Hardware Trojan Detection on PCBs Using Density Peak Clustering

Board-levelhardware Trojan (HT) is a malicious modification in the printed circuit board (PCB). It may lead the victim device to denial of service or leak confidential information. To tackle this issue, we propose an HT detection method which can detect and locate Trojan across layers. First, we obtain the thermal conductivity and temperature feature maps for a group of unauthentic PCBs. For each pixel in the maps, we construct a matrix which contains the above two feature pairs for all boards. Then, we apply the density peak clustering algorithm to distinguish Trojan-infected boards from Trojan-free ones in the feature space. Finally, we analyze the clustering result on each pixel and locate the Trojan position on the infected PCBs. The experimental results reveal that our method is effective. Generally, for the well-recognized PCBench and TRIT-PCB benchmarks, the Trojan detection accuracy is higher than 96.3% and the location error is less than 2.2%, even though the Trojan is as small as 3.5 mm$^{2}$. In addition, we demonstrate the advantages of our approach over two existing HT detection methods, namely, automated visual inspection and differential power monitoring. We also make a thorough discussion on how the PCB manufacturing process variation, feature map resolution, and Trojan area affect the detection results.