ResearchPreprint
HBF use cases expand beyond a single model copy
The paper explores capacity-oriented HBF uses in LLM serving, including model residency, expert replication, and multi-model deployment.
arXivPublic record / 2025—present
Chronology of High Bandwidth Flash
The public HBF record is young. This timeline follows the sequence of announcements, standards work, prototypes, and research—without treating them as the same kind of evidence.
2026
ResearchPreprint
A full-stack study draws a sharp boundary around HBF
The study argues that HBF performs poorly as a drop-in SSD replacement for transient KV-cache traffic, while identifying selective, more read-oriented uses where it can remain valuable.
arXivStandardsPrimary source
The first HBF specification package is announced
The companies announced an initial OCP HBF specification covering stack capacities up to 512 GB, three bandwidth grades, a UCIe host link, packaging and reliability, and software I/O guidance.
Sandisk + SK hynixResearchPeer reviewed
Hardware-managed tiering joins HBM and high-bandwidth flash
A hardware-managed heterogeneous memory proposal that studies how data can move between HBM and high-bandwidth flash during LLM inference.
IEEE Computer Architecture LettersResearchPreprint
FlashAccel explores a full-stack flash path for AI
A full-stack proposal for integrating high-bandwidth flash into AI systems, spanning media behavior, data movement, and model execution.
arXivResearchPreprint
TileLens adapts tensor layout to flash access granularity
TileLens explores data layouts intended to reduce the mismatch between accelerator tensor access and flash's comparatively large transfer granularity.
arXivResearchWorkshop paper
HotInfra paper questions whether HBF is the best middle tier
The paper compares HBF-like flash with on-package LPDDR and argues that writability and KV-cache behavior can favor DRAM-based alternatives.
HotInfra '26ResearchPeer reviewed
KV-cache research puts write endurance in focus
The paper examines the performance and lifetime consequences of placing write-intensive transformer KV caches in HBF.
IEEE Computer Architecture LettersResearchPeer reviewed
A custom base die pools HBM and HBF capacity
A system proposal for pooling HBM and HBF behind a custom logic die to support terabyte-scale inference memory footprints.
IEEE Computer Architecture LettersResearchPeer reviewed
A balanced systems study maps HBF opportunities and limits
The authors evaluate where HBF's density can help modern LLM serving and identify the architectural and software challenges that follow from NAND behavior.
IEEE Computer Architecture LettersIndustryPrimary source
SK hynix describes an HBF base-die architecture
SK hynix discussed AIN-B, a vertically stacked NAND architecture with a logic base die and high-bandwidth interface for AI workloads.
SK hynixResearchPeer reviewed
IEEE Computer introduces HBF to a broad audience
An accessible technical overview of HBF's proposed architecture, motivation, and place in next-generation AI systems.
IEEE ComputerResearchPreprint
MemExplorer searches heterogeneous memory designs
A design-space exploration framework for AI inference systems that includes HBF among possible capacity, bandwidth, and cost tradeoffs.
arXivIndustryPrimary source
Kioxia targets GPU-direct flash through Storage-Next
Kioxia announced an SSD family designed for NVIDIA's Storage-Next architecture and high-IOPS, direct data paths into accelerated systems.
KioxiaResearchPreprint
HAVEN applies HBF to billion-scale vector search
HAVEN proposes using HBF as a high-capacity tier for approximate nearest-neighbor search, moving beyond the dominant LLM-weight use case.
arXivStandardsPrimary source
The OCP HBF workstream formally begins
The companies announced the start of an Open Compute Project workstream to define a common HBF architecture and interface.
Sandisk + SK hynixResearchPeer reviewed
A broader map of the LLM inference memory problem
A survey and research agenda for inference hardware that situates emerging memory tiers, including HBF, inside the wider model-serving bottleneck.
IEEE Computer / arXivResearchPeer reviewed
H³ models a hybrid HBM–HBF inference system
The paper evaluates a heterogeneous HBM-HBF design for serving large language models under memory-capacity and cost constraints.
IEEE Computer Architecture Letters2025
ResearchPeer reviewed
Mem-Village studies HBF and HBM on glass
A package-level study of a hybrid HBM-HBF memory architecture using glass interconnect technology for large-model inference.
IEEE EDAPSIndustryPrimary source
SK hynix places HBF inside its AI-NAND portfolio
SK hynix described HBF as the bandwidth-oriented branch of its AI-NAND strategy and discussed vertically stacked NAND for AI systems.
SK hynixIndustryPrimary source
A 5 TB, 64 GB/s flash prototype explores another path
Kioxia and partners reported a PCIe 6.0-connected flash module prototype delivering 5 TB capacity and 64 GB/s aggregate bandwidth at under 40 W.
KioxiaNews & analysisPrimary source
Sandisk frames HBF as a capacity-first AI tier
A company-authored explanation of why model capacity and data movement motivate a new tier between HBM and conventional storage.
SandiskStandardsPrimary source
Sandisk and SK hynix align on HBF standardization
The companies announced an agreement to define interoperable HBF specifications and build a broader ecosystem around high-capacity, high-bandwidth flash memory.
Sandisk + SK hynixStandardsPrimary source
UCIe 3.0 doubles die-to-die data rates
UCIe 3.0 adds 48 and 64 GT/s operation plus manageability and system-in-package improvements for interoperable chiplets.
UCIe ConsortiumNews & analysisPrimary source
A technical advisory board is formed around HBF
Sandisk announced a group of academic and industry advisers intended to guide HBF architecture, software, standards, and ecosystem strategy.
SandiskIndustryPrimary source
The first compact public HBF architecture brief
A two-page overview of the HBF concept, its proposed system role, projected stack organization, and comparison with conventional high-bandwidth memory.
SandiskIndustryPrimary source
High Bandwidth Flash enters the public roadmap
Sandisk's investor presentation introduced HBF as a vertically integrated, capacity-oriented memory tier for AI inference and set out the earliest public product roadmap located by this project.
Sandisk