Samsung Electronics has introduced its 'CUBE' strategy, aimed at overcoming memory limitations in the era of artificial intelligence (AI) through three-dimensional structural innovation. The strategy seeks to enhance not only capacity but also data utilization, bandwidth, and power and thermal efficiency. The company is expanding its product lineup to include HBM5, next-generation zHBM, and zNAND-O, which bridges the gap between DRAM and NAND, to strengthen its leadership in AI memory.
On September 1, at the Memory Executive Summit during SEMICON Taiwan 2026 in Taipei, Choi Jang-seok, Senior Vice President of Samsung Electronics' Memory Business, presented this memory technology strategy. CUBE stands for Capacity, Utilization, Bandwidth, and Efficiency.
The core of the strategy is the recognition that the performance demands of AI cannot be met solely by miniaturizing existing planar structures, thus necessitating the active use of three-dimensional designs. Choi stated, "We will increase memory capacity by expanding vertically without being limited by PCB area, thus avoiding the need to increase board size."
Utilization of each layer is becoming increasingly important beyond simple stacking. Choi emphasized, "3D is not just about stacking; how each layer is utilized is key. Samsung is optimizing the placement of logic and memory to eliminate data processing delays."
The speed of data movement, which influences AI computational performance, is also a key aspect of the CUBE strategy. Choi noted, "By drastically shortening the distance between dies, we can create vertical highways that significantly improve bandwidth." This approach aims to reduce the distance data must travel by connecting chips vertically, thereby increasing throughput.
Power consumption and heat generation are also critical challenges. As the computational load of AI systems increases, so does the power required to move data. Choi remarked, "The ultimate goal of 3D is to reduce the energy needed to move a single bit of data, maximizing performance per watt by enhancing structural efficiency."
Samsung plans to apply this strategy to its next-generation HBM products. The HBM5 currently under development aims to double the performance of HBM4E, with a 20% improvement in performance per watt and a 20% reduction in thermal resistance.
Additionally, Samsung is developing zHBM, targeting eight times the performance and three times the performance per watt compared to HBM4E, with a goal of reducing thermal resistance by 75% to 90%. This reflects the need for memory to improve not only bandwidth but also power and heat management as AI accelerator performance rapidly increases.
Samsung is also preparing zNAND-O, which aims to bridge the performance and capacity gap between DRAM and NAND. zNAND-O is designed to achieve ten times the bit density of DRAM while increasing read bandwidth and power efficiency by seven times, catering to the large data requirements of large language models (LLMs). The company plans to begin sample supply in 2028.
Alongside the development of next-generation products, Samsung is leveraging its established production capabilities in the existing memory market as a competitive advantage in the AI memory sector. According to Counterpoint Research, Samsung held a 39% share of the global DRAM market by revenue in the second quarter of this year, leading over SK Hynix at 26% and Micron at 25%.
Samsung also maintained its top position in the NAND market, with a 29.3% share in the second quarter, as sales of enterprise solid-state drives (SSDs) for AI servers surged, resulting in a 70.7% increase in revenue compared to the previous quarter.
Its production capacity is among the largest in the industry. Omdia projects that Samsung's DRAM production capacity will reach approximately 8.175 million 300mm wafers annually this year. Samsung plans to expand its product portfolio in the AI memory market by integrating its production capabilities in general-purpose DRAM and NAND with HBM, enterprise SSDs, and next-generation 3D memory.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.