CN104330721A - Integrated circuit hardware Trojan horse detection method and integrated circuit hardware Trojan horse detection system - Google Patents

Integrated circuit hardware Trojan horse detection method and integrated circuit hardware Trojan horse detection system Download PDF

Info

Publication number
CN104330721A
CN104330721A CN201410598741.4A CN201410598741A CN104330721A CN 104330721 A CN104330721 A CN 104330721A CN 201410598741 A CN201410598741 A CN 201410598741A CN 104330721 A CN104330721 A CN 104330721A
Authority
CN
China
Prior art keywords
cluster
chip
measured
node
wooden horse
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201410598741.4A
Other languages
Chinese (zh)
Other versions
CN104330721B (en
Inventor
何春华
侯波
王力纬
恩云飞
谢少锋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fifth Electronics Research Institute of Ministry of Industry and Information Technology
Original Assignee
Fifth Electronics Research Institute of Ministry of Industry and Information Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fifth Electronics Research Institute of Ministry of Industry and Information Technology filed Critical Fifth Electronics Research Institute of Ministry of Industry and Information Technology
Priority to CN201410598741.4A priority Critical patent/CN104330721B/en
Publication of CN104330721A publication Critical patent/CN104330721A/en
Application granted granted Critical
Publication of CN104330721B publication Critical patent/CN104330721B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses an integrated circuit hardware Trojan horse detection method and an integrated circuit hardware Trojan horse detection system. The integrated circuit hardware Trojan horse detection method comprises the steps of: performing multi-parameter bypass test on all to-be-detected chips to obtain bypass data; extracting bypass data of partial to-be-detected chips for unsupervised learning clustering analysis to obtain a first cluster and a second cluster; extracting partial samples in the two clusters are extracted respectively for reverse engineering analysis, identifying the clusters to obtain a Trojan horse chip cluster and a non-Trojan horse chip cluster, and setting flag values respectively; performing supervised learning based on bypass data in both the Trojan horse chip cluster and the non-Trojan horse chip cluster and corresponding flag values; performing Trojan horse detection on to-be-detected chips which are not subjected to unsupervised learning clustering analysis based on a neural network; using the obtained through sampling analysis after clustering for supervised learning to improve learning accuracy, and accordingly improve hardware Trojan horse feature identification capability. In comparison with a traditional integrated circuit hardware Trojan horse detection method, detection accuracy is improved.

