| Summary: | Abstract Integrating multiple functions within a neuromorphic device is essential for simplifying circuit design in compact artificial vision applications. At the same time, there is a constant push to reduce the size of devices to improve integration. Nevertheless, decreasing the thickness of the active layer compromises photoelectric performance, affecting stability, uniformity, endurance, and photosensitivity. An optoelectronic memristor featuring an ultrathin AlOx/TiOy periodic heterostructure is proposed. This design minimizes the active layer thickness without compromising optoelectronic properties and enables multifunctionality as a photodetector, electric synapse, and optical synapse in a single device. The periodic heterostructure is successfully prepared by atomic layer deposition with a thickness of only ≈12 nm. The device enables electric synaptic behaviors, which are essential for neuromorphic computing. Notably, the dual‐functional photodetector and optical synapse facilitate the efficient acquisition and processing of visual information following specific application scenarios. It enables visual attention simulation for energy‐efficient object detection. Finally, a complete visual system is demonstrated, encompassing sensing, front‐end preprocessing, and back‐end computing. Based on the proposed system, a six‐layer convolutional neural network is built to recognize EMNIST patterns, and front‐end preprocessing improves recognition accuracy from 64% to 78%.
|
|---|
