Drawing inspiration from the human visual system’s exceptional capabilities in information processing and memory retention, optoelectronic neuromorphic devices have been considered a cutting-edge solution to mimic these key functions. These devices, particularly optoelectronic memristors, promise to revolutionize neuromorphic computing and visual biomimetic functions, holding significant potential to surpass the traditional von Neumann architecture. Herein, an optoelectronic memristor engineered from a MoS₂/WO₃ heterojunction is developed and integrated with optoelectronic synapses and optical perception capabilities. The device exhibits short/long-term synaptic plasticity under electrical and optical stimuli, effectively mimicking short/long-term memory and “learning–forgetting–relearning”. Leveraging its optical synaptic characteristics, the device successfully simulates complex synaptic behaviors, including Pavlovian conditioning, enabling visual associative learning similar to the biological brain. Through coordinated optoelectronic modulation of long-term potentiation/depression for synaptic weight, the system achieves 98.4% classification accuracy on the Modified National Institute of Standards and Technology (MNIST) handwritten digit recognition task. Moreover, a 4 × 4 optoelectronic memristor array demonstrates stable visual perception and memory functions under four distinct optical stimuli, facilitating adjustable image memory properties across different light wavelengths. This research advances the application of optoelectronic memristors in neuromorphic computing and bionic visual systems.

In current memristor-based neuromorphic computing research, several studies face the challenge of realizing only a single function at a time or having isolated functions. This limitation is particularly evident when simulating biological cognition, as the overall synergy between multiple cognitive functions is difficult to represent. In this work, a high-performance heterojunction memristor is presented at first. The memristor-based neural network and functional circuit are further implemented to realize and integrate multiple cognitive functions. Specifically, the proposed photoelectric memristor has the structure of Ag/ZnO-SnO₂/WO_3-x/ITO, it exhibits various synaptic behaviors under external modulations, which are characterized by good stability and repeatability. Based on this device, a neural network is built to realize the basic recognition function in biological cognition. The recognition results are translated into different labelled voltage signals and subsequently fed into a memristor-based functional circuit. By leveraging memory characteristics and tunable conductance of the memristor, and controlling the specific circuit functionalities, the input signals are processed to produce different outputs representing various cognitive functions. This methodology allows the realization and integration of recognition, memory, learning, association, relearning, and forgetting into one single system, thereby enabling a more comprehensive and authentic simulation of biological cognition. This work presents a novel memristor and a method for achieving and integrating multiple neuromorphic computing functions within a single system, providing a successful example for achieving complete biological function.

With the merits of non-contact, highly efficient, and parallel computing, optoelectronic synaptic devices combining sensing and memory in a single unit are promising for constructing neuromorphic computing and artificial visual chip. Based on this, a N:ZnO/ MoS₂-heterostructured flexible optoelectronic synaptic device is developed in this work, and its capability in mimicking the synaptic behaviors is systemically investigated under the electrical and light signals. Versatile synaptic functions, including synaptic plasticity, long-term/short-term memory, and learning-forgetting-relearning property, have been achieved in this synaptic device. Further, an artificial visual memory system integrating sense and memory is emulated with the device array, and the visual memory behavior can be regulated by varying the light parameters. Moreover, the optoelectronic co-modulation behavior is verified by applying mixed electric and light signals to the array. In detail, a transient recovery property is discovered when the electric signals are applied in synergy during the decay of the light response, of which property facilitates the development of robust artificial visual systems. Furthermore, by superimposing electrical signals during the light response process, a differentiated response of the array is achieved, which can be used as a proof of concept for the color perception of the artificial visual system.