Description

IC Hardware Trojan detecting method and system
Technical field
The present invention relates to electric circuit inspection technical field, particularly relate to a kind of IC Hardware Trojan detecting method and system.
Background technology
Along with the development of semiconductor technology and manufacturing technology, hardware outsourcing design becomes the trend of globalization.There is a kind of New Hardware attack pattern for integrated circuit (IC) chip in recent years, be called " hardware Trojan horse ".Hardware Trojan horse mainly refers at IC (integrated circuit, integrated circuit) malice is added some illegal circuit or distorts original design file artificially in Design and manufacture process, thus leave " time bomb " or " electronics back door " etc., for follow-on attack opens convenience.The harm of hardware Trojan horse mainly comprises steals the important information of chip, affects circuit performance and reliability, distorts chip functions even defective chip etc., therefore carries out hardware Trojan horse to integrated circuit and detects more and more by attention both domestic and external.
Traditional IC Hardware Trojan detecting method mainly comes whether there is wooden horse in decision circuitry by the bypass message detected in analysis circuit.But for the hardware Trojan horse circuit of special small size, wooden horse is usually very little to the contribution of bypass message, and be especially easily submerged in test noise, simple bypass data process is difficult to distinguish wooden horse chip and non-wooden horse chip smoothly.There is the low shortcoming of detection accuracy in traditional IC Hardware Trojan detecting method.
Summary of the invention
Based on this, be necessary for the problems referred to above, a kind of the IC Hardware Trojan detecting method and the system that improve detection accuracy are provided.
A kind of IC Hardware Trojan detecting method, comprises the following steps:
Multiparameter bypass test is carried out to all chips to be measured, obtains the bypass data of all chips to be measured;
The bypass data of Extraction parts chip to be measured carries out unsupervised learning cluster analysis, obtains the first cluster and the second cluster;
The part sample extracted respectively in described first cluster and the second cluster carries out converse works analyzing, carries out identification to described first cluster and the second cluster, obtains wooden horse chip cluster and non-wooden horse chip cluster and arranges mark value respectively;
Mark value according to the bypass data in described wooden horse chip cluster and non-wooden horse chip cluster and correspondence carries out supervised learning, obtains the neural network after training;
According to described neural network, trojan horse detection is carried out to the chip to be measured not being used as unsupervised learning cluster analysis.
A kind of IC Hardware Trojan horse detection system, comprising:
Data acquisition module, for carrying out multiparameter bypass test to all chips to be measured, obtains the bypass data of all chips to be measured;
Data categorization module, the bypass data for Extraction parts chip to be measured carries out unsupervised learning cluster analysis, obtains the first cluster and the second cluster;
Clustering recognition module, carries out converse works analyzing for the part sample extracted respectively in described first cluster and the second cluster, carries out identification to described first cluster and the second cluster, obtains wooden horse chip cluster and non-wooden horse chip cluster and arranges mark value respectively;
Learning training module, for carrying out supervised learning according to the mark value of the bypass data in described wooden horse chip cluster and non-wooden horse chip cluster and correspondence, obtains the neural network after training;
Wooden horse test module, for carrying out trojan horse detection according to described neural network to the chip to be measured not being used as unsupervised learning cluster analysis.
Said integrated circuit hardware Trojan horse detection method and system, carry out multiparameter bypass test to all chips to be measured, obtains the bypass data of all chips to be measured.The bypass data of Extraction parts chip to be measured carries out unsupervised learning cluster analysis, obtains the first cluster and the second cluster.The part sample extracted respectively in the first cluster and the second cluster carries out converse works analyzing, carries out identification to the first cluster and the second cluster, obtains wooden horse chip cluster and non-wooden horse chip cluster and arranges mark value respectively.Mark value according to the bypass data in wooden horse chip cluster and non-wooden horse chip cluster and correspondence carries out supervised learning, obtains the neural network after training.According to neural network, trojan horse detection is carried out to the chip to be measured not being used as unsupervised learning cluster analysis.Unsupervised learning and supervised learning algorithm are merged mutually, unsupervised learning is utilized to carry out cluster analysis to bypass data, utilize the mark value of supervised learning to the bypass data in wooden horse chip cluster and non-wooden horse chip cluster and correspondence to carry out training and obtain neural network, workload and the cost of converse works analyzing can be reduced by the method for sampling analysis after first cluster, improve analysis efficiency, data after cluster analysis are used for the accuracy that supervised learning can improve study, thus improve hardware Trojan horse feature recognition capability.Compared with traditional IC Hardware Trojan detecting method, improve detection accuracy.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of IC Hardware Trojan detecting method in an embodiment;
Fig. 2 is the structural drawing of self-organizing feature map neural network in an embodiment;
Fig. 3 is that the bypass data of Extraction parts chip to be measured in an embodiment carries out unsupervised learning cluster analysis, obtains the process flow diagram of the first cluster and the second cluster;
Fig. 4 is the structural drawing of radial base neural net in an embodiment;
Fig. 5 carries out supervised learning according to the mark value of the bypass data in wooden horse chip cluster and non-wooden horse chip cluster and correspondence in one embodiment, obtains the process flow diagram of the neural network after training;
Fig. 6 is the structural drawing of IC Hardware Trojan horse detection system in an embodiment;
Fig. 7 is the structural drawing of data categorization module in an embodiment;
Fig. 8 is the structural drawing of an embodiment learning training module.
Embodiment
For enabling above-mentioned purpose of the present invention, feature and advantage become apparent more, are described in detail the specific embodiment of the present invention below in conjunction with accompanying drawing.Set forth a lot of detail in the following description so that fully understand the present invention.But the present invention can be much different from alternate manner described here to implement, those skilled in the art can when without prejudice to doing similar improvement when intension of the present invention, therefore the present invention is by the restriction of following public specific embodiment.
Unless otherwise defined, all technology used herein and scientific terminology are identical with belonging to the implication that those skilled in the art of the present invention understand usually.The object of term used in the description of the invention herein just in order to describe specific embodiment is not be to limit the present invention.
A kind of IC Hardware Trojan detecting method, as shown in Figure 1, comprises the following steps:
Step S110: carry out multiparameter bypass test to all chips to be measured, obtains the bypass data of all chips to be measured.
Bypass data can be regarded as the multidimensional characteristic vectors be made up of information such as the maximum operation frequency of chip to be measured, time delay, power consumption, electromagnetism and thermal effect.By carrying out multiparameter bypass test to chip to be measured, obtain the bypass data of chip to be measured, as the data basis distinguishing wooden horse chip and non-wooden horse chip.
Particularly, can be using the multidimensional original number that records directly as the bypass data of chip to be measured, also can be carry out dimension-reduction treatment to the multidimensional original number recorded, using the characteristic that obtains after the dimension-reduction treatment bypass data as chip to be measured.
Step S120: the bypass data of Extraction parts chip to be measured carries out unsupervised learning cluster analysis, obtains the first cluster and the second cluster.
The bypass data of Extraction parts chip to be measured, it can be the bypass data of the chip to be measured obtaining predetermined number at random, also can be the data extracting chip to be measured by default ratio at random, such as, extract the data of the chip to be measured of in all chips to be measured 20% or 30%.Unsupervised learning cluster analysis is carried out to the bypass data extracted, obtains the first cluster and the second cluster.
Wherein in an embodiment, unsupervised learning cluster analysis is completed by self-organizing feature map neural network.In self-organizing feature map neural network, contiguous neuron can identify part contiguous in the input space, such self-organizing feature map neural network not only can learn the distribution situation inputted as Self-organizing Competitive Neutral Net, can also learn the topological structure carrying out training input vector.Self-organizing feature map neural network is a kind of unsupervised clustering, and the cluster centre that it is formed can be mapped in a curved surface or plane, and keeps topological structure constant.Because self-organizing feature map neural network comparatively meets the classification demand of IC Hardware trojan horse detection reality, precision and the speed of cluster analysis can be improved, thus improve detection accuracy and detection efficiency.
