Samsung Reportedly Reallocates Half of Its 4nm Foundry Capacity to HBM4 Base Dies
Samsung Foundry is reportedly dedicating approximately 50% of its overall 4nm (SF4) semiconductor manufacturing capacity to produce base dies for sixth-generation High Bandwidth Memory (HBM4). This strategic reallocation at its Pyeongtaek S5 and S6 production lines aims to address surging demand from custom AI chip and ASIC customers. Allowing clients to embed custom logic like memory controllers directly into the HBM4 base die frees up silicon area on main AI compute chiplets while boosting overall processing performance. This move reinforces Samsung's aggressive expansion in the memory market, where its HBM revenue share surged to 38% in Q2 2026 as customers look to diversify suppliers. The HBM4 custom base dies are fabricated on Samsung's 4nm node (SF4) at its P2 and P3 fabs in Pyeongtaek. By embedding memory controllers and PHY interface circuits onto the base die, custom ASIC and AI hardware designers can maximize compute density and power efficiency.
## BACKGROUND
High Bandwidth Memory (HBM) stacks multiple DRAM dies vertically connected by Through-Silicon Vias (TSVs), resting on top of a foundational silicon layer called a base die. While earlier HBM generations used mature semiconductor nodes primarily for signal routing, the transition to HBM4 requires advanced logic processes like 4nm to integrate complex computing components directly into the memory subsystem.