Cambridge memristor chip using hafnium oxide could cut AI energy use by 70%

University of Cambridge researchers led by Dr. Babak Bakhit developed a nanoelectronic memristor using modified hafnium oxide with strontium and titanium additions that mimics brain synapses, potentially reducing AI energy consumption by up to 70%. Published in Science Advances (DOI: 10.1126/sciadv.aec2324), the device creates p-n junctions that switch resistance smoothly via energy barrier height changes rather than unreliable conductive filaments. It achieves switching currents about a million times lower than conventional oxide-based devices and produces hundreds of distinct, stable conductance levels — a key requirement for analogue in-memory computing. Lab tests showed reliable endurance over tens of thousands of switching cycles with state retention of around one day. Current fabrication requires ~700°C temperatures; the team is working to reduce this for standard industry compatibility. The device addresses the von Neumann bottleneck where conventional chips waste energy moving data between memory and processing units.