Differential Power Analysis (DPA)
Project Overview
This project explores Side-Channel Attacks, specifically Differential Power Analysis (DPA), to extract the secret key from an AES-128 encryption implementation by analyzing power consumption traces.
The core objective was to move beyond standard byte-level attacks and implement granular Bit-Level strategies, comparing different statistical distinguishers to optimize attack efficiency.
Methodologies & Experiments
I implemented and benchmarked three distinct attack vectors:
1. Bit-Level Leakage Model
Instead of attacking full bytes, I targeted individual bits of the S-Box output (Intermediate Value).
- Result: Higher precision compared to byte-level guesses.
2. Correlation vs. Difference of Means (DoM)
I compared two statistical tools to distinguish the correct key guess:
- Pearson Correlation: Slower computation but highly efficient in terms of data. [cite_start]It successfully recovered the full key with just ~105 traces[cite: 1184, 1240].
- Difference of Means (DoM): Computed using the T-Statistic. [cite_start]It proved to be significantly faster (30 mins vs 4 hours) but “data-hungry”, requiring ~660-700 traces to reach stable key recovery[cite: 1077, 1181].
3. Linear vs. Non-Linear Targets
I attempted to attack the AddRoundKey (ARK) output directly (linear operation) versus the SubBytes output (non-linear).
- [cite_start]Finding: The attack on the linear ARK stage failed to recover most bytes (only 4/16 correct)[cite: 1119, 1174]. [cite_start]This empirically confirmed that non-linear operations (S-Box) are the primary source of exploitable leakage in block ciphers[cite: 645, 1120].
Key Takeaways
| Metric | Correlation Distinguisher | Difference of Means (DoM) |
|---|---|---|
| Computation Speed | Slow (High Complexity) | Fast (Statistical Binning) |
| Data Efficiency | High (~100 traces) | Low (~700 traces) |
| Attack Vector | Bit-Wise S-Box Output | Bit-Wise S-Box Output |
Tech Stack
- Language: Python 3.x
- Analysis: NumPy (for vectorised statistical computations), Matplotlib
- Target: TinyAES traces (Simulated/Captured power data)
Note
While this project focuses on Side-Channel attacks, the underlying cryptographic logic is based on my AES Implementation from Scratch.
