부산대학교 도서관

Remarked and recycled counterfeit integrated circuits (ICs) form a vast majority (80–90%) of the total number of counterfeit IC instances. Although different types of test strategies have been developed for recycled IC detection, techniques that detect remarked ICs are limited. In this paper, we develop a method to detect false remarking of commercial grade chips into industrial/automotive grade by distinguishing power supply rejection ratio (PSRR) of commercial and automotive grade low drop-out (LDO) regulators from four different vendors. In this process, we use supervised and unsupervised machine learning (ML) methods on PSRR measurements. Our results show a best-case accuracy of 90% for both commercial and industrial LDOs with supervised ML. On the other hand, unsupervised ML can detect commercial and industrial LDOs with a best-case accuracy of 75% for both types.