Side-Channel Analysis of DES and SPECK32

What is Side-Channel Analysis

Side-Channel Analysis (SCA) exploits physical leakage from cryptographic implementations rather than mathematical weaknesses. Power consumption is data-dependent and can reveal information about intermediate computations.

Leakage Model

The Hamming Weight (HW) model is used:

HW(x) = number of bits set to 1 in x

Power consumption is assumed to be proportional to the Hamming Weight of intermediate values.

SPECK32: Synthetic Trace Generation

Synthetic traces are generated by recording HW values of intermediate operations during encryption:

  • rotation output
  • modular addition result
  • XOR with round key
  • final mixing stage

Each round produces four leakage samples.

Trace length = 22 × 4 = 88 samples

Dataset characteristics:

  • 10,000 traces
  • fixed secret key
  • random plaintexts
  • Gaussian noise added
  • random desynchronization applied

This simulates realistic measurement conditions.

CNN-Based Key Recovery for SPECK32

A Convolutional Neural Network is trained to learn the mapping:

trace → key class

Key rank is used as the evaluation metric.
Rank 0 indicates successful key recovery.

alt text

DES S-box Side-Channel Attack

Instead of predicting the key directly, the model predicts the S-box output (16 classes).

Key recovery procedure:

  1. Compute S-box output for each key hypothesis
  2. Accumulate log-likelihoods across traces
  3. Select the key with maximum likelihood

This follows the template attack methodology with deep learning feature extraction.

alt text

Advantages of Deep Learning in SCA

  • automatic detection of leakage points
  • robustness to noise and misalignment
  • no manual feature engineering required