(697f) Automatic Stretchable Conductor Design and Fabrication Via Machine Learning

Stretchable conductors are essential for stretchable electronic circuits that can be applied to next-generation wearable devices and soft machines/robots. An ideal stretchable conductor can maintain strain-independent electrical conductivities during reversible stretching/relaxation processes. However, it is nearly impossible for traditional computational approaches to predict the performance of a stretchable conductor from its composition, morphology, and fabrication conditions. Machine learning is a versatile tool to uncover the non-linear correlations between fabrication recipes and stretchable conductors’ performance at the device level. Herein, a hybrid approach (wet-lab experiments and machine learning framework) is conducted to the automatic stretchable conductors’ design concept. Three functional materials (i.e., Ti₃C₂T_x MXene nanosheets, single-wall carbon nanotube, and gold nanoparticle), and one polymer binder (i.e., polyvinyl alcohol), are used for the fabrication of stretchable conductors. First, a support-vector machine classifier is trained with 286 different compositions to confirm a feasible fabrication regime. Second, hundreds of stretchable conductors with various recipes and five advanced structures (i.e., 1D, 2D, 2D1D and 2D2D) are fabricated through active learning loops followed by data augmentation, and >10,000 data points (virtual and real) can enrich the multidimensional database and build up a high-accuracy prediction model. Third, through data analyses, data-driven design principles for stretchable conductors can be uncovered and validated via in situ microscopic studies. Finally, we demonstrate how the implementation of collaborative robots can accelerate the prediction model construction.