A light-controlled artificial synapse, which mimics the human brain has been considered to be one of the ideal candidates for the fundamental physical architecture of a neuromorphic computing system owing to the possible abilities of high bandwidth and low power calculation. However, the low photosensitivity of synapse devices can affect the accuracy of recognition and classification in neuromorphic computing tasks. In this work, a planar light-controlled artificial synapse having high photosensitivity (I_on/I_off > 1000) with a high photocurrent and a low dark current is realized based on a ZnO thin film grown by radiofrequency sputtering. The synaptic functions of the human brain such as sensory memory, short-term memory, long-term memory, duration-time-dependent-plasticity, light-intensity-dependent-plasticity, learning-experience behavior, neural facilitation, and spike-timing-dependent plasticity are successfully emulated using persistent photoconductivity characteristic of a ZnO thin film. Furthermore, the high classification accuracy of 90%, 92%, and 86% after 40 epochs for file type datasets, small digits, and large digit is realized with a three-layer neural network based on backpropagation where the numerical weights in the network layer are mapped directly to the conductance states of the experimental synapse devices. Finally, characterization and analysis reveal that oxygen vacancy defects and chemisorbed oxygen on the surface of the ZnO film are the main factors that determine the performance of the device.

中文翻译：

---

用于神经形态计算的高光敏光控平面ZnO人工突触

由于可能的高带宽和低功耗计算能力，模仿人脑的光控人工突触被认为是神经形态计算系统基本物理体系结构的理想候选者之一。但是，突触设备的低光敏性会影响神经形态计算任务中识别和分类的准确性。在这项工作中，具有高光敏性的平面光控人工突触（I _on / I _off基于射频溅射生长的ZnO薄膜可实现高光电流和低暗电流（> 1000）。人脑的突触功能，例如感觉记忆，短期记忆，长期记忆，持续时​​间依赖于可塑性，光强度依赖于可塑性，学习经历行为，神经促进和尖峰定时ZnO薄膜的持久光电导特性成功模拟了依赖于可塑性的塑性。此外，通过基于反向传播的三层神经网络（其中网络层的数值权重），在三层神经网络的情况下，文件类型数据集，小数位和大数位在40个历元之后达到了90％，92％和86％的高分类精度。直接映射到实验突触设备的电导状态。最后，