As shown in Figure 2, the basic structure of self-organizing feature map neural network is made up of input layer and competition layer.Input neuron number is M, and competition layer is made up of m*n neuron, forms a two-dimensional planar array.Have two kinds of connection weights in network, a kind of is the connection weights of neuron to outside input reaction, and another kind is the connection weights between neuron, and its size controls the interactive size between neuron.For given input pattern, training process not only will regulate each connection weights of competition winning unit, and will regulate the weights of neighborhood unit of winning unit.
As shown in Figure 3, step S120 comprises step S122 to step S126.
Step S122: according to the competition winning unit of the bypass data determination competition layer of the part chip to be measured extracted.
Particularly, wherein in an embodiment, step S122 comprises step 12 to step 18.
Step 12: the connection weights of each unit of initialization competition layer.Initialization process is carried out to the connection weights of each unit of competition layer, gives the random value in [0,1] interval.
Step 14: using the bypass data of the part chip to be measured of extraction as the input of input layer, obtain input pattern.Using the input vector of the bypass data of each chip to be measured in the part chip to be measured of extraction as the input layer of self-organizing feature map neural network, each input vector of input layer forms fixing input pattern.
Step 16: calculate the distance between the connection weights of each unit of competition layer and input pattern.The distance between the connection weights of each unit of competition layer and input pattern can be calculated according to Euclidean distance formula.
Step 18: extract and connect weights and input pattern apart from minimum unit as competing winning unit.In each distance value step 16 calculated, connect weights and input pattern apart from minimum unit as competing winning unit.
Step S124: the connection weights of competition winning unit and neighborhood unit thereof are upgraded.Be specially
ω ji(k+1)=ω ji(k)+Δω ji
Wherein, ω jik () represents the connection weights before upgrading, ω ji(k+1) the connection weights after upgrading are represented, η cand η nrepresent the weights learning factor of competition winning unit and neighborhood unit respectively, span is (0,1), and η c> η n, x i, ω jirepresent the input amendment of competition winning unit respectively and connect weights.
Upgrade, until all input patterns are all by calculating one time according to the connection weights circulation of above formula to competition winning unit and neighborhood unit thereof.
Step S126: according to upgrading the competition winning unit after connecting weights and neighborhood unit thereof, the bypass data of the part chip to be measured extracted being classified, obtains the first cluster and the second cluster.
Each time to connection right value update, the position that the input vector of input layer is mapped to each unit in competition layer all can change, the neighborhood unit of the learning compete winning unit of self-organizing feature map neural network is encouraged, and unit far away outside neighborhood unit is suppressed.Competition winning unit after utilizing connection right value update to terminate and neighborhood unit thereof, classify to the bypass message of chip to be measured.Wherein in an embodiment, step S126 comprises step 22 to step 24.
Step 22: extract and be distributed in competition layer, and be less than or equal to the bypass data of predeterminable range with the distance of each unit of competition layer, obtain the first cluster.The value of predeterminable range can adjust according to actual conditions, specifically can in advance with each unit of competition layer for the center of circle, be that radius determines field with predeterminable range.The bypass data being positioned at field is divided into the first cluster.
Step 24: extract and be distributed in competition layer, and be greater than the bypass data of predeterminable range with the distance of each unit of competition layer, obtain the second cluster.The bypass data be positioned at outside field is divided into the second cluster.
After self-organizing feature map neural network cluster analysis, only need to carry out converse works analyzing to determine non-wooden horse chip cluster and wooden horse chip cluster to a few sampling chip, the number of chips of reverse-engineering can be made greatly to reduce, thus save time and cost.
Self-organizing feature map neural network amplifies the small different characteristic between wooden horse chip and non-wooden horse chip bypass message, common trait reduces, namely suppress the impact of common-mode signal, amplify difference mode signal, thus improve IC Hardware trojan horse detection resolution and accuracy of identification.In order to verify the correctness of self-organizing feature map neural network cluster analysis, multidimensional raw data corresponding respectively with 18 wooden horse sample chips for 18 non-wooden horse sample chips is used for cluster analysis, and the output obtained is [0 000000000000000 00 11111111111111111 1] t, visible unsupervised learning can carry out cluster accurately and effectively.
Step S130: the part sample extracted respectively in the first cluster and the second cluster carries out converse works analyzing, carries out identification to the first cluster and the second cluster, obtains wooden horse chip cluster and non-wooden horse chip cluster also arranges mark value respectively.
Converse works analyzing ensure that the accuracy of classification identification.Extracting the part sample in the first cluster and the second cluster, can be the sample of predetermined number in acquisition two clusters, equally also can be extract in setting ratio.The sample utilizing extraction to obtain carries out converse works analyzing, can be specifically that sample domain and original circuit design domain are compared, if domain unanimously, judges that the cluster belonging to it is non-wooden horse cluster, be wooden horse chip cluster by another cluster judgment, vice versa.Arrange mark value respectively to the wooden horse chip cluster obtained and non-wooden horse chip cluster, in the present embodiment, non-wooden horse chip cluster can be labeled as " 0 " class, and wooden horse chip cluster can be labeled as " 1 " class simultaneously.Just trojan horse detection is completed to the chip to be measured being used as unsupervised learning cluster analysis thus.
Step S140: the mark value according to the bypass data in wooden horse chip cluster and non-wooden horse chip cluster and correspondence carries out supervised learning, obtains the neural network after training.
Wherein in an embodiment, neural network is radial base neural net.Radial base neural net is the two-layer feedforward network of one with single hidden layer, and input layer is made up of signal node, and the unit number of hidden layer needs according to described problem and determines, and output layer makes response to input action.Input layer is nonlinear to the spatial alternation of hidden layer, transforming function transformation function is RBF (Radial Basis Function, radial basis function) function, it is a kind of local distribution, non-negative nonlinear function to the decay of central point radial symmetry, and hidden layer is linear to the spatial alternation of output layer.The output that is input to of network is nonlinear, and output is linear for adjustable parameter, therefore the weights of network just directly can be solved by system of linear equations or with LMS (Least mean square, least mean-square error) method calculating, thus greatly accelerate pace of learning and avoid local minimum problem.
As shown in Figure 4, suppose that network input x is M dimensional vector, exporting y is L dimensional vector, and sample is N to number.RBF functional form is more, adopts Gauss's activation function in ability embodiment, and namely the output of i-th hidden layer node can be expressed as:
u i ( x ) = exp [ - ( x - c i ) T ( x - c i ) 2 δ i 2 ] , i = 1,2 , . . . , q
Wherein u ibe the output valve of i-th hidden layer node, δ ibe the generalized constant of i-th hidden layer node, q is hidden layer node number, x=(x 1, x 2..., x m) tinput amendment, c ibe the center vector of i-th hidden layer node Gaussian function, this vector is the column vector identical with the dimension of input amendment x, i.e. a c i=(c i1, c i2..., c iM) t.By the visible node of above formula output area between zero and one, and input amendment is the closer to the center of node, and output valve is larger.Gaussian function advantage is adopted to be: form is simple, and radial symmetry is good, and slickness is good, there is arbitrary order derivative, and analyticity is good.
Hidden layer adopts linear mapping to output layer, namely
y k = Σ t = 1 q ω ki u i - θ k , k = 1,2 , . . . , L
Y kthe output of a kth output layer node, ω kithe weighting coefficient of hidden layer to output layer, θ kit is the threshold value of hidden layer.
Radial base neural net learning process will make total error function minimum exactly, and total error function is
J = Σ p = 1 N J p = 1 2 Σ p = 1 N Σ k = 1 L ( t k p - y k p ) 2 = 1 2 Σ p = 1 N Σ k = 1 L e k 2
J is total error function, t kand y kbe respectively and export expectation value and actual value.
Specifically as shown in Figure 5, step S140 comprises the following steps:
Step S142: initialization process is carried out to the center vector of each node of hidden layer.Give an initial center vector to each node of hidden layer, complete initialization process.
Step S144: using the bypass data in wooden horse chip cluster and non-wooden horse chip cluster as input amendment, calculates the distance of each node and input amendment in hidden layer.Be specially
d i ( k ) = | | x ( k ) - c i ( k - 1 ) | | , 1 ≤ i ≤ q d min ( k ) = min d i ( k ) = d r ( k )
