Resistive Switching Oxides: Mechanism, Performance, and Device-Algorithm Co-Design for Artificial Intelligence

The human brain is a natural computing platform composed of a vast number of neurons and synapses that excels in learning, memory, and parallel data processing with low energy consumption and high efficiency. Brain-inspired hardware is crucial in the application of neural networks, offering enhanced energy efficiency, parallel processing capabilities, and nonlinear dynamic properties, such as adaptability and plasticity. Complex oxides exhibit rich electrically metastable states, which enable diverse resistive switching dynamics in response to electrical stimuli. This allows them to achieve complex bioinspired behaviors at the single-device level and more advanced brain-like functions through multidevice integration. This review summarizes recent advances in resistive switching of complex oxides with a focus on the underlying physical mechanisms and application-driven device-algorithm co-design. It first elucidates the materials science and multiscale mechanisms of the switchable electrical characteristics of complex oxides. Next, the resistive switching behaviors of complex oxide devices and a survey of their state-of-the-art performance are discussed. Additionally, from the perspective of brain-inspired computing, device-level biomimetic applications and task-driven circuit-algorithm co-designs are explored. Finally, the current challenges of complex oxide devices are summarized, and an outlook for the development of complex oxide neuromorphic devices is provided.