Using active thermography and modified SVM for intelligent diagnosis of solder bumps

Solder bump technology has been used extensively in microelectronic packaging. But defect inspection becomes increasingly difficult due to the decrease of solder bumps in dimension and pitch. To overcome the shortages of traditional methods, we have developed an intelligent system using the active thermography for defects inspection of the solder bumps. A modified support vector machine (M-SVM) was investigated to solve the problem of small sample size in solder bumps classification. The chip SFA1 and SFA2 were chosen as the test vehicles. Captured thermal images were preprocessed using the improved wiener filter and moving average technique to remove the peak noise. The principal component analysis (PCA) algorithm was then adopted to reconstruct the thermal image, in which the hot spots were segmented. The statistical features corresponding to every solder bump were extracted and input into the M-SVM for solder bumps classification. The defective bumps w distinguished from the good bumps, which proves that the intelligent system using the modified SVM is effective for defects inspection in microelectronic packages.