Cambridge scientists build brainlike chip promising 70%+ AI energy savings
Technology
Cambridge scientists have built a new chip inspired by how our brains work, and it could help AI use more than 70% less energy.
Instead of keeping memory and processing separate (like most computers), this chip combines them into one, just like brain synapses do, making things way more efficient.
Hafnium oxide mix stabilizes chip learning
The secret is a special mix of materials (hafnium oxide with strontium and titanium) that keeps the chip's behavior steady and predictable, even at super-low power.
This design also helps the chip learn in a brain-like way.
The team is now working on lowering manufacturing temperatures so these chips can be made more easily, which could eventually help enable more energy-efficient AI tech.