Design of crack inspection system software for IC package with decision-making methods: Neural network & rule-based

Machine vision is widely used in the semiconductor industry. One of the applications is for the inspection system of an integrated circuit (IC) packaging. In the IC packaging process, an inspection error may occur when the defect packages are not detected by the inspector. This may due to fatigues and human are slower than the machines. In this research, five methods for the detection of crack defects on IC packages are proposed. There are two types of cracks covered by this study, i.e., apparent crack and hairline crack. These methods use blob analysis technique in image processing stage, and two decision-making methods to classify an IC package. These decision-making methods are rule-based decision and multi-layer perceptron (MLP) neural network classifier methods. This paper presents the various filters and operations employed in the blob analysis. Performance comparison was conducted for these methods. The first method using the MLP neural classifier produced an accuracy of 74.82% with 87.72 ms processing time, while the second method utilizing the same classifier achieved 86.17% accuracy with 119.45 ms processing time. The third method using rule-based classifier achieved 95.04% accuracy with a processing time of 133.19 ms. The forth and fifth utilised similar image processing method, but with different classifier. The forth method produce an accuracy of 96.1% with 188.44 ms processing speed, with the fifth method, using rule-based classifier, achieved 93.97% accuracy and processing speed of 193.50 ms. The results obtained in this study indicated that the forth method produced better performance, with higher accuracy and processing speed below 200 ms.