Wherein, d ik () represents the center vector c of input amendment x (k) and each node i(k-1) distance, q is node number; d rk () represents the minor increment of input amendment and node, vectorial c centered by r i(k-1) nearest with input amendment x (k) hidden node sequence number.
Step S146: the center vector of each node of hidden layer is upgraded and calculates clustering result quality, until clustering result quality is less than or equal to outage threshold.Be specially
c i ( k ) = c i ( k - 1 ) , 1 ≤ i ≤ q , i ≠ r c r ( k ) = c r ( k - 1 ) + β ( k ) [ x ( k ) - c r ( k - 1 ) ]
β(k)=β(k-1)/(1+int(k/q)) 1/2
J e = Σ i = 1 q | | x ( k ) - c i ( k ) | | 2 ≤ ϵ
Wherein, x (k) is input amendment, c i(k-1), c icenter vector k () is respectively other node updates except the node nearest with input amendment before and after upgrading, c r(k-1), c r(k) represent respectively with the nearest node updates of input amendment before and center vector after upgrading, β (k-1), β (k) represents respectively upgrade before with renewal after learning rate, int () function representation rounding operation, J efor clustering result quality, ε is outage threshold.
By carrying out to all input amendment the center vector that cluster tries to achieve each node of hidden layer, adopt C-means clustering method adjustment center vector, input vector is divided into some data group, in each data group, find out the center vector of a radial basis function, make each sample vector in this data group minimum apart from the distance at this data group center.
Step S148: using wooden horse chip cluster and mark value corresponding to non-wooden horse chip cluster as output sample, according to input amendment and output sample training by the weights between hidden layer to output layer, obtains the radial base neural net after training.Be specially
ω ki(k+1)=ω ki(k)+η(t k-y k)u i/(u Tu)
u=(u 1,u 2,...,u q) T
Wherein, η is learning rate, usually gets 0< η <1; t kand y kbe respectively and export expectation value and actual value, ui is the output valve of i-th hidden layer node, ω ki(k), ω ki(k+1) weights between the hidden layer after training front and training to output layer are represented respectively.
After weights between the center vector determining each node of hidden layer and hidden layer to output layer, output valve that equally can be corresponding by the bypass data in non-wooden horse chip cluster is labeled as " 0 " class, and wooden horse chip cluster is labeled as " 1 " class.In order to verify the correctness of radial base neural net pattern-recognition, multidimensional raw data corresponding respectively with 6 wooden horse chips to be measured for 6 non-wooden horse chips to be measured is used for analyzing, and the output obtained is [00 000011111 1] t, visible supervised learning can carry out pattern-recognition accurately and effectively.
Above-described self-organizing feature map neural network and radial base neural net not only may be used for the cluster analysis of the characteristic after dimensionality reduction, also can be used for the analysis of multidimensional raw data.In fact, self-organizing feature map neural network and radial base neural net have contained the process of feature extraction, and therefore data analysis is front without the need to carrying out the pre-service such as feature extraction, thus improves data-handling efficiency.Self-organizing feature map neural network and radial base neural net can realize fast based on Matlab software, and without the need to artificial experience intervention, automatic business processing improves recognition efficiency.
Be appreciated that in a comparatively detailed embodiment, in step S120, carry out unsupervised learning cluster analysis by self-organizing feature map neural network, and carry out supervised learning in step S140 and obtain radial base neural net.Training sample due to radial base neural net derives from the data after self-organizing feature map neural network cluster analysis, and therefore supervised learning effect can be more accurate.
Step S150: trojan horse detection is carried out to the chip to be measured not being used as unsupervised learning cluster analysis according to neural network.
Be not used for carrying out the input of data as neural network of unsupervised learning cluster analysis in the bypass data obtained by step S110, the output valve according to neural network judges.Specifically can by pre-setting judgment threshold, as 0.5.Output valve is more than or equal to 0.5, and when being less than or equal to 1, corresponding input value is judged to be the bypass data of wooden horse chip, i.e. " 1 " class; Output valve is more than or equal to 0, and when being less than 0.5, corresponding input value is judged to be the bypass data of non-wooden horse chip, i.e. " 0 " class.The trojan horse detection to residue chip to be measured can be completed thus.
Said integrated circuit hardware Trojan horse detection method, unsupervised learning and supervised learning algorithm are merged mutually, unsupervised learning is utilized to carry out cluster analysis to bypass data, utilize the mark value of supervised learning to the bypass data in wooden horse chip cluster and non-wooden horse chip cluster and correspondence to carry out training and obtain neural network, workload and the cost of converse works analyzing can be reduced by the method for sampling analysis after first cluster, improve analysis efficiency, data after cluster analysis are used for the accuracy that supervised learning can improve study, thus improve hardware Trojan horse feature recognition capability.Compared with traditional IC Hardware Trojan detecting method, improve detection accuracy.
A kind of IC Hardware Trojan horse detection system, as shown in Figure 6, comprises data acquisition module 110, data categorization module 120, clustering recognition module 130, learning training module 140 and wooden horse test module 150.
Data acquisition module 110, for carrying out multiparameter bypass test to all chips to be measured, obtains the bypass data of all chips to be measured.
Bypass data can be regarded as the multidimensional characteristic vectors be made up of information such as the maximum operation frequency of chip to be measured, time delay, power consumption, electromagnetism and thermal effect.By carrying out multiparameter bypass test to chip to be measured, obtain the bypass data of chip to be measured, as the data basis distinguishing wooden horse chip and non-wooden horse chip.
Particularly, can be using the multidimensional original number that records directly as the bypass data of chip to be measured, also can be carry out dimension-reduction treatment to the multidimensional original number recorded, using the characteristic that obtains after the dimension-reduction treatment bypass data as chip to be measured.
Data categorization module 120 carries out unsupervised learning cluster analysis for the bypass data of Extraction parts chip to be measured, obtains the first cluster and the second cluster.
The bypass data of Extraction parts chip to be measured, it can be the bypass data of the chip to be measured obtaining predetermined number at random, also can be the data extracting chip to be measured by default ratio at random, such as, extract the data of the chip to be measured of in all chips to be measured 20% or 30%.Unsupervised learning cluster analysis is carried out to the bypass data extracted, obtains the first cluster and the second cluster.
Wherein in an embodiment, unsupervised learning cluster analysis is for complete by self-organizing feature map neural network.As shown in Figure 7, data categorization module 120 comprises processing unit 122, updating block 124 and taxon 126.
Processing unit 122 is for the competition winning unit of the bypass data determination competition layer according to the part chip to be measured extracted.
Particularly, processing unit 122 determines that the step of the competition winning unit of competition layer comprises:
The connection weights of each unit of initialization competition layer.Initialization process is carried out to the connection weights of each unit of competition layer, gives the random value in [0,1] interval.
Using the bypass data of the part chip to be measured of extraction as the input of input layer, obtain input pattern.Using the input vector of the bypass data of each chip to be measured in the part chip to be measured of extraction as the input layer of self-organizing feature map neural network, each input vector of input layer forms fixing input pattern.
Calculate the distance between the connection weights of each unit of competition layer and input pattern.The distance between the connection weights of each unit of competition layer and input pattern can be calculated according to Euclidean distance formula.
Extract and connect weights and input pattern apart from minimum unit as competing winning unit.By in each distance value of calculating, connect weights and input pattern apart from minimum unit as competing winning unit.
Updating block 124 is for upgrading the connection weights of competition winning unit and neighborhood unit thereof.Be specially
ω ji(k+1)=ω ji(k)+Δω ji
Wherein, ω jik () represents the connection weights before upgrading, ω ji(k+1) the connection weights after upgrading are represented, η cand η nrepresent the weights learning factor of competition winning unit and neighborhood unit respectively, span is (0,1), and η c> η n, x i, ω jirepresent the input amendment of competition winning unit respectively and connect weights.
Upgrade, until all input patterns are all by calculating one time according to the connection weights circulation of above formula to competition winning unit and neighborhood unit thereof.
Taxon 126, for according to upgrading the competition winning unit and neighborhood unit thereof that connect weights, being classified to the bypass data of the part chip to be measured extracted, being obtained the first cluster and the second cluster.
