Side channel analysis method and device based on deep learning
By simulating reference leakage, the dependence problem on external physical leakage in the prior art is solved, and efficient side channel analysis without relying on external physical leakage is achieved, which is suitable for horizontal comparison of attack performance for different targets.
Patent Information
- Application Number
- CN202510498059.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-21
- Publication Date
- 2025-08-08
AI Technical Summary
The existing deep learning-based side channel analysis methods rely on external physical leakage, resulting in strong dependence on reference selection of attack results and difficulty in horizontal comparison.
The method of simulating reference leakage is adopted to enhance the deep learning modeling process through simulation leakage, and the simulation leakage is used for regularization and training, so as to realize side channel analysis that does not rely on external physical leakage.
It extends the practical scope of RePA's framework, allowing attackers to conduct effective side channel analysis without obtaining physical reference leaks, which is suitable for horizontal comparison of attack performance for different targets.
Smart Images
Figure CN120454968A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer hardware security, and in particular to a side channel analysis method and device based on deep learning. Background Art
[0002] With the development of the Internet of Things and embedded technology, hardware device security issues are increasingly attracting widespread attention from researchers. When previously secure and reliable cryptographic algorithms are applied to real-world devices, the physical properties generated during operation, such as power consumption and electromagnetic field, are directly correlated with the data used in the algorithm, posing a risk of sensitive data leakage. This approach, which exploits the signal characteristics generated during physical implementation to recover sensitive data, is called side-channel analysis. Modeling side-channel analysis is a two-stage analysis method that assumes the attacker can pre-acquire partial side-channel information to model the target side-channel information. With the recent convergence of deep learning and side-channel analysis, deep learning-based modeling side-channel analysis has become the most powerful analysis method. Within this field, a class of solutions has emerged that enhances the neural network-based modeling process by introducing reference devices or leaks, hereby referred to as RePA (Reference-enhanced Profiling Attack). RePA is an efficient and versatile attack whose greatest advantage is that it helps the neural network learn how to extract leak features, significantly promoting rapid network convergence. This raises the question of what the selection criteria for reference devices or leaks are, and whether this places higher demands on the attacker's capabilities.
[0003] The original RePA, known as CGAN-SCA, required that the reference device run a cryptographic algorithm implementation similar to that on the target device, meaning the attacker knew the target device's algorithm implementation. Subsequently, the newly proposed analysis method, LD-PA, removed this attacker's capability assumption and used a single-variable leak as the reference leak for enhanced modeling. Both approaches used actual, collected physical leaks as the reference leak, inevitably impacting the attack's outcome. Summary of the Invention
[0004] The inventors of this application have discovered through extensive research and practice that the existing methods CGAN-SCA and LD-PA both use actual collected physical leaks as reference leaks. Inevitably, the results of the attack will be affected by the selection of the reference. The reason for this is that the reference party is a new variable, and there are some device-related characteristics in the physical leak that make it beneficial to a specific target device. Therefore, the effect of initiating RePA with the same physical leak reference on different attack targets will be uncertain. If a horizontal comparison of the attack performance of different targets is required, selecting a specific physical leak as the reference for RePA will undoubtedly lead to distortion of the comparison results. If we can get rid of the dependence on external physical leaks and design a standard reference leak, then the previous problem can be easily solved.
[0005] Based on the above considerations, the present invention provides a side channel analysis method and device based on deep learning, which utilizes simulated reference leakage to enhance the deep learning modeling process, thereby realizing side channel analysis. It is a RePA using simulated reference leakage, referred to as SimRePA, which realizes efficient deep learning-based modeling side channel analysis that does not rely on external physical leakage.
[0006] In order to achieve the above objectives, the first aspect of the present invention provides a side channel analysis method based on deep learning, comprising:
[0007] Output simulated leakage according to the input intermediate variables and the selected simulation mode, wherein the intermediate variables are the real leakage values;
[0008] Regularize the transition leakage using the simulated leakage of the output, where the transition leakage is extracted from the input side channel curve by the basic neural network;
[0009] The intermediate value of the leakage is predicted based on the extracted transition leakage, and the cross entropy between the predicted intermediate value and the true leakage value is calculated as the classification loss. The regularization loss and the classification loss are combined to update the network parameters using back propagation to obtain the trained basic neural network for modeling side channel analysis based on deep learning.
[0010] The side channel curve of the unknown intermediate value is input into the trained basic neural network based on deep learning modeling side channel analysis, and the leakage intermediate value is inferred based on its output.
