SK Hynix has achieved a significant technological milestone by completing development and establishing the world’s first mass production system for HBM4, the next-generation high bandwidth memory designed for ultra-high performance artificial intelligence applications.
This breakthrough positions the South Korean memory manufacturer at the forefront of AI infrastructure development.
The HBM4 represents the sixth generation of High Bandwidth Memory technology, following the progression from the original HBM through HBM2, HBM2E, HBM3, and HBM3E.
This latest iteration addresses the escalating demands of AI systems and data centers, which require increasingly sophisticated memory solutions to handle massive data processing workloads while maintaining energy efficiency.
The technical specifications of HBM4 demonstrate substantial improvements over previous generations. The memory achieves doubled bandwidth performance through the implementation of 2,048 input/output terminals, representing a 100% increase from the previous generation.
Power efficiency gains exceed 40%, addressing critical concerns about data center energy consumption. Operating speeds surpass 10 gigabits per second, significantly exceeding the JEDEC standard of 8 gigabits per second.
SK Hynix projects that HBM4 implementation will enhance AI service performance by up to 69%, effectively resolving data bottlenecks while substantially reducing data center power costs.
The company has incorporated proven Advanced MR-MUF processing technology alongside their advanced 1b-nanometer manufacturing process to ensure production reliability and minimize manufacturing risks.
Company executives emphasized the strategic importance of this achievement. Joohwan Cho, Head of HBM Development, characterized the completion as an industry milestone that will enable timely delivery of products meeting customer requirements for performance, power efficiency, and reliability.
Justin Kim, President and Head of AI Infrastructure, positioned HBM4 as a solution for overcoming AI infrastructure limitations and reinforced the company’s commitment to becoming a comprehensive AI memory provider.