Samsung is developing the revolutionary zNAND-O flash-memory concept, promising microsecond latency and unprecedented bandwidth to accelerate on-device AI processing and challenge existing storage standards amidst industry-wide advancements.
Samsung is developing a new flash-memory concept designed to speed up artificial intelligence tasks run directly on devices rather than in the cloud. The project, called zNAND-O, revives the company’s earlier high-performance Z-NAND work and is aimed at cutting access times while raising throughput for local AI workloads, according to ComputerBase and Tom’s Hardware.
The new design appears to borrow ideas from high bandwidth memory used in DRAM systems. Samsung’s concept stacks multiple NAND dies and connects them with through-silicon vias, or TSVs, instead of the edge wiring used in conventional packages. The company says the architecture could use four or eight layers per stack and deliver read latency below 3 microseconds, compared with roughly 45 microseconds for Samsung’s 8th-generation V-NAND. Tom’s Hardware reported that Samsung is also targeting up to a 15-fold performance gain over standard NAND and power savings of as much as 80%.
Samsung has not given a launch date, and zNAND-O remains a concept. Even so, the figures it has published are ambitious: bandwidth of 200GB/s to 400GB/s per stack, with more than 1TB/s possible if multiple stacks are combined. That would place it far beyond today’s SSDs, including PCIe 6.0 models, although real-world results would still depend on the interface and system design.
The proposal arrives as the storage industry pushes several competing approaches for AI infrastructure. SanDisk and SK hynix have been advancing High Bandwidth Flash, or HBF, an open standard under the Open Compute Project that aims to combine NAND capacity with HBM-like bandwidth for inference workloads. Tom’s Hardware said HBF is being designed for up to 16 stacked NAND layers and bandwidth from 0.4TB/s to 3TB/s, though public latency data are still lacking. That makes Samsung’s zNAND-O look less like an isolated experiment and more like part of a wider race to build memory that sits between fast storage and working memory for AI systems.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
