National University of Singapore Innovates AI with New Transistor Design

The National University of Singapore has announced a breakthrough in neuromorphic computing with the invention of a new computing cell that can mimic the behavior of both electronic neurons and synapses. This innovation, led by Associate Professor Mario Lanza, utilizes a single conventional silicon transistor to replicate these functions, significantly reducing the size and cost of artificial neural networks.

Traditionally, implementing electronic neurons and synapses required multiple transistors, making them large and expensive. However, Professor Lanza's team has discovered a method to reproduce these electronic behaviors using a single transistor by adjusting the resistance of the bulk terminal. This approach leverages a physical phenomenon known as "impact ionisation," which was previously considered a failure mechanism in silicon transistors.

The new design allows for a reduction in the size of electronic neurons by a factor of 18 and synapses by a factor of 6. This advancement could lead to more efficient computing systems capable of processing more information while consuming less energy. The team has also developed a Neuro-Synaptic Random Access Memory (NSRAM) cell, which can switch between neuron and synapse modes, offering versatility in manufacturing without the need for advanced transistor fabrication processes.