SK hynix is moving to a generational shift in the AI accelerator memory market. SK hynix said on Wednesday it has supplied 12-layer samples of HBM4E, its next-generation high-performance DRAM for AI, to key customers.

HBM4E is the latest generation of HBM, a high-performance memory that boosts data processing speed by stacking multiple semiconductor chips. It improves both performance and power efficiency compared with the previous-generation HBM4. It delivers data processing speeds of up to 16 Gbps per pin and improves energy efficiency by more than 20 percent. The company said it focuses on boosting processing speed for AI training and inference tasks.

HBM4E reduces data transfer latency through the latest interface and design changes. It also delivers stable operation in high-bandwidth environments. The company expects processing efficiency to improve for next-generation AI data centers and large-scale computing systems.

SK hynix applied Advanced MR-MUF, a process that injects and hardens a protective material after chip stacking, to HBM4E to achieve a 48GB capacity based on a 12-layer stack. It lowered thermal resistance by about 17 percent compared with HBM4, improving memory stability in high-performance computing environments.

Ahn Hyun (안현), head of development at SK hynix, said, "By carrying forward the industry's best technological competitiveness and mass-production capabilities we have built up to the HBM4E product as well, we have laid the foundation to continue leading AI innovation." He added, "Based on cooperation with partners, we will proactively realize the value the market demands and solidify our technological leadership as a full-stack AI memory creator."