In the biological visual retina, the ON/OFF pathways formed by photoreceptors and bipolar cells can exploit incident light stimuli to generate excitatory and inhibitory signals in parallel, enabling integrated perception and preprocessing. Consequently, realizing optoelectronic synapses with bidirectional plasticity and nonvolatile characteristics under purely optical control remains a significant challenge. In this Letter, we propose a voltage-divider circuit architecture based on indium gallium zinc oxide (IGZO) optoelectronic synapses, which, in combination with the persistent photoconductivity effect in oxide semiconductors, enables bidirectional and nonvolatile all-optical weight modulation. Essential biological synaptic functions are successfully emulated, including excitatory/inhibitory postsynaptic potentials, paired-pulse facilitation and depression, as well as long-term potentiation and depression. Leveraging this all-optically modulated IGZO synaptic circuit, an artificial neural network is constructed for image classification tasks, achieving recognition accuracies of 88% and 98% on the Fashion-MNIST and MNIST datasets, respectively. This work provides a robust strategy for developing energy-efficient and biologically realistic visual perception systems.
Qian et al. (Mon,) studied this question.