A Side-Channel Attack Defense System and Method Based on Dynamic Noise Injection and Multipath Randomization

Abstract

This invention proposes a side-channel attack defense system and method based on dynamic noise injection and multi-path randomization, belonging to the field of network security technology. The system includes a noise injection module, a time equalization engine, a dynamic mapping unit, and a quantum key interface. The method includes six steps: noise masking, random source acquisition, path randomization, dynamic data mapping, equalization execution, and result output. This invention masks physical signals through dynamic noise injection, disrupts execution trajectories using multi-path randomization, and eliminates timing differences by combining time equalization operations, thereby constructing a collaborative defense system from the physical layer, logic layer, and timing layer. It can effectively resist various side-channel attacks such as power analysis, electromagnetic analysis, and time analysis, significantly improving security while ensuring system computational efficiency. It can be widely applied in scenarios such as encryption chips, IoT devices, and cloud computing containers.

Description

Technical Field

[0001] This invention relates to the field of network security technology, and in particular to a side-channel attack defense system and method based on dynamic noise injection and multi-path randomization. Background Technology

[0002] Side-channel attacks (SCA) pose a significant challenge to embedded security and hardware cryptography. Unlike traditional cryptanalysis, these attacks do not directly target the algorithm itself. Instead, they infer keys or sensitive data by collecting and analyzing physical characteristics unintentionally leaked during the operation of cryptographic devices. These physical characteristics include, but are not limited to, power consumption fluctuations, electromagnetic radiation, differences in instruction execution time, and even the sounds generated during device operation. Attackers can leverage techniques such as differential power analysis (DPA), electromagnetic analysis (EMA), and timing analysis to successfully extract key information from these seemingly unrelated physical leaks, posing a major security threat to encryption chips, smart cards, and IoT terminals.

[0003] To address these threats, existing defense technologies primarily focus on three levels, but some problems still remain: Firstly, at the physical signature protection level, traditional methods such as masking techniques attempt to sever the correlation between data and power consumption by operating on intermediate values ​​and random numbers. However, this method is exceptionally complex in algorithm design and implementation, and introduces huge computational and storage overheads, resulting in a significant degrade in encryption and decryption performance, making it difficult to meet the needs of applications with high real-time requirements. Simple noise injection, on the other hand, often lacks dynamism and adaptability; its fixed patterns can be filtered out by attackers through long-term monitoring and signal processing techniques, offering limited masking effectiveness. The correlation between key operations and power consumption, as well as electromagnetic signals, can still be exploited.

[0004] Secondly, regarding the protection of timing information, the inherent branching, table lookup, and conditional execution logic in encryption algorithms can lead to differences in execution time related to key values. Existing time equalization techniques attempt to eliminate these differences by inserting fixed delays or standardizing operation steps, but they struggle to completely cover all microarchitecture-level timing fluctuations (such as cache hits / misses, pipeline stalls, etc.). Consequently, incomplete equalization leaves behind analyzable timing features, allowing attackers to still infer key processing logic through high-precision timing analysis.

[0005] Finally, regarding defense efficiency, while existing randomization methods (such as instruction reordering and address randomization) increase uncertainty, their randomness quality depends on the pseudo-random number generator, making them predictable. Furthermore, many protective measures are implemented in isolation, lacking system-level collaborative design. This often leads to unacceptable negative impacts on other dimensions (such as performance, power consumption, and area) while improving security in one dimension. For example, complex masking and randomization operations significantly reduce encryption and decryption speeds, making it difficult to deploy the solution in resource-constrained or high-efficiency environments. Therefore, this invention proposes a side-channel attack defense system and method based on dynamic noise injection and multi-path randomization to address the problems existing in the prior art. Summary of the Invention

[0006] To address the aforementioned problems, the present invention aims to propose a side-channel attack defense system and method based on dynamic noise injection and multi-path randomization. The present invention provides a comprehensive, efficient and practical side-channel attack protection solution for scenarios such as encryption chips and IoT devices.

[0007] To achieve the objectives of this invention, the invention is implemented through the following technical solution: a side-channel attack defense system based on dynamic noise injection and multipath randomization, characterized in that it includes: The noise injection module is used to generate controllable random noise in real time when the device performs key-related operations, and inject the noise signal into the power consumption and electromagnetic signal path of the device. The time equalization engine, connected to the noise injection module, is used to perform time standardization scheduling on the instruction execution sequence in the encryption algorithm. By randomizing the instruction execution order and inserting equalization delay, it eliminates the execution time differences caused by different key values. The dynamic mapping unit, connected to the time equalization engine, is used to divide the intermediate state data in the key operation process into multiple data slices and allocate each data slice to a mutually isolated hardware processing channel for execution. The quantum key interface is used to receive a sequence of truly random numbers generated by an external quantum random number generator and input it as a randomization parameter to the dynamic mapping unit or time equalization engine.