Each time to connection right value update, the position that the input vector of input layer is mapped to each unit in competition layer all can change, the neighborhood unit of the learning compete winning unit of self-organizing feature map neural network is encouraged, and unit far away outside neighborhood unit is suppressed.Competition winning unit after utilizing connection right value update to terminate and neighborhood unit thereof, classify to the bypass message of chip to be measured.
Particularly, the step that the bypass data of taxon 126 to the part chip to be measured extracted is classified comprises:
Extraction is distributed in competition layer, and is less than or equal to the bypass data of predeterminable range with the distance of each unit of competition layer, obtains the first cluster.The value of predeterminable range can adjust according to actual conditions, specifically can in advance with each unit of competition layer for the center of circle, be that radius determines field with predeterminable range.The bypass data being positioned at field is divided into the first cluster.
Extraction is distributed in competition layer, and is greater than the bypass data of predeterminable range with the distance of each unit of competition layer, obtains the second cluster.The bypass data be positioned at outside field is divided into the second cluster.
After self-organizing feature map neural network cluster analysis, only need to carry out converse works analyzing to determine non-wooden horse chip cluster and wooden horse chip cluster to a few sampling chip, the number of chips of reverse-engineering can be made greatly to reduce, thus save time and cost.Self-organizing feature map neural network amplifies the small different characteristic between wooden horse chip and non-wooden horse chip bypass message, common trait reduces, namely suppress the impact of common-mode signal, amplify difference mode signal, thus improve IC Hardware trojan horse detection resolution and accuracy of identification.
Clustering recognition module 130 carries out converse works analyzing for the part sample extracted respectively in the first cluster and the second cluster, carries out identification to the first cluster and the second cluster, obtains wooden horse chip cluster and non-wooden horse chip cluster and arranges mark value respectively.
Converse works analyzing ensure that the accuracy of classification identification.Extracting the part sample in the first cluster and the second cluster, can be the sample of predetermined number in acquisition two clusters, equally also can be extract in setting ratio.The sample utilizing extraction to obtain carries out converse works analyzing, can be specifically that sample domain and original circuit design domain are compared, if domain unanimously, judges that the cluster belonging to it is non-wooden horse cluster, be wooden horse chip cluster by another cluster judgment, vice versa.Arrange mark value respectively to the wooden horse chip cluster obtained and non-wooden horse chip cluster, in the present embodiment, non-wooden horse chip cluster can be labeled as " 0 " class, and wooden horse chip cluster can be labeled as " 1 " class simultaneously.Just trojan horse detection is completed to the chip to be measured being used as unsupervised learning cluster analysis thus.
Learning training module 140, for carrying out supervised learning according to the mark value of the bypass data in wooden horse chip cluster and non-wooden horse chip cluster and correspondence, obtains the neural network after training.
Wherein in an embodiment, neural network is radial base neural net.As shown in Figure 8, learning training module 140 comprises initialization unit 142, computing unit 144, control module 146 and training unit 148.
Initialization unit 142 is for carrying out initialization process to the center vector of each node of hidden layer.Give an initial center vector to each node of hidden layer, complete initialization process.
Computing unit 144 for using the bypass data in wooden horse chip cluster and non-wooden horse chip cluster as input amendment, calculate the distance of each node and described input amendment in hidden layer.Be specially
d i ( k ) = | | x ( k ) - c i ( k - 1 ) | | , 1 &le; i &le; q d min ( k ) = min d i ( k ) = d r ( k )
Wherein, d ik () represents the center vector c of input amendment x (k) and each node i(k-1) distance, q is node number; d rk () represents the minor increment of input amendment and node, vectorial c centered by r i(k-1) nearest with input amendment x (k) hidden node sequence number.
Control module 146 is for upgrading and calculate clustering result quality the center vector of each node of hidden layer, until clustering result quality is less than or equal to outage threshold.Be specially
c i ( k ) = c i ( k - 1 ) , 1 &le; i &le; q , i &NotEqual; r c r ( k ) = c r ( k - 1 ) + &beta; ( k ) [ x ( k ) - c r ( k - 1 ) ]
β(k)=β(k-1)/(1+int(k/q)) 1/2
J e = &Sigma; i = 1 q | | x ( k ) - c i ( k ) | | 2 &le; &epsiv;
Wherein, x (k) is input amendment, c i(k-1), c icenter vector k () is respectively other node updates except the node nearest with input amendment before and after upgrading, c r(k-1), c r(k) represent respectively with the nearest node updates of input amendment before and center vector after upgrading, β (k-1), β (k) represents respectively upgrade before with renewal after learning rate, int () function representation rounding operation, J efor clustering result quality, ε is outage threshold.
By carrying out to all input amendment the center vector that cluster tries to achieve each node of hidden layer, adopt C-means clustering method adjustment center vector, input vector is divided into some data group, in each data group, find out the center vector of a radial basis function, make each sample vector in this data group minimum apart from the distance at this data group center.
Training unit 148 for using wooden horse chip cluster and mark value corresponding to non-wooden horse chip cluster as output sample, according to input amendment and output sample training by the weights between hidden layer to output layer, obtain the radial base neural net after training.Be specially
ω ki(k+1)=ω ki(k)+η(t k-y k)u i/(u Tu)
u=(u 1,u 2,...,u q) T
Wherein, η is learning rate, usually gets 0< η <1; t kand y kbe respectively and export expectation value and actual value, u ibe the output valve of i-th hidden layer node, ω ki(k), ω ki(k+1) weights between the hidden layer after training front and training to output layer are represented respectively.
After weights between the center vector determining each node of hidden layer and hidden layer to output layer, output valve that equally can be corresponding by the bypass data in non-wooden horse chip cluster is labeled as " 0 " class, and wooden horse chip cluster is labeled as " 1 " class.
Above-described self-organizing feature map neural network and radial base neural net not only may be used for the cluster analysis of the characteristic after dimensionality reduction, also can be used for the analysis of multidimensional raw data.In fact, self-organizing feature map neural network and radial base neural net have contained the process of feature extraction, and therefore data analysis is front without the need to carrying out the pre-service such as feature extraction, thus improves data-handling efficiency.Self-organizing feature map neural network and radial base neural net can realize fast based on Matlab software, and without the need to artificial experience intervention, automatic business processing improves recognition efficiency.
Be appreciated that data categorization module 120 carries out unsupervised learning cluster analysis by self-organizing feature map neural network in a comparatively detailed embodiment, and learning training module 140 is carried out supervised learning and is obtained radial base neural net.Training sample due to radial base neural net derives from the data after self-organizing feature map neural network cluster analysis, and therefore supervised learning effect can be more accurate.
Wooden horse test module 150 is for carrying out trojan horse detection according to neural network to the chip to be measured not being used as unsupervised learning cluster analysis.
Be not used for carrying out the input of data as neural network of unsupervised learning cluster analysis in the bypass data obtained by data acquisition module 110, the output valve according to neural network judges.Specifically can by pre-setting judgment threshold, as 0.5.Output valve is more than or equal to 0.5, and when being less than or equal to 1, corresponding input value is judged to be the bypass data of wooden horse chip, i.e. " 1 " class; Output valve is more than or equal to 0, and when being less than 0.5, corresponding input value is judged to be the bypass data of non-wooden horse chip, i.e. " 0 " class.The trojan horse detection to residue chip to be measured can be completed thus.
Said integrated circuit hardware Trojan horse detection system, unsupervised learning and supervised learning algorithm are merged mutually, unsupervised learning is utilized to carry out cluster analysis to bypass data, utilize the mark value of supervised learning to the bypass data in wooden horse chip cluster and non-wooden horse chip cluster and correspondence to carry out training and obtain neural network, workload and the cost of converse works analyzing can be reduced by the method for sampling analysis after first cluster, improve analysis efficiency, data after cluster analysis are used for the accuracy that supervised learning can improve study, thus improve hardware Trojan horse feature recognition capability.Compared with traditional IC Hardware Trojan detecting method, improve detection accuracy.
The above embodiment only have expressed several embodiment of the present invention, and it describes comparatively concrete and detailed, but therefore can not be interpreted as the restriction to the scope of the claims of the present invention.It should be pointed out that for the person of ordinary skill of the art, without departing from the inventive concept of the premise, can also make some distortion and improvement, these all belong to protection scope of the present invention.Therefore, the protection domain of patent of the present invention should be as the criterion with claims.

