Oxygen vacancy-mediated linear, long-term retention weight tuning in amorphous ITZO optoelectronic synapses for high-performance neuromorphic visual systems

Optoelectronic synaptic devices featuring linear weight updates are essential for high-precision visual information perception, storage, and processing, meeting high-efficiency and low-power demands in artificial visual systems. This study leverages the oxygen vacancy states in amorphous indium-tin-zinc oxide (ITZO) to design Al/ITZO/Al two-terminal optoelectronic synaptic devices with highly linear weight updates and long-term retention of light-encoded information. A large number of oxygen vacancy defect levels with a broad energy distribution in ITZO thin films provide efficient pathways for carriers to transition under illuminations with sub-bandgap energy and allow carriers to be stored in these defect levels for an extended period, providing conditions for linearly enhanced weight updates and long-term current retention. Devices with different oxygen vacancy concentrations exhibit stable synaptic characteristics, including paired-pulse facilitation (PPF) index exceeding 190%, highly linear pulsed light-modulated plasticity (adjusted R2 values of linear fitting ≥ 0.98), and learning-forgetting-relearning processes (standard deviation of ΔEPSC < 0.02%). In addition, the different “learning attitudes” (simulated by tuning oxygen vacancy concentrations to mimic neurotransmitter concentrations) are visualized via light-encoded images. Benefiting from the multi-level linearity and uniformly distributed photoresponse current, the devices are applicable for hexadecimal light-encoded information processing, image convolution processing, and image reconstruction with high fidelity. Notably, the ITZO(9:1:3) device simulates the fully opened eyelids and thus retains the most image information. When used for CIFAR-10 image classification in a convolutional neural network (CNN), it achieves an accuracy of 87.58%, verifying the potential of ITZO optoelectronic synapses in artificial vision systems.