The prevalence of counterfeit integrated circuits has become a problem for manufacturers, vendors, and end users in recent years. While several test methods aid in the identification of counterfeit devices, radiography offers a nondestructive technique to perform internal examination. With detailed manufacturer information, the task of identifying nonconforming devices reduces to identifying deviations from a known standard, the common quality control task of defect detection. Without manufacturer support, the task becomes one of grouping devices based on features whose values and distributions are not known a priori. In a batch of authentic devices, the features are uniform. Nonuniformity of internal construction despite uniformity of external appearance suggests that some or all of the devices are counterfeit. Currently, industrial radiographers manually perform this time intensive task. Focusing on the latter case, this research develops a system to automatically identify radiographically visible features of monolithic integrated circuits in dual-in-line packages. It then uses unsupervised classification to group the devices based on those features. The system’s decisions are compared to those of an experienced radiographer. If used in a routine inspection process, the system can free the radiographer to focus on borderline cases and reduce fatigue-induced operator error.

Related Publications