[0008] A further improvement is that the noise injection module includes: A noise generation circuit is used to generate noise signals with controllable spectrum and amplitude. The filtering unit, connected to the noise generation circuit, is used to filter out frequency components in the noise signal that interfere with normal operation. The sound acquisition unit is used to acquire sound wave signals caused by the charging and discharging of capacitors during equipment operation in real time.

[0009] A further improvement is that the dynamic mapping unit includes a data fragment scheduler and several sets of hardware processing channels. The data fragment scheduler is used to divide the intermediate data of key operation into multiple fragments and map each fragment to a different processing channel according to the random number sequence. The several sets of hardware processing channels are isolated from each other by physical or logical means.

[0010] A further improvement is that the time equalization engine includes a polynomial operation scheduling unit and a branch prediction flushing unit, wherein the polynomial operation scheduling unit is used to randomly sort the polynomial calculation steps involved in the encryption algorithm, and the branch prediction flushing unit is used to insert randomly selected redundant instructions into the execution pipeline when the branch prediction is successful.

[0011] A further improvement is that it also includes a noise intensity feedback control loop, which adjusts the noise intensity and spectral distribution of the noise injection module in real time based on the characteristics of the sound wave signal collected by the sound acquisition unit.

[0012] The side-channel attack defense method based on dynamic noise injection and multipath randomization includes the following steps: Step 1, Noise Masking: When the key operation is triggered, the noise injection process is started to generate controllable random noise and inject it into the power consumption and electromagnetic signal path of the device; Step 2, Random Source Acquisition: Obtain a true random number sequence from an external quantum random number generation device; Step 3, Path Randomization: Based on a true random number sequence, dynamically select one of the multiple preset encryption / decryption operation paths as the current execution path; Step 4, Dynamic Data Mapping: Based on the execution path and the true random number sequence, the intermediate state data of the key operation is segmented and randomly mapped, and then scheduled to multiple hardware processing channels; Step 5, Balanced Execution: In each hardware processing channel, encryption operations are performed according to the selected path, and the instruction execution order is randomly rearranged, balanced delays are inserted, and pipeline flushing operations are performed. Step 6: Output the result: After the operation is completed, output the final encryption result.

[0013] A further improvement is that step one includes collecting high-frequency acoustic signals caused by capacitor charging and discharging during device operation, and adjusting the intensity and spectrum of the injected noise in real time based on the time-frequency characteristics of the signal.

[0014] A further improvement is that, in step three, the multiple predefined encryption and decryption operation paths include at least: paths based on different permutations and combinations of S-boxes in the AES algorithm, paths based on different computation orders in modular exponentiation or elliptic curve dot product operations, and paths based on different scheduling strategies for parallelizable subtasks in the algorithm.

[0015] A further improvement is that, in step four, scheduling refers to mapping each data segment to different hardware processing channels based on a true random number sequence.

[0016] A further improvement is made in step five, where the pipeline flushing operation specifically involves inserting one or more randomly selected redundant arithmetic logic instructions into the execution pipeline after the branch prediction is hit.

[0017] The beneficial effects of this invention are as follows: By dynamically injecting noise and adaptively adjusting it, this invention masks the power consumption and electromagnetic characteristics caused by key operations at the physical level, making it difficult for attackers to extract valid signals, thereby significantly improving the ability to resist differential power analysis and electromagnetic analysis. Furthermore, by using time-equalized execution and multi-path randomization, the timing characteristics of key dependencies are eliminated at the algorithm execution level, making the execution path of each operation unpredictable, thus effectively defending against side-channel attacks based on time analysis and path analysis. Finally, by using dynamic data mapping and hardware channel isolation, intermediate states are dispersed and isolated at the data processing level, avoiding information leakage related to data dependencies, thereby enhancing robustness against attacks such as simple power analysis.

[0018] Furthermore, by introducing quantum random numbers as a random source and combining them with efficient hardware implementation, high-intensity unpredictability is provided while minimizing the impact of defense mechanisms on the original computing performance, thus achieving a balance between security and real-time performance. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the system structure framework of Embodiment 1 of the present invention.

