Tunable and Silicon-Compatible Memristive Devices Based on a Perovskite Nickelate System for Electrocardiogram Pattern Recognition

Perovskite oxides (ABO3) have received increasing attention due to their exotic and tunable physical properties, which make them ideal candidates for constructing high-performance memristors. However, the high processing temperatures and non-CMOS-compatible substrates typically employed in the growth of this material family pose significant challenges to the commercial deployment of these devices. Herein, a typical perovskite oxide, SmNiO3 (SNO), has been employed to construct a silicon-compatible memristor. The possible influence of defect content and the interfacial effect on the memristive performance has been systematically investigated. The optimized device demonstrates remarkable nonvolatile memory characteristics, including 2 orders of magnitude enhanced on/off ratio and programmable conductivity featuring 20 distinguishable resistance levels. Notably, the memristor exhibits bidirectional analog conductance modulation capabilities that enable faithful emulation of essential synaptic functionalities, including paired-pulse facilitation (PPF) and paired-pulse depression (PPD). Moreover, based on their stable pulse modulation plasticity performance, a convolutional neural network is constructed to identify the electrocardiogram (ECG) data sets to realize the diagnosis of diseases with a recognition accuracy of 83.5%. These findings provide valuable insights into the construction of silicon-compatible ABO3-based memristors and demonstrate their viability for neuromorphic computing applications.