Automatic Feature Selection and Improved Classification in SICADA Counterfeit Electronics Detection
Defense Microelectronics Activity (DMEA) McClellan United States
Pagination or Media Count:
Counterfeiters seeking financial gain can introduce misrepresented or recycled microelectronic components to both government and commercial supply chains. This reduces system reliability and trust, and currently has few comprehensive and practical solutions. The SICADA methodology was developed to detect such counterfeit microelectronics by collecting power side channel data and applying machine learning to identify counterfeits. This methodology has been extended to include a two-step automated feature selection process and now uses a one-class SVM classifier. We describe this methodology and show results for empirical data collected from several types of Microchip dsPIC33F microcontrollers.