The domestic semiconductor industry is expected to experience a prolonged boom, driven by increased investment in artificial intelligence (AI). As demand for AI inference surges, the high-performance memory market, including high bandwidth memory (HBM), is rapidly expanding. In response to China's technological advancements, domestic semiconductor companies are intensifying their competition in next-generation technology development.
On October 14, at the 28th Semiconductor Exhibition (SEDEX 2026) held at COEX in Gangnam, Seoul, SK Hynix Chief Development Officer (CDO) Ahn Hyun projected that "the performance scale of AI infrastructure will grow approximately 20 times by 2030 compared to this year."
Ahn noted that the complexity of AI inference tasks, along with the volume of information processed and the number of requests, is driving a significant increase in memory demand. He explained that the rise of 'KV cache,' which stores and recycles information processed by AI, is creating new performance bottlenecks in data movement between HBM, DRAM, and solid-state drives (SSD).
To address this, SK Hynix plans to expand its customized HBM offerings while pursuing the development of high bandwidth flash (HBF) with over 10 times the capacity of HBM, as well as three-dimensional stacked DRAM directly on top of computing semiconductors. Ahn stated, "While standard HBM and customized HBM will coexist, customized products will expand in the long term."
Additionally, China's memory technology advancements pose a competitive challenge for domestic firms. Chinese memory company ChangXin Memory Technologies (CXMT) has announced plans to apply '4F²' technology to next-generation server DRAM. This technology reduces the cell area of DRAM, allowing for more data to be stored in the same space.
On the same day, Samsung Electronics Chief Technology Officer (CTO) Song Jae-hyuk confirmed to reporters that the company is indeed developing the 4F² technology, responding affirmatively to inquiries about its progress. Although both Samsung and SK Hynix have been researching this technology, it has not yet been applied to mass production due to cost issues. As China's technological development accelerates, domestic companies are focusing on securing next-generation memory technologies.
At the exhibition, both companies showcased their next-generation AI semiconductor technologies and business strategies. Samsung highlighted its comprehensive semiconductor capabilities, covering memory, foundry, and advanced packaging. The company presented next-generation HBM4E, HBM4E base die based on a 4-nanometer process, and three-dimensional memory technology zHBM, introducing an integrated solution for AI semiconductors. Samsung also unveiled plans for infrastructure expansion at key production sites, including the Yongin National Industrial Complex, where it plans to invest approximately 360 trillion won. Construction is set to begin in 2027, with the first production line expected to be operational by October 2029.
SK Hynix focused on AI memory technology centered around HBM and emphasized its collaboration with Nvidia. The company displayed a large booth featuring models of HBM and HBF structures, high-performance DRAM for AI servers, and enterprise SSDs. Notably, the partnership with Nvidia was highlighted through the display of the 'Vera Rubin' superchip and an HBM wafer signed by CEO Jensen Huang. Various memory products, including a 256GB 3DS RDIMM for servers and LPDDR6 for mobile devices, were also showcased.
On October 14, at the 28th Semiconductor Exhibition (SEDEX 2026) held at COEX in Gangnam, Seoul, SK Hynix Chief Development Officer (CDO) Ahn Hyun projected that "the performance scale of AI infrastructure will grow approximately 20 times by 2030 compared to this year."
Ahn noted that the complexity of AI inference tasks, along with the volume of information processed and the number of requests, is driving a significant increase in memory demand. He explained that the rise of 'KV cache,' which stores and recycles information processed by AI, is creating new performance bottlenecks in data movement between HBM, DRAM, and solid-state drives (SSD).
To address this, SK Hynix plans to expand its customized HBM offerings while pursuing the development of high bandwidth flash (HBF) with over 10 times the capacity of HBM, as well as three-dimensional stacked DRAM directly on top of computing semiconductors. Ahn stated, "While standard HBM and customized HBM will coexist, customized products will expand in the long term."
Additionally, China's memory technology advancements pose a competitive challenge for domestic firms. Chinese memory company ChangXin Memory Technologies (CXMT) has announced plans to apply '4F²' technology to next-generation server DRAM. This technology reduces the cell area of DRAM, allowing for more data to be stored in the same space.
On the same day, Samsung Electronics Chief Technology Officer (CTO) Song Jae-hyuk confirmed to reporters that the company is indeed developing the 4F² technology, responding affirmatively to inquiries about its progress. Although both Samsung and SK Hynix have been researching this technology, it has not yet been applied to mass production due to cost issues. As China's technological development accelerates, domestic companies are focusing on securing next-generation memory technologies.
At the exhibition, both companies showcased their next-generation AI semiconductor technologies and business strategies. Samsung highlighted its comprehensive semiconductor capabilities, covering memory, foundry, and advanced packaging. The company presented next-generation HBM4E, HBM4E base die based on a 4-nanometer process, and three-dimensional memory technology zHBM, introducing an integrated solution for AI semiconductors. Samsung also unveiled plans for infrastructure expansion at key production sites, including the Yongin National Industrial Complex, where it plans to invest approximately 360 trillion won. Construction is set to begin in 2027, with the first production line expected to be operational by October 2029.
SK Hynix focused on AI memory technology centered around HBM and emphasized its collaboration with Nvidia. The company displayed a large booth featuring models of HBM and HBF structures, high-performance DRAM for AI servers, and enterprise SSDs. Notably, the partnership with Nvidia was highlighted through the display of the 'Vera Rubin' superchip and an HBM wafer signed by CEO Jensen Huang. Various memory products, including a 256GB 3DS RDIMM for servers and LPDDR6 for mobile devices, were also showcased.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.