This paper presents how to use machine learning to accelerate the development of lamellar block copolymer process. The first part introduces an automated algorithm to measure lamellar CD-SEM images oriented to allow fast images screening, identify and select the most relevant chemical composition. The second part is the use of machine learning in the process development which is a data driven approach. This part is also divided in two sections, the first one for the prediction of the process, where having a set of experiments in a process window, a model is created so that the outcome from is estimated from different parameters. The second part is the estimation of the process window. These tools are oriented to assist process engineers in the process optimization which must be driven under expert supervision.
The latest advances in Machine Learning (ML) produce results with unprecedented accuracy, and could signal a new era in the smart manufacturing field. We propose a framework designed to work alongside experts: learning from them and optimizing their knowledge. This framework must be considered as a tool to assist the experts in their daily work. The user creates a measurement recipe which includes an example of the feature as well as the measurements placed by the process engineer. Grouping the measurement recipes of the same object in an entity collection allows the user to train a machine learning recipe which includes a deformation model to handle variations in structure and contrast. The new images are analyzed following the machine learning pipeline which includes the detection of features, repositioning, measurement, quality evaluation and finally the results of measurement are given to the user. We discuss the pipeline and we focus on the metrics to validate the machine learning recipe, providing quantitative results for stability and robustness to variations.