Vertically Partitioned Interfacial Design in Self-Powered Perovskite Photomemristors Enables CMOS-Integrated Motion Vision System

Abstract Photomemristive artificial synapses, which integrate optical sensing, memorizing, and preprocessing, offer a new route for neuromorphic vision systems that avoid high computational power demands. However, for most perovskite photomemristors devices with a simple metal-semiconductor junctions, the interfacial origin of persistent optical memory remains poorly understood, which has limited the optimal design of device architecture. In this work, we report a vertically partitioned interfacial architecture for self-powered metal-halide perovskite photomemristors. For the buried interface, P3HT directs perovskite crystallization and establishes a hole selective photovoltaic pathway, whereas the polymer overlayer conventionally used for encapsulation or passivation is redefined as an active ion-electron relaxation interphase. To modify the top interface of the perovskite layer, a series of insulating polymers with different terminal groups were screened. Among them, PBMA was identified as the optimal top interphase, which is attributed to the fact that Pb coordinating ester carbonyls stabilize iodide redistribution, while the butyl side chain environment prolongs electron side recovery. This vertically partitioned architecture stabilizes photoinduced conductance states without external bias, yielding a 119% paired pulse facilitation index and the most persistent cumulative conductance modulation among the polymer series. Furthermore, when integrated with a 640 ⋅ 512 CMOS readout platform with 15 µm pixels, the optimized stack enables hardware level motion direction recognition and trajectory recording, increasing the downstream clustering score from 0.6011 to 0.9090. This work establishes a vertically partitioned interfacial architecture as a structural basis for self-powered perovskite photomemristors and interface-defined neuromorphic vision hardware.