SK hynix Unveils High Bandwidth Flash Specifications at FMS 2026

This title was summarized by AI from the post below.

Building the future of AI memory takes more than faster chips—it requires new architectures. At #FMS 2026, SK hynix unveiled the first standard specifications for High Bandwidth Flash (#HBF) with Sandisk through the Open Compute Project Foundation (#OCP) and showcased its vision for Tiered Memory. The company also revealed its tenth-generation 375-layer 4D NAND, offering up to 2.5× higher power efficiency than the previous generation. 🔗 Read more in our newsroom: https://lnkd.in/diyugDkJ #SKhynix #HighBandwidthFlash #FMS2026 #Semiconductor

  • No alternative text description for this image

This is a pivotal moment for the memory hierarchy. High Bandwidth Flash is not just another NAND spec – it's an architectural bridge between the raw capacity of SSDs and the bandwidth of DRAM, positioned to address the growing imbalance in AI data pipelines where storage throughput has become a bottleneck for training and inference workloads. By formalising HBF through OCP and partnering with Sandisk on standardisation, SK hynix is effectively creating a new tier in the memory stack that did not exist before. The 375-layer 4D NAND with 2.5× better power efficiency than its predecessor moves the industry closer to exascale storage density while reducing the energy footprint per bit – a critical variable as AI data centres scale toward gigawatt consumption. Together with the Tiered Memory vision, which proposes seamless data movement across DRAM, HBF and conventional NAND, SK hynix is not just advancing a product line; it is redefining how the industry thinks about data placement, latency and cost across the entire AI storage stack. For hyperscalers and enterprise customers, this could unlock new architectural choices.

This is exactly why I don’t think HBM alone is the long-term answer. The future will likely be a tiered architecture where HBM handles the hottest data, HBF provides greater capacity and bandwidth, and SSDs support the deeper storage layer. Standardizing HBF through OCP could be just as important as the hardware itself, especially as AI inference grows and power efficiency becomes a bigger constraint.

The next AI bottleneck is shifting from compute to memory architecture. Innovations like HBF and Tiered Memory could prove just as transformative as next-generation GPUs, enabling more scalable, power-efficient, and cost-effective AI infrastructure.

Like
Reply

Great to see SK hynix advancing the future of AI memory with HBF and next-generation 4D NAND. Congratulations on this impressive achievement at FMS 2026!

Like
Reply
See more comments

To view or add a comment, sign in

Explore content categories