Anomaly Detection of Cdsem Images

In semiconductor manufacturing factories, Scanning Electron Microscope (SEM) are widely used for critical dimension metrology, process control inspection and defects analysis. There are lots of times to measure the after developing inspection (ADI) and after etching inspection (AEI) in the process. And it's a waste time and energy thing for engineers to distinguish between normal and abnormal images. Here, we used unsupervised anomaly detection to analyze CD-SEM images. We used an autoencoder which has an encoder-decoder-encoder pipeline, and capture the training data distribution within both image and latent vector space. And the result indicated that this method could save engineers' time on distinguishing normal and abnormal CD-SEM images