“Demand for improvement in memory bandwidth is increasing following a spike in data traffic in the era of AI and high-performance computing,” Lee Kang-wook, head of package development at SK hynix, said during a session at Semicon Taiwan, an annual international semiconductor exhibition, in Taipei, Taiwan.
“HBM has been widely adopted as the memory product for AI servers and high-performance computing,” he added.
HBM is a high-performance dynamic random access memory (DRAM) that is seeing high demand, particularly via U.S. AI chip giant Nvidia’s GPUs, which are key components of AI computing.
SK hynix is leading the market with fifth-generation HBM3E. Its latest 12-layer HBM3E products are scheduled for mass production in the third quarter.
Lee also highlighted that the company’s sixth-generation HBM4 will offer higher memory capacity, faster data processing speeds and improved efficiency compared with HBM3E due to technological advancements such as the adoption of the logic process on the base die.
Technological and performance improvements in HBM have fueled growing demand, particularly in the generative AI market, which is projected to grow at an average rate of 27 percent between 2023 and 2032. As a result, the HBM market is expected to expand by an average of 109 percent between 2022 and 2025.
To keep pace, SK hynix is focusing on packaging technologies, specifically mass reflow molded underfill (MR-MUF). This technology, which SK hynix introduced with HBM2E in 2019, injects a liquid protective material between stacked chips, significantly reducing heat.
Building on MR-MUF technology, SK hynix plans to ship 12-high HBM4 products next year and aims to launch 16-high HBM4 products by 2026.
In addition, SK hynix announced plans to strengthen its cooperation with global tech giants, such as Nvidia, Apple, Microsoft and Alphabet, to solidify its leadership position in the HBM market.
“SK hynix has also been enhancing its collaboration with global partners in creating the ecosystem to meet various customer demands in a more efficient way as customization becomes a trend from HBM4E,” Lee said.
Yonhap