Researchers at the University of Cambridge have developed a new hafnium oxide material that could dramatically reduce AI's energy consumption by mimicking how neurons work in the brain. The breakthrough memristor technology uses strontium and titanium to create tiny electronic gates that switch states using extremely low currents, achieving switching currents about a million times lower than conventional devices.

Current AI systems waste massive amounts of electricity shuttling data between memory and processing units. The Cambridge team's approach stores and processes information in the same place, potentially cutting energy use by up to 70 percent. The devices produced hundreds of distinct, stable conductance levels essential for analogue 'in-memory' computing and could reliably endure tens of thousands of switching cycles.

Lead author Dr. Babak Bakhit said the technology overcomes the random behavior that plagues most existing memristors by switching at interfaces rather than forming unpredictable filaments. The devices also reproduced fundamental learning rules observed in biology, such as spike-timing dependent plasticity. While the current fabrication process requires temperatures around 700°C - higher than standard semiconductor manufacturing - Bakhit believes solving this challenge could make the technology game-changing for energy-efficient AI hardware.