Skip to content

Citrini View: Korean Manufacturers Adopt Cautious Approach to 3D IC Commercialization, while Chinese ChangXin Expedites Market Positioning through Customization Strategy

Jul 25, 17:12

July 25th, Citrini analyst Jukan quoted the latest industry analysis from ZDNet Korea, pointing out that the AI semiconductor performance competition is shifting from process miniaturization to advanced stacking technology. Vertical integration of logic and memory in 3D IC is becoming the key technology of the next generation. Samsung Foundry has recently seen a surge in inquiries about 3D IC, with nearly every potential customer asking about this technology. However, Jukan believes that 3D IC is essentially a single custom DRAM, which fundamentally conflicts with the core business model of large-scale standardized production of Samsung and SK Hynix. Both Korean fabs are expected to be limited to preliminary research and remain cautious about full commercialization.

This structural hesitancy has opened a window for niche players. China's ChangXin Memory, facing difficulties entering the HBM market due to U.S. semiconductor restrictions, is using 3D IC as a differentiation breakthrough. They are targeting custom memory instead of directly competing with Korean fabs in commodity DRAM. South Korean semiconductor industry officials have indicated that China has already begun producing multiple 3D DRAM sample chips, and niche market participants like ChangXin and Winbond are likely to be the first to establish a foothold in this market.

Analyst Jukan had previously pointed out that NVIDIA's CUDA moat is coming to an end. The breakdown of software ecosystem barriers and the trend towards customization in 3D IC are happening simultaneously, potentially leading to a structural reshaping of the next-generation AI memory competitive landscape.

Note: A niche market refers to a niche in a large market that mainstream players ignore or are unwilling to enter. Niche players refer to companies focusing on such niche markets.

View source