News New Memory Research Teases 100x Density Jump, Merged Compute and Memory

Given it's limited switching endurances of >10^4 cycles = 10,000 cycles.

It's best used as a replacement for NAND Flash, can it get to the cheap costs that NAND Flash is currently at?

How long can it hold data in a off-line state?

Many people would be happy with 10,000 cycles Endurance given how abysmal QLC is right now.
nfg95Bk.jpg
 
Given it's limited switching endurances of >10^4 cycles = 10,000 cycles.

It's best used as a replacement for NAND Flash, can it get to the cheap costs that NAND Flash is currently at?

How long can it hold data in a off-line state?

Many people would be happy with 10,000 cycles Endurance given how abysmal QLC is right now.
nfg95Bk.jpg
Even with 1k cycles that is still plenty for even data center drives. With the increase in storage amount you get a big increase in endurance. The Solidigm 7.68TB QLC drive has a 5.9PB endurance. https://www.servethehome.com/solidigm-has-a-61-44tb-ssd-coming-this-quarter/
 
It's the compute on memory that I'm most excited about. Merging the two is for sure the future. How much cache is being slapped on top on AMD chips now? Might as well build a tech that combines the two right? But this seems to be an "either-or" process/memory tech huh?
 
It's the compute on memory that I'm most excited about. Merging the two is for sure the future. How much cache is being slapped on top on AMD chips now? Might as well build a tech that combines the two right? But this seems to be an "either-or" process/memory tech huh?
There's no hope for a memory replacement as this research is currently described, but a NAND replacement is already desperately needed. And that can even be relevant to GPUs: see AMD's "SSG" concept.
 
Tell me on the desktop will you notice any difference from 10PB or write endurance vs 1PB? No. These new QLC drives probably have more write endurance than the 80GB old SLC drives from 2010 even with fewer write cycles.
But one SSD will last significantly longer than the other.

That has value to many folks, not creating more e-waste faster is good for everybody.

Buy once, use for the rest of your life; you can even pass it down to the next generation once you pass away and it'd still be useful since there is plenty of "Write Endurance" left.
 
Tell me on the desktop will you notice any difference from 10PB or write endurance vs 1PB? No. These new QLC drives probably have more write endurance than the 80GB old SLC drives from 2010 even with fewer write cycles.
I see 100-200:1 ratios on some of these cheaper drives, e.g. 100 TB endurance for a 1 TB drive.

Another related problem is unpowered data retention, which continues to get worse. That might not be a huge problem for most internal SSDs, but it would suck to lose terabytes because you didn't plug in a particular flash drive for 6 months.

The situation is expected to get worse. PLC NAND (5 bpc) is on the menu for sure, maybe 6/7/8 bpc later.
 
I see 100-200:1 ratios on some of these cheaper drives, e.g. 100 TB endurance for a 1 TB drive.
Most cheap SSD's are 300-600TBW for a 1TB drive. That puts it from 1/6-1/3 DWPD. How many people are going to be writing 160-330GB/every day for 5 years? Not many. Those that are, using the SSD as a scratch drive for video editing for example, are not using bargain basement SSDs for their scratch drives. They will be using the drives that offer greater endurance.

Another related problem is unpowered data retention, which continues to get worse. That might not be a huge problem for most internal SSDs, but it would suck to lose terabytes because you didn't plug in a particular flash drive for 6 months.
Yes that is a problem. However, even with the older drives you were told that SSD as cold storage isn't a good long term usage. You needed to power it on every year because after that the loss of data was possible. Another thing is why would you want to pay the price premium to use SSD as cold storage. A 4TB external HDD costs A LOT less than the same SSD and if it is going to be sitting around the HDD doesn't need to be powered on nearly as often to retain the data.
 
But one SSD will last significantly longer than the other.
Yes the one with the higher TBW should last longer. Just because the cells have fewer cycles doesn't mean that it won't last as long. Tell me which should last longer. 1 cell with 10k cycles or 100 cells with 1k cycles? This is an accurate depiction of where SSDs are today. Each cell doesn't have the same endurance as the older ones but you have 100x the cells. That means the wear is more spread out so you end up getting greater endurance.
 
Yes the one with the higher TBW should last longer. Just because the cells have fewer cycles doesn't mean that it won't last as long. Tell me which should last longer. 1 cell with 10k cycles or 100 cells with 1k cycles? This is an accurate depiction of where SSDs are today. Each cell doesn't have the same endurance as the older ones but you have 100x the cells. That means the wear is more spread out so you end up getting greater endurance.
But we could be heading into a new storage memory with more cells & write cycles.

Shouldn't that make you happier?
 
Would a new design with more cell and write cycles be nice? Of course. However, this is what we have now.
But lets say the researchers work the details and kinks out in the next 5-10 years and bring out QLC style (4-bits per cell) RSM (Resistive Switching Memory) Flash.

But with 10x the existing QLC NAND Flash cells with 10k P/E cycles per cell.

That should be a joyous occaision.
 
But lets say the researchers work the details and kinks out in the next 5-10 years and bring out QLC style (4-bits per cell) RSM (Resistive Switching Memory) Flash.

But with 10x the existing QLC NAND Flash cells with 10k P/E cycles per cell.
Looks like the better way to describe it would be 40-500 bits per cell:
These synapse devices can access a lot more intermediate voltage states than even the densest NAND technology in today's roomiest USB sticks can - by a factor of 10 or 100.