A multi-layer memristive recurrent neural network for solving static and dynamic image associative memory

Abstract Recent years have seen increased attention being given to recurrent neural networks in associative memory applications. The activation function is the core of the recurrent neural network for associative memory, but its hardware implementation is quite complicated. A novel multi-layer memristive recurrent neural network (MMRNN) is proposed, in which nano-scaled element memristor is employed as activation function. Its stability is proved by numerical analysis, which provides the theoretical premise for its application. There are three advantages: (i) Since the memristor is a passive device, the activation function circuit will be low power consumption; (ii) It can not only simplify the hardware circuit, but also facilitate the implementation of VLSI; (iii) MMRNN can realize the associative memory of both static and dynamic images. A series of theoretical analysis and numerical simulations demonstrated the effectiveness of the proposed MMRNN. The study helps to explore the dynamic image associative memory, and provides a new effective way to realize the associative memory of video.