Application of neural network technologies to identify objects in images of topological layers of integrated circuits 

author:Adiya Biyekenova

Abstract

Abstract

The work is devoted to research of neural network technologies to identify objects on the images of topological layers of integrated circuits used in space information technology for data transmission and processing by space technology, modeling and control of space systems. The problem of a shortage of technology to check for defects in the topological layers of integrated circuits in Kazakhstan is a pressing one. In this regard, the purpose of this paper is to develop neural network technology to identify objects in the images of topological layers of integrated circuits of space information technology. Defect control, which ensures the quality of microchips, is usually the largest cost item in PCB manufacturing. Detecting defects in microchips using neural network algorithms is a complex process involving several inspection steps.  

This paper attempts to apply deep learning algorithms in combination with ConvNets, CASAE, machine vision algorithm, etc. to solve the problem of defect detection.  To train the neural network, some images of microarrays in the laboratory were inspected. In addition, a Henon map was used to train the neural network. A convolutional neural network for image classification is created using the Keras library. The result was an algorithm for detecting integrated circuit objects, which allows us to determine the causes of damage to the microchips. In the future, these results may be acceptable for topology recognition to prevent unauthorized objects, defects, or earlier detection of wear and tear in integrated circuits.