SK Hynix supplies 12-layer HBM4E samples, achieving up to 16Gbps per pin

SK Hynix is stepping up its game in the high-bandwidth memory (HBM) market with the launch of its HBM4E 12-layer sample, targeting advanced AI applications. The company announced on June 18 that it has started supplying these next-generation chips to key customers.

The HBM4E represents a significant leap forward from its predecessor, HBM4, boasting enhanced performance and power efficiency. It achieves data processing speeds of up to 16Gbps per pin, a 20% improvement in energy efficiency, crucial for the demanding workloads of AI training and inference. The chip's design optimizes data transfer latency and ensures stable operation in high-bandwidth environments, promising to boost the efficiency of next-generation AI data centers and large-scale computing systems.

A key innovation in the HBM4E is SK Hynix's proprietary 'Advanced MR-MUF' (Mass Reflow Mold Underfill) process. This technology enables the stacking of 12 layers, achieving a 48GB capacity while significantly improving structural stability. Notably, the thermal resistance has been reduced by approximately 17% compared to HBM4, ensuring reliable performance even under intense computing conditions. The MR-MUF process involves injecting a protective liquid material between stacked chips and solidifying it to shield the circuits.

SK Hynix aims to leverage its proven quality and supply capabilities to address bottlenecks in AI systems alongside its customers, supporting the development of next-generation infrastructure. Ahn Hyun, SK Hynix's Chief Development Officer, expressed confidence, stating, "We have laid the foundation to continuously lead AI innovation by carrying over our industry-leading technological competitiveness and mass production capabilities to the HBM4E product."

“

We have laid the foundation to continuously lead AI innovation by carrying over our industry-leading technological competitiveness and mass production capabilities to the HBM4E product.