Analysis of Memristor Neural Networks for fault Tolerant Computing

A memristor, is a two-terminal non-volatile memory device that exhibits a unique relationship between electrical charge and magnetic flux linkage. The unique advantages of memristors, including non-volatility, low power consumption, compatibility with CMOS technology, and high scalability, have positioned them as promising components in the field of Neuromorphic Computing Systems, To enable the quality and reliability of Neuromorphic Computing Systems, efficient test solutions are necessary fault models are caused by open, short, bridge defects,Stuck-At-Faults (SAFs), which significantly reduce the computational accuracy of crossbar based Computing Systems.This analysis explores diverse applications and innovations in the field of neuromorphic systems, memristor-based circuits, and spiking neural networks. In order to minimize recognition rate loss and optimize crossbar programming in neural networks, the integration of partial-gated training with asymmetrical training is investigated. Additionally, a novel leaky integrate and fire neuron design is proposed for a Artificial Neural Network (ANN), achieving high frequencies on FPGA implementations and Synthesis using Cadence Genus software tool. The use of memristor-based inhibitory synapses is suggested to enhance lateral inhibition and homeostasis in a spiking neural network, leading to improved learning efficiency, methodologies, nonneuromorphic comptutation using memristors, and multifault detection approaches.