Intel and Micron Produce Breakthrough Memory Technology
jmg
Posts: 15,148
or so the banner headlines claim.... They claim this
Now in production
but then say this
3D XPoint technology will sample later this year with select customers,
and Intel and Micron are developing individual products based on the
technology.
Hmm. maybe they really mean
"Production Qualified", which is still significant, and real commercial
products later this year, is some great stealth work.
three-dimensional checkerboard where memory cells sit at the
intersection of word lines and bit lines, allowing the cells to be
addressed individually. As a result, data can be written and read in
small sizes, leading to faster and more efficient read/write processes.
More details about 3D XPoint technology include:
["Memory cells are accessed and written or read by
varying the amount of voltage sent to each selector."]
So they can Read, WriteZero and WriteOne, just with varying voltage ? Impressive.
Now in production
but then say this
3D XPoint technology will sample later this year with select customers,
and Intel and Micron are developing individual products based on the
technology.
Hmm. maybe they really mean
"Production Qualified", which is still significant, and real commercial
products later this year, is some great stealth work.
-
Intel and Micron begin production on new class of non-volatile memory,
creating the first new memory category in more than 25 years.
New 3D XPoint™ technology brings non-volatile memory speeds up to
1,000 times faster1 than NAND, the most popular
non-volatile memory in the marketplace today.
The companies invented unique material compounds and a cross point
architecture for a memory technology that is 10 times denser than
conventional memory
three-dimensional checkerboard where memory cells sit at the
intersection of word lines and bit lines, allowing the cells to be
addressed individually. As a result, data can be written and read in
small sizes, leading to faster and more efficient read/write processes.
More details about 3D XPoint technology include:
-
Cross Point Array Structure – Perpendicular conductors connect
128 billion densely packed memory cells. Each memory cell stores a
single bit of data. This compact structure results in high performance
and high-density bits.
Stackable – In addition to the tight cross point array
structure, memory cells are stacked in multiple layers. The initial
technology stores 128Gb per die across two memory layers. Future
generations of this technology can increase the number of memory
layers, in addition to traditional lithographic pitch scaling, further
improving system capacities.
Selector – Memory cells are accessed and written or read by
varying the amount of voltage sent to each selector. This eliminates
the need for transistors, increasing capacity while reducing cost.
Fast Switching Cell – With a small cell size, fast switching
selector, low-latency cross point array and fast write algorithm, the
cell is able to switch states faster than any existing non-volatile
memory technology today.
["Memory cells are accessed and written or read by
varying the amount of voltage sent to each selector."]
So they can Read, WriteZero and WriteOne, just with varying voltage ? Impressive.
Comments
10 times denser than conventional memory ... what is conventional here? Nand Flash or maybe they mean DRAM?
1,000 times faster than NAND ... doesn't say if they mean read or write speed. Seems to imply no erase cycle.
128Gb per die across two memory layers ... that's a lot, roughly comparable to nand, certainly not 10x as much. They do state single bit per cell, whereas Nand is two or three bits per cell. That's still not 10x. Maybe they are comparing to Nor Flash on this one.
All forms of RAM are a grid of perpendicular conductors that have various junction types. A transistor doesn't make it any bigger necessarily.
Then the write speed claim may or may not include comparison with say MRAM. Since MRAM can be considered as nominally not in use yet. MRAM certainly hasn't reached it's claimed abilities nor any significant production.
-Tor
Intel is big on FinFETs lately, however they call it "tri-gate". I wonder how this "stack-able" memory will integrate or merge into current transistor design.
Currently the commercial CMOS technology is at 22nm, released in 2012, that's almost 67 times more dense than the 180nm technology being used for the Propeller II.
By the end of 2015 they are expecting to have 10nm commercial processes available, that's 324 times more dense than 180nm technology ... So if memory follows a similar path and is stack-able I would expect to see about a 10x increase simply going from 22nm to 10nm .... 484 /100 = 4.8x then if you stack it at least once ... that's about 9.7X ... so 10X does not seem out of the question.
And while I might point out, YES, these new processes are more expensive, but for instance going from 180nm to 22nm with a density gain of 67X ... 22nm vs 180nm is NOT going to be 67 times the cost if you are just jumping process and not adding any features to grow a current design. Also true that some components do not scale well such as capacitors, and they must remain the same size. However, in many cases the design can be changed, and would most likely need to anyway to meet certain timing requirements, so that smaller capacitors can be utilized.
This is all a moot point since Intel is already into the single digit nm technology where the common units now are in Angstroms .... not a huge jump though.... 10 Angstroms = 1 nm .... At that level though things are very different and Moore's law shall be interesting to see. i.e. The width of a stable covalent bond of 2 Aluminum atoms is about 3 Angstroms.
Memristor reference:
https://en.wikipedia.org/wiki/Memristor
FinFET reference:
https://en.wikipedia.org/wiki/Multigate_device
Here's an interesting read: http://www.dailytech.com/Exclusive+If+Intel+and+Microns+Xpoint+is+3D+Phase+Change+Memory+Boy+Did+They+Patent+It/article37451.htm
Yes, better than the worst-case FLASH they chose to compare with, but a long way short of unified memory spin.
It could be usefully used for write-rarely memory, like code memory, but will never compete with DRAM for Data stacks.
It may drive a renaissance of separate CODE and DATA.
I can see a new method of virus attack emerging from this, wayward software that seeks out XPoint memory and writes it to death...
At their claimed speeds, it could be all over in under 1 second.
Its going to be a much faster, lower power flash.
I really wonder if a simple 32GB module could be put on every motherboard, and simply used as the default boot partition to RAM.
End of the day, shutdown changes only those files which changed.
Or, hibernation file written to Xpoint
Not that the same couldn't be done now with Flash.
On the plus side everyone, watch SSD prices plummet in 5.4.3.2, even with the 300-400% markup Xpoint will initially have.
So, that leaves premium server grade PCIe expansion cards as the target market.
In the longer term, maybe the density will improve, more stacked layers for example, there is a decent chance of wiping Flash from existence. If that happens then it's significantly improved endurance, compared to Flash, will make it very hard to knock from top position.
On the other hand, I'm not convinced there is a generic solution for extensive layering. All the 3D talk is limited to a few layers at best, as far as I know.