[0011] In one embodiment, the simulated leakage includes a signal part and a noise part, and the selected simulation modes include a pure simulation mode CS assuming that the leakage is noise-free, a full simulation mode FS assuming that the noise is of a specific distribution, and a semi-simulation mode PS assuming that the noise is filtered from the physical leakage.
[0012] In one embodiment, outputting simulation leakage according to the input intermediate variables and the selected simulation mode includes:
[0013] When the selected simulation mode is the pure simulation mode CS in which the leakage is assumed to be noise-free, a simulation signal is obtained according to the input intermediate variable through the preset leakage model, and as the signal part, it is used as the simulated leakage;
[0014] When the selected simulation mode is the full simulation mode FS in which the noise is assumed to be a specific distribution, the simulation signal obtained according to the input intermediate variable through the preset leakage model is used as the signal part, and the leakage noise is obtained based on the Gaussian distribution simulation, and then the leakage noise is superimposed on the signal part to obtain the simulated leakage.
[0015] When the selected simulation mode is the semi-simulation mode PS, which assumes that the noise is filtered from physical leakage, the simulated signal obtained according to the input intermediate variable through the preset leakage model is used as the signal part, and the side channel curve is used as input. The actual physical noise of the sample is obtained through high-pass filtering and then superimposed on the signal part to obtain the simulated leakage.
[0016] In one embodiment, regularizing the output simulation leakage and transition leakage includes:
[0017] The mean squared error between the simulated leakage and the transition leakage is calculated as the regularization loss.
[0018] In one embodiment, back propagation is used to update network parameters based on the regularized results and the classification loss to obtain a trained basic neural network for modeling side channel analysis based on deep learning, including:
[0019] The regularization loss and the classification loss are superimposed, back propagation is performed, the gradient is calculated and the neural network parameters are updated to obtain a trained basic neural network for modeling side channel analysis based on deep learning.
[0020] In one embodiment, the basic neural network for deep learning-based modeling side channel analysis is selected as a multi-layer perceptron MLP, which includes an encoder and a classifier.
[0021] In one embodiment, the side channel curve of the unknown intermediate value is input into a trained basic neural network for modeling side channel analysis based on deep learning, and the leakage intermediate value is inferred based on its output, including:
[0022] The encoder extracts features from the side channel curve of the unknown intermediate value input to generate transition leakage;
[0023] The transition leakage is analyzed by the classifier and the leakage value prediction is output.
[0024] Based on the same inventive concept, the second aspect of the present invention provides a side channel analysis device based on deep learning, comprising:
[0025] A simulation module is used to output simulated leakage according to an input intermediate variable and a selected simulation mode, wherein the intermediate variable is a real leakage value;
[0026] A regularization module for regularizing the transition leakage using the output simulated leakage, where the transition leakage is extracted from the input side channel curve by the basic neural network;
[0027] The basic neural network module is used to predict the intermediate value of the leakage based on the extracted transition leakage, calculate the cross entropy between the predicted intermediate value and the actual leakage value as the classification loss, and use backpropagation to update the network parameters by combining the regularization loss and the classification loss to obtain a trained basic neural network for modeling side channel analysis based on deep learning;
[0028] The attack module is used to input the side channel curve of the unknown intermediate value into the trained basic neural network based on deep learning modeling side channel analysis, and infer the leaked intermediate value based on its output.
[0029] Based on the same inventive concept, the third aspect of the present invention provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, it implements the side channel analysis method based on deep learning described in the first aspect.
[0030] Based on the same inventive concept, the fourth aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the program, the side channel analysis method based on deep learning described in the first aspect is implemented.
[0031] Compared with the prior art, the advantages and beneficial technical effects of the present invention are as follows:
[0032] The present invention provides a side channel analysis method based on deep learning, which includes a modeling stage and an attack stage. In the modeling stage, the intermediate variable is first used as input, the simulation mode is selected, and the corresponding side channel leakage (simulated leakage) is output; then, the transition leakage extracted by BNNM (basic neural network module for modeling side channel analysis based on deep learning) is regularized with the help of the output simulated leakage; during the training process, BNNM receives the side channel curve, extracts features and predicts the intermediate value of the leakage, and then calculates the loss between the predicted intermediate value and the true intermediate value, and uses back propagation to update the network parameters. In the attack stage, the side channel curve of the unknown intermediate value is input into the trained BNNM, and the leakage intermediate value is inferred based on its output. The present invention does not introduce external physical leakage, but uses simulated leakage as a reference, which expands the practical scope of the RePA framework. When the attacker cannot obtain physical reference leakage or needs to compare the attack performance of different targets horizontally, SimRePA is a general solution. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0034] Figure 1 is a flowchart of a side channel analysis method based on deep learning according to an embodiment of the present invention;
[0035] Figure 2 This is a framework diagram of a side channel analysis device based on deep learning provided by an embodiment of the present invention;
[0036] Figure 3 4 is a flowchart of the simulation module in an embodiment of the present invention. DETAILED DESCRIPTION
[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.
