SK Hynix Begins Supplying 7th-Generation HBM4E Samples to Customers
HBM4E 12-layer Samples Supplied to Major Clients
Achieves up to 16 Gbps per chip… Improved performance and power efficiency
"We Will Drive AI Innovation with Industry-Leading Technology and Mass Production Capabilities"
[Edaily Reporter Gong Ji-yu] SK Hynix is supplying samples of its 12-layer 7th-generation High Bandwidth Memory (HBM4E)—a new ultra-high-performance DRAM product designed for next-generation artificial intelligence (AI) accelerators—to major clients. The company plans to accelerate the supply of next-generation products by leveraging its accumulated HBM development capabilities and production know-how. SK Hynix’s 7th-generation High Bandwidth Memory (HBM4E). (Photo: SK Hynix) This new product is characterized by a significant improvement in both performance and power efficiency compared to the previous generation, HBM4. It achieves a maximum data transfer rate of 16 gigabits per second (Gbps) per pin and improves energy efficiency by more than 20%, thereby substantially enhancing the data processing performance essential for AI training and inference.
Furthermore, by utilizing the latest interfaces and design optimizations, the company has reduced data transmission latency and ensured stable operation even in high-bandwidth environments. The company expects this to further enhance the processing efficiency of next-generation AI data centers and large-scale computing systems. SK hynix’s 7th-generation High Bandwidth Memory (HBM4E). (Photo: SK hynix) SK Hynix applied the Advanced Mass Reflow Molded Underfill (MR-MU) process to HBM4E, achieving a capacity of 48 gigabytes (GB) based on a 12-layer stack while enhancing structural stability. In particular, thermal resistance was reduced by approximately 17% compared to HBM4, ensuring stable memory operation even in high-performance computing environments.
SK hynix explained that, based on its mass production and supply experience with HBM3, HBM3E, and HBM4, it has consistently provided memory solutions optimized for customer needs in a timely manner. With HBM4E, the company plans to continue working with customers to resolve bottlenecks in AI systems and support the implementation of next-generation infrastructure.
Ahn Hyun, President and Chief Development Officer (CDO) of SK hynix, stated, “By carrying forward the industry-leading technological competitiveness and mass production capabilities we have built up over the years into our HBM4E products, we have laid the foundation to continue leading AI innovation.” He added, “Based on collaboration with our partners, we will proactively deliver the value the market demands and solidify our technological leadership as a full-stack AI memory creator.”
The biggest bottleneck for AI servers is memory. As the KV cache—where large language models (LLMs) store past computations—accumulates, the required memory capacity increases exponentially. This prob…
Concerns about “tax risks” have been raised in some quarters of the financial investment, pharmaceutical, and biotech markets regarding Genosco, a subsidiary of OSCOTEC Inc.(039200)specializing in new…
As July began (June 29–July 3), the pharmaceutical and biotech industries turned their attention to Celltrion Pharm Inc.’s large-scale investment in production facilities and AriBio’s successful fundr…