SanDisk and SK hynix Publish First High Bandwidth Flash Specification

SanDisk and SK hynix have published the first technical specification for High Bandwidth Flash (HBF), a memory tier designed to sit between HBM and SSDs, through the Open Compute Project. The release lands six months after the HBF consortium began work in February, and two days before the two companies present the technology at the Future of Memory and Storage show (FMS) in Santa Clara.

HBF stacks NAND dies to deliver HBM-like bandwidth at a fraction of the cost. The spec covers packages up to 512GB (8-high and 16-high die stacks) with three bandwidth grades running from roughly 0.4TB/s to 3.0TB/s, connected to processors over UCIe, the open chiplet interconnect standard.

“AI infrastructure has entered a new phase where the ability to store, move, and access data efficiently has become as important as compute itself,” said Khurram Ismail, chief product officer at Sandisk. “Building on decades of flash innovation, Sandisk is advancing next-generation NAND, and emerging technologies such as HBF, to help advance the industry.”

AI inference models have grown faster than the memory budgets around them. HBM is fast and expensive; HBF is NAND underneath, so it trades a little bandwidth for far more capacity — Sandisk has previously claimed 8–16x HBM’s capacity at similar cost. Google and Tenstorrent are already in the consortium, which suggests the industry wants the standard set before the hardware exists.

At FMS, Sandisk is also showing BiCS10 QLC NAND — the first QLC 3D flash to reach a 4.8Gb/s interface, per Kioxia and Sandisk — plus enterprise SSDs aimed at KV cache workloads, with PCIe Gen 6 and E3.S designs in the pipeline. The company’s keynote, “NAND: The Versatile & Scalable Foundation of the AI Era”, runs August 5 at 11:40am PT; the HBF panel with Google and SK hynix follows the next morning.

This is the third approach to the same problem in a year. Samsung is pushing processing-in-memory chips, and Seagate is attacking AI storage from the hard-drive side. HBF’s bet is that the answer is NAND pretending to be HBM, with an open spec so that everyone builds it the same way.