The recent trend of utilizing machine learning to address challenging problems in various fields exists in the material science area; there have been attempts to apply deep learning in designing molecules’ structures, analyzing spectral data, and even sampling more accurate free energy landscapes. Despite the unlimited potential of deep learning technology, one of the crucial requirements that hinder its widespread use in material science is the limitation of ensuring sufficient training datasets. Here, we apply deep learning algorithms developed for image classification and object detection, named YOLOv3, to identify defects and calculate defect densities in self-assembled block copolymer films. Conventional defect inspection involves performing either manual defect detections or using inefficient image processing tools. We propose a modified YOLO network optimized for defect inspection of the scanning electron microscopy (SEM) images. A sufficient training dataset is prepared by mixing the small numbers of experimental SEM images, and the artificially generated data using Monte Carlo simulation with a coarse-grained model. Factors that affect the network’s performance to detect and classify the defects in block copolymer films such as distribution, quality, and mixing ratio of the simulated dataset are investigated.

Identification of defects in block copolymer films with deep learning network

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.