Behavioral Modeling and STDP Learning Characteristics of a Memristive Synapse

In this work, we study the feasibility of using memristor as a synapse in spiking neural networks. A simple voltage controlled model is selected and the effect of several parameters on the I-V characteristics and memristance are studied. Next, the ability of the device to perform synaptic behaviors has been investigated. The Spike-Timing-Dependent Plasticity (STDP) learning algorithm is applied to a circuit model which contains LIF pre and post-synaptic neurons and the memristor as a synapse. Components are modeled in verilog-A, and the circuit is simulated in HSPICE. Continuous and analog weight change of synapse is demonstrated. The two most important characteristics of the synapse, i.e the maximum voltage over the memristor and the STDP characteristics, have been extracted from simulations. These results confirm that the proposed memristor model is an attractive candidate for complex spiking neural networks.