[0038] Example 1
[0039] This embodiment discloses a side channel analysis method based on deep learning, see Figure 1 ,include:
[0040] S101: Outputting simulated leakage according to the input intermediate variables and the selected simulation mode, wherein the intermediate variables are actual leakage values;
[0041] S102: Regularizing the transition leakage using the output simulated leakage, wherein the transition leakage is extracted from the input side channel curve by the basic neural network;
[0042] S103: Predicting an intermediate value of leakage based on the extracted transition leakage, calculating the cross entropy between the predicted intermediate value and the true leakage value as the classification loss, and combining the regularization loss and the classification loss to update the network parameters using back propagation to obtain a trained basic neural network for modeling side channel analysis based on deep learning;
[0043] S104: Input the side channel curve of the unknown intermediate value into the trained basic neural network for modeling side channel analysis based on deep learning, and infer the leakage intermediate value based on its output.
[0044] Specifically, a side-channel attack is a technique for inferring key information by analyzing the physical or electromagnetic leakage of cryptographic devices during cryptographic operations. These leakages include energy consumption and electromagnetic radiation. A key concept in side-channel attacks is the intermediate value, a specific value associated with the key and ciphertext during the encryption process. The intermediate value plays a crucial role in side-channel attacks because it is directly related to the key. By observing changes in the intermediate value, an attacker can infer key information.
[0045] In one embodiment, the simulated leakage includes a signal part and a noise part, and the selected simulation modes include a pure simulation mode CS assuming that the leakage is noise-free, a full simulation mode FS assuming that the noise is of a specific distribution, and a semi-simulation mode PS assuming that the noise is filtered from the physical leakage.
[0046] Specifically, the simulated leakage output consists of two components: signal and noise. The signal component is derived from the leakage model mapping specified intermediate variables. Preferably, the Hamming weight model (HW) is selected as the preset leakage model. Three simulation modes are defined based on the source of the noise: a pure simulation mode (CS) assuming the leakage is noise-free; a full simulation mode (FS) assuming the noise has a specific distribution; and a semi-simulation mode (PS) assuming the noise is filtered from physical leakage. Preferably, the noise in the default FS mode follows a Gaussian distribution.
[0047] In one embodiment, outputting simulation leakage according to the input intermediate variables and the selected simulation mode includes:
[0048] When the selected simulation mode is the pure simulation mode CS in which the leakage is assumed to be noise-free, a simulation signal is obtained according to the input intermediate variable through the preset leakage model, and as the signal part, it is used as the simulated leakage;
[0049] When the selected simulation mode is the full simulation mode FS in which the noise is assumed to be a specific distribution, the simulation signal obtained according to the input intermediate variable through the preset leakage model is used as the signal part, and the leakage noise is obtained based on the Gaussian distribution simulation, and then the leakage noise is superimposed on the signal part to obtain the simulated leakage.
[0050] When the selected simulation mode is the semi-simulation mode PS, which assumes that the noise is filtered from physical leakage, the simulated signal obtained according to the input intermediate variable through the preset leakage model is used as the signal part, and the side channel curve is used as input. The actual physical noise of the sample is obtained through high-pass filtering and then superimposed on the signal part to obtain the simulated leakage.
[0051] In one embodiment, regularizing the output simulation leakage and transition leakage includes:
[0052] The mean squared error between the simulated leakage and the transition leakage is calculated as the regularization loss.
[0053] In one embodiment, back propagation is used to update network parameters based on the regularized results and the classification loss to obtain a trained basic neural network for modeling side channel analysis based on deep learning, including:
[0054] The regularization loss and the classification loss are superimposed, back propagation is performed, the gradient is calculated and the neural network parameters are updated to obtain a trained basic neural network for modeling side channel analysis based on deep learning.
