Amorphous Erbium Oxide (Er2O3) Memristors with Highly Linear Weight Updates for Artificial Synapses

Abstract An analog memristor with multilevel conductance states is essential for scalable neuromorphic computing, but achieving linear and symmetric SET/RESET switching characteristics for well-defined conductance states remains a major challenge. Here, we report an amorphous erbium oxide (Er2O3)-based memristor that addresses this challenge through stable low-voltage bipolar analog resistive switching, highly linear synaptic weight updates, and biologically realistic plasticity emulation. The Au/Er2O3/FTO memristor exhibits more gradual resistive switching with ultralow SET and RESET voltages of –0.49 and 0.54 V, respectively, while maintaining low cycle-to-cycle and device-to-device variability, robust endurance exceeding 104 switching cycles, and a fast switching speed of approximately 50 ns. Physics-based electrothermal simulations reveal that the analog switching behavior originates from spatially distributed oxygen vacancy migration and the gradual formation and rupture of conductive filaments within the amorphous oxide matrix, which collectively suppress stochastic variability. The memristor further demonstrates excellent synaptic plasticity with low nonlinearity factors of long-term potentiation (NLP) = 0.74 and long-term depression (NLD = 0.47), a high symmetricity of approximately 89%, large dynamic conductance windows of 9.2 and 6.5, and an effective conductance-state utilization of nearly 91.4%. In addition, system-level evaluation within the artificial neural network and convolutional neural network implementations yields inference accuracies of approximately 95 and 89% for the MNIST and CIFAR-10 datasets, respectively, confirming the suitability of Er2O3 memristors as synaptic elements. These results establish amorphous Er2O3 as a promising resistive switching material for next-generation analog memory and large-scale neuromorphic computing systems.