Ferroelectric tunnel junctions (FTJs) as the artificial synaptic devices have been considered promising for constructing brain-inspired neuromorphic computing systems. However, the memristive synapses based on the flexible FTJs have been rarely studied. Here, we report a flexible FTJ memristor grown on a mica substrate, which consists of an ultrathin ferroelectric barrier of BiFeO₃, a semiconducting layer of ZnO, and an electrode of SrRuO₃. The obtained flexible FTJ memristor exhibits stable voltage-tuned multi-states, and the resistive switchings are robust after 10³ bending cycles. The capability of the FTJ as a flexible synaptic device is demonstrated by the functionality of the spike-timing-dependent plasticity with bending, and the accurate conductance manipulation with small nonlinearity (−0.24) and low cycle-to-cycle variation (1.77%) is also realized. Especially, artificial neural network simulations based on experimental device behaviors reveal that the high recognition accuracies up to 92.8% and 86.2% are obtained for handwritten digits and images, respectively, which are close to the performances for ideal memristors. This work highlights the potential applications of FTJ as flexible electronics for data storage and processing.

铁电隧道结（FTJ）作为人工突触装置被认为有希望构建受大脑启发的神经形态计算系统。然而，很少有人研究基于柔性 FTJ 的忆阻突触。这里，我们报告生长的云母基板，它由的BiFeO的超薄铁电屏障在柔性FTJ忆阻器₃，ZnO构成的半导体层，和的SrRuO的电极₃。获得的柔性 FTJ 忆阻器表现出稳定的电压调谐多态，并且电阻开关在 10 ³弯曲循环。FTJ 作为灵活突触装置的能力通过具有弯曲的尖峰时间依赖性可塑性的功能以及具有小非线性 (-0.24) 和低循环间变化 (1.77%) 的精确电导操纵来证明也实现了。特别是，基于实验设备行为的人工神经网络模拟表明，手写数字和图像的识别准确率分别高达 92.8% 和 86.2%，接近理想忆阻器的性能。这项工作突出了 FTJ 作为用于数据存储和处理的柔性电子设备的潜在应用。