AutoDetect: Novel Autoencoding Architecture for Counterfeit IC Detection

• Published in: Journal of hardware and systems security 2024-06, Vol.8 (2), p.113-132
• Main Authors: Bhure, Chaitanya, Nicholas, Geraldine Shirley, Ghosh, Shajib, Asadi, Navid, Saqib, Fareena
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Authenticity; Automation; Circuit reliability; Circuits and Systems; Classification; Computer Hardware; Computer vision; Counterfeit; Counterfeiting; Datasets; Deep learning; Electronic components; Electronic devices; Engineering; Global economy; Human error; Information Systems Applications (incl.Internet); Integrated circuits; Machine learning; Neural networks; Packaging; Physical properties; Property rights; Quality control; Semiconductor devices; Semiconductors; Supply chains; Systems and Data Security
• ISSN: 2509-3428; 2509-3436
• DOI: 10.1007/s41635-024-00149-3

The global electronic supply chain industry is a complex and interconnected network of companies, organizations, and individuals that collaborate to produce and distribute electronic devices and components. It plays a critical role in the global economy, as electronic devices have become an essential part of modern life. Electronic components and devices are produced and distributed across the world, with different regions and countries specializing in different aspects of the supply chain. This global nature of the electronic supply chain industry also poses challenges, particularly in terms of quality control and supply chain transparency involving counterfeit components owing to its interconnected nature. Counterfeiting of integrated circuits (ICs) and semiconductor devices is a significant challenge that poses threats to the safety and reliability of electronic devices, the global economy, intellectual property rights, and the overall sustainability of the electronic supply chain industry. To address this challenge, we propose a novel autoencoding architecture for counterfeit IC detection using a labeled database. The proposed architecture achieves an overall accuracy of 83% which is ≈ 20% more than the existing transfer learning approaches.