Research And Implementation Of Machine Learning Detection Methods For Hardware Trojan

The security and reliability of integrated circuits are under serious threat from hardware Trojans,and the detection of hardware Trojans hidden in integrated circuits has become one of the key concerns for researchers.In this paper,we will investigate hardware Trojans and their detection methods to address the problems of cost and accuracy of detection,over-reliance on pure chips,and poor generality of detection.This paper first designed a hardware Trojan circuit for cryptographic chips and verified the design specifications of the circuit,which functioned as expected and achieved the design requirements of hardware Trojan stealth and low triggering.The design of the hardware Trojan circuit deepens the understanding of hardware Trojans and lays a good foundation for the subsequent research on detection methods.The ideas of classification decision making in machine learning are then applied to hardware Trojan detection,and the design of hardware Trojan detection methods for support vector machines and anomaly detection are presented respectively.In the final experimental stage,the FPGA-based experimental results show that the support vector machine hardware Trojan detection method can detect different types of hardware Trojans hidden in the IC with an average detection rate of 99.43% under the Advanced Encryption Standard(AES)circuit.The hardware Trojan detection method with anomaly detection detects different types of hardware Trojans with an average detection accuracy of 86.4% to 100% under SM3 and SM4 background circuits respectively.Both detection method designs combined with machine learning have the advantages of high detection accuracy,low cost and good generality.