Machine-learning micropattern manufacturing

•Machine learning algorithms are used to define experimental boundary conditions and facilitate micropattern optimization.•Integrating classification and regression models in the learning loops, we improve the accuracy of model predictions.•Through active learning, we accelerate the optimization process of fabricating micropatterns with the least experiments.•The optimal micropattern has a diameter range of 27-470 nm, much wider than the widest diameter range ever reported.•The optimal gradient micropattern further shows superior performance in high-throughput screening for biomedical use. [Display omitted] Micropatterning has been widely applied in electronics, biomaterials engineering, and microfluidics studies. A key challenge in using bipolar electrochemistry for fabricating titanium dioxide (TiO2) nanotube micropatterns (TNMs) with desired properties is to balance interrelated experimental parameters and define experimental boundary conditions. For example, it is challenging to determine the anodization voltage boundary as high anodization voltage with certain conditions might induce titanium foils rupture. Here, we utilize active learning to facilitate the optimization process of fabricating TNMs with a wide dimension range within one sample using bipolar electrochemistry. Starting with a small dataset, the decision tree model differentiates normal data from abnormal data (i.e., titanium foils ruptured), which helps define the experimental boundaries. Then gradient boosted regression tree (GBRT) model analyzes the data and provides predictions and directions for optimizing TNMs. Then predictions are verified by experiments, and new results update the training dataset for the next learning loop. Results show that ML algorithms well define the experimental boundary conditions. And only within several iterations, we obtained the optimal TNMs with a diameter range of 27–470 nm, expanding the gradient to the largest extend without tedious experiments. Those results indicate that machine learning algorithms are effective in accelerating materials manufacture and optimization. Further silver nanoparticle doping demonstrates that large-scale TNMs are effective platforms for high-throughput screening.