Strong PUF (Physical Unclonable Function) machine learning attack resisting method based on dynamic matrix encryption

Abstract

The invention discloses a strong PUF (Physical Unclonable Function) machine learning attack resisting method based on dynamic matrix encryption, which comprises the following steps of: when incentives are confused, performing transformation operation on a magic cube matrix based on the current incentives to generate an encryption matrix, and constructing an incentive matrix based on the current incentives; determining a matrix multiplication mode of the encryption matrix and the excitation matrix based on the encryption matrix, multiplying the encryption matrix and the excitation matrix to obtain a confusion matrix, converting the confusion matrix into 0 or 1 according to the parity of elements in the confusion matrix to obtain an encryption excitation matrix, and finally extracting the elements in the encryption excitation matrix to form encryption excitation, if the current is the first PUF response generating the PUF response sequence, the magic cube matrix generating the encryption excitation is generated randomly, and if the current is not the first PUF response generating the PUF response sequence, the magic cube matrix generating the encryption excitation is the encryption matrix generated last time; the method has the advantages of being high in machine learning attack resistance and capable of ensuring that the strong PUF has high stability.

Description

Technical Field

[0001] The present invention relates to a strong PUF anti-machine learning attack method based on dynamic matrix encryption, and in particular to a strong PUF anti-machine learning attack method. Background Art

[0002] Physically Unclonable Function (PUF) uses tiny inhomogeneities in the chip manufacturing process to provide a unique identity for each device. It has the advantages of low cost, low power consumption, and no need for additional storage space, so it has broad application prospects in the field of IoT security.

[0003] According to the number of stimulus response pairs (CRPs) generated, PUFs can be divided into two categories: strong PUFs and weak PUFs. Weak PUFs usually only generate a small number of CRPs, which can be used as keys for encryption systems. Strong PUFs can provide a large number of unique CRPs, making them suitable for lightweight device authentication. However, the current strong PUFs are vulnerable to machine learning attacks, that is, attackers can build a PUF model by collecting a certain number of CRPs. Taking the 64-bit APUF disclosed in the literature 1 (X Ma, P Wang, G Li, Z Zhou, "Machine Learning Attacks Resistant Strong PUF Design Utilizing Response Obfuscates Challenge with Lower Hardware Overhead" [J]. Microelectronics Journal, vol. 142, 2023.) as an example, when about 1000 CRPs are used for modeling attacks, the prediction accuracy of the training model exceeds 95%; when 10,000 CRPs are used, the prediction accuracy is as high as 99%.

[0004] In recent years, in order to resist ML modeling attacks, many strong PUF anti-machine learning attack methods have been proposed. These strong PUF anti-machine learning attack methods confuse the stimulus response and complicate the mapping relationship between stimulus and response, thereby preventing attackers from collecting valid CRP to perform ML modeling attacks on PUF. The XOR-APUF disclosed in document 2 (NN Anandakumar, M S Hashmi and M A Chaudhary, "Implementation of Efficient XOR Arbiter PUF on FPGA With Enhanced Uniqueness and Security" [J] IEEE Access, vol. 10, 2022 pp. 129832-129842.), the MPUF disclosed in document 3 (DP Sahoo, D. Mukhopadhyay, RS Chakraborty and P H Nguyen, "A Multiplexer-Based Arbiter PUF Composition with Enhanced Reliability and Security" [J] IEEE Transactions on Computers, vol. 67, no. 3, 2018, pp. 403-41.), and the MPUF disclosed in document 4 (W Xu, L Pang, Y Tang, and M Chen, "Security Evaluation of Feed-Forward Interpose PUF Against Modelling Attacks" [C] 2024 IEEE 4th The iPUF disclosed in International Conference on Power, Electronics and Computer Applications (ICPECA), Shenyang, China, 2024, pp.871-877.) is a typical example.

