Hardware Trojan Detection Based On Word-level Signal Property And Node Betweenness Centrality

In recent years,with the development of IC industry,there has been an emerging security threat,namely Hardware Trojan(HT)in the chip.Third-party IP adoption and fabless manufacturing have become a trend in IC industry.However,such trend provides an opportunity for adversary to insert HTs during the IC supply chain.This raises serious concerns about trustworthiness of ICs used in mission critical applications.Currently,researchers pay more attention to the following research fields:IP trust verification techniques in pre-silicon stage and test pattern-based HT activation techniques in post-silicon stage.Generally,HTs are located in the rare nodes in the circuit and how to find the rare nodes is the key to detect the HT.Therefore,this paper first utilizes the word level statistical parameters to measure the transition activity of the internal nodes in the circuit.Basically,nodes with low transition activity are considered as rare nodes.Then by analyzing the signal propagation property of different operators,we finally obtain the requirements for signal monotonous control.Moreover,based on the different structures of the circuit,we propose a traceback approach to generate test patterns for HT activation.The results show that the transition activity of rare bits can be dramatically increased by up to 166.7 times and Trojan activation time can be reduced by up to 121 times.On the other hand,since current logic analysis-based IP verification techniques are too time-consuming to identify the HT-related signals,we transform the logic analysis of the circuit to the data flow graph-based analysis.We first demonstrate the relationship between the rareness and the betweenness of rare nodes.Then we use betweenness centrality as the metric to identify the HT-related suspicious signals in the circuits.The experimental results show that 100% detection capacity can be obtained with low false positive rate and the detection time ranges from 2 seconds to145 seconds for the tested benchmarks.