Su8 out-of-plane stress reduction via design of experiment and machine learning

Stress-induced bending in a stand-alone SU8 shaft structure lowers the yield and prolongs the development of brain-machine interface (BMI) that is packaged with SU8. The stress and bending can be reduced by properly adjusting several experiment parameters. In this work, design of experiment(DoE) used to systematically plan factors and levels to get conditions with minimal bending and high fidelity. The data bending results of the released SU8 shaft structure are processed with analysis of variance (ANOVA) and the parameter space is mapped out for the first time by Machine Learning algorithm (Support Vector Machine with RBF kernel). The algorithm predicts the optimal range of conditions for low shaft deflection and is experimentally validated.