NEO Semiconductor, recently announced the launch of the world’s first Extreme High Bandwidth Memory (X-HBM) architecture designed specifically for AI chips. Developed to address the escalating performance requirements of generative AI and high-performance computing (HPC) applications, X-HBM delivers unprecedented capabilities with a 32K-bit data bus and the potential to achieve 512 Gbit per die.
This breakthrough significantly outperforms the limitations of conventional High Bandwidth Memory (HBM), offering up to 16 times greater bandwidth or 10 times higher density, setting a new benchmark for memory performance in next-generation AI systems.
“X-HBM is not an incremental upgrade, it’s a fundamental breakthrough,” said Andy Hsu, Founder & CEO of NEO Semiconductor. “With 16X the bandwidth or 10X the density of current memory technologies, X-HBM gives AI chipmakers a clear path to deliver next-generation performance years ahead of the existing roadmap. It’s a game-changer for accelerating AI infrastructure, reducing energy consumption, and scaling AI capabilities across industries.”
Built on NEO’s proprietary 3D X-DRAM architecture, X-HBM represents a major leap in memory technology by eliminating long-standing limitations in bandwidth and density. In contrast, HBM5, still in development and expected to reach the market around 2030, is projected to support only 4K-bit data buses and 40 Gbit per die. A recent study from the Korea Advanced Institute of Science and Technology (KAIST) projects that even HBM8, expected around 2040, will offer just 16K-bit buses and 80 Gbit per die. In comparison, X-HBM delivers 32K-bit buses and 512 Gbit per die, allowing AI chip designers to bypass a full decade of incremental performance bottlenecks associated with traditional HBM technology.
Key Features and Benefits:
- Scalable – Enables faster data transfer between GPUs and memory for more efficient AI scaling.
- High-Performance – Unlocks untapped GPU capabilities to boost AI workloads.
- Sustainable – Reduces power and hardware needs by consolidating AI infrastructure.
COMMENTS