Wednesday, September 27th 2017

Mobiveil, Crossbar Partnership Aims to Bring ReRAM to SSDs

Looking to bridge the gap between current SSDs and resistive memory technologies for the consumers that can actually afford it, Mobiveil and Crossbar have recently announced that they are working in conjunction to bring Crossbar's ReRAM technology to an SSD form-factor. ReRAM is a new type of non-volatile memory (meaning it stores data even when it's not being powered). It's based on a simple three-layer structure of top electrode, switching medium and bottom electrode, where the resistance switching mechanism is based on the formation of a filament in the switching material when a voltage is applied between the two electrodes. Crossbar in particular (this type of resistive memory is also being pursued by other companies, such as HP) says their ReRAM cell is very stable, capable of withstanding temperature swings from -40°C to 125°C, 1M+ write cycles, and managing data retention of 10 years at 85°C. As an upside, it is 3D scalable, and its production can be achieved in standard CMOS manufacturing fabs.

This isn't the next evolution on SSD's - at least, not for the general consumer. ReRAM production and implementation costs will be leagues ahead of what current 3D NAND memory production entails, thus making this a niche product that is there for the customers that absolutely require the fastest throughput possible across a standard interface. In this case, NVMe is the choice - particularly, Mobiveil's NVMe, PCIe and DDR3/4 controllers can easily be adapted to accommodate the Crossbar ReRAM architecture, which is capable of six-million 512B IOPS below 10us latency.
The companies expect this partnership to capture potential customers on Intel and Micron's 3D XPoint technology, which left users wondering where lofty, 1000x improvement performance claims had gone to. However, there's not much the partnership can achieve in that are, through that particular interface: at the type of speeds and IOPS that Crossbar's ReRAM can achieve, "(...) the NVMe interface becomes a large part of the delay", said Jim Handy, principal analyst at Objective Analysis.
The key benefit of using ReRAM in an SSD is that it reduces storage controller complexity - and bottlenecks. It does so by removing large portions of the background memory accesses needed for garbage collection, and also by providing independent, atomic erasure by eliminating the need to build large-block memory arrays in flash designs. The companies know the market will be small: "It's going to be pretty expensive," said Handy, comparing it to the NV-DIMM market. "The current NV-DIMMS are more expensive than DRAM, they're way more expensive than SSDs, but offer blazing speed for people who want to pay for it," Handy said.
Handy said the Crossbar ReRAM-based SSDs will find a niche with customers willing to pay top dollar for persistence and performance, adding that Intel is selling Optane at a loss because it helps the company sell more expensive processors. SSDs with 3D NAND are not in any danger, said Handy. "They will be far more economical than anything made of out Crossbar ReRAM," he said. Sources: EETimes, Independent.co.uk
Add your own comment

11 Comments on Mobiveil, Crossbar Partnership Aims to Bring ReRAM to SSDs

#1
Gasaraki
6 million IOPS... not bad.
Posted on Reply
#2
R-T-B
Raevenlord said:
(...) the NVMe interface becomes a large part of the delay
To be frank, I'm surprised to be reading that sentence so soon... Though this will surely cost a small fortune. Still, progress is progress, and it's always good.
Posted on Reply
#3
R0H1T
R-T-B said:
To be frank, I'm surprised to be reading that sentence so soon... Though this will surely cost a small fortune. Still, progress is progress, and it's always good.
Tbf do we really need such high speed storage ~ https://www.techpowerup.com/237384/eight-nvme-drives-raided-on-amd-x399-break-the-28-gb-s-barrier

The RAM densities are constantly going up & we'll be having DDR5 standard within the next two years, I can understand HPC but even data centers may not need such ultra high speed storage with density less than the current (or future) SSD's like ~ https://www.theregister.co.uk/2017/08/09/samsungs_128tb_ssd_bombshell/
Posted on Reply
#4
R-T-B
R0H1T said:
Tbf do we really need such high speed storage
No, not at all (at least consumers don't).

But that wasn't really what I was commenting on either.
Posted on Reply
#5
R0H1T
R-T-B said:
No, not at all (at least consumers don't).

But that wasn't really what I was commenting on either.
I wasn't inferring consumers by "we" but even within the industry there's really only a handful of streams which may need performance exceeding 20GBps for a single drive.
I can understand HPC & it's a rapidly expanding market, but I'm unsure if there's really a big demand for ReRAM or 3D xpoint unless they can match the best SSD densities out there.
Posted on Reply
#6
theoneandonlymrk
R0H1T said:
I wasn't inferring consumers by "we" but even within the industry there's really only a handful of streams which may need performance exceeding 20GBps for a single drive.
I can understand HPC & it's a rapidly expanding market, but I'm unsure if there's really a big demand for ReRAM or 3D xpoint unless they can match the best SSD densities out there.
Thats underestimating the stagnation in simulation techniques due to the physical constraints of current platforms , exascale computing isn't just about the processing it will need just as extraordinary data handling to be useful.
This like optane is of limited to marginal use to consumers unless they have stuck to Hdd based drives and for some reason Hate ssds.
Posted on Reply
#7
Boosnie
R0H1T said:
I wasn't inferring consumers by "we" but even within the industry there's really only a handful of streams which may need performance exceeding 20GBps for a single drive.
I can understand HPC & it's a rapidly expanding market, but I'm unsure if there's really a big demand for ReRAM or 3D xpoint unless they can match the best SSD densities out there.
pick any database that comes out of your mind.
That database needs an ultrafast access medium.
Posted on Reply
#8
CheapMeat
theoneandonlymrk said:

This like optane is of limited to marginal use to consumers unless they have stuck to Hdd based drives and for some reason Hate ssds.
You do know you can use Optane like any other drive without the special proprietary sauce marketed on slides by Intel? You could run in all Optane system without issue.
Posted on Reply
#9
theoneandonlymrk
CheapMeat said:
You do know you can use Optane like any other drive without the special proprietary sauce marketed on slides by Intel? You could run in all Optane system without issue.
My wallet would definitely have issues with that :)
Posted on Reply
#10
lexluthermiester
R-T-B said:
No, not at all (at least consumers don't).

But that wasn't really what I was commenting on either.
Gotta disagree. While I'm not in the "general consumer" level of user, I can tell you that even with my SSD Raid setup, my system still spends a decent amount of time waiting on the storage array. The level of performance offered by this technology will be most welcome in not just in the enterprise, workstation, desktop and laptop sectors but also the mobile sector. Extra performance is always welcome.
Posted on Reply
#11
R0H1T
Boosnie said:
pick any database that comes out of your mind.
That database needs an ultrafast access medium.
The question atm is speed vs density, can any of these alternative technologies match the speed of an SSD RAID array (for argument's sake) & also beat them in capacity?
Over time they might replace SSD but in the near to medium term I don't see any of these taking a major chunk of business away from NAND.
Posted on Reply
Add your own comment