Unsupervised recycled FPGA detection using exhaustive nearest neighbor residual analysis

Measuring and analyzing aging-induced delay degradation of ring oscillators (ROs) is an effective method to detect recycled field-programmable gate arrays (FPGAs). However, detection methods of conventional recycled FPGAs detection methods assume the existence of known fresh FPGAs (KFFs) as training data for machine-learning-based classification, which is an unrealistic assumption. In this paper, we propose an unsupervised recycled FPGA detection method, where little information on KFF is available. In the proposed method, estimated frequency is calculated from neighboring ROs, and then the residual frequency between the measured and estimated frequencies is used for the detection. Because of the systematic component of process variation, the frequencies of neighboring ROs should be similar when the target FPGA is fresh. Therefore, if the residual is high, the target FPGA is determined as recycled. Experiments using 25 commercial FPGAs under various aging scenarios demonstrate that the proposed method successfully distinguishes between recycled and fresh FPGAs.