Claims (10)

1. an IC Hardware Trojan detecting method, is characterized in that, comprises the following steps:
Multiparameter bypass test is carried out to all chips to be measured, obtains the bypass data of all chips to be measured;
The bypass data of Extraction parts chip to be measured carries out unsupervised learning cluster analysis, obtains the first cluster and the second cluster;
The part sample extracted respectively in described first cluster and the second cluster carries out converse works analyzing, carries out identification to described first cluster and the second cluster, obtains wooden horse chip cluster and non-wooden horse chip cluster and arranges mark value respectively;
Mark value according to the bypass data in described wooden horse chip cluster and non-wooden horse chip cluster and correspondence carries out supervised learning, obtains the neural network after training;
According to described neural network, trojan horse detection is carried out to the chip to be measured not being used as unsupervised learning cluster analysis.
2. IC Hardware Trojan detecting method according to claim 1, it is characterized in that, using the bypass data of the multidimensional raw data of described chip to be measured as described chip to be measured, maybe using the characteristic that obtains after carrying out dimension-reduction treatment to the described multidimensional raw data bypass data as described chip to be measured.
3. IC Hardware Trojan detecting method according to claim 1, is characterized in that, described unsupervised learning cluster analysis is completed by self-organizing feature map neural network; The bypass data of described Extraction parts chip to be measured carries out unsupervised learning cluster analysis, obtains the step of the first cluster and the second cluster, as follows:
According to the competition winning unit of the bypass data determination competition layer of the part chip to be measured extracted;
The connection weights of described competition winning unit and neighborhood unit thereof are upgraded, is specially
ω ji(k+1)=ω ji(k)+Δω ji
Wherein, ω jik () represents the connection weights before upgrading, ω ji(k+1) the connection weights after upgrading are represented, η cand η nrepresent the weights learning factor of competition winning unit and neighborhood unit respectively, span is (0,1), and η c> η n, x i, ω jirepresent the input amendment of competition winning unit respectively and connect weights;
According to upgrading the competition winning unit after connecting weights and neighborhood unit thereof, the bypass data of the part chip to be measured extracted being classified, obtains described first cluster and the second cluster.
4. IC Hardware Trojan detecting method according to claim 3, is characterized in that, the step of the competition winning unit of the bypass data determination competition layer of the described part chip to be measured according to extracting, comprises the following steps:
The connection weights of each unit of competition layer described in initialization;
Using the bypass data of the part chip to be measured of extraction as the input of input layer, obtain input pattern;
Calculate the distance between the connection weights of each unit of described competition layer and described input pattern;
Extract and connect weights and described input pattern apart from minimum unit as described competition winning unit.
5. IC Hardware Trojan detecting method according to claim 3, it is characterized in that, described according to upgrading the competition winning unit and neighborhood unit thereof that connect weights, the bypass data of the part chip to be measured extracted is classified, obtain the step of described first cluster and the second cluster, specifically comprise the following steps:
Extraction is distributed in described competition layer, and is less than or equal to the bypass data of predeterminable range with the distance of each unit of described competition layer, obtains described first cluster;
Extraction is distributed in described competition layer, and is greater than the bypass data of predeterminable range with the distance of each unit of described competition layer, obtains described second cluster.
6. IC Hardware Trojan detecting method according to claim 1, is characterized in that, described neural network is radial base neural net; The described mark value according to the bypass data in described wooden horse chip cluster and non-wooden horse chip cluster and correspondence carries out supervised learning, obtains the step of the neural network after training, as follows:
Initialization process is carried out to the center vector of each node of hidden layer;
Using the bypass data in described wooden horse chip cluster and non-wooden horse chip cluster as input amendment, calculate the distance of each node and described input amendment in hidden layer, be specially
d i ( k ) = | | x ( k ) - c i ( k - 1 ) | | , 1 &le; i &le; q d min ( k ) = min d i ( k ) = d r ( k )
Wherein, d ik () represents the center vector c of input amendment x (k) and each node i(k-1) distance, q is node number; d rk () represents the minor increment of input amendment and node, vectorial c centered by r i(k-1) nearest with input amendment x (k) hidden node sequence number;
The center vector of each node of hidden layer is upgraded and calculates clustering result quality, until described clustering result quality is less than or equal to outage threshold, is specially
c i ( k ) = c i ( k - 1 ) , 1 &le; i &le; q , i &NotEqual; r c r ( k ) = c r ( k - 1 ) + &beta; ( k ) [ x ( k ) - c r ( k - 1 ) ]
β(k)=β(k-1)/(1+int(k/q)) 1/2
J e = &Sigma; i = 1 q | | x ( k ) - c i ( k ) | | 2 &le; &epsiv;
Wherein, x (k) is input amendment, c i(k-1), c icenter vector k () is respectively other node updates outside the node nearest with input amendment before and after upgrading, c r(k-1), c r(k) represent respectively with the nearest node updates of input amendment before and center vector after upgrading, β (k-1), β (k) represents respectively upgrade before with renewal after learning rate, int () function representation rounding operation, J efor clustering result quality, ε is outage threshold;
Using described wooden horse chip cluster and mark value corresponding to non-wooden horse chip cluster as output sample, according to described input amendment and output sample training by the weights between hidden layer to output layer, obtain the radial base neural net after training, be specially
ω ki(k+1)=ω ki(k)+η(t k-y k)u i/(u Tu)
u=(u 1,u 2,...,u q) T
Wherein, η is learning rate, usually gets 0< η <1; t kand y kbe respectively and export expectation value and actual value, u ibe the output valve of i-th hidden layer node, ω ki(k), ω ki(k+1) weights between the hidden layer after training front and training to output layer are represented respectively.
7. an IC Hardware Trojan horse detection system, is characterized in that, comprising:
Data acquisition module, for carrying out multiparameter bypass test to all chips to be measured, obtains the bypass data of all chips to be measured;
Data categorization module, the bypass data for Extraction parts chip to be measured carries out unsupervised learning cluster analysis, obtains the first cluster and the second cluster;
Clustering recognition module, carries out converse works analyzing for the part sample extracted respectively in described first cluster and the second cluster, carries out identification to described first cluster and the second cluster, obtains wooden horse chip cluster and non-wooden horse chip cluster and arranges mark value respectively;
Learning training module, for carrying out supervised learning according to the mark value of the bypass data in described wooden horse chip cluster and non-wooden horse chip cluster and correspondence, obtains the neural network after training;
Wooden horse test module, for carrying out trojan horse detection according to described neural network to the chip to be measured not being used as unsupervised learning cluster analysis.
8. IC Hardware Trojan horse detection system according to claim 7, it is characterized in that, described data acquisition module using the bypass data of the multidimensional original number of described chip to be measured as described chip to be measured, maybe using the characteristic that obtains after carrying out dimensionality reduction to the described multidimensional original number bypass data as described chip to be measured.
9. IC Hardware Trojan horse detection system according to claim 7, is characterized in that, described unsupervised learning cluster analysis is completed by self-organizing feature map neural network; Described data categorization module comprises:
Processing unit, for the competition winning unit of the bypass data determination competition layer according to the part chip to be measured extracted;
Updating block, for upgrading the connection weights of described competition winning unit and neighborhood unit thereof, is specially
ω ji(k+1)=ω ji(k)+Δω ji
Wherein, ω jik () represents the connection weights before upgrading, ω ji(k+1) the connection weights after upgrading are represented, η cand η nrepresent the weights learning factor of competition winning unit and neighborhood unit respectively, span is (0,1), and η c> η n, x i, ω jirepresent the input amendment of competition winning unit respectively and connect weights;
Taxon, for according to upgrading the competition winning unit and neighborhood unit thereof that connect weights, classifying to the bypass data of the part chip to be measured extracted, obtaining described first cluster and the second cluster.
10. IC Hardware Trojan horse detection system according to claim 7, is characterized in that, described learning training module comprises:
Initialization unit, for carrying out initialization process to the center vector of each node of hidden layer;
Computing unit, for using the bypass data in described wooden horse chip cluster and non-wooden horse chip cluster as input amendment, calculate the distance of each node and described input amendment in hidden layer, be specially
d i ( k ) = | | x ( k ) - c i ( k - 1 ) | | , 1 &le; i &le; q d min ( k ) = min d i ( k ) = d r ( k )
Wherein, d ik () represents the center vector c of input amendment x (k) and each node i(k-1) distance, q is node number; d rk () represents the minor increment of input amendment and node, vectorial c centered by r i(k-1) nearest with input amendment x (k) hidden node sequence number;
Control module, for upgrading the center vector of each node of hidden layer and calculate clustering result quality, until described clustering result quality is less than or equal to outage threshold, is specially
c i ( k ) = c i ( k - 1 ) , 1 &le; i &le; q , i &NotEqual; r c r ( k ) = c r ( k - 1 ) + &beta; ( k ) [ x ( k ) - c r ( k - 1 ) ]
β(k)=β(k-1)/(1+int(k/q)) 1/2
J e = &Sigma; i = 1 q | | x ( k ) - c i ( k ) | | 2 &le; &epsiv;
Wherein, x (k) is input amendment, c i(k-1), c icenter vector k () is respectively other node updates except the node nearest with input amendment before and after upgrading, c r(k-1), c r(k) represent respectively with the nearest node updates of input amendment before and center vector after upgrading, β (k-1), β (k) represents respectively upgrade before with renewal after learning rate, int () function representation rounding operation, J efor clustering result quality, ε is outage threshold;
Training unit, for using described wooden horse chip cluster and mark value corresponding to non-wooden horse chip cluster as output sample, train by the weights between hidden layer to output layer according to described input amendment and output sample, obtain the radial base neural net after training, be specially
ω ki(k+1)=ω ki(k)+η(t k-y k)u i/(u Tu)
u=(u 1,u 2,...,u q) T
Wherein, η is learning rate, usually gets 0< η <1; t kand y kbe respectively and export expectation value and actual value, u ibe the output valve of i-th hidden layer node, ω ki(k), ω ki(k+1) weights between the hidden layer after training front and training to output layer are represented respectively.
CN201410598741.4A 2014-10-29 2014-10-29 IC Hardware Trojan detecting method and system Active CN104330721B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410598741.4A CN104330721B (en) 2014-10-29 2014-10-29 IC Hardware Trojan detecting method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410598741.4A CN104330721B (en) 2014-10-29 2014-10-29 IC Hardware Trojan detecting method and system

