SK hynix has shipped samples of HBM4E, its next-generation DRAM for AI, to major customers. The company said it delivered the 12-stack HBM4E on schedule, citing its HBM development and production capabilities, and plans to work with partners toward timely mass production.
The 12-layer HBM4E is designed to improve performance and power efficiency for AI training and inference. It offers a maximum data processing speed of 16 Gbps per pin, while power efficiency is up more than 20 percent from previous models.
SK hynix said the memory reduces data transfer latency through a new interface and design optimization while maintaining stable operation in high-bandwidth environments. The company uses Advanced MR-MUF technology to achieve a 48 GB capacity in a 12-layer stack and said heat resistance has improved by 17 percent compared with the preceding HBM4, supporting use in AI datacenters and large-scale computing systems.
