A Leaky-Integrate-and-Fire Neuron Analog Realized with a Mott Insulator

Significance Statement

Neuromorphic engineering has been implemented to simulate neuro-biological architectures like that of a nervous system with the use of an artificial neural network which consists of connected neurons by means of synapse to convey signal to other neuron.

Available neuromorphic computers at present need to be enhanced powerfully in order to completely simulate reflection of network information like that of a human brain. However, the resultant effects of these potential neuromorphic computers which would definitely exhibit increased convolution and high power loss needs to be considered as well. At present, neuromorphic computing system that makes use of memristors, transistors and other switching devices have shown certain positive potentials.

For the first time, researchers led by Dr. Laurent Cario put forward a single-component device based on a Mott insulator compound which performs the simple function of spiking neurons by a phenomenon known as ‘leaky integrate and fire’. The research work is now published in Advanced Functional Materials.

Experimental setup included a simple circuit of series connection between generator, narrow-gap Mott insulator of GaTa4Se8 crystal and a load resistor. Experiments were performed at 74 K, and a means for initiating the resistive switching was provided.

The authors observed that during a delay time of 89µs, a swift decrease in voltage resulted to a rise in current intensity known as the firing event. A volatile resistive transition was also noticed with GaTa4Se8 crystal sample when the transmission pulse was ended, and results indicated an abrupt increase resistance. The long-time relaxation period when fitted with an exponential equation form gave a relaxation time related to the energy barrier of the metastable metallic state.

Preceding these results, the team observed that at the transmission of short pulses at duration and separation period at 20µs and 30µs, lower than that of the firing event (89µs) and the relaxation time (518µs) led to a resistive transition in the set crystal sample after six voltage pulses were applied. Further results also revealed the leaky integrate and fire functionality with shorter pulses observed during the duration and separation time.

They also provided a modeling of resistive switching from several short pulses. Similar features where observed at the duration and separation pulses in relation to the number of applied voltage pulses. Results from snapshots of the microscopic state of the resistor network at four different simulated pulses indicated quick formation of a conductive bridge which occurred during a single pulse.

Further simulation results showed that Mott devices system has comparable features to that of an artificial neuron with the leaky integrate and fire functionality. The Mott device system developed by the way of the authors simulates basic functions of spiking neurons with the phenomenon of the leaky integrate and fire.

Reference

Stoliar, P., Tranchant, J., Corraze, B., Janod, E., Besland, M.P., Tesler, F., Rozenberg, M., Cario, L. A Leaky-Integrate-and-Fire Neuron Analog Realized with a Mott Insulator, Advanced Functional Materials 27 (2017) 1604740.

Check Also

3D-printed short carbon fibre reinforced perforated structures with negative Poisson's ratios: Mechanisms and design - Advances in Engineering

3D-printed short carbon fibre reinforced perforated structures with negative Poisson’s ratios