ANALYZING AES POWER TRACES FOR SIDE-CHANNEL ATTACK: GENERATION, CLASSIFICATION, KEY DEDUCTION, AND MITIGATION STRATEGIES

Abstract

Side-channel attacks (SCAs) exploit leakages from a system’s physical implementation, like acoustic signals, electromagnetic, and power emissions, to deduce sensitive information, thus bypassing traditional security measures. SCAs appeal to hackers due to their non-invasive nature and low cost and thereby necessitate robust countermeasures. The Advanced Encryption Standard (AES) is a widely used symmetric key cryptography known for its robustness, but it remains susceptible to SCAs. This research analyzes power traces to identify vulnerabilities, classifies power traces based on AES implementation, employs Deep Learning(DL) models to deduce cryptographic keys, and develops mitigation strategies against SCAs. We generated and analyzed real and synthetic power traces from masked AES implementations using a Syscomp waveform generator and a Python script. Techniques like Fast Fourier Transform (FFT), Wavelet transform, and linear regression were used to correlate the traces. Power traces from the AES Power Trace (AES PT) dataset were classified into three AES implementations using feature extraction techniques and Support Vector Machine (SVM) classification based on statistical properties from Principal component analysis (PCA).We used hashing and metadata techniques retrieved original power traces from the feature set. The study used ANSSI Side-channel Analysis Database (ASCAD) and adopted deep learning models for key deduction: Residual Networks were transformed into ResTraceNet using 1D convolutional layers, and Gated Recurrent Units (GRUs) were modified into GRUTrace to process 1D power traces. These models deduced one key byte using only 100 power traces, achieving testing accuracies of 96.68% and 96.28%. We proposed a mitigation strategy involving structured masking and Gaussian noise to obscure relationships between cryptographic keys and power consumption patterns. Our proposed research provides a comprehensive analysis of AES power traces, using DL models to perform feature extraction, classify and deduce cryptographic keys, and proposes mitigation techniques to enhance defenses against SCAs.