It has been reported that Samsung Electronics is developing new packaging technology for next-generation HBM (Wide Band Memory) for smartphones, marking the start of a full-scale “memory race” for the era of AI smartphones.
Samsung upgrades packaging for mobile HBM ‘ultra-high aspect ratio copper pillar’
🟧 Samsung to develop “Mobile HBM” for smartphones
Samsung is reportedly developing the “Multi Stacked FOWLP” technology, which evolves existing VCS (Vertical Cu-post Stack) technology by combining ultra-high aspect ratio copper pillars (Cu-post) with FOWLP (Fan-Out Wafer Level Packaging). The goal is to realize high-performance on-device AI even in smartphones and XR devices. Traditional LPDDR has limitations in bandwidth and heat efficiency, and improving AI processing performance requires new memory implementation technologies.
- The aspect ratio of the copper columns has been improved from the conventional 3–5:1 to 15–20:1, enhancing wiring density and bandwidth.
- By combining FOWLP, the strength and mounting stability of thin Cu-posts are improved.
- Bandwidth increased by 15–30%, stacked capacity exceeding 1.5 times, anticipating future Galaxy AI and XR applications
🟧In the era of AI smartphones, “memory bandwidth” becomes the main competitive axis.
The background to this announcement is the rapid expansion of “on-device AI,” which operates generative AI and AI agents on smartphones. In recent years, while NPU performance has improved, the “memory bandwidth” required for high-speed transfer of large amounts of data required for AI processing has become a new bottleneck. Especially in LLMs and multimodal AI, memory transfer performance is considered even more important than computational performance.
Furthermore, across the entire industry, the focus is beginning to shift from mere miniaturization competition to the “advanced packaging competition.” Just like TSMC’s CoWoS and Intel’s Foveros, we live in an era where “how you connect” determines performance.
Additionally, SK Hynix is reportedly developing ‘HBS (High Bandwidth Storage),’ which is likely to further accelerate the competition for high-bandwidth memory for mobile AI.
🟧 Conclusion
The mobile HBM technology that Samsung is developing is attracting attention as not only speeding up smartphone memory but also achieving the “ultra-high bandwidth and low latency” needed in the AI era.
By combining ultra-fine Cu-post and FOWLP, high-density implementation is possible even in limited smartphone spaces, potentially leading to improved performance in future Galaxy AI and XR devices.
