Cambridge Memristor Design Could Cut AI Energy Use by 70%

The rapid expansion of generative AI has intensified pressure to curb the power consumed by data movement and computation. A University of Cambridge team has developed a low-energy memristor based on modified hafnium dioxide. The neuromorphic device combines memory and processing functions, potentially reducing the repeated transfer of data between separate storage and computing units that makes conventional AI hardware energy-intensive.

As of Aug. 5, 2026, the researchers said the device offered strong uniformity, low switching currents and multiple stable conductance levels suited to analog computing. Their design could reduce AI system energy use by as much as 70%, according to the reported findings. Commercial adoption remains uncertain, however, because fabrication still requires high temperatures that are difficult to integrate with established semiconductor manufacturing processes and complete AI chip systems.