Side-Channel Power Analysis Based on SA-SVM

Round 1

Reviewer 1 Report

In this paper, authors have proposed a simulated annealing optimization SVM parameter (SA-SVM) model in the side-channel power analysis. The accuracy is used as the optimization objective of simulated annealing to find the optimal penalty factor ? and the optimal parameters in different kernels. The results show that the SA-SVM model improves the accuracy of the optimization search by 0.25% to 3.25% and reduces the time required by 39.96%-98.02% compared to the SVM model when the interest point is 53. Moreover, the SA-SVM model recovery key only requires three power traces. This manuscript seems interesting, however, there are major concerns which should be addressed carefully. Please find comments as below:

1. Abstract should elaborate on novelty and best findings of the study.

2. Introduction should be improved in light with the most important and recent articles in the field.

3. Novelty should be improved substantially.

4. Each reference should be discussed separately.

5. The accuracy and validity should be improved to convince the usefulness. 

6. Results should be supported scientifically.

7. Comprehensive proofread is essential.

8. Quality of figures should be improved.

9. What percentage of data is selected for validation purpose?

 

 

Author Response

Dear reviewer,

Re: Manuscript ID: applsci-2334046 and Title: Side-channel power analysis based on SA-SVM

 

Thanks for your comments concerning our manuscript. Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper. The responds to the reviewer’s comments are as following:

 

1. Comment: Abstract should elaborate on novelty and best findings of the study.

Response: In order to address your suggestion, we have revised the abstract as follows:

 

“Support Vector Machine (SVM) has been widely used in Side-channel power analysis. The selection of penalty factor and kernel parameter heavily influences how well support vector machines work. Setting reasonable SVM hyperparameters is a key issue in Side-channel power analysis. The novel side channel power analysis method SA-SVM, which combines simulated annealing (SA) and support vector machines (SVM) to analyze the power traces and crack the key, is pro-posed in this paper as a solution to this issue. This method differs from other approaches in that it integrates SA and SVM, enabling us to more effectively explore the search space and produce su-perior results. In this paper, we conduct experiments on SA-SVM and SVM models from three different aspects: the selection of kernel functions, the number of parameters, and the number of eigenvalues. To compare with previous research, we perform experimental evaluations on open datasets. The results indicate that, compared to the SVM model, the SA-SVM model improves the accuracy by 0.25% to 3.25% and reduces the required time by 39.96% to 98.02% when the point of interest is 53, recovering the key using only three power traces. The SA-SVM model outperforms existing methods in terms of accuracy and computation time.”

 

2. Comment: Introduction should be improved in light with the most important and recent articles in the field.

Response: 

1. We have further revised the introduction. Too long to display here.
2. b) We now provide some more up-to-date references including:

“

Reference replacement:

[2] The EM side-channel(s)(2002)→Far Field EM Side-Channel Attack on AES Using Deep Learning(2020)

[7] A New Class of Collision Attacks and Its Application to DES(2003)→An Efficient Collision Power Attack on AES Encryption in Edge Computing(2019)

[8] Correlation Power Analysis with a Leakage Model(2004)→ Improved Correlation Power Analysis on Bitslice Block Ciphers（2022）

[9] Template Attacks(2002)→ Efficient, Portable Template Attacks.(2018)

[13] K-Nearest Neighbors Algorithm in Profiling Power Analysis Attacks(2016)→ Machine Learning-Based Similarity Attacks for Chaos-Based Cryptosystem(2021)

[15] Optimization of Power Analysis Using Neural Network(2011)→Deep Learning Side-Channel Attack against Hardware Implementations of AES(2021)

 

References added:

[21] Efficient Framework for Genetic Algorithm-Based Correlation Power Analysis

[22] A Neural Network Trojan Detection Method Based on Particle Swarm Optimization

[31] Comparing Support Vector Machines with Gaussian Kernels to Radial Basis Function Classifiers

”

 

3. Comment: Novelty should be improved substantially.

Response: Thank you for pointing that out. Previous works have conducted extensive work on support vector machine attacks in side-channel power analysis. However, previous works did not specifically address the problem of parameter selection in support vector machines. Our work fills this gap.

 

4. Comment: Each reference should be discussed separately.

Response: I have revised the introduction to provide a discussion and description of every reference cited.

