AI Can Now Design Computer Chips 2.6 Times Faster

Every computer chip — the brain inside your phone, laptop, car, and microwave — has to be designed before it can be built. Traditionally, engineers write incredibly detailed instructions describing millions of tiny connections, like drawing a house by specifying the position of every single brick. It works, but it's slow and expensive. This new research asks a simple question: what if AI starts with a rough sketch instead of every brick?

The team, from UCLA, tested AI 'agents' — software that can plan, try things, check results, and try again — at four different levels of detail. The winning combination, called AHRR, has the AI first describe what the chip should do at a high level, then drop down and fine-tune the small details. Across 11 different test designs, this finished 2.6 times faster than making the AI grind through the lowest-level instructions from the start.

The insight is that each level of detail is good at different things. The high-level version bundles decades of engineering know-how into reusable building blocks, so the AI doesn't reinvent the wheel. But those shortcuts hide small opportunities for speed and efficiency. Dropping back down to the detailed level lets the AI recover those wins. Think of it like planning a road trip with a map first, then checking individual street turns.

Here's the honest catch: this was tested on FPGAs (chips you can reprogram after manufacturing) using 11 benchmark designs, not on the cutting-edge chips inside a new iPhone. So it's a promising proof of concept, not a finished product. Still, it points to a real shift. If AI can shave months off chip design, the companies making your next device save money and time — and some of that usually reaches your wallet. It also raises a question for chip engineers: which parts of the job become automated, and which become more valuable?