Low-Cost Yet Effective Trojan Mitigation Techniques for Approximate Systems

Abstract

Approximate Computing (AC) is a promising computing paradigm that provides efficiency and speed while maintaining acceptable accuracy in error-tolerant applications. However, most of the existing AC designs are prone to Hardware Trojan (HT) attacks. Moreover, the existing HT mitigation techniques have been designed for exact systems and due to the architectural differences in the approximate systems, these conventional mitigation techniques have limited efficacy when used in the context of AC. Therefore, this brief presents two novel methods, named Partial Comparison (PC) and Hybrid Comparison (HC), that effectively mitigate HTs in approximate systems while achieving a 30% and 55% improvement in the area and the power efficiency, compared to the existing state-of-the-art technique. PC focuses on LSBs while HC considers both LSBs and MSBs for HT masking. The effectiveness of proposed techniques is validated through a practical case study of image processing; three IPs with different HTs are successfully masked with a PSNR and SSIM of 33dB and 0.92 respectively. In conclusion, this research provides practical solution to enhance security and reliability of approximate systems with optimal performance benefits.