Memristors offer advantages as a hardware solution for neuromorphic computing, however, their non-ideal property makes the weight update difficult and reduces the accuracy of a neural network. Also, a large amount of personal data has raised great concern about the privacy preservation of neural networks. Thus, the performance of memristor-based neural networks gets worse when considering non-ideal property and introducing a privacy preservation mechanism. This dissertation focuses on improving the performance of a memristor-based privacy-preserving neural network. A piecewise linear (PL) method is proposed to mitigate the nonlinear effect of memristors by calculating the weight update parameters along a piecewise line, which reduces errors in the weight update process. It mitigates the nonlinearity impact without reading the precise conductance of the memristor in each updating step, thereby avoiding complex peripheral circuits. What’s more., the PL method is proved to be an effective technique that can prevent accuracy loss and increase privacy preservation space for privacy-preserving ANN. Also, we propose a Noise Distribution Normalization (NDN) method to add Gaussian distributed noise through hardware implementation, thereby achieving differential privacy in edge AI. Instead of using traditional algorithmic noise-insertion methods, we take advantage of inherent cycle-to-cycle variations of memristors during the weight-update process as the noise source, which does not incur extra software or hardware overhead.