Samsung moved the base layer of its memory stacks to a 4nm logic process — and found spare room on it

At Hot Chips 2026 Samsung described building the base die underneath its HBM4 and HBM4E memory stacks on a 4nm logic process instead of a memory process. That held down the power growth each memory generation brings, and left unused area on the die, because the base die cannot shrink below the memory stacked on top of it. Samsung set out three phases for filling that space, starting with moving the memory controller itself down onto the stack.

What it means Memory bandwidth and memory power are what actually cap how cheaply a big model can be served; moving logic into the memory stack is where the next headroom comes from.