Hardware startups lose months to mistakes nobody catches until a prototype comes back dead. Circuit Mind, a London company automating board design, raised $25 million to attack the gap.
A new piece of hardware can eat more than a year, much of it spent in datasheets thousands of pages long. Faults surface only after a board is built and tested, and fixing them costs a redesign.
Why general models struggle here
Off-the-shelf language models fall short here, Circuit Mind argues, having never been trained on real circuits. Its agents sit on a patented physics engine that holds each decision to electrical limits, fed by machine-readable component models and the customer’s own engineering rules.
A block diagram comes back minutes later as a schematic, bill of materials and analysis, with variants ranked by cost, power or another priority. The winner exports into the CAD tools already in use.
Checking the schematic before the prototype
The Agentic Verification Engine, launched alongside the funding, targets design review. Swarms of agents crawl datasheets and review checklists, tracking connections and buses until a conflict with a pin table or voltage limit surfaces. Only after the system argues with itself does an engineer see anything.
Los Alamos National Laboratory cleared a report in which its staff finished a medium-difficulty board in a little over four hours; the hand route had been costed at 60 to 80. Daniel Gheno, Schneider Electric’s chief technology officer for energy management, said early evaluations suggest more than 90% less effort exploring architectures.
The real market is defense and industrials
The largest piece is an $18.5 million Series A led by Norwest Venture Partners, joined by newcomer Lightscape Partners and returning backers Third Prime, Differential Ventures and Blackhorn Ventures. A seed round Third Prime led, never announced until now, provides the remainder.