Publications (2)

Publication Number Publication Date
CN104330721A true CN104330721A (en) 2015-02-04
CN104330721B CN104330721B (en) 2017-03-08

Family

ID=52405484

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410598741.4A Active CN104330721B (en) 2014-10-29 2014-10-29 IC Hardware Trojan detecting method and system

Country Status (1)

Country Link
CN (1) CN104330721B (en)

Cited By (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104636687A (en) * 2015-03-02 2015-05-20 中国电子科技集团公司第五十八研究所 Circuit design method capable of improving hardware Trojan horse detection distinguishability and hardware Trojan horse detection method
CN104950246A (en) * 2015-06-11 2015-09-30 工业和信息化部电子第五研究所 Hardware trojan detection method and system based on time delay
CN105158674A (en) * 2015-08-27 2015-12-16 工业和信息化部电子第五研究所 Hardware Trojan detection method by means of parasitic effect and system thereof
CN105807204A (en) * 2016-03-08 2016-07-27 天津大学 Spectrum refinement-based hardware Trojan detection method
CN105893876A (en) * 2016-03-28 2016-08-24 工业和信息化部电子第五研究所 Chip hardware Trojan horse detection method and system
CN106250378A (en) * 2015-06-08 2016-12-21 腾讯科技(深圳)有限公司 Public identifier sorting technique and device
CN106706546A (en) * 2016-12-28 2017-05-24 中山市腾创贸易有限公司 Analysis method for artificial intelligence learning materials on basis of infrared and Raman spectrum data
CN106841987A (en) * 2017-01-25 2017-06-13 天津大学 Hardware Trojan horse side channel detection method based on electromagnetism and electric current
CN107703186A (en) * 2017-09-26 2018-02-16 电子科技大学 Hardware Trojan horse detection method based on chip temperature field-effect
CN108052840A (en) * 2017-11-13 2018-05-18 天津大学 Hardware Trojan horse detection method based on neutral net
CN108154051A (en) * 2017-11-23 2018-06-12 天津科技大学 A kind of hardware Trojan horse detection method of discrimination based on support vector machines
CN108446555A (en) * 2018-02-11 2018-08-24 复旦大学 The method that hardware Trojan horse is monitored in real time and is detected
CN108460058A (en) * 2017-02-22 2018-08-28 北京京东尚科信息技术有限公司 Data processing method and system
CN108846283A (en) * 2018-06-15 2018-11-20 北京航空航天大学 A kind of hardware Trojan horse real-time detecting system and its design method
CN109740348A (en) * 2019-01-29 2019-05-10 福州大学 A kind of hardware Trojan horse localization method based on machine learning
CN110287735A (en) * 2019-07-04 2019-09-27 电子科技大学 Wooden horse based on chip netlist feature infects circuit identification method
CN110363033A (en) * 2018-04-09 2019-10-22 国民技术股份有限公司 A kind of chip security appraisal procedure and device
CN113553630A (en) * 2021-06-15 2021-10-26 西安电子科技大学 Hardware Trojan horse detection system based on unsupervised learning and information data processing method
US11170106B2 (en) 2018-05-10 2021-11-09 Robotic Research, Llc System for detecting hardware trojans in integrated circuits
US11321463B2 (en) 2020-01-09 2022-05-03 Rockwell Collins, Inc. Hardware malware profiling and detection system
US11372981B2 (en) 2020-01-09 2022-06-28 Rockwell Collins, Inc. Profile-based monitoring for dual redundant systems
CN117034374A (en) * 2023-08-28 2023-11-10 绍兴龙之盾网络信息安全有限公司 LM-BPNN hardware Trojan detection method and system based on PSO

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5530753A (en) * 1994-08-15 1996-06-25 International Business Machines Corporation Methods and apparatus for secure hardware configuration
CN102654560A (en) * 2012-04-20 2012-09-05 南开大学 Nondestructive detection system for hardware trojan in integrated circuit
CN103198251A (en) * 2013-03-28 2013-07-10 哈尔滨工业大学(威海) Hardware Trojan horse recognition method based on neural network
US20130204553A1 (en) * 2011-08-03 2013-08-08 President And Fellows Of Harvard College System and method for detecting integrated circuit anomalies
CN103488941A (en) * 2013-09-18 2014-01-01 工业和信息化部电子第五研究所 Hardware Trojan horse detection method and hardware Trojan horse detection system
CN103698687A (en) * 2013-12-18 2014-04-02 工业和信息化部电子第五研究所 Method and system for processing signals of hardware Trojan detection in integrated circuit

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5530753A (en) * 1994-08-15 1996-06-25 International Business Machines Corporation Methods and apparatus for secure hardware configuration
US20130204553A1 (en) * 2011-08-03 2013-08-08 President And Fellows Of Harvard College System and method for detecting integrated circuit anomalies
CN102654560A (en) * 2012-04-20 2012-09-05 南开大学 Nondestructive detection system for hardware trojan in integrated circuit
CN103198251A (en) * 2013-03-28 2013-07-10 哈尔滨工业大学(威海) Hardware Trojan horse recognition method based on neural network
CN103488941A (en) * 2013-09-18 2014-01-01 工业和信息化部电子第五研究所 Hardware Trojan horse detection method and hardware Trojan horse detection system
CN103698687A (en) * 2013-12-18 2014-04-02 工业和信息化部电子第五研究所 Method and system for processing signals of hardware Trojan detection in integrated circuit

