Local Brand Awareness
A conversation with Dr. Babak Bakhit, postdoctoral researcher at Cambridge and DTU, on how neuromorphic chips could cut AI energy consumption by up to 70%.
Overview
Current AI hardware separates memory and processing, forcing constant data movement between the two and driving up energy use — the Von Neumann bottleneck. Traditional transistors are also nearing their physical scaling limits.
Key Topics
Dr. Bahit's research centres on a co-doped hafnium oxide memristor that switches at the interface rather than through a conductive filament, giving lower power draw, smoother analog switching and no electroforming step, using an industry-compatible two-step sputtering process.
Outlook
Bringing neuromorphic chips from lab prototypes to commercial hardware depends on closer collaboration across materials science, neuroscience, electrical engineering and industry, alongside applications reaching beyond AI into sensing, solar energy management and neurological diagnostics.
Potential AI energy reduction
