Samsung Electronics and SK Hynix intensify competion over HBM4 production standards at Semicon Korea 2026, driven by surging AI demand for high-performance memory.
Samsung’s HBM4 Speed Milestone
Samsung Electronics Chief Technology Officer Song Jong-ho announced at the event that the company achieved the highest HBM4 data transfer speed of 11.7Gbps. This surpasses JEDEC standards by 37% and exceeds the prior HBM3E generation (9.6Gbps) by 22%.
Song highlighted the prototype’s business viabily, stating, “Samsung advances memory, foundry, and packaging to meet current AI chip requirements, optimizing the supply chain for collaboration.” He added that the firm leads development for next-gen HBM4E and HBM5 products.
Samsung plans mass production of HBM4 after partner verification on the smallest enterprise NVIDIA platforms. The prototype features a 12-Hi stack reaching 36GB capacy and up to 3TB/s bandwidth per stack—2.4 times denser than conventional single-stack memory. Innovations include a ’10nm-class 6-Hi DRAM stack’ and ‘4nm foundry-based power delivery’ for superior performance.
Song emphasized HBM’s role as the ‘AI system archecture platform,’ enhancing overall densy, structure, packaging, and logic to meet AI workloads.
SK Hynix Accelerates AI-Focused R&D
SK Hynix R&D Executive Vice President Kim Young-il presented on ‘The New Era of Change: Leading Memory Innovation Through Convergence.’ He stressed breakthroughs in AI collaboration ecosystems, noting, “DRAM currently faces 10nm process densy lims and internal high-integration challenges, prompting bold R&D advances.”
The executive outlined priories like structure and materials for upcoming paradigm shifts, including VG (Vertical Gate), 3D DRAM, and high-densy NAND fabrication. “We maintain development momentum through structural evolution and advanced semiconductor materials,” he said.
SK Hynix optimizes NVIDIA collaboration simulations and pushes boundaries in human resources and resources. Kim explained, “AI-driven cooperation introduces novel solutions differing from conventional approaches.” He affirmed the company’s capacy to lead new structure explorations and material integrations for AI memory demands.
Further, he noted, “Running AI models requires vast resources beyond tradional manpower; even minimal experiments yield optimal joint outcomes.” This shift from ‘Man-Month Based R&D’ to ‘AI-Based R&D’ maximizes efficiency.
Market Leadership in HBM
SK Hynix dominates HBM supply, handling two-thirds of NVIDIA’s volume for HBM3 and beyond. Both companies posion HBM4 (6th gen) as pivotal for AI expansion, wh Samsung leveraging foundry-DRAM synergies and SK Hynix focusing on AI-centric innovations.
