Samsung is shifting its HBM4 memory strategy toward more advanced manufacturing, a move that could strengthen its position in the high-performance AI chip market. The company plans to produce the base logic dies for these memory stacks using 2-nanometer technology at its new facility in Giheung, Korea. This change matters because the base die determines the core performance and efficiency of the memory, which is critical for NVIDIA's upcoming AI accelerators. Buyers of next-generation AI hardware will see the impact of this manufacturing shift in the final chip capabilities.
Samsung plans 2nm base logic dies for HBM4 memory stacks destined for NVIDIA AI accelerators
The HBM4 memory chips themselves will continue to use 4-nanometer technology, a plan Samsung announced earlier. This creates a hybrid approach where the base die and the stack dies are manufactured on different process nodes. The base dies are expected to be destined for NVIDIA's AI chips, linking Samsung's memory production directly to the dominant GPU maker. Samsung intends to use the NRD-K Line 2 at its under-construction R&D facility for this purpose.
The 2-nanometer process for the base dies represents a significant jump in transistor density compared to the 4-nanometer node used for the rest of the chip. This advanced manufacturing node enables higher bandwidth and reduced power consumption in the memory module's core logic. Competitors SK hynix and Micron have partnered with TSMC for their HBM4 base die manufacturing, putting Samsung in a different position by using its own facilities. The NRD-K Line 2 is specifically identified as the production line for this 2nm base die work.
These plans are subject to revision as the facility opening is roughly scheduled in two years. The projected timeline provides Samsung with flexibility to adapt its manufacturing strategy in response to market demand and technical challenges. We looked at the last HBM4 update, where several of the same balance and stability themes came up. Samsung's success in executing this 2nm base die plan will be a key factor in its competitiveness against rivals linked to TSMC.
Samsung aims to deliver HBM4 memory with 2nm base dies for NVIDIA AI chips by leveraging its new NRD-K Line 2 facility. The company will use 4nm technology for the remaining HBM4 memory components. This manufacturing split defines the current roadmap for Samsung's next-generation high-bandwidth memory.
Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.