Synaptic and neural behaviours in a standard silicon transistor - Nature

Hardware-based artificial neural networks (ANNs) have shown the potential to outperform traditional computers in energy efficiency. This is attributed to their ability to compute and store data at the same location, thus minimizing energy consumption and latency associated with data transfer. Essential components of ANNs include electronic neurons that generate nonlinear output signals and synapses that modify their resistance to strengthen or weaken connections, a process integral to learning. Dynamic behaviours in these networks contribute significantly to lower energy usage, particularly when electrical pulses are utilized instead of constant biases.

Hardware-based artificial neural networks (ANNs) outperform traditional computers in energy efficiency by computing and storing data locally, minimizing energy consumption and delays.

Electronic neurons in ANNs generate output signals that resemble nonlinear mathematical operations when excited by voltage or current inputs, facilitating computational tasks.

Synapses in ANNs change their electrical resistance to alter connections between neurons, a key process for learning features, characterized by long or short-term plasticity.

Dynamic neural and synaptic behaviours enhance energy efficiency and synchronize complex systems, enabling reduced energy consumption through the application of electrical pulses.

Read at Nature