Computer chip material inspired by the human brain could slash AI energy use

16 hours ago

Researchers from the University of Cambridge developed a new nanoelectronic device using hafnium oxide that mimics neural connections.
The device acts as a memristor, enabling energy-efficient neuromorphic computing by storing and processing data in the same place.
Current AI hardware consumes high energy due to data shuttling between memory and processing units.
The new memristor achieves switching currents a million times lower than conventional oxide-based devices.
It provides hundreds of distinct conductance levels with excellent stability and uniformity, essential for analogue in-memory computing.
The device replicates biological learning mechanisms like spike-timing dependent plasticity.
Key fabrication challenge is the high temperature requirement (~700°C), but efforts are underway to reduce it for industry compatibility.
If integrated into chips, this technology could drastically cut AI energy use by up to 70% and enable adaptive hardware.
The breakthrough followed years of experimentation, with a patent application filed by Cambridge Enterprise.
Research was supported by various institutions, including the Swedish Research Council and UKRI.

Hasty Briefsbeta