Computer Vision for Hardware Trojan Detection on a PCB Using Siamese Neural Network

Piliposyan, Gor ORCID: 0000-0001-7182-0717 and Khursheed, Saqib (2022) Computer Vision for Hardware Trojan Detection on a PCB Using Siamese Neural Network. In: 2022 IEEE Physical Assurance and Inspection of Electronics (PAINE), 2022-10-25 - 2022-10-27, USA.

PDF
PAINE-2022_CR ver.pdf - Author Accepted Manuscript
Download (1MB) | Preview

Official URL: http://dx.doi.org/10.1109/paine56030.2022.10014967

Abstract

With advances in technology Hardware Trojan (HT) attacks on printed circuit boards (PCB) are becoming more sophisticated and the need for more effective HT detection methods is becoming crucial. Automated visual inspection (AVI) is one of the most promising solutions in detecting malicious implants on a PCB. It is non-destructive, effective in testing PCBs on an industrial scale, demands minimum human involvement, and can potentially identify malicious inclusions and modifications on PCBs at all stages of production and thereafter. In recent years, machine learning algorithms have been successfully applied, significantly improving the effectiveness of AVI methodologies. In this paper, an AVI methodology is proposed for detecting HTs on a PCB, using input data from a low-cost digital optical camera. It is based on a combination of conventional computer vision techniques and a dual tower Siamese Neural Network (SNN), modelled in a three stage pipeline. Further, a dataset of PCB images has been developed in a controlled environment of a photographic tent. The results show that the methodology has an average 95.6% classification accuracy for PCBs with HT inclusions with surface area between 4mm2 and 280 mm2.

Item Type:	Conference or Workshop Item (Unspecified)
Uncontrolled Keywords:	PCB Inspection, Hardware Trojan Detection, Deep Learning, Automated Visual Inspection, Siamese Neural Network, Computer Vision
Depositing User:	Symplectic Admin
Date Deposited:	03 Oct 2022 08:15
Last Modified:	26 Apr 2024 20:50
DOI:	10.1109/PAINE56030.2022.10014967
Related URLs:	Author Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3164978