Research On Hardware Trojan Detection Technology Based On Machine Learning

Hardware Trojans,abbreviated as HTs,refer to the purposeful modification or tampering of the original circuit without authorization during the chip design or manufacturing process,so as to achieve the purpose of changing the system function unconditionally or conditionally,leaking confidential information or destroying the system.Under the guidance of Moore's law and Dennard's scaling law,the integration of chips is getting higher and higher.In VLSI,a single chip can hold hundreds of millions of transistors.With the rapid growth of IC scale and the more flexible manufacturing model,the main security problems of IC have changed from the defects of the original process in the immature period to the hardware Trojan horse caused by disputes of interest.The development trend of IC has made them extremely vulnerable to hardware.Trojan attack.The detection of hardware Trojan horses essentially starts from the working principle of HTs,and establishes a mathematical model based on the physical parameters or physical characteristics of the hardware Trojan horses,and turns the detection of hardware Trojan horses into a process of data classification problems.The accuracy of the model largely depends on the effective parameter input scale and function fitting ability of the model.In order to better establish the detection model,this paper,based on the research of the hardware Trojan attack mechanism and detection principle,combined with the convolutional neural network algorithm,explores the design and detection of the hardware Trojan implanted IP,and carried out machine learning based Algorithmic hardware Trojan detection technology research work.This article first reviews the research work of hardware Trojan horses at home and abroad,and discusses the benefits of machine learning algorithms for hardware Trojan horse detection based on the shortcomings of existing mainstream technologies.In the experiment,the GDMA carrier circuit was first designed and implemented,and based on the IP security and Bus security of the system-level design,three hardware Trojan horses were specifically designed to attack the carrier circuit.Secondly,taking the hardware Trojan detection technology of the ring vibrator network as the research object,proposes optimization strategies from circuit hardware,data measurement,software algorithm,etc.,designs and implements a special detection circuit structure,and builds a two-dimensional detection data collection hardware platform.Finally,the detection process was designed according to the hardware Trojan detection ideas and the convolutional neural network classification model for the problem was established.The designed Trojan was detected on the FPGA development board and the experimental results were analyzed.The detection platform reached about 93% of the data.The detection accuracy rate is about 9%higher than that of the mean dimensionality reduction method.