0
SK Hynix HBM4E. / Photo courtesy of SK Hynix

SK Hynix announced on June 18 that it has supplied 12-layer samples of HBM4E, its next-generation ultra-high-performance D-RAM new product for AI, to major customers.

SK Hynix explained, "Based on our accumulated HBM advanced development capabilities and production know-how, we were able to showcase the 12-layer HBM4E samples to customers," adding, "We will work closely with core customers to ensure timely mass production."

The new product features an upgrade in both performance and power efficiency compared to the previous generation, HBM4. It realizes a maximum data processing speed of 16 Gbps per pin and improves energy efficiency by over 20%, drastically increasing the data processing performance essential for AI training and inference.

In addition, HBM4E reduces data transmission latency through the latest interface and design optimization, achieving stable operation even in high-bandwidth environments. The company expects that this will further enhance the processing efficiency of next-generation AI data centers and large-scale computing systems.

SK Hynix applied the Advanced MR-MUF process to HBM4E to realize a 48 GB capacity based on 12-layer stacking while increasing structural stability. In particular, it lowered thermal resistance by approximately 17% compared to HBM4, ensuring that the memory operates stably even in high-performance computing environments.

Ahn Hyun, head of development at SK Hynix, said, "By carrying over our industry-leading technological competitiveness and mass-production capabilities built over the years to the HBM4E product, we have laid the foundation to continuously lead AI innovation," adding, "Based on cooperation with our partners, we will preemptively implement the value demanded by the market, thereby solidifying our technological leadership as a full-stack AI memory creator."

                                                                                                            Ahn So-yeon