News 3D X-DRAM aims for 10x capacity of today's memory — NEO Semiconductor's memory has up to 512 Gb per module

[Admin]

NEO Semiconductor has announced progress on its new 3D X-DRAM technology, with new 1T1C and 3T0C chip designs increasing DRAM speeds by an order of magnitude for tomorrow’s top-end computing.

3D X-DRAM aims for 10x capacity of today's memory — NEO Semiconductor's memory has up to 512 Gb per module : Read more

 

[ekio]

That would be nice, because the ram price/GB didn't improve much over the past 10 years.

For example: the MacBook Pros are still sold with the exact same default amount of Ram as more than 10 years ago... If I was told that ten years ago, I would have thought it's a joke.

 

[helper800]

That would be nice, because the ram price/GB didn't improve much over the past 10 years.

RAM is considerably cheaper now than it was 10 years ago... If I recall properly, 10 years ago, a cheap DDR3 16gb kit (2x8gb) was 30-40 dollars around its cheapest. You can get 32gb (2x16) of DDR4 for 40-50 dollars now, or 60 dollars for 32GB (2x16) of DDR5.

For example: the MacBook Pros are still sold with the exact same default amount of Ram as more than 10 years ago... If I was told that ten years ago, I would have thought it's a joke.

Tha is Apple fattening their margins, not RAM not getting cheaper...

 

[usertests]

M is considerably cheaper now than it was 10 years ago... If I recall properly, 10 years ago, a cheap DDR3 16gb kit (2x8gb) was 30-40 dollars around its cheapest. You can get 32gb (2x16) of DDR4 for 40-50 dollars now, or 60 dollars for 32GB (2x16) of DDR5.

This is of course, not much improvement in 10 years. Significant stagnation set in after around 2012, with prices swinging up before dropping down to new lows.

Today's $/GB is about 1/3rd of what it was in 2012, and 1/2th of 2016. And we are forecasted to be in the low point before another upswing.

the new cells are advertised as being able to hold 512 Gb (64 GB) on a single module; at least 10x more than any modules currently commercially available.

Is the 512 Gb "module" intended to be compared to the 32 Gb dies being used in products today? That would mean a 16x improvement. The newly launched 64 GB DDR5 modules are double-sided using 16 of the 32 Gb dies. If the sizes are the same, you could have a 1 TB DDR5 module using this technology, with no die stacking.

It's vaporware until proven otherwise, but if they can beat Samsung's timeline for 3D DRAM by 5+ years, that could be great for everybody.

 

[bit_user]

That would be nice, because the ram price/GB didn't improve much over the past 10 years.

It did improve, but it seems you're on to something.

RAM is considerably cheaper now than it was 10 years ago... If I recall properly, 10 years ago, a cheap DDR3 16gb kit (2x8gb) was 30-40 dollars around its cheapest. You can get 32gb (2x16) of DDR4 for 40-50 dollars now, or 60 dollars for 32GB (2x16) of DDR5.

It's difficult to work out by comparing specific examples, because the market undergoes fluctuations and there are different premiums or discounts that can be applied to memory in certain performance tiers.

Luckily, being a commodity market, there are people who look at this stuff in aggregate. I found a couple of decent articles and they do confirm that the cost curve of DRAM is flattening.

Source: https://semianalysis.com/2024/09/03/the-memory-wall/
​

Among other things, that article looks at the impact of density increases on pricing:

​

It's a very detailed and comprehensive article, with a major thrust being the prospects of 3D DRAM. Definitely give it a look, if you're interested in this subject.

Here's another, older one I ran across, which includes a comparison of $/bit between both DRAM and NAND, over a similar timescale.

Source: https://semiengineering.com/high-performance-memory-at-low-cost-per-bit/
​

 

[helper800]

It did improve, but it seems you're on to something.


It's difficult to work out by comparing specific examples, because the market undergoes fluctuations and there are different premiums or discounts that can be applied to memory in certain performance tiers.

Luckily, being a commodity market, there are people who look at this stuff in aggregate. I found a couple of decent articles and they do confirm that the cost curve of DRAM is flattening.

​

​

Source: https://semianalysis.com/2024/09/03/the-memory-wall/​

​

Among other things, that article looks at the impact of density increases on pricing:

​

​

It's a very detailed and comprehensive article, with a major thrust being the prospects of 3D DRAM. Definitely give it a look, if you're interested in this subject.

Here's another, older one I ran across, which includes a comparison of $/bit between both DRAM and NAND, over a similar timescale.

​

​

Source: https://semiengineering.com/high-performance-memory-at-low-cost-per-bit/​

​

I took issue with the claim, "...ram price/GB didn't improve much over the past 10 years," because I believe that the colloquially understood understanding of the word 'much' has been satisfied. Has the cost per GB started to flatten out? Yes, and so has the cost per unit proposition for almost all other tech products, but to say they "didn't improve much over the past 10 years," is laughable. Increased inflation alone could account for a lot of that flattening, though this is pure speculation. I will read through the articles thoroughly later. In 2012 I bought a cheap, and on sale, SSD that was 120GB for 130 dollars before tax. That same cost will get you a 2TB NVMe SSD that is 12 times faster in sequential speeds with a multi gigabyte DRAM cache. That seems like a much improved cost per GB to me.

 

[bit_user]

BTW, one more thought I had about this is that it might be viable only as some sort of on-package memory. The issue is that you generally need to scale up interface speeds in proportion with capacity. Once the two get sufficiently mismatched, it limits the way you can use the storage or memory technology.