"AI Memory Breaks New Limits Again"... SK hynix Begins Supplying HBM4E 12-Stack Samples

Achieves up to 16Gbps per pin
Improves power efficiency by 20% and reduces thermal resistance by 17%
Delivers 48GB capacity with Advanced MR-MUF process

SK hynix has once again demonstrated its leadership in the artificial intelligence (AI) semiconductor market by rapidly delivering samples of its 7th generation High Bandwidth Memory (HBM4E)—which surpasses the limitations of previous products—to major global big tech clients. Building on the quality reliability and robust supply chain capabilities it has accumulated since the HBM3 era, the company’s strategy for this round of HBM4E competition is to meet the demanding infrastructure requirements of big tech customers in a timely manner.

On June 18, SK hynix announced that it has begun supplying samples of the next-generation, ultra-high-performance DRAM product for AI, the 12-layer ‘HBM4E’, to key customers. According to the company, this marks the launch of a new generation lineup, leveraging its overwhelming preemptive development capabilities in HBM and advanced production know-how accumulated over time.

SK hynix 7th Generation High Bandwidth Memory (HBM4E). SK hynix

The newly unveiled HBM4E stands out for dramatically enhancing both computing performance and power efficiency compared to the previous generation, HBM4 (6th generation). In terms of data processing speed, it consistently achieves up to 16Gbps per pin, while energy efficiency has improved significantly—by more than 20% over the previous generation. This specification can revolutionize the astronomical data processing speeds required during large-scale AI training and inference.

In addition, by applying the latest interface and conducting precise design optimization, SK hynix has minimized latency issues that can occur during ultra-fast data transmission. Even in high-bandwidth environments with massive traffic, the product supports stable operation without malfunctions, raising expectations that it will significantly enhance the overall processing efficiency of next-generation AI data centers and large-scale computing systems.

Furthermore, the company has implemented the ‘Advanced MR-MUF’ process, which has proven its competitiveness in previous generations. Thanks to this, SK hynix has achieved an overwhelming capacity of 48GB (gigabytes) with 12-layer stacking, while maximizing the stability of the chip’s internal physical structure. Most notably, the company succeeded in lowering the thermal resistance—a key indicator of heat dissipation efficiency—by about 17% compared to HBM4. This effectively eliminates critical bottlenecks, ensuring that the memory operates coolly and stably without overheating, even in extremely high-power computing environments.

Jensen Huang, CEO of NVIDIA, and Chey Tae-won, Chairman of SK Group, raised their glasses during a chicken and beer meeting at Kanbu Chicken Samsung Branch in Gangnam-gu, Seoul on the 7th. Photo by Yonhap News

Industry experts agree that SK hynix’s HBM4E sample supply means more than a simple new product launch. At the recent COMPUTEX 2026 event in Taiwan, Jensen Huang, CEO of NVIDIA, publicly praised SK hynix’s technology and expressed strong trust in the company. With SK hynix responding to Huang’s continued interest by shipping HBM4E samples, it is expected to further accelerate NVIDIA’s efforts to secure the supply chain for its next-generation AI accelerators.

An Hyun, President and Chief Development Officer (CDO) of SK hynix, expressed confidence, stating, “The industry-leading technological competitiveness and mass production capabilities we have built up through previous generation lineups are fully embodied in this HBM4E product, providing a solid foundation for us to continue leading global AI innovation.” He added, “We will continue to proactively realize the future value demanded by the market through close technological collaboration with global partners, solidifying our unparalleled leadership as a ‘full-stack AI memory creator’ that goes beyond simply being a component supplier.”

This content was produced with the assistance of AI translation services.