SK hynix Showcases Leading AI Memory Technology for the AI Era

SK hynix has highlighted its AI memory leadership at Taiwan's COMPUTEX 2026, positioning memory as the core of the AI era. The company participated in 'COMPUTEX 2026,' held from June 2nd to 5th at the Taipei Nangang Exhibition Center, with the concept 'At the core of the AI wave – Memory!', emphasizing memory's crucial role in the AI era.

The centerpiece of SK hynix's booth was the 'AI Factory Zone,' showcasing its collaboration with NVIDIA. This zone visually conveyed the role of memory in actual AI systems by displaying NVIDIA products alongside SK hynix memory solutions. Featured items included NVIDIA's latest supercomputer 'DGX Spark' and SK hynix's LPDDR5X memory integrated into it, along with the next-generation AI accelerator 'GB300' and SK hynix's HBM3E, demonstrating the technological synergy between the two companies. NVIDIA CEO Jensen Huang visited SK hynix's booth on June 2nd and met with SK Group Chairman Chey Tae-won, leaving his signature on collaborative products. The exhibition also displayed a physical model of NVIDIA's latest AI accelerator 'GB300' with SK hynix's HBM3E, and a 'NVIDIA Partner Sign' bearing Jensen Huang's signature, symbolically representing their partnership. The GB300 model facilitated visitor understanding of the internal structure and memory layout of AI accelerators. Additionally, a model of NVIDIA's upcoming superchip 'Vera Rubin' was exhibited alongside SK hynix's LPCAMM2 and HBM4, with the slogan 'Scalable AI with Confidence' underscoring their ongoing collaboration for future AI infrastructure.

As AI models expand and service scopes widen, memory's bandwidth, capacity, power efficiency, and packaging technologies become critical for system scalability. SK hynix supported the scalability of AI infrastructure through its HBM, specialized modules, and server memory products. The 'Chronicle of SK hynix's HBM' section at the booth provided a historical overview of HBM development, showcasing it as a high-value memory product that increases bandwidth by stacking multiple DRAMs vertically, emerging as a key component in the competitive AI accelerator market. SK hynix presented its HBM development journey chronologically, emphasizing that its current market position is a result of extensive technological accumulation. An internal structure model of HBM4E was also displayed, offering insights into the direction of advanced packaging and stacking technologies.

The 'Product Portfolio Zone' at the center of the booth showcased SK hynix's AI memory lineup, divided into two sections: 'AI Frontier & Industry Standards' and 'Infrastructure & Next-Gen.' This setup allowed visitors to grasp the company's main products driving the current AI ecosystem and its next-generation solutions for future infrastructure. The 'AI Frontier & Industry Standards' section featured HBM3E 36GB 12-layer, HBM4 48GB 16-layer, and HBM4E 48GB 12-layer as its core lineup, providing the bandwidth and capacity required for AI accelerators and high-performance computing environments. Also introduced were 3DS RDIMM 256GB for high capacity catering to AI workloads, RDIMM 64GB based on the 10nm-class 6th generation process, and DDR5 MRDIMM 128GB for high-bandwidth server memory. As the data volume processed by AI servers increases, the performance and efficiency of server DRAM have become a key factor in data center competitiveness. The eSSD lineup also garnered attention. SK hynix presented the PEB210 E1.S supporting liquid cooling, as well as the high-capacity, low-power PS1110 E3.S in NVMe E3.S specification, and the QLC-based PS1101 E3.S. AI data centers require repetitive reading and writing of large datasets during training and inference, making storage devices crucial elements that influence AI infrastructure performance beyond mere data retention. SK hynix emphasized its comprehensive memory and storage portfolio, covering the entire AI infrastructure, by presenting both DRAM and NAND-based products.

'Infrastructure & Next-Gen' focused on next-generation memory solutions. Notably, HBF was introduced as a novel memory concept involving vertical stacking of NAND flash using TSV technology, similar to HBM. As AI infrastructure evolves towards greater data processing capabilities and higher efficiency, HBF is being recognized as a technology poised to drive changes in the architecture of next-generation data centers and AI systems. This section also featured ZUFS 4.1 and LPCAMM2. ZUFS 4.1 is designed to enhance random read performance and reduce loading times for large language models (LLMs). LPCAMM2 bundles multiple LPDDR5X chips into a single module, offering high speeds even in low-power environments.