Memristors are two-terminal reminiscence gadgets that may change conductance state and retailer analog values. Due to 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 have often suffered 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 completely CMOS-compatible HfO2-based memristor is mentioned, by utilizing a second-order memristor impact with a heating pulse and a voltage divider composed of a sequence resistor and two diodes. We additionally show that the improved system traits allow energy-efficient and quick coaching of a memristor crossbar array-based neural community with excessive accuracy via a sensible model-based simulation. By enhancing the memristor system’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 identical time.