[0020] Figure 2 This is a schematic diagram of the method flow of Embodiment 2 of the present invention. Detailed Implementation

[0021] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0022] Example 1 according to Figure 1 As shown, this embodiment proposes a side-channel attack defense system based on dynamic noise injection and multi-path randomization, which can be integrated into an encryption chip or security module.

[0023] It includes a noise injection module, a time equalization engine, a dynamic mapping unit, and a quantum key interface. Specifically, the noise injection module generates controllable random noise in real time when the device performs key-related operations and injects this noise signal into the device's power consumption and electromagnetic signal paths, thereby masking the true physical characteristics. Accordingly, the noise injection module internally includes a noise generation circuit, a filtering unit, and a sound acquisition unit. The noise generation circuit generates a noise signal with controllable spectrum and amplitude; the filtering unit is connected to the noise generation circuit and is used to filter out frequency components in the noise signal that interfere with normal operations; the sound acquisition unit acquires sound wave signals caused by capacitor charging and discharging during device operation in real time, providing feedback for dynamic adjustment of noise intensity.

[0024] The time equalization engine, connected to the noise injection module, performs time-normalized scheduling of the instruction execution sequence in the encryption algorithm. It eliminates execution time differences caused by different key values ​​by randomizing the instruction execution order and inserting equalization delays. Internally, the time equalization engine includes a polynomial operation scheduling unit and a branch prediction flushing unit. The polynomial operation scheduling unit is responsible for randomly ordering the polynomial computation steps involved in the encryption algorithm; the branch prediction flushing unit, when a branch prediction is successful, inserts randomly selected redundant instructions into the execution pipeline to eliminate timing characteristics caused by branch jumps.

[0025] The dynamic mapping unit is connected to the time equalization engine and is used to divide the intermediate state data during the key operation process into multiple data slices. Each data slice is then assigned to an isolated hardware processing channel for execution, thereby achieving randomization of the operation path and data isolation. It includes a data slice scheduler and several sets of hardware processing channels. Specifically, the data slice scheduler divides the intermediate key operation data into multiple segments and maps each segment to a different processing channel based on a random number sequence. The several sets of hardware processing channels are physically or logically isolated to prevent side-channel information leakage across channels.

[0026] The quantum key interface is used to receive true random number sequences generated by an external quantum random number generator and input them as randomization parameters to the dynamic mapping unit or time equalization engine, providing a high-intensity random source for the system's randomization operations.

[0027] In addition, the system also includes a noise intensity feedback control loop, which adjusts the noise intensity and spectrum distribution of the noise injection module in real time according to the characteristics of the sound wave signal collected by the sound acquisition unit, so as to achieve adaptive noise masking.

[0028] Example 2 like Figure 2As shown, this embodiment provides a side-channel attack defense method based on dynamic noise injection and multi-path randomization, applicable to scenarios where encryption chips, IoT devices, and other devices perform encryption and decryption operations. Specifically, it includes the following steps: Step 1: Noise Masking When the device triggers key operations, it performs a noise masking step, which initiates a noise injection process. This generates controllable random noise and injects it into the device's power consumption and electromagnetic signal paths. Simultaneously, it collects high-frequency acoustic signals generated by capacitor charging and discharging during device operation. Based on the time-frequency characteristics of this signal, it adjusts the intensity and spectrum of the injected noise in real time to achieve adaptive matching between noise and changes in the physical environment.

[0029] Step 2: Obtaining a random source A sequence of truly random numbers is obtained from an external quantum random number generator. This sequence will serve as the source for all subsequent randomization operations, ensuring the unpredictability and high intensity of randomness. Step 3: Path Randomization Based on the acquired true random number sequence, a predefined encryption / decryption operation path is dynamically selected as the current execution path. These predefined operation paths include at least paths based on different permutations and combinations of S-boxes in the AES algorithm, paths based on different computation orders in modular exponentiation or elliptic curve multiplication operations, and paths based on different scheduling strategies for parallelizable subtasks in the algorithm. The random selection of the path ensures that the execution trajectory of each encryption operation is different, effectively blocking path-dependent side-channel attacks. Step 4: Dynamic Data Mapping Based on the selected execution path and the true random number sequence, the intermediate state data of the key operation is segmented and randomly mapped, and then scheduled to multiple isolated hardware processing channels. Specifically, scheduling refers to mapping each data segment to different hardware processing channels according to the true random number sequence, thereby achieving data-level isolation and randomization. Step 5: Balanced Execution In each hardware processing channel, encryption operations are performed according to the selected path, and the instruction execution order is randomly rearranged, latency is balanced, and pipeline flushing is performed. Specifically, after a branch prediction hit, one or more randomly selected redundant arithmetic logic instructions are inserted into the execution pipeline. This series of operations ensures that the timing characteristics during the computation are balanced, eliminating time-side channel leakage. Step Six: Output Results Once the encryption operation is complete, the final encryption result is output. At the same time, the system can filter the injected noise signal to ensure the purity of the output signal.

