| With the rapid development of network and information technology,the need of information security improve continuously,and people have paid more concern and attention on it.Hardware Trojans are more dangerous than software malicious codes.Since they are usually in the core of the system,the main feature of hardware Trojans is strong concealment.Up to now the mian hazard of integrated circuit comes from hardware Trojans.And with the increasing of the scale of integrated circuit,there are more and more types of the hardware Trojans,the application environment becomes more and more complex,the chip design becomes cleverer and clever,and the implanting technology becomes diverser and diverser.All these factors leads that the impact of the hardware Trojans with respect to the original circuit becomes smaller and smaller.Because the side channel information is relatively weak,the existing detection and identification methods are increasingly unable to effectively detect the presence of small hardware Trojans.In this paper the hardware Trojans theory and circuit design are described firstly,then thefeature extraction and machine learning pattern classification theory are studied,and ultimately the two classification detection system of the hardware Trojans will be set up.The Distance Measure method is applied in the first classification level,and the Support Vector Machine classifier is applied in the second level.The classification detection model based on Distance Measure can achieve the roughly classification and identification of hardware Trojans in the first level.Three methods are set up in the first level,including Distance Measurement distribution method,(K)Nearest Neighbor method and improved(K)Nearest Neighbor method.When the Trojan circuit of different area(3.333%,1.523% and 0.69%)is implanted into the standard circuit,the detection and recognition rate can reach the value of 87.2245%,81.068 and 61.2245% according to the Trojan area.The classification detection model based on Support Vector Machine can achieve the accurate classification and identification of hardware Trojans in the second level.The key problem to build the SVM is the selection and optimization of super parameters and kernel function.So the Cross-validation and other heuristic algorithm is used to improve and optimize the parameters of SVM.When the Trojan circuit of area 0.69% is implanted into the standard circuit,the detection and recognition rate can reach the value of 98.64% and 99.17% according to the CV algorithm and heuristic algorithm respectively. |
