AI EncyclopediaChips & deep techChips

HBM (high bandwidth memory)

HBM(고대역폭 메모리)

AI memory built by stacking memory chips in layers so that data moves in and out very fast.

Overview

An AI chip (GPU) has to keep pulling the data it will calculate from memory. If the memory is slow, even the best GPU waits around and cannot reach its full speed.

HBM stacks memory chips in layers like an apartment building and links the layers directly with a great many thin passages. This greatly increases the amount of data that can move at once. SK hynix, Samsung Electronics and Micron make it.

An everyday example

No matter how fast a kitchen (the GPU) is, food comes out late if the passage that carries the ingredients is narrow. HBM widens that passage into a highway with many lanes.

Why it matters

Several HBM chips are attached to every AI chip. It is a part that runs short as AI investment grows, so it ties directly to the earnings of memory chip makers.

Also calledhigh bandwidth memory, AI memory

On the timelines

These concept articles were written by AI and checked again for accuracy before publishing. The English edition is translated from the Korean articles. They leave out things that change quickly, such as prices and rankings. Tell us if something is wrong. Report an errorLast updated

You