Reliability of Deep Learning Models for Scanning Electron Microscopy

Scanning electron microscopy (SEM) provides high-resolution nanoscale imaging crucial for advanced materials characterization. Recent advancements in deep learning have significantly enhanced SEM analysis by automating the identification of nanoscale features. However, the reliability of these models remains insufficiently explored. We investigate two popular deep learning architectures, ResNet-50 and Swin Transformer, by systematically injecting faults into their weight parameters for SEM characterization. Our analysis demonstrates how performance degrades under varying fault scenarios, and also pinpoints which layers and bit positions exhibit heightened vulnerability. These findings provide insights for dependable, high-throughput automated materials characterization, thereby paving the way for more reliable deployment of SEM-based deep learning in industrial and research environments.