SK Hynix is flexing some future memory tech you will not be able to afford. The company showed off its next-level HBM4 with a 16-layer stack packing 48 gigabytes, claiming it as an industry first. This follows their existing 12-layer HBM3E product that is actually hitting the market soon. They also displayed a new server memory module called SOCAMM2, designed for better power efficiency in AI systems. For regular devices, they are working on LPDDR6 memory, which promises faster speeds and lower power use for on-device AI tasks.
Their NAND flash presentation included a 321-layer quad-level cell product aimed at massive data center solid-state drives. This thing holds two terabits per die, boasting better power and performance than older tech. They also had a big physical model of something called custom HBM. This concept moves some computing functions directly into the memory stack itself, a change from the standard design where the GPU handles everything. The whole exhibit was a speculative look at components for future AI hardware, with most products still in development.
Their NAND flash presentation included a 321-layer quad-level cell product aimed at massive data center solid-state drives. This thing holds two terabits per die, boasting better power and performance than older tech. They also had a big physical model of something called custom HBM. This concept moves some computing functions directly into the memory stack itself, a change from the standard design where the GPU handles everything. The whole exhibit was a speculative look at components for future AI hardware, with most products still in development.