[0055] In one embodiment, the basic neural network for deep learning-based modeling side channel analysis is selected as a multi-layer perceptron MLP, which includes an encoder and a classifier.
[0056] In one embodiment, the side channel curve of the unknown intermediate value is input into a trained basic neural network for modeling side channel analysis based on deep learning, and the leakage intermediate value is inferred based on its output, including:
[0057] The encoder extracts features from the side channel curve of the unknown intermediate value input to generate transition leakage;
[0058] The transition leakage is analyzed by the classifier and the leakage value prediction is output.
[0059] Example 2
[0060] Based on the same inventive concept, this embodiment discloses a side channel analysis device based on deep learning, comprising:
[0061] A simulation module is used to output simulated leakage according to an input intermediate variable and a selected simulation mode, wherein the intermediate variable is a real leakage value;
[0062] A regularization module is used to regularize the output simulation leakage and transition leakage, where the transition leakage is extracted from the input side channel curve by the basic neural network;
[0063] The basic neural network module is used to predict the intermediate value of the leakage based on the extracted transition leakage, calculate the cross entropy between the predicted intermediate value and the actual leakage value as the classification loss, and use backpropagation to update the network parameters based on the regularized result and the classification loss to obtain a trained basic neural network for modeling side channel analysis based on deep learning;
[0064] The attack module is used to input the side channel curve of the unknown intermediate value into the trained basic neural network based on deep learning modeling side channel analysis, and infer the leaked intermediate value based on its output.
[0065] like Figure 2 As shown in Figure 1, in the modeling phase, the training samples (side channel samples / curves used in training) are generated by the encoder to generate transition leakage, and then the leakage value is predicted by the classifier. The classification loss between the leakage prediction and the true leakage value (i.e., the intermediate variable) is calculated. At the same time, the simulator in the SM (simulation) receives the true leakage variable, selects the simulation working mode (CS / FS / PS), and Figure 2 The workflow diagram outputs the simulated leakage. The LRM (Regularization Module) calculates the mean squared error between the simulated leakage and the transition leakage as the regularization loss. The classification loss is superimposed on the regularization loss, and backpropagation is performed to calculate the gradient and participate in the neural network parameter update process. The attack phase does not require the participation of the SM or LRM. The side channel sample used in the attack is directly input into the BNNM neural network, and the secret information is inferred based on the predicted leakage value.
[0066] like Figure 3 As shown in Figure 1, SM includes three simulation modes. The intermediate value of leakage is a mandatory input variable of SM, and the output is the simulated leakage. In CS mode, the simulated leakage is only the effective leakage obtained based on the Hamming weight model, that is, the signal. In FS mode, it is necessary to superimpose the leakage noise simulated based on Gaussian distribution on the simulated signal. When the simulation mode is set to PS, the simulator also needs to receive training samples as input (see Figure 2 The dotted arrow in the upper left corner and Figure 3 The actual physical noise of the sample is obtained through high-pass filtering and then superimposed on the signal part to obtain the simulated output of the leakage.
[0067] Since the device described in Example 2 of the present invention is used to implement the deep learning-based side-channel analysis method described in Example 1 of the present invention, those skilled in the art will be able to understand the specific structure and variations of the device based on the method described in Example 1 of the present invention, and therefore will not be described in detail here. All devices used in the method described in Example 1 of the present invention are within the scope of protection of the present invention.
[0068] Example 3
[0069] Based on the same inventive concept, the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described in the first embodiment is implemented.
[0070] Since the computer-readable storage medium described in Example 3 of the present invention is used to implement the deep learning-based side-channel analysis method described in Example 1 of the present invention, the specific structure and variations of the computer-readable storage medium are readily understood by those skilled in the art based on the method described in Example 1 of the present invention, and thus will not be further described here. All computer-readable storage media used in the method of Example 1 of the present invention fall within the scope of protection of the present invention.
[0071] Example 4
[0072] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first embodiment when executing the program.
[0073] Since the computer device described in Example 4 of the present invention is the computer device used to implement the deep learning-based side channel analysis method described in Example 1 of the present invention, those skilled in the art will be able to understand the specific structure and variations of the computer device based on the method described in Example 1 of the present invention, and therefore will not be described in detail here. All computer devices used in the method of Example 1 of the present invention fall within the scope of protection of the present invention.
