Adaptation of neural network and application of digital ultrasonic image processing for the pattern recognition of defects in semiconductor

In this study, the classification of artificial defects in semiconductor devices are performed by using pattern recognition technology. For this target, a pattern recognition algorithm including user made software was developed and the total procedure including image processing and self-organizing map was treated by a backpropagation neural network, where image processing was composed of ultrasonic image acquisition, equalization filtering, binary processing and edge detection. Image processing and self-organizing map were compared as preprocessing methods for the reduction of dimensionality as input data into multi-layer perceptron or backpropagation neural networks. Also, the pattern recognition technique has been applied to classify two kinds of semiconductor defects: cracks and delamination. According to these results, it was found that the self-organizing map provided recognition rates of 83.4% and 75.7% for delamination and cracks, respectively, while BP provided 100% recognition rates for the results.