News Researchers Create Non-Volatile Memory as Fast as DRAM and 100x as Efficient

DavidC1

Distinguished
May 18, 2006
391
6
18,785
0
The problem is neither performance nor the nonvolatile aspect of it.

The problem is cost. DRAM has advantages of absolutely massive scale. Every computing system in the world uses it, be it notebooks, desktops, tablets, smartphones, IoT devices, services, even your routers and Alexa devices.

That scale allowed it to hit other "DRAM killers" with the speed and brute force of a speeding freight train. That's why they all died off. Not because they were technically inferior.
 

bit_user

Splendid
Herald
There will be a niche market for non-volatile high speed memory though.
What about using it for cache, in mobile chips? Currently, when mobile CPU cores drop into low-power state, they flush their caches and turn them off. However, if you have some nonvolatile memory, then you'd get the same benefit without the penalty of having to re-read data when the core wakes up.
 

DavidC1

Distinguished
May 18, 2006
391
6
18,785
0
The real cost is in the materials - silicon is cheaper than III-V semiconductors.
There will be a niche market for non-volatile high speed memory though.
When the volume is high enough, it won't matter as much. Material costs are only significant when the volume starts being comparable.

A newcomer needs a significant advantage to beat the incumbent. Heck, NAND flash used to be more expensive than DRAM per bit until the past 15 years. Yet its 1-2 orders of magnitude better in density.

Yeah, I was wondering how the density compares.

If the density is good enough, perhaps this stuff could challenge Intel/Micron's 3D XPoint.
There are always trade-offs. Western Digital had a graph of the tradeoffs of various different memory types.

Basically, DRAM is fastest because its volatile, and it has extremely high endurance. Flash is slow, but its very persistent and very low persistence. Even for 3D XPoint Micron had a presentation where they said a version could be made where its more expensive/fast/less persistent vs less expensive/slower/greater persistency.

Many will fail, as it has in the past.
 

Olle P

Distinguished
Apr 7, 2010
555
15
19,015
23
The problem is cost. DRAM has advantages of absolutely massive scale. ...
You don't use DRAM for SSDs, do you?
What I see here is essentially an SSD with the speed of DRAM.
Given that there isn't a (significantly) limited number of writings possible, which we don't know yet, it might be possible to use this new memory as a combined storage an RAM unit!
 

mradr

Distinguished
Oct 12, 2011
10
1
18,515
0
The trouble with this sort of News, is we get reports of this type of fast storage every few years and yet still no actual product.
Let's wait until it's launched? :eek:
Not true, SSDs, Intel Opt, etc are some of those faster memory that was talk about. Now we even have a faster interconnection with Gen4 and 5 on the way.

With that said - this new memory to make it to market fast enough still has to meet:
  1. Easy to scale
  2. Cost of materials
  3. Easy manufacturing
  4. Performance Scale ( long as it is with in reason of DRAM )
Mostly, for scale of manufacturing, we can just replace it in small scale areas such as for cache drives or for the DRAM found on SSDs/HDD. The big cost though will be what does it cost and the yields to creating it. If it isn't too complex to create a layer of them and if they can be stack - then you already have a winner that will start replacing DRAM/NAND products pretty quickly. As far as production goes - I am sure companies would love to manufacture the stuff if it offers more for less. Sure it wont be a over night thing - but if it can use today's manufacturing equipment then you could see it in 6 months far faster if they have to create a new plant or 3-5 years. It wouldn't even have to beat DRAM in terms of being the fastest - long it is with in reason - it still would be faster than going over a wire to read data from another device thus could see its head pop up almost anywhere on board for the MB, CPU, and GPU. They will continue to use DRAM I am sure for main memory use - but having a way to store something longer terms does translate to way faster performance than something you have to fetch for.

For SSD replacement:
  1. Needs to be cheap or little higher or near about the same current cost per Gig (obs, its faster and new tech will make it start off a lot higher)
  2. Scale and materials have to be easy to double up or scale down to require less materials needs
  3. Complexity is just base off time - the problem is making sure yields are high enough for that given amount of time. Thus, it needs to be either simple - or the yields to storage mass just needs to be better than current flash memory can provide.
 
Last edited:

ASK THE COMMUNITY

TRENDING THREADS