A Memristor-Based Neural Network Design for Associative Learning

Various types of memristor-based network circuits have been reported to imitate multifunctional associative learning, while at the price of complex control circuits. In this work, we construct a compact memristive network for associative learning, in which memristors serve as interconnected synapses to extract relevance and timing information. A spice-compatible dynamic model of memristor is established and all the learning functions of Pavlov's dog are successfully demonstrated through circuit simulation. By incorporating the actual characteristics of the memristor into the network, multiple functions such as associative learning, acquisition, backward conditioning, and extinction are completely implemented. Moreover, the spike timing coding scheme and learning ability versus signal interval (Delta t) are in line with biological behavior, making this design a promising solution for future brain-inspired associative learning systems.