AHD-LAM: A New Mitigation Method against Voltage-Drop Attacks in Multi-tenant FPGAs
Mashrafi Alam Kajol, Sandeep Sunkavilli, Qiaoyan Yu · 2023
Multi-tenant Field-Programmable Gate Arrays (FPGAs) are vulnerable to voltage-drop attacks, which could jeopardize the safety and reliability of FPGA utilization. Existing countermeasures deploy numerous sensors to detect fault attacks and mitigate attacks by sacrificing other FPGA tenants’ normal operations. To address those limitations, we propose an accumulated Hamming distance-based local attack mitigation (AHDLAM) method to precisely pinpoint attack locations with a new assessment metric and multi-stage localization process. Furthermore, a dynamic partial reconfiguration-based attack mitigation technique is proposed to reduce the cost of eliminating malicious FPGA logic. Experimental results show that our method reduces the number of on-chip sensors by 92.7% and lowers the attack mitigation granularity by up to 87.5%, compared to existing works. The partial reconfiguration technique reduces the attack mitigation time by 78% than the baseline.