Stretchable synaptic transistors show great promise in mimicking brain activities in soft robotics and skin electronics applications. However, the fabrication of such device arrays on elastic substrates with high stability, throughput, and yield remains challenging. Here, we have developed an approach to fabricate stretchable synaptic transistors directly on elastic substrates, in which carbon nanotubes and SU-8 are used as channel and dielectric, respectively. This method employs a fully photolithography-based microfabrication process that operates at relatively low temperatures. The devices exhibit an average on–off ratio of 2 × 10⁶ and show minimal degradation when stretched up to 40%. Single-pulse, paired-pulse, and repetitive-pulse responses are also demonstrated, showing their ability to work as artificial synapses. The devices exhibit a high linearity of ≤1 with 100 distinct conductance states in long-term plasticity and a dynamic range of 15. Furthermore, we conducted a handwritten digit recognition simulation, achieving a learning accuracy of over 90%. We believe our work can serve as a guide for developing high-performance stretchable synaptic devices for various applications.

中文翻译：

---

通过直接微纳加工实现基于应变不敏感、空气稳定的可拉伸碳纳米管突触晶体管阵列，用于神经形态计算

可伸缩突触晶体管在软机器人和皮肤电子应用中模拟大脑活动方面显示出巨大的前景。然而，在具有高稳定性、吞吐量和产量的弹性基板上制造此类器件阵列仍然具有挑战性。在这里，我们开发了一种直接在弹性衬底上制造可拉伸突触晶体管的方法，其中碳纳米管和 SU-8 分别用作通道和电介质。该方法采用完全基于光刻的微纳加工工艺，在相对较低的温度下运行。这些器件的平均通断比为 2 × 10⁶，并且在拉伸高达 40% 时表现出最小的退化。还展示了单脉冲、成对脉冲和重复脉冲反应，显示了它们作为人工突触工作的能力。这些器件表现出 ≤1 的高线性度，在长期塑性下具有 100 种不同的电导状态，动态范围为 15。此外，我们进行了手写数字识别模拟，实现了超过 90% 的学习准确率。我们相信我们的工作可以作为为各种应用开发高性能可伸缩突触装置的指南。