Researchers have unlocked the "Holy Grail" of memory technology

[Shawn Knight]

Forward-looking: Researchers at the Korea Advanced Institute of Science and Technology (KAIST) in South Korea have developed a new type of phase change memory that isn't subject to the shortcoming of earlier iterations.

Phase change memory, or PCM for short, works by shifting between two physical states: crystallized (with low resistance) and amorphous (with high resistance). Think of it as an optimal blend of DRAM and NAND flash.

DRAM is speedy but volatile, meaning the data stored in it disappears when power is cut (like when you turn off your computer). NAND flash memory, like what is used in SSDs, can retain data even when power is pulled but is significantly slower than DRAM.

PCM is both fast and non-volatile, but has traditionally been expensive to manufacture and is power hungry (heat is needed to melt the phase change material into an amorphous state, which hampers energy efficiency).

Earlier efforts to address high power consumption focused on reducing the physical size of an entire device through cutting-edge lithography techniques. Improvements were nominal, and the increased cost and complexity involved in fabricating on smaller tech was not justifiable.

Professor Shinhyun Choi and team devised a method to shrink only the components directly involved in the phase change process to create a phase-changeable nano filament.

The novel approach cut power consumption by 15 times compared to traditional phase change memory made using expensive lithography tools, and is also far less expensive to manufacture.

The new phase change memory retains many characteristics of traditional memory such as fast speed, a large on / off ratio, small variations, and multi-level memory properties.

Choi said they expect the results of their study to become the foundation of future electronic engineering, and could benefit applications including high density 3D vertical memory, neuromorphic computing systems, edge processors, and in-memory computing systems.

The team's research was published in the journal Nature earlier this month in a paper titled, Phase-Change Memory via a Phase-Changeable Self-Confined Nano-Filament.

Image credit: Shaine Tsou

Permalink to story:

Researchers have unlocked the "Holy Grail" of memory technology

 

[user556]

The single biggest shortcoming of all phase-change type memories is that they claim to be a RAM tech when they're really all ROM techs. Having a known limited write endurance is an outright fail for anything claiming to be a RAM tech.

 

[human7]

The single biggest shortcoming of all phase-change type memories is that they claim to be a RAM tech when they're really all ROM techs. Having a known limited write endurance is an outright fail for anything claiming to be a RAM tech.

The article is behind a paywall, so it is hard to verify, but the abstract says "The device maintains favourable memory characteristics such as a large on/off ratio, fast speed, small variations and multilevel memory properties.".

I too am curious about the endurance of this new memory type, but, like battery tech and 100 TB/s network tech that have been reported on, it won't matter to us it doesn't make it out of the lab. It's a long road from the lab to consumer products.

 

[user556]

It got my hackles up because Intel's failure with Optane was also talked about as if it were competing against DRAM too. It really wasn't. Optane, and all phase-change memories, should be compared to HDDs ... just like Flash memory is.

 

[Uncle Al]

I'm guessing this will take another 8-10 years before it makes it to the market in any meaningful way and perhaps longer for the price point to be affordable for the everyday computer user .....

 

[user556]

I'm guessing this will take another 8-10 years before it makes it to the market in any meaningful way and perhaps longer for the price point to be affordable for the everyday computer user .....

It'd need to be sold at on-par pricing to Flash ... in other words, not a chance. I guess that's why such techs never get marketed against Flash.

It'll get opportunities in harsh environments where Flash is not meeting application requirements.

 

[Jack77]

PCM could also mean instant-on features for devices. OTOH, you could lose the imo large benefit of simply turning the system off to clear the memory. E.g. in case of a crash, malware ..

 

[user556]

PCM could also mean instant-on features for devices.

Not really - For the very reason of limited write endurance. MRAM is probably the only memory type that could fill such a position without battery backup ... or the complex way with DRAM backed by PCM/Flash/whatever that's faster than paging to HDD.

 

[ChipBoundary]

The single biggest shortcoming of all phase-change type memories is that they claim to be a RAM tech when they're really all ROM techs. Having a known limited write endurance is an outright fail for anything claiming to be a RAM tech.

Endurance issue aside, comparing it to ROM is nonsensical. ROM doesn't have write cycles. It stands for read-only memory. If anything, in a weird way, ROM and RAM are more comparable than this tech. You can't actually write anything to RAM, because that isn't how it works. You can't write anything to ROM, either.

This would be a NVM tech, but if what is stated is true, then it will reach speeds in the region of what RAM does and circumvent the need for RAM in the first place. HP years ago was working on a similar concept that they dubbed "The Machine". In their case, they were trying to develop a system that used something called memristors. It would have allowed the elimination of RAM or any other such tech, because the permanent storage was as fast as, or faster than, any RAM. It also would have reduced the size of an entire server rack down to a cube that fit in the palm of your hand. Unfortunately memristors are extremely hard to manufacture and the yields are abysmal. So, HP shelved the project.

