Ultra-high density perovskite nanowire array memristor-based multi-layer perceptron

Passive memristor crossbars present a high-density and low-power platform for neuromorphic computing. Although perovskite memristors present advantages such as re-configurability between electrochemical metallization and valence change mechanisms compared to traditional oxides, they face issues with uniformity, variability, and filamentary instability, which impede their large-scale integration. This work demonstrates that these limitations are overcome by a three-dimensional perovskite nanowire array architecture, where a precisely engineered ITO barrier is instrumental in maintaining controlled, analog conductance modulation. The resulting memristors exhibit 139 non-overlapping Ohmic conduction states within an optimal analog range (10–100 µS), long retention ( > 10⁶ s), endurance exceeding 4 × 10⁵ cycles, an asymmetric non-linearity factor of 0.3, and minimal variability (device-to-device <5%, cycle-to-cycle <1.5%). The memristors are further integrated into a 64 × 64 crossbar array and utilized in developing a fully integrated multi-layer perceptron for zebrafish head and jaw movement analysis, thereby establishing a scalable pathway for robust perovskite-based neuromorphic hardware. Poddar et al. report perovskite nanowire array-based memristors. A thin ITO layer is introduced as a semi-permeable barrier between electrode and perovskite for controllable analog switching. A physical multi-layer perceptron based on a 64×64 crossbar array enables the analysation of zebrafish strike patterns.