[0005] However, the current strong PUF anti-machine learning attack methods have certain limitations. For obfuscated stimuli, when the obfuscation algorithm and structure are made public, attackers can establish a relationship between the obfuscated stimulus and the response, thereby bypassing the obfuscation and directly attacking the strong PUF. For obfuscated responses, the final response is obtained by obfuscating the responses of multiple independent strong PUFs. However, each strong PUF will be affected by certain noise, usually including temperature changes, voltage changes, etc. When the responses of multiple strong PUFs are obfuscated, the instability of the response of any strong PUF may lead to the instability of the final response, so the noise will be amplified, resulting in a decrease in the stability of the strong PUF. In addition, for obfuscated stimuli and obfuscated responses, the current response is determined by the current stimulus, which is similar to the characteristics of combinational logic circuits. The complexity of obfuscation is limited, so it is difficult to resist high-order machine learning attacks. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a strong PUF anti-machine learning attack method based on dynamic matrix encryption, which has strong anti-machine learning attack capability and can ensure that the strong PUF has high stability.

[0007] The technical solution adopted by the present invention to solve the above technical problems is: a strong PUF anti-machine learning attack method based on dynamic matrix encryption, when the strong PUF needs to generate a PUF response sequence, before each PUF response of the PUF response sequence is generated, the current stimulus of the strong PUF is first confused to generate an encrypted stimulus, so that the strong PUF generates a PUF response under the action of the encrypted stimulus, and each encrypted stimulus is generated in the following manner: first, a magic cube matrix is ​​transformed based on the current stimulus to generate an encryption matrix, and an incentive matrix is ​​constructed based on the current stimulus, and then the matrix multiplication method of the encryption matrix and the incentive matrix is ​​determined based on the encryption matrix, and then the encryption matrix and the incentive matrix are multiplied to obtain a confusion matrix, and the parity of the elements in the confusion matrix is ​​converted to 0 or 1 to obtain an encrypted incentive matrix, and finally the elements in the encrypted incentive matrix are extracted to form an encrypted incentive, wherein, if this is the first PUF response to generate the PUF response sequence, the magic cube matrix for generating the encrypted incentive this time is randomly generated, and if this is not the first PUF response to generate the PUF response sequence, the magic cube matrix for generating the encrypted incentive this time is the encryption matrix generated last time.

[0008] The method for resisting machine learning attacks by strong PUF based on dynamic matrix encryption comprises the following steps:

[0009] Step 1: When the strong PUF needs to generate a PUF response sequence, first randomly generate 6 8*8 magic cube matrices for stimulus confusion on the computer side, and record the i-th magic cube matrix as M i , i=1,2,3,4,5,6, the i-th magic cube matrix Mi Each element in is 0 or 1, and the i-th magic cube matrix M i The element in the jth column and the pth row is recorded as j=1,2,3,4,5,6,7,8,p=1,2,3,4,5,6,7,8,then start a stimulus confusion;

[0010] Step 2: Generate a 64-bit stimulus C{c0,c1,…,c 63}, that is, the current stimulus, calculate the Hamming weight of c0 to c4 in stimulus C, and record the Hamming weight as n;

[0011] Step 3. Select the Rubik's Cube Matrix M from the computer. n+1 , and according to the magic cube matrix M n+1 Elements in Determine the i-th magic cube matrix M i The direction of rotation, if is equal to 0, then the magic cube matrix M i Rotate 90 degrees clockwise, if is equal to 1, then the magic cube matrix M i Rotate 90 degrees counterclockwise to make the magic cube matrix M i The matrix obtained after rotation is called its encryption matrix, denoted by M i ', the encrypted matrix M i The element in the jth column and the pth row is recorded as

[0012] Step 4: Generate an incentive matrix based on the current incentive on the computer, where the element in the jth column and the pth row of the incentive matrix is ​​c (j-1)*8+(p-1) , determine the encryption matrix M n+1 ' Is it 0? If is 0, the encrypted matrix M n+1 'Pre-multiply with the excitation matrix, if If it is not 0, the encryption matrix M n+1 'Multiply it with the activation matrix to get the confusion matrix, denoted as H;

[0013] Step 5: Convert the confusion matrix H on the computer to obtain the encrypted excitation matrix H'. Specifically, judge each element in the confusion matrix H. If the element is an odd number, update the value of the element to 1, otherwise update it to 0.

