Can Physics Expedite Artificial Intelligence?

Artificial Intelligence (AI) requires intensive optimization calculations, which now demands nuclear power for the data centers. Logic functions, storage, and data communications each consume ~1pJ/bit, about 6 orders of magnitude higher than the Landauer Limit. This gives rise to 25MW in a Data Center versus the 25Watts of the human brain. This 6 order of magnitude shortfall is owing to the transistor being a flawed switch. Overcoming this shortfall should be deemed a “Grand Challenge of Condensed Matter Physics”. New approaches are beginning to emerge in the context of Artificial Intelligence.

Speaker: Eli Yablonovitch, UC Berkeley