Neuromorphic computing seeks to match the brain's 20-watt efficiency

The human brain operates on roughly 20 watts, while conventional AI systems require far more energy due to the separation of memory and processing. Neuromorphic computing aims to close this gap by integrating memory and computation, using sparse data representations, and activating only when events occur. This event-driven approach could dramatically reduce power consumption in future computing systems.

The brain's efficiency stems from two design principles: energy is expended only where needed, and memory is encoded directly in synaptic connection strengths, eliminating the need to shuttle data between separate locations. Conventional chips suffer from the "Von Neumann bottleneck," a term coined by computer scientist John Backus in 1978, where transferring data between processor and memory consumes substantial energy.

Event-driven computing is already producing practical hardware. Event cameras, modeled on the retina, trigger individual pixels only when scene changes occur, unlike conventional cameras capturing full frames continuously. This enables low-power operation, blur-free fast motion capture, and functionality in extreme lighting—valuable for autonomous vehicles and space debris tracking. Australia's BrainChip sells a commercial neuromorphic processor for low-power camera and sensor applications, suggesting these chips will complement rather than replace conventional GPUs.