[0014] Step 6: On the computer side, take out the elements in the encrypted incentive matrix H' in the order of the 1st column to the 8th column, and take out the elements of each column in the order of the 1st row to the 8th row, and arrange them from high to low to form 64-bit data. The 64-bit data is the encrypted incentive, recorded as C', and an incentive confusion is completed;

[0015] Step 7: On the computer side, the encrypted stimulus C' obtained by the stimulus confusion is input into the strong PUF as the stimulus signal of the strong PUF, and the strong PUF generates a PUF response output of the PUF response sequence;

[0016] Step 8. On the computer, insert the Rubik's Cube Matrix M i Update to be equal to the encrypted matrix M obtained by this incentive confusion i ', return to step 2 for the next stimulus confusion until the strong PUF generates the required PUF response sequence.

[0017] Compared with the prior art, the advantage of the present invention is that each encrypted excitation is generated in the following manner: first, the magic cube matrix is ​​transformed based on the current excitation to generate an encryption matrix, and an excitation matrix is ​​constructed based on the current excitation. Then, after determining the matrix multiplication method of the encryption matrix and the excitation matrix based on the encryption matrix, the encryption matrix and the excitation matrix are multiplied to obtain a confusion matrix, and the elements in the confusion matrix are converted into 0 or 1 according to the parity of the elements in the confusion matrix to obtain an encrypted excitation matrix. Finally, the elements in the encrypted excitation matrix are extracted to form an encrypted excitation. If this is the first PUF response to generate a PUF response sequence, the magic cube matrix for generating the encrypted excitation is randomly generated. If this is not the first PUF response to generate a PUF response sequence, the magic cube matrix for generating the encrypted excitation is the encryption matrix generated last time. Thus, only the excitation of the strong PUF is confused, and the confusion matrix is ​​not generated. In the obfuscation process, there is no element affected by noise, such as the PUF response output by the strong PUF. Therefore, the obfuscation process of the excitation is stable and reliable, will not be affected by noise, and will not have an adverse effect on the stability of the strong PUF. The stability of the strong PUF is relatively high. At the same time, since the accuracy of machine learning attack modeling is related to the complexity of the mapping relationship between the excitation response, the more complex the mapping relationship, the lower the modeling accuracy. The present invention adopts a method of sequential logic obfuscation, and uses the encryption matrix generated by the previous obfuscation as the magic cube matrix in the next obfuscation, so that the current PUF response of the strong PUF depends not only on the current excitation, but also on the previous excitation. In addition, the matrix multiplication method is used in the obfuscation process, which increases the complexity of the mapping relationship between the excitation and the response. Therefore, the present invention can effectively resist machine learning attacks and has good anti-machine learning attack capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1It is a flow chart of the method for strong PUF anti-machine learning attack based on dynamic matrix encryption of the present invention;

[0019] FIG2( a ) is a graph showing the prediction rate (FPGA) of an APUF with a confusion stimulus and several common strong PUFs against logistic regression attacks using the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention;

[0020] FIG2( b ) is a graph showing the prediction rate (FPGA) of an APUF using the dynamic matrix encryption-based strong PUF anti-machine learning attack method of the present invention and several common strong PUFs against lightweight gradient boosting machine attacks using a confused excitation;

[0021] FIG2( c ) is a diagram showing the prediction rate (FPGA) of an APUF with obfuscated excitation and several common strong PUFs against support vector machine attacks using the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention;

[0022] FIG2( d ) is a diagram showing the prediction rate (FPGA) of an APUF using the dynamic matrix encryption-based strong PUF anti-machine learning attack method of the present invention and several common strong PUFs against artificial neural network attacks;

[0023] Figure 3 The diagram is a statistical distribution diagram of the inter-chip Hamming distance and the intra-chip Hamming distance between PUF responses output by a strong PUF using the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention. DETAILED DESCRIPTION

[0024] The present invention is further described in detail below with reference to the accompanying drawings.

