On September 4, Lee Jung-bae, president of Samsung Electronics’ Memory Business Division, and Kim Joo-sun, president of SK Hynix’s AI Infrastructure Division, delivered speeches at Semicon Taiwan 2024 in Taipei, Taiwan. The two executives highlighted their companies’ advancements and strategies in the next-generation memory market, emphasizing the critical role of high-speed memory (HBM) and other emerging technologies in the AI era.
Chairman Lee Jung-bae began his speech by discussing the limitations of improving HBM performance with existing memory processes alone. He then focused on the strengths of Samsung’s HBM, emphasizing the need to integrate logic technology to maximize performance. “To maximize HBM performance, logic technology must be integrated,” Lee said. He also highlighted Samsung’s strong market position due to its capabilities in both foundry (contract semiconductor manufacturing) and semiconductor design. “Samsung, which has both foundry and semiconductor design capabilities, has a strong market position,” he added.
On the same day, SK Hynix President Kim Joo-sun discussed the company’s progress and future plans. He revealed that SK Hynix is smoothly progressing the development of the 6th generation HBM, HBM4, in collaboration with TSMC to ensure timely supply to customers. “Starting from the 6th generation HBM, HBM4, we are smoothly progressing the development work in collaboration with TSMC to supply customers in a timely manner,” Kim said. He also mentioned that SK Hynix plans to outsource the production of the core die to TSMC for the 6th generation products to further improve the features.
Lee and Kim’s speeches highlighted Samsung and SK Hynix’s pride in competing in the next-generation memory market. While SK Hynix currently holds a dominant position in the HBM memory market, the semiconductor industry is predicting potential changes in the future memory landscape. The two companies outlined their plans to captivate component companies and customers in Taiwan and gain a head start in the AI ecosystem.
The two executives expressed confidence in their next-generation memory products after Samsung President Lee HBM explained Samsung’s advantage in custom HBM production, explaining that the company can handle everything from design to production of the logic die, which acts as a control tower between the GPU and memory in AI accelerators. SK Hynix, on the other hand, refers to this component as a core die.
SK President Kim said, “We are continuously preparing for the development of HBM4, Low Power Compression Attached Memory (LPCAMM), Compute Express Link (CXL) and others.” These products are attracting increasing interest as demand for ultra-high-performance memory in the AI era increases.
LPCAMM, a product that addresses the high power consumption and heat issues of HBM, improves power efficiency while improving performance by aggregating multiple low-power DRAMs (LPDDRs). Samsung developed the industry’s first LPCAMM2 last September, and SK Hynix unveiled an LPCAMM2 supporting a transfer speed of 9.6 GB per second two months later.
CXL, another innovative product, breaks the conventional wisdom about existing memory by allowing separate (pooled) memory spaces to be shared like drawers. This technology is crucial as the parameters of recently released AI models have reached hundreds of billions. While HBM increases the computational speed by stacking memory vertically, CXL increases the absolute capacity itself.
The two executives pointed out that the biggest challenges in the AI memory market are improving energy efficiency and increasing memory capacity. Chairman Lee explained, “Training generative AI consumes a considerable amount of energy,” and added, “Although GPU computing power has increased significantly, memory bandwidth has not kept pace, which could limit the performance of AI models.”
Chairman Kim stressed the need to increase both speed and capacity while reducing power consumption in the AI era. “In the AI era, memory faces the challenge of increasing both speed and capacity while reducing power consumption,” he said. He also mentioned SK Hynix’s plans to strengthen infrastructure such as glass substrates and immersion cooling in collaboration with partners to become a global leader in AI chips. “To become a global leader in AI chips, we plan to strengthen infrastructure such as glass substrates and immersion cooling in collaboration with partners,” Kim concluded.
As competition between Samsung and SK Hynix intensifies, the semiconductor industry is eagerly watching for future developments that could reshape the memory market and drive advances in AI technology.
