A Novel Window Function Enables Memristor Model With High Efficiency Spiking Neural Network Applications

Memristor, a nanoscale device with the advantages of simple structure, excellent scalability, and complementary metal–oxide–semiconductor (CMOS) process compatibility, has drawn extensive research attention for various applications. An appropriate memristor model is essential for researcher to explore potential applications of memristor-based systems. Hewlett-Packard (HP) memristor model with the simple computation complexity is a favorable choice for simulation works. Incorporating a window function into the HP memristor model is essential for emulating the fealty nonlinearity response of practical device. However, a window function, with the ability of controllable resistance evolution speed and range for memristor model, has not been reported. In this work, we introduced two parameters ( ${G}$ and ${S}$ ) into window function for controlling the evolution speed and range of memristor model, respectively. The results indicate that the memristor model based on this window function exhibits high flexibility in resistance evolution speed and range. Furthermore, the memristor model based on this window function is used to implement the spiking neural network (SNN) simulation. The results demonstrated that memristor model based on this novel window function shows higher computation efficiency for SNN than previous window functions.