Cambridge Researchers Develop Brain-Inspired Chip That Could Cut AI Energy Use by 70%
University of Cambridge researchers have created a nanoelectronic device that mimics how neurons process information, potentially reducing the energy consumption of AI systems by up to 70%. The breakthrough centers on a modified hafnium oxide component called a memristor that combines memory and processing in a single location, similar to biological brains.
Current AI systems waste substantial energy by constantly moving data between separate memory and processing units. A more efficient architecture, called neuromorphic computing, processes and stores information simultaneously in the same place.
How the New Device Works
Most existing memristors operate by forming tiny conductive filaments inside metal oxide materials. These filaments behave unpredictably and require high voltages, limiting their use in large-scale computing.
The Cambridge team engineered a hafnium-based thin film using a different mechanism. By adding strontium and titanium through a two-step growth process, they created small electronic gates called p-n junctions at layer interfaces. The device changes its resistance by adjusting the energy barrier at these junctions rather than forming and breaking filaments.
This approach eliminates the random behavior that plagues filament-based devices. Laboratory tests showed the new memristors switch at currents roughly a million times lower than conventional oxide-based alternatives and achieve hundreds of stable conductance levels needed for analog in-memory computing.
The devices also demonstrated spike-timing dependent plasticity-a biological learning process where neural connections strengthen or weaken based on timing. This property allows hardware to learn and adapt rather than simply store data.
Manufacturing Challenges Remain
The current fabrication process requires temperatures around 700°C, higher than standard semiconductor manufacturing allows. Researchers are working to lower this temperature to make the technology compatible with existing industry processes.
The research team spent several years developing the technology before achieving stable results in late 2025. A patent application has been filed by Cambridge Enterprise, the University's innovation arm.
If the temperature issue is resolved, the technology could be integrated into practical chip-scale systems that would significantly reduce energy demands for AI applications across industries.
Related reading: AI for Science & Research and Generative AI and LLM
Your membership also unlocks:
