Accidental Lab Discovery Could Revolutionize Computing Efficiency

Researchers have accidentally discovered a way to make ordinary transistors mimic the behavior of biological neurons and synapses, potentially revolutionizing computing efficiency, particularly for artificial intelligence (AI). Current AI systems rely on Graphics Processing Units (GPUs) that consume vast amounts of energy, with individual GPUs using up to 1,000 watts, comparable to household appliances but operating continuously. This inefficiency stems from simulating neural networks using software and billions of transistors, which requires significant energy for data movement. The human brain, in contrast, is millions of times more energy-efficient at comparable tasks. Neuromorphic engineering aims to create electronic components that function like biological neurons and synapses, but previous attempts have been either unreliable experimental devices or required complex circuits of many transistors, limiting scalability. The breakthrough involves utilizing a standard Complementary Metal-Oxide-Semiconductor (CMOS) transistor, specifically by adding resistance to its bulk terminal. This modification allows the transistor to exhibit a sudden, nonlinear current spike when a voltage threshold is crossed, mimicking a neuron's action potential, and then relax to a resting state. This single device can also be adjusted to act as a synapse with stable, linearly adjustable conductance. This discovery promises to drastically reduce the energy footprint of AI by enabling the creation of more efficient, brain-inspired hardware.