Enhancing Dermatological Image Segmentation via In‐Sensor Computing With a Neuromorphic Ga‐Doped ZnO Nanorod‐Based Optoelectronic Memristors

ABSTRACT

Conventional computing architectures with separate sensing and processing units are plagued by latency and energy inefficiencies. By fusing local data and computation, in‐sensor computing enables real‐time localized data processing and decision‐making, approaching the energy efficiency of biological neural systems. Here, a neuromorphic optoelectronic memristor based on gallium doped zinc oxide (GZO) nanorod arrays (NRAs) is fabricated. Benefiting from the gallium doping and the ordered NRAs structure, the conductive filaments of the resistive switching layer can be modulated by the oxygen vacancy concentration and Schottky barrier height. The GZO NRAs memristor exhibits low switching voltages, high cycle‐to‐cycle consistency, and excellent retention time. Furthermore, the GZO NRAs memristor demonstrates dual‐mode electrical and optical modulation with key synaptic plasticity behaviors, including long‐term potentiation/depression, paired‐pulse facilitation, and spike‐timing‐dependent plasticity. The GZO device is integrated with a U‐Net neural network to enable adaptive lesion segmentation. This sensor‐computing fusion strategy markedly improves the diagnostic accuracy for dermatological images from 88.89% to 99.24%. These findings indicate a device‐to‐system pathway for applying GZO NRAs memristors to intelligent healthcare and neuromorphic computing.