[0030] Example 3 This embodiment uses an IoT security chip with an integrated AES-256 encryption engine as the implementation example. Specifically: When the instruction decoding unit of the encryption chip receives the AES encryption operation instruction and the plaintext data block to be encrypted from the system bus, the defense process is immediately initiated. The noise injection module within the chip is triggered synchronously. This module's noise generation circuit is based on a waveform synthesizer driven by a hybrid linear feedback shift register (LFSR) and a true random number generator, generating a pseudo-random noise signal with a frequency range of 1MHz to 50MHz and an amplitude fluctuating within 5% of the chip's power supply voltage. This signal is then directly injected into the power distribution network of the encryption engine's core power domain through a dedicated analog switch array. Simultaneously, a piezoelectric acoustic wave sensor integrated into the chip's package substrate begins operation, continuously acquiring acoustic vibration signals generated during the high-speed switching process of the chip's internal decoupling capacitors, with frequencies primarily concentrated in the 1-10MHz range. The amplitude envelope of this acoustic wave signal is extracted and fed into an analog-digital hybrid feedback control circuit. The circuit adjusts the output power of the waveform synthesizer in real time based on the amplitude of the sound wave. When it detects a change in the sound wave characteristics caused by the approach of an external electromagnetic probe, it automatically increases the amplitude of the noise signal by 20%, thus achieving adaptive noise intensity adjustment.

[0031] In the same clock cycle after noise injection is initiated, the chip obtains a 256-bit true random number seed from an externally mounted random number generator based on photonic quantum noise through its quantum key interface (a dedicated pin group compliant with the SPI protocol). This seed is loaded into the random number extension module within the chip to generate a series of random parameters required for subsequent processes.

[0032] Next, the dynamic mapping unit begins operation. Its internal data fragmentation scheduler first uses the first 32 bits of the acquired true random number seed as a selection signal to choose one of the four equivalent AES encryption paths fixed in the chip hardware. These four paths correspond to: Path A uses standard S-box sequential and column mixing operations; Path B delays the S-box lookup operation by one round; Path C decomposes the column mixing operation into two interleaved sub-steps; and Path D uses different round keys and timing sequences. After selecting a path, the scheduler logically interleaves and divides the 128-bit plaintext data block to be encrypted and the 128-bit round key of the current round according to the subsequent bits of the true random number seed, forming four 32-bit sub-data segments. These four sub-segments are allocated to four physically isolated Single Instruction Multiple Data (SIMD) operation channels within the chip through a crossbar switch matrix. Each channel has an independent 32-bit register and a dedicated S-box lookup table to ensure that the data segments do not generate power line coupling in subsequent operations.

[0033] The chip then enters the encryption operation phase, where the time balancing engine deeply intervenes in the micro-operation scheduling of each operation channel within each clock cycle. The polynomial operation scheduling unit within the engine dynamically rearranges the execution order of the 16-byte processing subtasks in each round of AES operation based on a pseudo-random sequence driven by truly random numbers. Simultaneously, the engine inserts no-op (NOP) bubbles, determined by random numbers and ranging in length from 1 to 3 clock cycles, into the pipeline of each operation channel to ensure that the total number of clock cycles required to complete one round of encryption operation remains constant when processing arbitrary data. Furthermore, the branch prediction flushing unit monitors the instruction prefetch buffer. When a fixed loop branch in the encryption algorithm is predicted to be hit, this unit does not immediately jump, but instead forces the current pipeline to execute two redundant instructions generated by random number indices (e.g., performing a circular left shift and an XOR operation on a temporary register consecutively), before executing the correct jump instruction. This mechanism completely disrupts the correspondence between program flow and power timing.

[0034] After all 14 rounds of encryption operations are completed, the ciphertext fragments output from the four SIMD channels are reassembled into a complete 128-bit ciphertext. Before the ciphertext is sent to the chip's output buffer, the filtering unit in the noise injection module is activated. This unit consists of a programmable switched-capacitor filter with a cutoff frequency set to 50MHz. This filter attenuates noise components injected earlier with frequencies higher than 50MHz by more than 40dB, while completely preserving the fundamental frequency component of the digital signal generated by the encryption operations. Finally, the clean and timing-compliant ciphertext data is output through the chip's parallel data interface.

