Sci. Adv. 2, e1501326 (2016)

The highly interconnected network of neurons in the human brain consumes only tens of watts of power to perform highly complex algorithms. Since a reasonable estimate for the number of synapses transmitting signals between neurons approaches 1014, this remarkable efficiency is made possible by the tiny amount of energy required by each synaptic event, on the order of a few femtojoules. Artificial synapses must have comparable performance to their biological counterparts if neuromorphic hardware architectures are to be used in future novel computational approaches. However, the efficiency limits set by nature are as yet unequalled.

Now, Tae-Woo Lee and co-workers at the Pohang University of Science and Technology in Korea report on artificial synapses that outperform the energy efficiency of their biological counterparts. The researchers grew devices based on core–shell organic nanowires to mimic a nerve fibre. The synapse's functionality was reproduced in a transistor with an ion gel to provide signals analogous to both inhibitory and excitatory presynaptic pulses. The resulting differential dynamics of anions and cations inside the gel and their interaction with the nanowires allowed neuronal signal propagation, mimicking the different specific synaptic functions. Through device optimization, the researchers achieved an average energy consumption of around 1 femtojoule per synaptic event.