A wiring mistake revealed a simpler artificial neuron
Researchers found that an ordinary MOSFET can mimic neuron and synapse behaviour, opening a possible route to smaller, lower-energy AI hardware.
A student in a Singapore laboratory forgot to connect one terminal of a transistor. The mistake exposed an electrical behaviour researchers had spent years trying to build with far more complicated devices.
The component was an ordinary MOSFET, the basic transistor used throughout modern electronics. With its bulk terminal left floating, charge built up inside until the device produced a sudden current spike and then relaxed. The pattern resembled the way a biological neuron gathers signals, fires and returns to rest.
The overlooked fourth terminal
Once the team understood the effect, it used resistance at the bulk terminal to control when the transistor fired. The same kind of device could also hold an adjustable level of conductance, allowing it to act like the synapse connecting two neurons. The researchers call the combined approach neurosynaptic random-access memory, or NSRAM.
The devices completed more than 10 million test cycles without a reported failure, and the behaviour was reproduced on chips from a second foundry. Because the approach relies on standard silicon manufacturing rather than an exotic new material, it could be easier to produce at scale.
A device, not yet a computer
Turning the discovery into useful hardware will take much more work. A complete neuromorphic chip still needs control circuits, system-level simulation and several rounds of fabrication. Low-power sensors and other edge devices may be the first realistic use. Matching today's GPUs would require the design to scale far beyond the experiments reported so far.
Source: IEEE Spectrum, June 29, 2026.