[0035] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the present invention without departing from its framework and scope of application, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A side-channel attack defense system based on dynamic noise injection and multi-path randomization, characterized in that: include: The noise injection module is used to generate controllable random noise in real time when the device performs key-related operations, and inject the noise signal into the power consumption and electromagnetic signal path of the device. The time equalization engine, connected to the noise injection module, is used to perform time standardization scheduling on the instruction execution sequence in the encryption algorithm. By randomizing the instruction execution order and inserting equalization delay, it eliminates the execution time differences caused by different key values. The dynamic mapping unit, connected to the time equalization engine, is used to divide the intermediate state data in the key operation process into multiple data slices and allocate each data slice to a mutually isolated hardware processing channel for execution. The quantum key interface is used to receive a sequence of truly random numbers generated by an external quantum random number generator and input it as a randomization parameter to the dynamic mapping unit or time equalization engine.

2. The side-channel attack defense system based on dynamic noise injection and multi-path randomization according to claim 1, characterized in that: The noise injection module includes: A noise generation circuit is used to generate noise signals with controllable spectrum and amplitude. The filtering unit, connected to the noise generation circuit, is used to filter out frequency components in the noise signal that interfere with normal operation. The sound acquisition unit is used to acquire sound wave signals caused by the charging and discharging of capacitors during equipment operation in real time.

3. The side-channel attack defense system based on dynamic noise injection and multi-path randomization according to claim 1, characterized in that: The dynamic mapping unit includes a data fragment scheduler and several sets of hardware processing channels. The data fragment scheduler is used to divide the intermediate data of key operation into multiple fragments and map each fragment to a different processing channel according to the random number sequence. The several sets of hardware processing channels are isolated from each other by physical or logical means.

4. The side-channel attack defense system based on dynamic noise injection and multi-path randomization according to claim 1, characterized in that: The time equalization engine includes a polynomial operation scheduling unit and a branch prediction flushing unit. The polynomial operation scheduling unit is used to randomly sort the polynomial calculation steps involved in the encryption algorithm, and the branch prediction flushing unit is used to insert randomly selected redundant instructions into the execution pipeline when the branch prediction is successful.

5. The side-channel attack defense system based on dynamic noise injection and multi-path randomization according to claim 2, characterized in that: It also includes a noise intensity feedback control loop, which adjusts the noise intensity and spectral distribution of the noise injection module in real time based on the characteristics of the sound wave signal collected by the sound acquisition unit.

6. A side-channel attack defense method based on dynamic noise injection and multipath randomization, characterized in that: Includes the following steps: Step 1, Noise Masking: When the key operation is triggered, the noise injection process is started to generate controllable random noise and inject it into the power consumption and electromagnetic signal path of the device; Step 2, Random Source Acquisition: Obtain a true random number sequence from an external quantum random number generation device; Step 3, Path Randomization: Based on a true random number sequence, dynamically select one of the multiple preset encryption / decryption operation paths as the current execution path; Step 4, Dynamic Data Mapping: Based on the execution path and the true random number sequence, the intermediate state data of the key operation is segmented and randomly mapped, and then scheduled to multiple hardware processing channels; Step 5, Balanced Execution: In each hardware processing channel, encryption operations are performed according to the selected path, and the instruction execution order is randomly rearranged, balanced delays are inserted, and pipeline flushing operations are performed. Step 6: Output the result: After the operation is completed, output the final encryption result.

7. The side-channel attack defense method based on dynamic noise injection and multipath randomization according to claim 6, characterized in that: Step one also includes collecting high-frequency acoustic signals caused by capacitor charging and discharging during device operation, and adjusting the intensity and spectrum of the injected noise in real time based on the time-frequency characteristics of the signal.

8. The side-channel attack defense method based on dynamic noise injection and multipath randomization according to claim 6, characterized in that: In step three, the multiple predefined encryption and decryption operation paths include at least: paths based on different permutations and combinations of S-boxes in the AES algorithm, paths based on different computation orders in modular exponentiation or elliptic curve dot product operations, and paths based on different scheduling strategies for parallelizable subtasks in the algorithm.

9. The side-channel attack defense method based on dynamic noise injection and multipath randomization according to claim 6, characterized in that: In step four, scheduling refers to mapping each data segment to different hardware processing channels based on a true random number sequence.

10. The side-channel attack defense method based on dynamic noise injection and multipath randomization according to claim 6, characterized in that: In step five, the pipeline flushing operation specifically involves inserting one or more randomly selected redundant arithmetic logic instructions into the execution pipeline after the branch prediction is hit.

Images