 

5. Comment: The accuracy and validity should be improved to convince the usefulness.

Response: We used public datasets to ensure experiment reproducibility and supplemented some dataset details in the paper.

 

6. Comment: Results should be supported scientifically.

Response: We conducted a comparative analysis of two optimization methods for SVM, grid search and simulated annealing. And discussed our findings in Results and Analysis. Furthermore, to reinforce our conclusions, we compared them with those of existing studies. We aim to rephrase the experimental results in a clear and concise manner.

 

7. Comment: Comprehensive proofread is essential.

Response:  We have carefully considered your suggestion and have proofread our manuscript accordingly.

 

8. Comment: Quality of figures should be improved.

Response: We will bundle and send images in PDF format.

 

9. Comment: What percentage of data is selected for validation purpose?

Response: In the article, we divided the dataset into two-thirds as the training set and one-third as the test set. However, due to our negligence, we did not emphasize that cross-validation instead of validation was used in the training set. We now add more detail with respect to our dataset:

“We have chosen 1000 power traces from DPAv4 for our experimental dataset, and two-thirds of them are used as the learning set while the remaining one-third is used for the testing set. The learning set is partitioned into three equal subsets by 3-fold cross-validation. We train using two subsets at a time, and the other subset for validation, repeating this process until three subsets have been used for validation. This process generates three models, and we choose the parameters of the model with the highest accuracy and evaluated its effectiveness on the testing set.”

 

We are extremely grateful to Reviewer for pointing out these problems. If you have any questions, please contact us without hesitate.

 

Yours sincerely,

Ying Zhang

Reviewer 2 Report

applsci-2334046

Title: Side-channel power analysis based on SA-SVM

Indeed, the manuscript is well-written and easy to follow. Some points need to be known.

-It will be good to include brief information on SA-SVM.

-Please explain Table 1 more clearly.

-It will be good to include the reference for equation 11 also.

-The novelty of the work should be clearly highlighted (in the Abstract and in the conclusions).

 

-It is better to list a comparison table to compare results with previous work.

Author Response

Thanks for your comments concerning our manuscript. Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper. The responds to the reviewer’s comments are as following:

 

1. Comment: It will be good to include brief information on SA-SVM.

Response: We have added the following information about SA-SVM in the introduction:

“The SA-SVM model uses a certain probability of accepting negative increments to jump out of local optima and find optimal parameters more easily.”

 

2. Comment: Please explain Table 1 more clearly.

Response: We add it as follows:

“We recorded the peak points in Figure 1 in Table 1. There are 192 points of interest with absolute values of correlation coefficient larger than 0.5, and 4 points of interest with absolute values greater than 0.9. Having fewer number of interest points (POIs) is not necessarily better. Although reducing the number of POIs speeds up computation, decreases information redundancy, it also results in the loss of useful information. Choosing appropriate number can help us more effectively crack the key.”

 

3. Comment: It will be good to include the reference for equation 11 also.

Response: We have updated our manuscript by including references for Equation 11. The specific references are as follows:

" [31] Comparing Support Vector Machines with Gaussian Kernels to Radial Basis Function Classifiers."

 

4. Comment: The novelty of the work should be clearly highlighted (in the Abstract and in the conclusions).

Response: I have translated the modified Abstract and Conclusion as follows:

“Abstract: Support Vector Machine (SVM) has been widely used in Side-channel power analysis. The selection of penalty factor and kernel parameter heavily influences how well support vector machines work. Setting reasonable SVM hyperparameters is a key issue in Side-channel power analysis. The novel side channel power analysis method SA-SVM, which combines simulated annealing (SA) and support vector machines (SVM) to analyze the power traces and crack the key, is pro-posed in this paper as a solution to this issue. This method differs from other approaches in that it integrates SA and SVM, enabling us to more effectively explore the search space and produce su-perior results. In this paper, we conduct experiments on SA-SVM and SVM models from three different aspects: the selection of kernel functions, the number of parameters, and the number of eigenvalues. To compare with previous research, we perform experimental evaluations on open datasets. The results indicate that, compared to the SVM model, the SA-SVM model improves the accuracy by 0.25% to 3.25% and reduces the required time by 39.96% to 98.02% when the point of interest is 53, recovering the key using only three power traces. The SA-SVM model outperforms existing methods in terms of accuracy and computation time.”

