<p>Memristors crossbar arrays are an attractive option for use in neuromorphic computing systems and future memory architectures. It has caught the attention of researchers due to their small size, non-volatility, low power consumption, compactness, and high density, also naturally performs matrix vector multiplication. Memristor a non-linear passive electrical device is utilized to realize the synapse of neural networks and its resistance denotes the weight of synapse. However, stuck at faults, resistance variations, operational faults, conductance variation and fabrication defects present in the memristor crossbar arrays can lead to errors and dramatically reduce the computing accuracy of neuromorphic computing system and chip yield. In this paper, we explore different software and hardware based solution and various fault diagnose and fault tolerant techniques to find stuck at faults and large resistance variation in neural network so that to minimize the impact of faulty memristors. The work presents survey of several methods to handle performance degradation due to faults in memristor and loss recovery techniques. Fault tolerant methods are evaluated using different datasets like MNIST and CIFAR. It has been observed that fault tolerance technique is more efficient if it consumes low test time, reduced computational latency, power consumption, and high fault coverage. Additionally, the amount of fault tolerance depends on many features like depth of neural network, memristor precision, and layer-wise fault percentage among others.</p>