AI memory chips help servers and data centres run AI workloads faster, process more data, and use less power.
SK hynix has shipped samples of its 12-layer HBM4E, its next-generation high-bandwidth memory (HBM) for AI applications, to major customers. The company plans to work with customers to prepare the product for mass production.
The HBM4E memory delivers data transfer speeds of up to 16Gbps per pin while improving power efficiency by more than 20% compared to the previous generation. The higher speed and lower power consumption are designed to support AI training, AI inference, and other data-intensive workloads.
The memory is intended for AI chipmakers, server manufacturers, cloud service providers, and companies building AI data centres and high-performance computing (HPC) systems. They can use it to build AI accelerators and servers that process larger AI models, train models faster, run AI inference more efficiently, and handle larger volumes of data while reducing power consumption.
The memory also reduces data transfer latency through an updated interface and design improvements while maintaining stable performance in high-bandwidth computing environments. This helps improve data processing efficiency in AI data centres and large-scale computing systems.
To build the 12-layer stack, SK hynix uses its Advanced MR-MUF (Mass Reflow Molded Underfill) packaging process, which injects protective material between stacked memory chips. The technology enables a 48GB capacity while maintaining structural stability. The company also said the new memory offers 17% better heat resistance than its HBM4 product, allowing stable operation in high-performance computing systems.
“SK hynix has laid the foundation to strengthen its AI leadership with HBM4E based on its market-leading technological capabilities and manufacturing expertise,” said Ahn Hyun, President and Chief Development Officer, adding, “Through close collaboration with our partners, we will deliver the value needed in the market while reinforcing our technology leadership as a full-stack AI memory creator.”
Click here for the original announcement.