“Conclusions: This paper presents a SA-SVM model for side-channel power analysis. The approach looks for continuous decision variables and optimizes SVM parameter values to achieve superior classification results. Experiments are conducted on DPAv4.1, testing the model under various POIs and kernel functions. Compared with the SVM grid search method, SA-SVM improves accuracy by approximately 0.25-3.25% while reducing running time by up to 98.02%. The study compares linear, RBF, and three kinds of wavelet kernels, revealing that wavelet kernels have higher accuracy than RBF while requiring only 3 traces to recover the key. The combination of SA and SVM provides a new method for addressing the parameter selection challenge in side-channel power analysis.”

 

5. Comment: It is better to list a comparison table to compare results with previous work.

Response: In the discussion, we compared previous works under similar conditions with the same dataset and similar research interests.

Table 4. Machine learning, DPAv4

Ref.	MLSCA	No. of traces	No. of POIs	Acc (%)
[14]	NB	1000	50	37.6
[14]	MLP	1000	50	44.8
[14]	XGBoost	1000	50	52.0
[14]	RF	1000	50	49.2
[14]	CNN	1000	50	60.4
[12]	RF_SMOTE	2000	-	93.0
this Article	SA-SVM	1000	53	94.5

 

 

We are extremely grateful to Reviewer for pointing out these problems. If you have any questions, please contact us without hesitate.

 

Yours sincerely,

Ying Zhang

Reviewer 3 Report

The topic is novel, and with fewer articles talking about this area, more clarification should be included about side-channel power analysis first in the introduction. Regarding the methods and materials section, it is recommended to have a little discussion about the PDA context versions, why you chose V1 and not V2 in the fourth version, and if there are any alternatives for databases for experiment testing because PDA v4 was created in 2014.

You mentioned in Figure 3 (the side-channel power attack process based on SA-SVM) that you pre-processed the figure; could you exhibit the original figure or at least talk about the circumstances under which you processed it?

A list of abbreviations should be included.

 

Many references are relatively new; we need to add a few more extra new references to make them recent (>2019) and old.

* Comments included in the original attached manuscript. 

Comments for author File: Comments.pdf

Author Response

Dear reviewer,

Re: Manuscript ID: applsci-2334046 and Title: Side-channel power analysis based on SA-SVM

 

Thanks for your comments concerning our manuscript. Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper. The responds to the reviewer’s comments are as following:

 

1. Comment: The topic is novel, and with fewer articles talking about this area, more clarification should be included about side-channel power analysis first in the introduction.

Response: Provide further elaboration on the subject of side-channel power analysis in the introduction. The specific content is as follows:

 

“Kocher et al. [5] suggested power consumption assaults attack for the first time in the late 1990s. It is a branch of side-channel attacks that targets devices by measuring power consumption. Kocher proposed that the classic Differential Power Analysis (DPA)had successfully cracked the DES algorithm key. They reveal that there is some relation between power consumption and data when the device is encrypted. And the relation contains encrypted device key data that can be used to crack the key. By analyzing the power consumption of a device during encryption or decryption, it is possible to de-duce the key used. In order to carry out this type of attack, a computer uses an encryption device and inputs a set of known plaintexts into the device for encryption. As the device performs encryption, an oscilloscope measures power consumption, thus obtaining power traces.”

 

2. Comment: Regarding the methods and materials section, it is recommended to have a little discussion about the PDA context versions, why you chose V1 and not V2 in the fourth version, and if there are any alternatives for databases for experiment testing because PDA v4 was created in 2014.

Response:

1. a) The difference between V1 and V2: V1 uses AES-256 algorithm and tracks only the beginning of the first and second rounds of AES, with only 435,000 sampling points per traces. V2 uses AES-128 algorithm. The trace contains a complete encryption, which has 1,704,402 samples of the trace. Dataset V1 is similar to dataset V2, but choosing V1 will improve efficiency
2. b) SCA has relatively few public datasets, among which DPA contest is the most widely used and can be compared with previous work.
3. c) Not all datasets can be replaced. Because some datasets adopt protection measures such as random delay. This method only discusses unprotected usage scenarios.

