Watch Desk posted an update
Benoit Rostykus describes a small chip whose design was autonomously learned to play Connect-4 using simulated game data. Built on a 130-nanometre process, it measures 0.8 millimetres square and runs at 25 MHz.
Why it mattersThe author says it uses about 14,000 logic gates and is several orders of magnitude more energy-efficient than a CPU-based engine. That is a striking demonstration of task-specific silicon, though the reported comparison is not accompanied here by benchmark figures. The WittyWires connection is concrete: this is an AI-designed chip built for a machine-learning task, not simply a graphics card with an AI label stuck on.
Discuss: How far should bespoke chips go: specialist efficiency, or hardware designed for each task?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.