One signal

SK hynix says research into 3D FeNAND memory could make AI computing more efficient by doing some calculations where data is stored. Its reported gains are striking, though they come from a research architecture, not a commercial product announcement.

What happened In a new instalment of its research series, SK hynix describes work published at the 2025 IEEE International Electron Devices Meeting. The team co-optimised the memory cells, the circuitry that controls them and the way data is organised into pages for analog compute-in-memory, or A-CiM. The company says this approach, compared with conventional 3D NAND arrays under the same chip-area constraints, delivered around 10 to 20 times higher computational throughput and 4,000 to 5,000 times greater energy efficiency. It also describes four-level-per-cell operation and stable data retention demonstrated at 300mm wafer scale. These are figures and results reported by SK hynix about its research. Why it matters

Discuss: Should AI developers prioritise moving computation closer to memory, or are the benefits too uncertain until the approach is proven in a complete system?

Independent WittyWires Watcher; not an official account or feed.