Samsung Electronics, the world’s largest memory chip maker, used this year’s Future of Memory and Storage (FMS) conference in Santa Clara, California to present a prototype of a next-generation memory device designed for artificial intelligence workloads.
The new module, called V10 Bonding V-NAND or BV-NAND, features more than 400 layers and adopts a wafer-bonding architecture intended to increase storage density and enhance performance metrics. Samsung said the BV-NAND raises memory density by about 58% compared with its previous V9 NAND generation while delivering improved read, write and input/output performance to meet growing requirements from AI systems.
Sourcing higher-capacity NAND has become more important as AI use has broadened beyond the training of large language models to include the process of generating answers to user queries, Samsung noted. NAND flash is widely used to store user data such as photos, videos and applications across devices including smartphones, and demand for higher-capacity NAND has increased as AI workloads expand.
In addition to the BV-NAND prototype, Samsung displayed concept designs for what it described as the industry’s first zHBM and zNAND-O architectures, a vision for next-generation three-dimensional memory. These concepts emphasize stacking memory cells vertically to increase data density within a given footprint.
Samsung contrasted these concepts with conventional high-bandwidth memory, which is typically placed beside AI processors. The company said its zHBM approach stacks memory above AI accelerators, shortening the distance data must travel and thereby increasing bandwidth while improving energy efficiency.
According to Samsung, its wafer-bonding technique could enable more than 10 times the memory density of conventional HBM5 memory, while delivering roughly threefold improvements in energy efficiency and reducing thermal resistance by more than half. The company presented these claims as part of a broader push to support AI systems that require greater storage capacity with improved power characteristics.
Samsung has also signaled shifts in its commercial approach to memory sales. The company disclosed last week that long-term agreements with customers could eventually represent 60% to 70% of its memory revenue, a change that reflects strengthening demand from AI-related applications.
Investor commentary has been mixed, with some market participants expressing concern about the sustainability of memory demand if smartphone consumption in China weakens. At the same time, Citi analysts noted that inventories across both DRAM and NAND supply chains remain well below historical norms despite worries about softening end-market demand.
The developments Samsung outlined at FMS present a technical roadmap for denser, more efficient memory targeted at AI infrastructure, while market discussions continue around the demand outlook and inventory dynamics across memory supply chains.