Linearly separable pattern classification using memristive crossbar circuits

This paper presents a practical approach for the classification of linearly separable patterns using a single-layer perceptron network implemented with a memristive crossbar circuit (synaptic network) and a CMOS Op-Amps based neuron. Memristors (resistors with memory) promise the efficient implementation of synapses in artificial neural networks, as they bears astonishing resemblance to the biological synapses in its functionality, performance and integration capability. The proposed design of memristive perceptron is implemented in HSPICE and trained using the Matlab software by applying the perceptron learning rule. In order to analyze the performance of the proposed memristive crossbar circuit based perceptron design, a comparison is made with the existing MOS technology based synaptic network design. The simulation results thus obtained motivate the efficient implementation of sophisticated multi-layer neuromorphic systems with memristive crossbar circuits in the near future.