Scientists develop brain-inspired chip to cut AI energy use by 70%

Researchers at the University of Cambridge have developed a novel brain-inspired memristor chip using a specialized hafnium oxide thin film, which could reduce AI energy consumption by up to 70%. Unlike conventional computer chips that waste energy transferring data between memory and processors, this memristor mimics the brain’s synapse by processing and storing information in the same place. The device achieves this through stable, low-energy switching at an interface regulated by internal p-n junctions created with strontium and titanium, avoiding the unpredictable conductive filaments used in older memristors. This results in outstanding uniformity, reliability, and the ability to support hundreds of distinct electrical states necessary for advanced analog in-memory computing. The new memristor operates with switching currents a million times smaller than previous technologies and demonstrates durability over tens of thousands of cycles while retaining data for about a day. It also replicates biological learning processes like spike-timing dependent plasticity, enabling hardware that can learn and adapt rather than merely store information