Linearly Separable Pattern Classification Using Memristive Crossbar Circuits

Komal Singh,  Chitrakant Sahu,  Jawar Singh
Indian Institute of Information Technology Design and Manufacturing, Jabalpur, India


This paper presents a practical approach for linearly separable pattern classification using a single-layer perceptron network implemented with memristive crossbar circuit and CMOS neurons. Memristors (resistors with memory) promise the efficient implementation of synapse in artificial neural network, since, resistance of memristor depends on the electric current passed through it in past and present. Synaptic weights that correspond to memristors conductance are first calculated with the MATLAB software using perceptron learning rule and then imported to memristor crossbar circuit implemented for hardware implementation in SPICE. The simulation and analytical results motivate the efficient implementation of artificial neural networks and encourage implementations of more sophisticated multi-layer neuromorphic systems with memristive crossbar circuits.