AI That Designs Better Computer Chips, Making Tech Faster and Cheaper

Think of AI like an artist that's learned from millions of paintings. Normally, it can only create new paintings in a similar style. But this new system, from researchers at multiple institutions, does something different: it takes an existing electronic design and "mutates" it — like evolution, but with careful guidance. The AI slowly adds noise to a design, then cleans it back up in a new way, creating a brand-new variant that still works.

These mutations aren't random guesses. The system is paired with professional engineering tools that test every idea for correctness and performance. If a mutation doesn't work, it's thrown away early. This means the AI can explore thousands of designs that no human or old-style AI would ever think of. In tests, it discovered a 32-bit adder — a key part of every computer processor — that is 17% faster and 18% smaller than a top industry design called Kogge-Stone. It also found seven new amplifier circuits for analog electronics, none of which existed in its training data.

Why does this matter to you? Your devices rely on designs like these. Faster, smaller components mean phones that process photos quicker, laptops that don't overheat, and chips that cost less to produce. The same approach could one day design batteries, medical implants, or even whole processors. The paper was released on arXiv, a public research site, so the trickle-down to consumer products is still years away — but this is a clear glimpse of what AI can do in engineering.