Micro integrated circuits are vulnerable to external factors such as tiny scratches, dust particles, and electrostatic discharges, which human eye cannot trace and which can degrade their performance or lead to complete failure with potentially severe consequences. To address this issue, recent studies have started applying data mining and machine learning methods, such as Decision Trees, Support Vector Machines, Random Forests, Long Short-Term Memory, Convolutional Neural Networks, to tabular data containing signals recorded from the sensors, to ultrasound data, and to optical image data. However, the data is not adequately diverse and contains noise that impacts the deep learning models’ accuracy rate. Whereas, our work employs the integrated circuit data collected by using scanning electron microscopy. The original dataset consists of 25,160 normal and 116 anomalous high resolution images. Additionally, we approach by deploying a deep learning ensemble model for anomaly detection based on Reconstruction-Based Neural Network, which utilizes multiple Convolutional Neural Networks and Autoencoders. Our experimental results showed that the method successfully identifies and locates anomalies in integrated circuits with an 86% accuracy rate.