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
赵崇征等: "基于旁路分析的集成电路芯片硬件木马检测", 《微电子学与计算机》 *

Cited By (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104636687A (en) * 2015-03-02 2015-05-20 中国电子科技集团公司第五十八研究所 Circuit design method capable of improving hardware Trojan horse detection distinguishability and hardware Trojan horse detection method
CN104636687B (en) * 2015-03-02 2017-12-22 中国电子科技集团公司第五十八研究所 Improve the circuit design method and hardware Trojan horse detection method of hardware Trojan horse detection resolution
CN106250378A (en) * 2015-06-08 2016-12-21 腾讯科技(深圳)有限公司 Public identifier sorting technique and device
CN106250378B (en) * 2015-06-08 2020-06-02 腾讯科技(深圳)有限公司 Public identification classification method and device
CN104950246A (en) * 2015-06-11 2015-09-30 工业和信息化部电子第五研究所 Hardware trojan detection method and system based on time delay
CN104950246B (en) * 2015-06-11 2017-09-19 工业和信息化部电子第五研究所 Hardware Trojan horse detection method and system based on delay
CN105158674B (en) * 2015-08-27 2018-07-20 工业和信息化部电子第五研究所 Utilize the hardware Trojan horse detection method and system of ghost effect
CN105158674A (en) * 2015-08-27 2015-12-16 工业和信息化部电子第五研究所 Hardware Trojan detection method by means of parasitic effect and system thereof
CN105807204A (en) * 2016-03-08 2016-07-27 天津大学 Spectrum refinement-based hardware Trojan detection method
CN105807204B (en) * 2016-03-08 2018-07-27 天津大学 Hardware Trojan horse detection method based on frequency spectrum refinement
CN105893876A (en) * 2016-03-28 2016-08-24 工业和信息化部电子第五研究所 Chip hardware Trojan horse detection method and system
CN106706546A (en) * 2016-12-28 2017-05-24 中山市腾创贸易有限公司 Analysis method for artificial intelligence learning materials on basis of infrared and Raman spectrum data
CN106841987A (en) * 2017-01-25 2017-06-13 天津大学 Hardware Trojan horse side channel detection method based on electromagnetism and electric current
CN106841987B (en) * 2017-01-25 2019-09-13 天津大学 Hardware Trojan horse side channel detection method based on electromagnetism and electric current
CN108460058A (en) * 2017-02-22 2018-08-28 北京京东尚科信息技术有限公司 Data processing method and system
CN107703186A (en) * 2017-09-26 2018-02-16 电子科技大学 Hardware Trojan horse detection method based on chip temperature field-effect
CN108052840A (en) * 2017-11-13 2018-05-18 天津大学 Hardware Trojan horse detection method based on neutral net
CN108154051A (en) * 2017-11-23 2018-06-12 天津科技大学 A kind of hardware Trojan horse detection method of discrimination based on support vector machines
CN108446555A (en) * 2018-02-11 2018-08-24 复旦大学 The method that hardware Trojan horse is monitored in real time and is detected
CN110363033A (en) * 2018-04-09 2019-10-22 国民技术股份有限公司 A kind of chip security appraisal procedure and device
US11170106B2 (en) 2018-05-10 2021-11-09 Robotic Research, Llc System for detecting hardware trojans in integrated circuits
CN108846283A (en) * 2018-06-15 2018-11-20 北京航空航天大学 A kind of hardware Trojan horse real-time detecting system and its design method
CN108846283B (en) * 2018-06-15 2021-11-02 北京航空航天大学 Hardware trojan real-time detection system and design method thereof
CN109740348A (en) * 2019-01-29 2019-05-10 福州大学 A kind of hardware Trojan horse localization method based on machine learning
CN109740348B (en) * 2019-01-29 2022-06-14 福州大学 Hardware Trojan horse positioning method based on machine learning
CN110287735A (en) * 2019-07-04 2019-09-27 电子科技大学 Wooden horse based on chip netlist feature infects circuit identification method
CN110287735B (en) * 2019-07-04 2021-05-04 电子科技大学 Trojan horse infected circuit identification method based on chip netlist characteristics
US11321463B2 (en) 2020-01-09 2022-05-03 Rockwell Collins, Inc. Hardware malware profiling and detection system
US11372981B2 (en) 2020-01-09 2022-06-28 Rockwell Collins, Inc. Profile-based monitoring for dual redundant systems
CN113553630A (en) * 2021-06-15 2021-10-26 西安电子科技大学 Hardware Trojan horse detection system based on unsupervised learning and information data processing method
CN117034374A (en) * 2023-08-28 2023-11-10 绍兴龙之盾网络信息安全有限公司 LM-BPNN hardware Trojan detection method and system based on PSO

Also Published As

Publication number Publication date
CN104330721B (en) 2017-03-08

Similar Documents

Publication Publication Date Title
CN104330721A (en) Integrated circuit hardware Trojan horse detection method and integrated circuit hardware Trojan horse detection system
Ferreira et al. Time series clustering via community detection in networks
CN105389583A (en) Image classifier generation method, and image classification method and device
CN103473540B (en) The modeling of intelligent transportation system track of vehicle increment type and online method for detecting abnormality
US10627446B2 (en) Importance sampling method for multiple failure regions
WO2016091084A1 (en) Complex network-based high speed train system safety evaluation method
CN102629305B (en) Feature selection method facing to SNP (Single Nucleotide Polymorphism) data
CN110175434A (en) A kind of rail fastener system injury detection method based on convolutional neural networks
CN104316861B (en) integrated circuit hardware Trojan detection method and system
CN101404071B (en) Electronic circuit fault diagnosis neural network method based on grouping particle swarm algorithm
CN107579846B (en) Cloud computing fault data detection method and system
CN109934269A (en) A kind of opener recognition methods of electromagnetic signal and device
CN109272502B (en) PCB hardware safety detection method based on temperature field effect
CN102842043B (en) Particle swarm classifying method based on automatic clustering
CN103268431A (en) Cancer hypotype biomarker detecting system based on student t distribution
CN102509123A (en) Brain functional magnetic resonance image classification method based on complex network
CN107679453A (en) Weather radar electromagnetic interference echo recognition methods based on SVMs
CN104850867A (en) Object identification method based on intuitive fuzzy c-means clustering
CN104777418A (en) Analog circuit fault diagnosis method based on depth Boltzman machine
CN108877947A (en) Depth sample learning method based on iteration mean cluster
CN104809476A (en) Multi-target evolutionary fuzzy rule classification method based on decomposition
CN108470163B (en) Rail turnout plate gap defect detection method and terminal equipment
CN103455612A (en) Method for detecting non-overlapping network communities and overlapping network communities based on two-stage strategy
CN112231775B (en) Hardware Trojan horse detection method based on Adaboost algorithm
CN111983569A (en) Radar interference suppression method based on neural network

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant