Breaking the 100M Token Limit: EverMind's MSA Architecture Achieves Efficient End-to-End Long-Term Memory for LLMs

The research introduces a novel memory architecture called MSA (Memory Sparse Attention). Through a combination of the Memory Sparse Attention mechanism, Document-wise RoPE for extreme context ...

New memory architecture targets AI inference bottlenecks

Lightbits Labs Ltd. today is introducing a new architecture aimed at addressing one of the most stubborn bottlenecks in large-scale artificial intelligence inference: the growing mismatch between the ...

Phison Rescales Local AI Inferencing with Flash Memory Expansion

Phison Electronics (8299TT), a global leader in NAND flash controllers and storage solutions, today announced its GTC ...

AI’s memory chip shortage is quietly taxing the entire economy

AI's insatiable appetite for memory chips is crowding out all other buyers — and the consequences will ripple through every ...
Semiconductor Engineering

HBM4E Raises The Bar For AI Memory Bandwidth

The Rambus HBM4E memory controller extends a long-standing Rambus leadership position in HBM controller IP. Being first to ...
Semiconductor Engineering

Freeing Up Near-Memory Capacity For Cache Using Compression Techniques In A Flat Hybrid-Memory Architecture

A technical paper titled “HMComp: Extending Near-Memory Capacity using Compression in Hybrid Memory” was published by researchers at Chalmers University of Technology and ZeroPoint Technologies.
TweakTown

Intel teams with SoftBank on next-gen ZAM memory, prototypes to compete against HBM in 2028

TL;DR: Intel partners with SoftBank's SAIMEMORY to advance next-generation DRAM bonding technology using Z-Angle Memory (ZAM), enabling higher capacity, lower power consumption, and improved ...