Spiking Neural Networks (SNNs) are offering significant advantages in energy efficiency and temporal information processing. However, when deployed on hardware platforms, SNNs are inevitably exposed to hardware level faults. Understanding effects of these faults on network performance, particularly during the training phase, is critical for developing reliable neuromorphic systems. This thesis investigates the impact of various fault types on the training behavior and accuracy of SNNs. A fault injector was developed, enabling the injection of both transient and permanent faults into different components of the network, including weights, gradients, inputs, activations, membrane potentials, membrane decay constants, and firing thresholds during the training phase of SNNs.