AI 학습 및 추론을 지원하는 고성능·고대역폭 반도체 수요가 급격히 증가하고 있다. 특히 SK하이닉스가 경쟁우위를 확보한 HBM을 바탕으로 올해 D램 시장 규모가 지난해 대비 65% 가까이 성장한 117조원에 이르면서 SK하이닉스는 선도 경쟁력과 미래 비전을 논의하는 자리를 마련했다.

▲SK Hynix executives attending the symposium. From left, Vice President Kwon Eon-oh (HBM PI), Vice President Kim Ki-tae (HBM S&M), Vice President Lee Dong-hoon (321-layer NAND PnR), Vice President Oh Hae-soon (NAND Advanced PI), Vice President Gil Deok-shin (Material Development), Vice President Son Ho-young (Adv. PKG Development), and Vice President Lee Jae-yeon (Global RTC) / (Photo: SK Hynix Newsroom)
SK Hynix Executive Roundtable Held, HBM Highlighted
AI Memory Future Market Trend: Heterogeneous Convergence
Expanding AI utilization areas = Memory performance and capacity ↑
The demand for high-performance, high-bandwidth semiconductors that support AI learning and inference is rapidly increasing. In particular, with the DRAM market size reaching KRW 117 trillion this year, up nearly 65% from last year, based on HBM, in which SK Hynix has secured a competitive edge, SK Hynix has prepared a place to discuss its leading competitiveness and future vision.
SK Hynix recently held a roundtable discussion with new executives to discuss the background and competitiveness of its recent rise to become a leading global AI memory company and its future vision.
The symposium was attended by Vice President Kwon Eon-oh (HBM PI), Vice President Gil Deok-shin (Material Development), Vice President Kim Ki-tae (HBM S&M), Vice President Son Ho-young (Adv. PKG Development), Vice President Oh Hae-soon (NAND Advanced PI), Vice President Lee Dong-hoon (321-layer NAND PnR), and Vice President Lee Jae-yeon (Global RTC). The symposium was moderated by Vice President Won Jeong-ho (Global PR).
The new executives cited SK Hynix's long-term development of HBM before the market was formed and collaboration with customers for cross-species fusion as the reasons for its market leadership.
Vice President Kwon Eon-oh said, “Our persistent investment and research in AI memory, which has continued for a long time even before the market opened, has become the foundation for the company’s growth,” adding, “Based on this, we have preemptively secured the technological prowess and mass production know-how in HBM and various high-performance memories essential to AI infrastructure.”
As a result, the company was able to supply products faster than its competitors and gain experience in large-scale mass production, which seems to have created a virtuous cycle and a chain effect that relatively increased product reliability.
Vice President Son Ho-young added, “To meet market demands, we must build next-level partnerships with customers and prepare for collaboration for cross-disciplinary fusion that breaks down the boundaries between memory and systems, front-end and back-end processes.”
SK Hynix was preparing in advance for the next-generation AI memory competitiveness. This is a high-capacity NAND-based product.
Vice President Lee Dong-hoon said, “NAND-based high-capacity SSD products are essential to accumulate the large amount of data required for AI.” He added, “SK hynix is developing Multi-Plug technology that stably implements NAND’s ultra-high-density data storage area and All PUC (Peri. “We are enhancing our competitiveness in this field by securing technologies such as Cell Under Cell (Cell Under Cell) first in the industry,” he said.
Executives diagnosed that the reason AI memory is receiving so much attention is because high-performance solutions such as HBM, CXL, eSSD, and PIM are solving the data bottleneck of existing memory and increasing the operating speed of AI.
As the scope of AI utilization expands in the future, the adoption of high-performance, high-capacity memory is bound to increase. Vice President Kim Ki-tae predicted, “Generative AI technology is being widely used not only in public services such as education and healthcare, but also in major consumer analysis, product/service development and marketing, and supply chain management in the B2C market, so the utilization of memory will increase even more.”
Vice President Gil Deok-shin said, “It is important to see how generative AI technology can be integrated with many application fields and reach out to consumers,” and predicted, “In the future, the integration of on-device, autonomous driving, and the robotics industry with AI will drive changes and growth in the semiconductor market.”
Vice President Lee Dong-hoon predicted, “As AI technology advances further, we can imagine a future where the metaverse becomes so advanced that it is close to actual reality, and as big tech companies seek to dominate this field, the demand for memory will increase.”
It has been reported that the types of memory that will attract attention in the future market include M-RAM (magnetoresistive memory), R-RAM (resistive change memory), and PCM (phase change memory). Vice President Lee Jae-yeon emphasized the need to continuously strengthen research and development, saying, “In addition to MRAM, RRAM, and PCM, we are also paying attention to emerging memories that also have low-power characteristics, such as SOM (Selector Only Memory), Spin Memory, and Synaptic Memory.”
SOM is a memory that implements the functions of the capacitor (storage area) and transistor (control area) of existing memory in a single element, and is known as a memory that overcomes the limitations of the micro-process of existing PCM, and spin memory is a memory that distinguishes whether or not data is stored by applying the spin motion characteristics of electrons, a representative example of which is MRAM. Synaptic Memory is a memory based on artificial neural network elements, and it is expected to solve the limitations of existing computing structures (serial processing method) such as data bottlenecks by implementing a high-efficiency computing structure similar to the human brain.
In particular, the next-generation HBM4 is said to be the first to introduce a logic semiconductor process to memory and will be mass-produced, and SK Hynix is expecting that the introduction of the new process will lead to an improvement in HBM specifications and an increase in orders.
Accordingly, Vice President Son Ho-young emphasized the strengthening of cooperation with global academia, industry, and research institutes, saying, “In order to implement heterogeneous integration such as memory and non-memory, advanced packaging technology must be advanced.” Vice President Gil Deok-shin also added, “In the packaging field, which has recently become increasingly important, it is important to develop materials that enhance product performance and quality by increasing yield and heat dissipation characteristics.”