The HBF standard, now disclosed via the Open Compute Project, marks a significant shift toward open industry architecture. By positioning HBF as a memory layer between traditional HBM and SSDs, the consortium—which includes Google and Tenstorrent—seeks to solve the capacity and bandwidth bottlenecks that currently plague AI inference tasks. The specifications support capacities up to 512GB and offer three performance grades ranging from 0.4TB/s to 3.0TB/s.
At the FMS 2026 exhibition in Santa Clara, SK hynix is also showcasing its 375-layer 4D NAND (V10) technology. This tenth-generation NAND delivers a 2.5-fold increase in power efficiency compared to its predecessor, with mass production of high-capacity eSSDs slated for early next year. Executive Vice President Kim Chun-sung emphasized that the move toward 'Tiered Memory' architectures is essential as Agentic AI demands more autonomous and efficient data processing capabilities. The company will further detail these developments through a series of keynotes and panel discussions featuring leadership from Google DeepMind and Sandisk throughout the event.




Comments (0)
No comments yet. Be the first!