SK Hynix and SanDisk Release First HBF Standard for AI Memory

SK Hynix and SanDisk have released the first industry standard specification for High-Bandwidth Flash (HBF), a move that reshapes how we think about memory in AI hardware. This standardization effort matters because it creates a new storage tier positioned between High-Bandwidth Memory (HBM) and traditional Solid State Drives (SSD). Buyers and engineers can now expect more flexible configurations that balance speed and capacity for AI inference tasks.

New memory tier combines HBM speed with NAND flash capacity

The technology relies on NAND flash to deliver data transfer speeds comparable to HBM while offering far higher storage capacity. Unlike HBM, which is limited in size, HBF connects to GPUs, CPUs, and TPUs through logic chips and interposers. This architecture allows system designers to build customized HBF plus HBM combinations that address specific bandwidth and capacity challenges.

The HBF Alliance formed approximately six months before this standard release to support the technology ecosystem. Major players like Google and Tenstorrent have joined the alliance to help drive adoption and compatibility. This collaborative approach ensures that the new memory tier integrates smoothly with existing processor architectures.

HBF sits in a unique spot in the hardware hierarchy, bridging the gap between ultra-fast but small HBM and slower but large SSDs. By combining HBF with HBM, manufacturers can create systems that handle larger datasets without sacrificing the speed needed for inference. This hybrid approach offers a practical path forward for AI hardware scaling.