Mapping neural network computations onto memristor crossbar

The problems of programming memristor arrays (memristor crossbar) are considered. An estimate for the pulse width to set the desired memristor resistance (memristance) value is obtained. An algorithm for mapping the neurons layer weighting matrix onto the memristor crossbar with the given memristor conductivities boundary values is proposed. WTA (“Winner Takes All”) neural network implementation on the memristor crossbar and NMOS transistors is proposed for binary images recognition. The proposed WTA-network implementation was approved by simulation on the LTspice IV software.