Correlation between linear conductance variability and accuracy in neuromorphic computing using AuNP-DNA/HfO2 bilayer memristor devices

Abstract

Memristor-based neuromorphic computing needs to improve energy efficiency using eco-friendly materials, but a single active layer with traditional materials still requires an improvement in reliability and performance. We propose a bilayers-memristor with gold nanoparticle (AuNP)-DNA for artificial synapse applications. As a result, the HfO₂/AuNP-DNA-based memristor exhibits excellent linear weight update and large conductance ratio characteristics with high reliability and tunability. The high linearity of synaptic weights, achieved through simple programming according to pulse amplitude, can particularly enhance the device’s energy efficiency and learning accuracy in neural network applications. We assess the accuracy of the modified national institute of standards and technology simulations to compare the classification accuracy depending on linearity and the ratio of weight update. As a result, the smallest validation loss is observed at –7 V depression voltage, which has the best linearity and optimal conductance ratio, suggesting the potential application of the proposed memristor in neuromorphic computing.