[0025] Embodiment 1: A method for strong PUF to resist machine learning attacks based on dynamic matrix encryption. When the strong PUF needs to generate a PUF response sequence, before each PUF response in the PUF response sequence is generated, the current stimulus of the strong PUF is first confused to generate an encrypted stimulus, so that the strong PUF generates a PUF response under the action of the encrypted stimulus. Each encrypted stimulus is generated in the following manner: first, a magic cube matrix is ​​transformed based on the current stimulus to generate an encryption matrix, and an incentive matrix is ​​constructed based on the current stimulus. Then, after determining the matrix multiplication method of the encryption matrix and the incentive matrix based on the encryption matrix, the encryption matrix and the incentive matrix are multiplied to obtain a confusion matrix, and the elements in the confusion matrix are converted to 0 or 1 according to the parity of the elements, to obtain an encrypted incentive matrix. Finally, the elements in the encrypted incentive matrix are extracted to form an encrypted incentive. If this is the first PUF response to generate the PUF response sequence, the magic cube matrix for generating the encrypted incentive is randomly generated. If this is not the first PUF response to generate the PUF response sequence, the magic cube matrix for generating the encrypted incentive is the encryption matrix generated last time.

[0026] In this embodiment, when performing excitation obfuscation, only the excitation of the strong PUF is obfuscated, and no elements such as the PUF response output by the strong PUF that may be affected by noise participate in the obfuscation process. Therefore, the excitation obfuscation process is stable and reliable, will not be affected by noise, and will not have an adverse effect on the stability of the strong PUF. The stability of the strong PUF is relatively high. At the same time, since the accuracy of machine learning attack modeling is related to the complexity of the mapping relationship between the excitation response, the more complex the mapping relationship, the lower the modeling accuracy. The present invention adopts a method of sequential logic obfuscation to update the magic cube matrix in the next obfuscation to be equal to the encryption matrix generated by the previous obfuscation, so that the current PUF response of the strong PUF depends not only on the current excitation, but also on the previous excitation. In addition, the matrix multiplication method is used in the excitation obfuscation process, which increases the complexity of the mapping relationship between the excitation and the response. Therefore, the present invention can effectively resist machine learning attacks and has good anti-machine learning attack capabilities.

[0027] Embodiment 2: Figure 1 As shown, a strong PUF anti-machine learning attack method based on dynamic matrix encryption includes the following steps:

[0028] Step 1: When the strong PUF needs to generate a PUF response sequence, first randomly generate 6 8*8 magic cube matrices for stimulus confusion on the computer side, and record the i-th magic cube matrix as M i , i=1,2,3,4,5,6, the i-th magic cube matrix M i Each element in is 0 or 1, and the i-th magic cube matrix M i The element in the jth column and the pth row is recorded as j=1,2,3,4,5,6,7,8,p=1,2,3,4,5,6,7,8,then start a stimulus confusion;

[0029] Step 2: Generate a 64-bit stimulus C{c0,c1,…,c 63}, that is, the current stimulus, calculate the Hamming weight of c0 to c4 in stimulus C, and record the Hamming weight as n;

[0030] Step 3. Select the Rubik's Cube Matrix M from the computer. n+1 , and according to the magic cube matrix M n+1 Elements in Determine the i-th magic cube matrix M i The direction of rotation, if is equal to 0, then the magic cube matrix M i Rotate 90 degrees clockwise, if is equal to 1, then the magic cube matrix M i Rotate 90 degrees counterclockwise to make the magic cube matrix Mi The matrix obtained after rotation is called its encryption matrix, denoted by M i ', the encrypted matrix M i The element in the jth column and the pth row is recorded as

[0031] Step 4: Generate an incentive matrix based on the current incentive on the computer, where the element in the jth column and the pth row of the incentive matrix is ​​c (j-1)*8+(p-1) , determine the encryption matrix M n+1 ' Is it 0? If is 0, the encrypted matrix M n+1 'Pre-multiply with the excitation matrix, if If it is not 0, the encryption matrix M n+1 'Multiply it with the activation matrix to get the confusion matrix, denoted as H;

[0032] Step 5: Convert the confusion matrix H on the computer to obtain the encrypted excitation matrix H'. Specifically, judge each element in the confusion matrix H. If the element is an odd number, update the value of the element to 1, otherwise update it to 0.

