Synaptic and neural behaviours in a standard silicon ...
Researchers have developed a method to mimic spiking neural behavior using just two standard silicon transistors, a breakthrough in neuromorphic computing. This approach leverages a phenomenon known as "punch-through conditions," where charges build up in a semiconductor, allowing bursts of current to pass through even when the transistor is in the off state. This behavior closely resembles the spiking activity of biological neurons.
The team, a collaboration between researchers in Saudi Arabia and Singapore, configured the transistors to operate on the verge of punch-through mode. By adjusting the gate voltage, they could control the charge build-up, enabling the transistors to mimic neuronal spiking. The spiking frequency could vary by up to a factor of 1,000, and the system remained stable for over 10 million clock cycles.
This innovation has significant advantages:
However, there are challenges. The system requires additional hardware to control and reset the transistor states frequently. Additionally, spiking neural networks may not always match the performance of non-spiking networks in certain applications, and converting inputs into spiking signals can be complex.
Despite these hurdles, this research represents a promising step toward more energy-efficient AI hardware, which is critical as the energy demands of AI continue to grow. The study was published in Nature in 2025 (DOI: 10.1038/s41586-025-08742-4).