AI Applications for Failure Analysis - AI-Driven Anomaly Detection in Electron Microscopy Images of Periodic Circuit Structures
AI Applications for Failure Analysis - AI-Driven Anomaly Detection in Electron Microscopy Images of Periodic Circuit Structures
Wednesday, November 19, 2025
Summary:
This study investigates periodic circuit structures imaged using electron microscopy, employing a two-step analytical approach. Our research integrates cross-correlation techniques with autoencoder neural networks, effectively bridging traditional computer vision methods and modern AI capabilities. The methodology commences with comprehensive electron microscopy image acquisition and the identification of a reference pattern. Subsequently, cross-correlation analysis precisely locates instances of this pattern throughout the scanned image. Each identified pattern then undergoes evaluation by an autoencoder neural network to characterize the quality of the circuit features of interest. Utilizing this integrated approach, patterns exhibiting abnormal circuit features are systematically identified and quantified, providing an objective assessment of anomalies within the examined periodic structures. This framework, combining signal processing with deep learning, enables rapid and accurate detection of circuit anomalies, thereby enhancing failure diagnostics and reliability assessment in semiconductor failure analysis.
This study investigates periodic circuit structures imaged using electron microscopy, employing a two-step analytical approach. Our research integrates cross-correlation techniques with autoencoder neural networks, effectively bridging traditional computer vision methods and modern AI capabilities. The methodology commences with comprehensive electron microscopy image acquisition and the identification of a reference pattern. Subsequently, cross-correlation analysis precisely locates instances of this pattern throughout the scanned image. Each identified pattern then undergoes evaluation by an autoencoder neural network to characterize the quality of the circuit features of interest. Utilizing this integrated approach, patterns exhibiting abnormal circuit features are systematically identified and quantified, providing an objective assessment of anomalies within the examined periodic structures. This framework, combining signal processing with deep learning, enables rapid and accurate detection of circuit anomalies, thereby enhancing failure diagnostics and reliability assessment in semiconductor failure analysis.