Self-Rectifying Memristors Based on Dimensionally Graded Halide Perovskites

Neuromorphic in-memory computing has emerged as one of the forerunners in addressing the data deluge problem in this age of smart electronics and artificial intelligence. Memristor crossbar arrays are fundamental storage and processing hardware frameworks that enable in-memory computing. Halide perovskites have been examined for memristors, owing to their mixed ionic-electronic conduction and solution processability. However, such studies so far have not addressed the challenge of sneak paths, which can result in erroneous computation. Self-rectifying memristors, which can be integrated into a passive crossbar array, are the most efficient solution to the sneak-path problem in terms of circuit complexity and device footprint. This work introduces a new approach to realizing a self-rectifying halide memristor by creating a 2D to 3D dimensionally graded perovskite. Through a careful selection of 2D spacer cations based on the energy level alignment with methylammonium lead iodide, a favorable heterojunction is created that achieves a rectification ratio > 10³. Moreover, the memristor displayed robust synaptic characterization (endurance > 4 × 10⁴ pulses) with high linearity in weight update. By suppressing the sneak currents, a far larger 140 × 140 crossbar array could be supported. Using this, 93% accuracy is achieved in an image classification task despite introducing write noise.