Conventional and deep-learning-based side-channel analysis on embedded devices
The security of modern cryptographic devices is increasingly threatened by Side-Channel Attacks (SCA), particularly as traditional profiling attacks face limitations in scalability, adaptability, and noise handling. Manual feature selection, data acquisition, and fine-tuning are labor-intensive, esp...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2025
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/181998 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | The security of modern cryptographic devices is increasingly threatened by Side-Channel Attacks (SCA), particularly as traditional profiling attacks face limitations in scalability, adaptability, and noise handling. Manual feature selection, data acquisition, and fine-tuning are labor-intensive, especially when encountering advanced countermeasures such as masking and operation randomization. To address these challenges, this study integrates the Advanced Encryption Standard (AES-128) in ECB mode on the NXP i.MX RT1050 development board within a FreeRTOS environment, applying a deep learning-based side-channel analysis framework (SCAAML). Compared to conventional methods, this approach leverages deep learning models to automatically extract leakage features from electromagnetic and power consumption data, achieving greater attack accuracy and automation. The robustness of the SCAAML framework was demonstrated through comprehensive testing on various datasets, simulating noise and multitasking conditions. The SCAAML model demonstrated a 75% improvement in key recovery success over traditional Correlation Power Analysis (CPA) methods. Our deep learning approach achieved complete AES key recovery with as few as 400 traces, compared to the 6 key bytes recovered by CPA. The results indicate that deep learning techniques outperform traditional methods in adaptability, efficiently recovering AES keys with minimal traces even under complex multitasking environments. This study offers a more scalable and automated approach to SCA, providing valuable insights for enhancing the security assessment of cryptographic devices in real-world conditions. |
|---|