Radio Frequency-Based Device Discrimination of Mixed-Signal Integrated Circuits and Counterfeit Detection
Abstract
The research presented here focused on applying radio frequency-distinct native attributes (RF-DNA) feature extraction combined with various types of machine learning such as: multiple discriminant analysis/maximum likelihood (MDA/ML), generalized relevance learning vector quantized-improved (GRLVQI), quadratic discriminant analysis (QDA), and random forest (RndF) to discriminate mixed-signal integrated circuit (IC) devices and perform counterfeit detection. Unintentional RF emissions (URE) were collected from the device under test (DUT), Maxim MAX526CCWG digital to analog converter (DAC), that were independently screened into two categories of authentic and counterfeit. A subset of these devices were used to generate a model and new collections from all devices were used to verify the model. Additionally, RF-DNA combined with (MDA/ML) was used to develop a model to discriminate between the MAX526CCWG devices and the update devices MAX526CCWG , a lead free version of the MAX526CCWG. This research also explored feature and sampling rate reduction as a means to reduce complexity.