketed for commercial / enterprise space with an extreme price premium which will require a certain release / patch level O/S to utilize.

The only way this makes it to the consumer space is starting out with small drives with lower performance like how they pushed 32Gb SSD's onto the market as a start at a high price to milk the consumer cash cow with intentions for future upgrades through planned obsolescence and incremental releases. It will probably be 3-5 years until the average Joe Blow has a decent sized affordable drive that will perform on par with the first gen commercial offerings. Maybe they won't be available in the consumer space until Windows 11?

Nothing personal it's just business, and they will execute a plan for long term profitability. It always has been this way in the computer industry.. got to keep the wheels of commerce greased.
 
Last edited:
So is this a step in the direction of cartridge games for PC? That is I buy a game on a small chip in a small cartridge and plug it into some sort of port (with chaining) and access the game directly without installing it and it works about as fast as if on a RAM disk. And of course it will be protected from copying so good-bye Steam? And it can be updated via software patches since no one seems able to write a game that works out of the box.
 
Plus the multi-billion dollars fabs have to paid for somehow

Yes surely Intel with no considerable competition to speak of, prices the products fairly out of it's kindness. The stock owners are content and are not asking for more.
 
So is this a step in the direction of cartridge games for PC? That is I buy a game on a small chip in a small cartridge and plug it into some sort of port (with chaining) and access the game directly without installing it and it works about as fast as if on a RAM disk. And of course it will be protected from copying so good-bye Steam? And it can be updated via software patches since no one seems able to write a game that works out of the box.

If it can be read and written it can be copied as well. Game developers do not want old style console games because those can be sold freely in physical form. Plus it costs a lot of money to produce vs a digital download.

It's much more profitable for them to distribute digitally and prevent customers from selling their old games to other players.
 
Yes surely Intel with no considerable competition to speak of, prices the products fairly out of it's kindness. The stock owners are content and are not asking for more.

Intel is not the sole developer of this technology. And in the field of storage, which this is, there is a good bit of competition.

If this were a new cpu tech the argument would have more weight.
 
Back
Top