Researchers at the University of Southern California (USC) have developed artificial neurons that physically replicate the complex electrochemical behaviour of biological brain cells. The innovation, detailed in Nature Electronics, could lead to chips that reduce energy consumption and size by “orders of magnitude” and advance artificial general intelligence (AGI).
The new approach, led by Professor Joshua Yang, uses a device called a “diffusive memristor” which relies on the movement of atoms rather than electrons. Unlike existing neuromorphic chips that simulate neural activity through mathematical equations, these artificial neurons emulate the analogue dynamics of biological counterparts. They use chemicals to initiate computation, mirroring how neurochemicals function in the human brain.
The human brain uses ions like potassium or sodium to generate electrical signals. The USC team’s device, developed at the Centre of Excellence on Neuromorphic Computing at USC, uses silver ions in oxide to generate an electrical pulse and successfully emulate the biological process. Yang noted that while silver ions are not the same as those in the brain, “the physics governing the ion motion and the dynamics are very similar”.
Mimicking the brain’s wetware
The team chose to mimic the brain’s ion-based “wetware” because it is the “most efficient intelligent engine” developed through evolution. Yang explained that the primary problem with modern computing is not a lack of power, but a lack of efficiency, particularly for energy-intensive AI models. While electrons are faster, ions are a better medium for “hardware-based learning,” which is how the brain operates, rather than the “software-based learning” of conventional chips.
This hardware-based learning allows the human brain to consume only about 20 watts of power, compared to the megawatts required by supercomputers. The new artificial neuron achieves its efficiency in both energy and size, requiring the footprint of only one transistor, whereas conventional designs use thousands.
Yang acknowledged that the silver used in the experiment is not readily compatible with current semiconductor manufacturing, and alternative ionic materials will need to be investigated. The next step is to integrate large numbers of these artificial synapses and neurons to test their efficiency and capabilities at scale.
“Even more exciting is the prospect that such brain-faithful systems could help us uncover new insights into how the brain itself works,” says Yang,
