Automatic detection of solder joint defects on integrated circuits

In this paper an automatic inspection method for the diagnosis of solder joint defects on integrated circuits mounted in Surface Mounting Technology is presented. The diagnosis is handled as a classification problem with a neural network approach. Two classes of solder joints have been fixed with respect to the amount of the soldering paste. These correspond to acceptable and non acceptable solder joints to assure the correct working of the integrated circuits. The images of the boards under test are acquired by an acquisition system; then they are pre-processed to extract the region of interest for the diagnosis. A "geometric" features vector is extracted from this region and it feeds a Multi Layer Perceptron neural network. Experimental results show that these networks perform high recognition rate and the robustness of the proposed method.