[0033] Step 6: On the computer side, take out the elements in the encrypted incentive matrix H' in the order of the 1st column to the 8th column, and take out the elements of each column in the order of the 1st row to the 8th row, and arrange them from high to low to form 64-bit data. The 64-bit data is the encrypted incentive, recorded as C', and an incentive confusion is completed;

[0034] Step 7: On the computer side, the encrypted stimulus C' obtained by the stimulus confusion is input into the strong PUF as the stimulus signal of the strong PUF, and the strong PUF generates a PUF response output of the PUF response sequence;

[0035] Step 8. On the computer, insert the Rubik's Cube Matrix M i Update to be equal to the encrypted matrix M obtained by this incentive confusion i ', return to step 2 for the next stimulus confusion until the strong PUF generates the required PUF response sequence.

[0036] In order to verify the performance of the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention, logistic regression attack, lightweight gradient boosting machine attack and support vector machine attack are used to compare and verify the performance of the APUF and several common strong PUFs that are confused and stimulated by the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention in Example 2.

[0037] Figure 2(a) to Figure 2(d)The following are the prediction rate diagrams of APUF, 2XOR-APUF, 4XOR-APUF, 8XOR-APUF and APUF with confusion incentives using the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention under four machine learning attacks. Figure 2(a) to Figure 2(d) It can be seen that the anti-machine learning attack capability of the APUF using the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention is better than that of the traditional APUF, 2XOR-APUF, 4XOR-APUF and 8XOR-APUF, and the attack results show that when the number of stimulus response pairs is as high as 1 million groups, the prediction rates of the four machine learning attacks are all around 50%, which is close to the ideal performance. Therefore, the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention has good anti-machine learning attack capability.

[0038] Figure 3 The statistical distribution diagram of the inter-chip Hamming distance and intra-chip Hamming distance between the responses of the APUF outputted by the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention, wherein the intra-chip Hamming distance collects 50 groups of 2000 CRPs, and the excitation of each group is kept constant; the inter-chip Hamming distance collects 50 groups of 2000 CRPs, and the excitation of each group is kept constant, and each group uses different strong PUFs of the same type. The average value and standard deviation of the inter-chip Hamming distance are 49.84% and 0.000435, respectively, which are close to the ideal values ​​of 50% and 0, indicating that the strong PUF using the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention has good uniqueness. The average value and standard deviation of the intra-chip Hamming distance are 0.295% and 0.000458, respectively, which are close to the ideal value of 0, indicating that the strong PUF using the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention has high reliability.

[0039] The NIST randomness test was performed on the traditional APUF and the APUF using the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention according to the second embodiment. The test data is shown in Table 1:

[0040] Table 1

[0041]

[0042] Table 1 shows the randomness test suite (SP 800-22) proposed by the National Institute of Standards and Technology (NIST). NIST contains 15 tests, and whether each test passes is determined by the relationship between the p-value and the pre-set test significance level: a specified p-value ≥ α indicates that the sequence is random, and vice versa. It is generally recommended to set it to 0.01, and the p-value interval is [0,1]. In the experiment, 1 million randomly generated stimuli are applied to the traditional APUF and the APUF that uses the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention to confuse the stimuli, generating 1 million responses, which are divided into 50 groups (20,000 in each group) for NIST testing.

[0043] From the analysis of Table 1, it can be seen that the APUF with obfuscated excitation using the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention passed 10 random number tests, and the traditional APUF passed 6 tests, indicating that the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention can improve the randomness of the strong PUF.

[0044] The performance of the APUF using the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention and the traditional strong PUF is compared, and the comparison data is shown in Table 2:

[0045] Table 2

[0046]

[0047] In Table 2, (2,1)-MPUF is a 2-to-1 MPUF, which consists of three APUFs and a 2-to-1 path selector, where the outputs of two APUFs are the two input signals of the 2-to-1 path selector, and the output of one APUF is the selection signal of the 2-to-1 path selector. (3,3)-iPUF consists of two different 3XOR-APUFs, one of which is a 64-level 3XOR-APUF and the other is a 65-level 3XOR-APUF. The output of the 64-level 3XOR-APUF is inserted into the middle of the 64-bit stimulus to generate a new 65-bit stimulus, which acts on the 65-level 3XOR-APUF to obtain the final response.

