Machine learning based real-time assessment of fabrication deviation induced mechanical performance variations in stretchable silicon arrays

Microelectromechanical system fabrication represents a promising approach for silicon-based flexible electronics, leveraging its scalability and miniaturization merits. However, fabrication-induced geometric deviations stretchable microstructures can result in significant variations in mechanical performances. Current assessment methods lack sufficient accuracy for these precision-sensitive manufacturing processes. This work proposes a machine-learning (ML)-based assessment methodology for accurately and rapidly predicting the mechanical performances, including equivalent Young's modulus and the maximum elastic stretchability, of Parylene three-dimensional micro-Kirigami stretchable structures in a stretchable silicon array affected by the fabrication-induced geometric deviations. By applying the dimensionality reduction technique specifically designed for few-shot ML modeling, the framework achieves prediction accuracies exceeding 95% on the test set. SHapley Additive exPlanations (SHAP) analysis is further utilized to quantify the impact of various geometric features on mechanical performances. This ML-based assessment methodology successfully facilitates real-time feedback from process-induced geometric deviations to the qualification probability of mechanical performances. This proposed approach supports design-for-manufacturability (DFM) of silicon-based stretchable arrayed devices manufacturing and lays the foundation for high-consistency wafer-scale manufacturing of high-performance stretchable silicon electronics.