Artificial synapse topologies using arbitrary-order memristors

With the recent advances in the study of the memristive elements along with the theory of arbitrary-order calculus, a wide amount of opportunities has emerged in manufacturing and modeling of devices, as well as in the development of applications of arbitrary-order memristive circuits. This paper addresses the derivation of novel electronic bridge circuits for emulating the behavior of artificial synapses, which are the cornerstone in the design of on-chip neural networks. Bridge topologies are based on arbitrary-order memristors and can perform positive, negative and zero synaptic weights. According the characteristics and type of configuration, incremental or decremental, of each charge-or flux-controlled arbitrary-order memristor, sixty-four topologies are generated. The dynamical behavior of each memristive bridge is also derived. Features of the arbitrary-order memristive bridge synapses are investigated via simulations. The results show a significant improvement in terms of symmetry, control and update of the arbitrary-order memristive synaptic weights.

Automated summary

This paper proposes the derivation of novel electronic bridge circuits for emulating artificial synapse behavior using arbitrary-order memristors. These bridge circuits serve as the cornerstone for on-chip neural network design. Sixty-four topologies are generated based on the characteristics and configuration type (incremental or decremental) of each charge or flux-controlled arbitrary-order memristor. Simulations are conducted to investigate the features of the arbitrary-order memristive bridge synapses, which show significant improvement in terms of symmetry, control, and update of synaptic weights.