Memristors are two-terminal reminiscence units that may change the conductance state and retailer analog values. Because of their easy construction, suitability for high-density integration, and non-volatile traits, memristors have been intensively studied as synapses in synthetic neural community techniques. Memristive synapses in neural networks have theoretically higher power effectivity in contrast with typical von Neumann computing processors. Nevertheless, memristor crossbar array-based neural networks often endure from low accuracy due to the non-ideal components of memristors corresponding to non-linearity and asymmetry, which forestall weights from being programmed to their focused values. On this article, the development in linearity and symmetry of pulse replace of a totally CMOS-compatible HfO2-based memristor is mentioned, through the use of a second-order memristor impact with a heating pulse and a voltage divider composed of a collection resistor and two diodes. We additionally show that the improved gadget traits allow energy-efficient and quick coaching of a memristor crossbar array-based neural community with excessive accuracy via a practical model-based simulation. By enhancing the memristor gadget’s linearity and symmetry, our outcomes open up the potential for a trainable memristor crossbar array-based neural community system that possesses nice power effectivity, excessive space effectivity, and excessive accuracy on the similar time.