Sandisk and SK hynix released the first HBF (High Bandwidth Flash) technical specification through the Open Compute Project (OCP), advancing standardization for AI inference systems. Developed in six months with Google and Tenstorrent contributing, the spec provides a framework for high-bandwidth, high-capacity memory near compute cores, critical for large language models. It defines system interfaces, reliability guidelines, and software tools to improve power efficiency and reduce costs. Alper Ilkbahar, Sandisk's CTO, emphasized its role in enabling flexible AI architectures. The OCP framework ensures open access, positioning HBF as a de facto standard.

The specification addresses growing demands for near-compute memory in AI inference, allowing system designers to integrate HBF alongside High Bandwidth Memory. Technical guidelines cover performance expectations, x PU-HBF interfaces, and packaging. This standardization supports ecosystem readiness by offering a common framework for adoption. Sandisk and SK hynix aim to accelerate market expansion through open collaboration, hosting keynotes and panels at The Future of Memory and Storage Conference (FMS) to discuss HBF’s impact on memory hierarchy and AI infrastructure.

Events at FMS in Santa Clara highlight HBF’s potential to redefine memory density and persistence. Discussions will focus on technical challenges, standardization timelines, and economic benefits. The initiative reflects a proactive strategy to dominate the AI storage market via ecosystem growth and technological maturity.