AI EncyclopediaChips & deep techChips
ASIC (application-specific integrated circuit)
ASIC(맞춤형 칩)
A chip designed to do just one job well. The best-known examples are chips that big tech companies design for their own AI.
Overview
A GPU is a chip that can be used for many kinds of calculation. An ASIC is the opposite: it is made to fit one set job exactly. By cutting out what is not needed, it can do that one job more cheaply and with less electricity.
A well-known example is the TPU, which Google made for its own AI. Amazon, Meta and Microsoft also design such chips themselves, which is why they are also called in-house chips. The downside is that designing one takes a long time, and if AI technology changes, it is hard to repurpose it for other work.
An everyday example
If a GPU is a kitchen knife that can cut anything, an ASIC is a machine that only peels apples. If all you do is peel apples, it is much faster and cheaper, but it cannot peel anything else.
Why it matters
To cut the money and electricity spent on AI chips, big companies are making more in-house chips. News that a company uses a custom chip instead of a GPU connects to how the AI chip market is being divided.
Also calledcustom chip, application-specific integrated circuit