[0074] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0075] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0076] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, the present invention is intended to include such changes and modifications to the embodiments of the present invention if they fall within the scope of the claims and their equivalents.
Claims
1. A side channel analysis method based on deep learning, characterized in that: include: Output simulated leakage according to the input intermediate variables and the selected simulation mode, wherein the intermediate variables are the real leakage values; Regularize the transition leakage using the simulated leakage of the output, where the transition leakage is extracted from the input side channel curve by the basic neural network; The intermediate value of the leakage is predicted based on the extracted transition leakage, and the cross entropy between the predicted intermediate value and the true leakage value is calculated as the classification loss. The regularization loss and the classification loss are combined to update the network parameters using back propagation to obtain the trained basic neural network for modeling side channel analysis based on deep learning. The side channel curve of the unknown intermediate value is input into the trained basic neural network based on deep learning modeling side channel analysis, and the leakage intermediate value is inferred based on its output.
2. The side channel analysis method based on deep learning according to claim 1, characterized in that The simulated leakage includes signal part and noise part. The selected simulation modes include pure simulation mode CS which assumes the leakage is noise-free, full simulation mode FS which assumes the noise is of specific distribution, and semi-simulation mode PS which assumes the noise is filtered from physical leakage.
3. The side channel analysis method based on deep learning according to claim 2, characterized in that: Based on the input intermediate variables and the selected simulation mode, the simulation leakage is output, including: When the selected simulation mode is the pure simulation mode CS in which the leakage is assumed to be noise-free, a simulation signal is obtained according to the input intermediate variable through the preset leakage model, and as the signal part, it is used as the simulated leakage; When the selected simulation mode is the full simulation mode (FS) that assumes a specific noise distribution, a simulated signal obtained by a preset leakage model based on the input intermediate variable is used as the signal part, and leakage noise is simulated based on a Gaussian distribution. The leakage noise is then superimposed on the signal part to obtain simulated leakage. When the selected simulation mode is the semi-simulation mode PS, which assumes that the noise is filtered from physical leakage, the simulated signal obtained according to the input intermediate variable through the preset leakage model is used as the signal part, and the side channel curve is used as input. The actual physical noise of the sample is obtained through high-pass filtering and then superimposed on the signal part to obtain the simulated leakage.
4. The side channel analysis method based on deep learning according to claim 1, characterized in that Regularize the output simulation leakage and transition leakage, including: The mean squared error between the simulated leakage and the transition leakage is calculated as the regularization loss.
5. The side channel analysis method based on deep learning according to claim 4, characterized in that: Based on the regularized results and classification loss, backpropagation is used to update the network parameters to obtain the trained basic neural network for deep learning-based modeling side channel analysis, including: The regularization loss and the classification loss are superimposed, back propagation is performed, the gradient is calculated and the neural network parameters are updated to obtain a trained basic neural network for modeling side channel analysis based on deep learning.
6. The side channel analysis method based on deep learning according to claim 1, characterized in that The basic neural network for modeling side channel analysis based on deep learning is selected as the multi-layer perceptron MLP, which includes an encoder and a classifier.
7. The side channel analysis method based on deep learning according to claim 6, characterized in that: Input the side channel curve of the unknown intermediate value into the trained basic neural network for deep learning-based modeling side channel analysis, and infer the leaked intermediate value based on its output, including: The encoder extracts features from the side channel curve of the unknown intermediate value input to generate transition leakage; The transition leakage is analyzed by the classifier and the leakage value prediction is output.
8. A side channel analysis device based on deep learning, characterized in that: include: A simulation module is used to output simulated leakage according to an input intermediate variable and a selected simulation mode, wherein the intermediate variable is a real leakage value; A regularization module for regularizing the transition leakage using the output simulated leakage, where the transition leakage is extracted from the input side channel curve by the basic neural network; The basic neural network module is used to predict the intermediate value of the leakage based on the extracted transition leakage, calculate the cross entropy between the predicted intermediate value and the true leakage value as the classification loss, and use backpropagation to update the network parameters by combining the regularization loss and the classification loss to obtain a trained basic neural network for modeling side channel analysis based on deep learning; The attack module is used to input the side channel curve of the unknown intermediate value into the trained basic neural network based on deep learning modeling side channel analysis, and infer the leaked intermediate value based on its output.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the side channel analysis method based on deep learning as described in any one of claims 1 to 7 is implemented.
10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the side channel analysis method based on deep learning is implemented as described in any one of claims 1 to 7.