Finite write cycles are fine, if the speed and storage capacity exceed current modern technology and the price is lower than the combined cost of VM+NVM setup. As with SSD's they have a finite write cycle, but it is a simple matter of replacement at that point. If PCM can fill a similar role, great. Options are always welcome. And it is likely that memristors and PCM could have different applications such as physical environment and the like.

 

[DanteSmith]

KAIST again? Hm, either these guys have been quietly working out all the problems in hardware computing tech and are finally bringing it all to light at once, or the check is late and they want to show off some pretty papers.

 

[Gars]

KAIST again? Hm, either these guys have been quietly working out all the problems in hardware computing tech and are finally bringing it all to light at once, or the check is late and they want to show off some pretty papers.

its like the military, they dont show of something if they dont have at least one in the pocket

 

[user556]

Endurance issue aside, comparing it to ROM is nonsensical. ROM doesn't have write cycles. It stands for read-only memory. If anything, in a weird way, ROM and RAM are more comparable than this tech. You can't actually write anything to RAM, because that isn't how it works. You can't write anything to ROM, either.

Flash, like EEPROM before it, has always been classed as a ROM tech. There's no reason not to place other NVM in that class also.

This would be a NVM tech, but if what is stated is true, then it will reach speeds in the region of what RAM does and circumvent the need for RAM in the first place.

Except that it doesn't replace RAM at all. Best it could do is be a backing store. Which is what a filesystem is.

 

[Athlonite]

PCM could they not have thought to themselves oh there's already something using that acronym and choose a different name for it Why they didn't use PCFM ie: phase change filament memory

 

[ferrellsl]

Yeah, and fusion energy is gonna work any day now.

 

[SixTymes]

I recall reading about Phase-change memory 10 years ago. Did it take them this long to further R&D? OR, did this company take the original R&D engineering from another group?

 

[JCChua]

I really want and hope Intel's Optane Memory will have a glorious comeback as a viable replacement to the antiquated NAND/NOR Flash Memory. That's because the NAND Flash Memory of conventional SSDs aren't suited/fit to make the most out of NVME Interface especially with PCI-E 4.0 & 5.0. They will just wear out relatively fast. I strongly believe Intel's Optane Memory with proper investments & refinements can easily handle the blistering Sequential & Random Read & Write Speeds of PCI-E 4.0 & 5.0 NVME Interface while still lasting at least 10 Years.

 

[LemmingO]

Again? We've been going on about breakthroughs in MRAM for the past 20 years. What makes this any more relevant than the last iteration?
Intel's Optane was supposed to be MRAM. They carefully steered into an assistive SSD technology because phase change wasn't fit for purpose. Then they gave up on it altogether. And this was at a time that Intel still did something worthwhile.

From the KAIST site it sounds like they're likening the technology to next-gen CMOS. Not RAM or SSD.

 

[user556]

Intel's Optane was supposed to be MRAM.

It was never MRAM. Intel may have compared Optane/Xpoint to MRAM in it's press releases.

From the KAIST site it sounds like they're likening the technology to next-gen CMOS. Not RAM or SSD.

That'll be an AI marketing spin. Neural-network inference is almost entirely about memory.

 

[captaincranky]

I'm still stuck in the old ways where, "PCM" means, "Pulse Code Modulation".

 

[user556]

That's still very much current terminology. PCM, PWM, PDM are all modulations that won't ever go away. Same as pure analogue won't ever go away either. Different uses for the same TLAs is fine. Context matters is all.

 

[user556]

From the KAIST site it sounds like they're likening the technology to next-gen CMOS. Not RAM or SSD.

That'll be an AI marketing spin. Neural-network inference is almost entirely about memory.

Maybe more importantly, that memory use is ROM in nature. Its knowledge is taught and set by other systems.

 

[ChipBoundary]

Flash, like EEPROM before it, has always been classed as a ROM tech. There's no reason not to place other NVM in that class also.


Except that it doesn't replace RAM at all. Best it could do is be a backing store. Which is what a filesystem is.

Flash has never been classed as ROM tech. Anybody that did that was frankly, well, an *****. I've literally never heard anyone, in decades of IT work, refer to it as ROM technology. By definition, it is impossible for it TO be ROM. If it can be written to, it is automatically not ROM. This was even the case with disc-based technology. Rewritable CD's and DVD's were never called CD/DVD-ROM.

ROM means a very specific thing, and misusing it muddies the water. PCM as it stands now, may not replace RAM, but that is the intent. Same with memristor technology. Given enough development these things can and will replace RAM. Or a similar technology will. Hell, as it is graphics cards are bypassing RAM to access storage directly because it so insanely fast, and ends up being faster than feeding things through RAM.

 

[user556]

A memory tech can only transition from being a ROM to a RAM when it achieves unlimited write endurance.
All filesystems are a backing store. That's all ROM from a memory point of view.