A Japanese memory initiative, which has been working quietly for more than a decade, is on the agenda with a big claim. Neo Semiconductor has announced that it wants to produce DRAM chips that are 8 times denser than existing 16Gbit DRAM, using a technology called 3D X-DRAM.
As the name suggests, a total of 230 layers are planned to be used, which will help achieve the 128Gbit DRAM chip. For comparison, Samsung is aiming to launch a 32 Gbit DDR5 DRAM chip with 1TB memory modules in 2023.
According to a statement by the company’s co-founder and CEO Andy Hsu, the first prototypes are expected to be presented next year if negotiations to license the technology to DRAM manufacturers (Micron, Samsung Semi, SK Hynix, Kingston Technology) go well.
Similar to 3D NAND, 3D X-DRAM can reach 1Tb before 2024 as memory densities increase exponentially. It took more than a decade for the DRAM industry to transition from 4Gb to 16Gb memory chips, in what could be said to be a relatively minor improvement.
While memory modules are relatively cheap on the lower end of the options, looking at the top end you can see they’re almost unbelievably expensive. A single 256GB DDR4 server RAM module, the maximum memory size on the market, retails for around $2,500 (or $10 per GB). By comparison, you can get 32GB of RAM for less than $60 (or less than $2 per GB).
Neo’s technology solution can significantly reduce the cost of memory as 3D NAND does for solid-state storage. Another feature that makes the solution even more attractive is that it uses existing fabrication techniques, similar to 3D NAND, to achieve layering.
Popular applications such as machine learning or artificial intelligence (e.g. ChatGPT) using LLM (Large language models) require access to very large memory pools and this comes at a significant financial and power/latency cost.
While Intel 3D XPoint (or Optane) technology hasn’t managed to find its place as the level between system memory and SSD, it may take some time for this technology to reach end users. But it certainly has the potential to change the memory-storage pyramid.
From the perspective of end users, considering that system storage has reached a plateau, it may be possible to see RAM-only devices in the distant future. Such a future could create a shift that could change the computing environment in general and the way modern operating systems work.
52185
