The increasing demand for mobile artificial intelligence applications has elevated edge computing to a prominent research area. Silicon materials, renowned for their excellent electrical properties, are extensively utilized in traditional electronic devices. However, the development of silicon materials for flexible neuromorphic computing devices encounters great challenges. To address these limitations, ultrasoft silicon nanomembranes have emerged as a focal point due to their capability to preserve the superior electrical properties of silicon while providing substantial mechanical flexibility and interfacial tunability. Despite these advantages, difficulties remain in the transfer process of silicon nanomembranes and their integration for flexible synaptic transistors. In this work, an organic-inorganic hybrid polyimide-Al2O3 dielectric layer has been designed for synaptic behavior grown by an atomic layer deposition process, and integrated with a silicon nanomembrane to realize highly flexible synaptic transistors. These transistors demonstrate stable electrical performance even after undergoing 10 000 bending cycles at an extreme curvature radius of 2.2 mm. Furthermore, the silicon nanomembrane transistors effectively emulate synaptic functions, exhibiting exceptional linearity in their long-term characteristics, making them suitable for the application scenarios of detecting subtle signals. When applied to handwritten digit recognition simulations, these synaptic transistors have achieved a high accuracy rate of 93.2%.
Keywords: flexible electronics; high bendability; neuromorphic computing; silicon nanomembrane; synaptic transistors.
© 2025 Wiley‐VCH GmbH.