Designing a new microchip is really difficult, you've got to figure out how to best place thousands of components into an area of only about 100 square millimeters and how to connect them all together with miles of super thin wire in a way that maximizes speed and minimizes power usage. Typically it takes a team of engineers months to do this, but Google now has an AI program that can do it in 6 hours.
A graph placement methodology for fast chip design <https://www.nature.com/articles/s41586-021-03544-w> Google trained the AI program by showing it 10,000 successful chip floor plans already made by humans and then just told the AI to learn how to do that sort of thing. It did learn how to do it and it has already been used by Google in the design of its newest Tensor processing unit. In a editorial in the same issue of Nature the editors expressed alarm that this new type of automation could lead to the massive unemployment of human chip designers: Google’s AI approach to microchips is welcome — but needs care <https://www.nature.com/articles/d41586-021-01507-9> Hmm ..., computers making better computers, that sounds sort of familiar, where have I heard that before? Oh yeah, positive feedback and the Singularity. John K Clark See what's on my new list at Extropolis <https://groups.google.com/g/extropolis> uy88 -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/everything-list/CAJPayv1oh%2BE7KgqBYZgcpuyy0-76LAyU32_rtNnXFag4AcFWng%40mail.gmail.com.