We have added some explanations as follows:

 

“DPA Contest is a globally recognized standard competition in the field of crypto-graphic security, and the latest version is DPA Contest V4. Since complete encryption will not be used in this experiment, DPA Contest V4.1 has been selected as the dataset.”

 

3. Comment: You mentioned in Figure 3 (the side-channel power attack process based on SA-SVM) that you pre-processed the figure; could you exhibit the original figure or at least talk about the circumstances under which you processed it?

Response: Data pre-processing is completed in section 2.1, so it is not shown in figure 3. Figure 1 shows the raw data and the processed data. To prevent any possible misunderstandings, we have included a description here. We have added some explanations as follows:

 

“The data processing section has been explained in section 2.1.”

       

4. Comment: A list of abbreviations should be included.

Response: We have carefully considered your advice and compiled the following list of abbreviations. The location to add the abbreviation table will depend on the editor.

 

Abbreviation Comparison Table

ACC	Accuracy
ACO	Ant Colony Optimization
AES	Advanced Encryption Standard
CA	Collision Attacks
CNN	Convolutional Neural Network
CPA	Correlation Power Analysis
DES	Data Encryption Standard
DPA	Differential Power Analysis
DPAv4	DPA Contest v4
GA	Genetic Algorithm
GE	Guessing Entropy
KNN	K-Nearest Neighbor
HW	Hamming Weight
MIA	Mutual Information Analysis
MLP	Multi-layer Perceptron
MLSCA	Machine Learning-based Side Channel Analysis
PoIs	Points of Interest
PSO	Particle Swarm Optimization
RBF	Radial Basis Function
RF	Random Forest
SA	Simulated annealing
S-Box	Substitution-Box
SCA	Side Channel Analysis
SPA	Simple Power Analysis
SRM	Structural Risk Minimization
SVM	Support Vector Machine
TA	Template Attacks
XGBoost	eXtreme Gradient Boosting

 

5. Comment: Many references are relatively new; we need to add a few more extra new references to make them recent (>2019) and old.

Response: We have updated the references with the latest sources.

“

Reference replacement:

[2] The EM side-channel(s)(2002)→Far Field EM Side-Channel Attack on AES Using Deep Learning(2020)

[7] A New Class of Collision Attacks and Its Application to DES(2003)→An Efficient Collision Power Attack on AES Encryption in Edge Computing(2019)

[8] Correlation Power Analysis with a Leakage Model(2004)→ Improved Correlation Power Analysis on Bitslice Block Ciphers（2022）

[9] Template Attacks(2002)→ Efficient, Portable Template Attacks.(2018)

[13] K-Nearest Neighbors Algorithm in Profiling Power Analysis Attacks(2016)→ Machine Learning-Based Similarity Attacks for Chaos-Based Cryptosystem(2021)

[15] Optimization of Power Analysis Using Neural Network(2011)→Deep Learning Side-Channel Attack against Hardware Implementations of AES(2021)

 

References added:

[21] Efficient Framework for Genetic Algorithm-Based Correlation Power Analysis

[22] A Neural Network Trojan Detection Method Based on Particle Swarm Optimization

[31] Comparing Support Vector Machines with Gaussian Kernels to Radial Basis Function Classifiers

”

We are extremely grateful to Reviewer for pointing out these problems. If you have any questions, please contact us without hesitate.

 

Yours sincerely,

Ying Zhang

 

Round 2

Reviewer 1 Report

Accept.

Reviewer 2 Report

applsci-2334046

Side-channel power analysis based on SA-SVM

Thank you for allowing me to revise resubmitted manuscript titled " Side-channel power analysis based on SA-SVM" I believe the submitted manuscript and presented work is suitable for publishing in Applied Sciences, except for one minor revision.

 

1-      Please improve the quality of all the figures.

 

Author Response

Dear reviewer,

Re: Manuscript ID: applsci-2334046 and Title: Side-channel power analysis based on SA-SVM

 

Comment: Please improve the quality of all the figures.

Response: We have improved the quality of figures 2 and figures 3, and changed the colors of figure 4 to make it more distinguishable.

 

Regards,

Ying Zhang