[0048] From the analysis of Table 2, it can be seen that the prediction rates of the four machine learning attacks of the APUF with confusion excitation using the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention are lower than those of other traditional strong PUFs, and are all around 50%. At the same time, its randomness, uniqueness and stability are close to the ideal values, indicating that the strong PUF anti-machine learning attack method based on dynamic matrix encryption of the present invention has good performance.

Claims

1. A method for strong PUF to resist machine learning attacks based on dynamic matrix encryption. When the strong PUF needs to generate a PUF response sequence, each time before generating a PUF response in the PUF response sequence, the current stimulus of the strong PUF is first obfuscated to generate an encrypted stimulus, so that the strong PUF generates a PUF response under the action of the encrypted stimulus, characterized in that Each encryption stimulus is generated in the following way: first, the magic cube matrix is ​​transformed based on the current stimulus to generate an encryption matrix, and an incentive matrix is ​​constructed based on the current stimulus. Then, the matrix multiplication method of the encryption matrix and the incentive matrix is ​​determined based on the encryption matrix. The encryption matrix and the incentive matrix are multiplied to obtain a confusion matrix, and the elements in the confusion matrix are converted to 0 or 1 according to the parity of the elements. The encryption incentive matrix is ​​obtained, and finally the elements in the encryption incentive matrix are extracted to form an encryption stimulus. If this is the first PUF response to generate a PUF response sequence, the magic cube matrix for generating the encryption stimulus is randomly generated. If this is not the first PUF response to generate a PUF response sequence, the magic cube matrix for generating the encryption stimulus is the encryption matrix generated last time.

2. A strong PUF anti-machine learning attack method based on dynamic matrix encryption according to claim 1, characterized in that The following steps are involved: Step 1: When the strong PUF needs to generate a PUF response sequence, first randomly generate 6 8*8 magic cube matrices for stimulus confusion on the computer side, and record the i-th magic cube matrix as M i , i=1,2,3,4,5,6, the i-th magic cube matrix M i Each element in is 0 or 1, and the i-th magic cube matrix M i The element in the jth column and the pth row is recorded as j=1,2,3,4,5,6,7,8,p=1,2,3,4,5,6,7,8,then start a stimulus confusion; Step 2: Generate a 64-bit stimulus C{c0,c1,…,c 63 }, that is, the current stimulus, calculate the Hamming weight of c0 to c4 in stimulus C, and record the Hamming weight as n; Step 3. Select the Rubik's Cube Matrix M from the computer. n+1 , and according to the magic cube matrix M n+1 Elements in Determine the i-th magic cube matrix M i The direction of rotation, if is equal to 0, then the magic cube matrix M i Rotate 90 degrees clockwise, if is equal to 1, then the magic cube matrix M i Rotate 90 degrees counterclockwise to make the magic cube matrix M i The matrix obtained after rotation is called its encryption matrix, denoted by M i ', the encrypted matrix M i The element in the jth column and the pth row is recorded as Step 4: Generate an incentive matrix based on the current incentive on the computer, where the element in the jth column and the pth row of the incentive matrix is ​​c (j-1)*8+(p-1) , determine the encryption matrix M n+1 ' Is it 0? If is 0, the encrypted matrix M n+1 'Pre-multiply with the excitation matrix, if If it is not 0, the encryption matrix M n+1 'Multiply it with the activation matrix to get the confusion matrix, denoted as H; Step 5: Convert the confusion matrix H on the computer to obtain the encrypted excitation matrix H'. Specifically, judge each element in the confusion matrix H. If the element is an odd number, update the value of the element to 1, otherwise update it to 0. Step 6: On the computer side, take out the elements in the encrypted incentive matrix H' in the order of the 1st column to the 8th column, and take out the elements of each column in the order of the 1st row to the 8th row, and arrange them from high to low to form 64-bit data. The 64-bit data is the encrypted incentive, recorded as C', and an incentive confusion is completed; Step 7: On the computer side, the encrypted stimulus C' obtained by the stimulus confusion is input into the strong PUF as the stimulus signal of the strong PUF, and the strong PUF generates a PUF response output of the PUF response sequence; Step 8. On the computer, insert the Rubik's Cube Matrix M i Update to be equal to the encrypted matrix M obtained by this incentive confusion i ', return to step 2 for the next stimulus confusion until the strong PUF generates the required PUF response sequence.

Images