Accidental Discovery of Single-Transistor Neurons Could Slash AI Energy Use

Researchers found that a single flawed CMOS transistor can act as an artificial neuron or synapse, potentially replacing the thousands of transistors and bulky capacitors currently needed. This neuromorphic approach could make AI hardware one million times more energy-efficient than GPUs, which consume up to 1,000 watts each. The discovery, hiding in plain sight, promises to drastically reduce the environmental footprint of data centers running AI workloads.