[Session 1: Alternative MI] E-Beam Metrology, Inspection and Deep Learning
The deep learning revolution this decade owes much of its success to the advances in convolutional neural networks (CNNs), enabling applications including computer vision and image analysis. Such techniques benefit from a large amount of well-structured data (often image data), which make them a perfect fit for SEM-related applications.
For several years, the vast amount of data provided by high-throughput SEMs (such as the eP5 from HMI) and improved contour extraction algorithms have enabled deep learning applications in OPC modeling, replacing part of the OPC model with CNNs. More recently, similar applications have been extended to the modeling of stochastic effects which are of particular interest in EUV lithography, including prediction of stochastic edge placement errors (EPEs), and prediction of contact hole failure rate from optical images (simulated) or resist images (measured).
In addition, benefits of deep learning are demonstrated for inspection applications, improving the accuracy or throughput, by enhancing the quality of SEM images. Additional use cases include simulation of SEM images from design layout, automatic defect classification, and pattern identification.
We will discuss the current synergies between e-beam metrology, inspection, and deep learning, and provide an outlook to the future.