Samsung previews zHBM and zNAND-O 3D memory concepts targeting 10× higher density and 8× HBM5 performance by 2030

Samsung unveiled a multi-generation memory roadmap at the Future of Memory and Storage Summit, previewing breakthrough technologies designed to overcome what the company calls the digital memory wall facing AI infrastructure. The roadmap centers on zHBM—a concept that stacks high-bandwidth memory directly on top of AI accelerators (rather than beside them), and zNAND-O, a next-generation NAND architecture for edge AI, alongside V10 Bonding V-NAND with over 400 layers.

The zHBM design shortens the physical distance data must travel between processor and memory, promising approximately 8× the performance of HBM5, more than 10× the memory density, 3× the energy efficiency, and less than half the thermal resistance compared to HBM5. Samsung expects to require close co-design with accelerator partners (notably NVIDIA, Google, and others) to realize those gains. Separately, V10 BV-NAND uses wafer bonding to stack over 400 memory layers, boosting density by ~58% versus the previous generation while improving I/O performance.

In the near term, Samsung is already in mass production of HBM4 and has begun shipping HBM4E samples (14 Gbps stable, up to 16 Gbps scalable). HBM4E offers 20%+ speed uplift over HBM4 and will expand to 32GB (8-layer), 48GB (12-layer), and 64GB (16-layer) configurations. The company expects HBM4E to enter mass production aligned with customer schedules. Samsung also plans to improve HBM energy efficiency by 2.5× by 2030, a critical metric as power consumption becomes the limiting factor in data-center density.

For architects: The zHBM vertical-stacking concept signals a structural shift in memory-compute co-design. Memory is no longer peripheral to accelerator planning—it's becoming the constraint that defines per-GPU performance ceilings. Practitioners should track HBM supply, qualification timelines, and thermal integration strategies as closely as compute roadmaps; memory lead times now dictate AI infrastructure deployment velocity.