Open Source Datasets

We identify a critical gap in open-source datasets for the implementation security testing of post-quantum cryptography. The following are our guidelines for generating this dataset for side-channel analysis. Fault injection datasets would also be useful, but need more research maturity before discussing specifics.

Dataset Scope and Strategy

• Full algorithm execution traces may result in large, impractical datasets.
• A more effective approach may involve generating side-channel traces for unit-level testing.

Proposed Units for Trace Collection

• Targeted units include:
  • Number Theoretic Transform (NTT)
  • Samplers (e.g., binomial, ternary, Gaussian, rejection sampling)
  • Hash functions (particularly those used in PRFs for PQC)
  • Comparison functions (e.g., for rejection sampling)
  • Arithmetic-to-Boolean and Boolean-to-Arithmetic conversions used in masking
• Each unit can be further subdivided based on algorithmic variations and implementation strategies.

Implementation Variants

• Each unit should ideally be implemented in four forms:
  1. Baseline (no countermeasures)
  2. First-order masking
  3. Shuffling
  4. Higher-order masking

Hardware Measurement Emphasis

• While software-based measurements are acceptable, there is a strong community preference for hardware-based side-channel measurements, particularly from FPGA implementations

Metadata Requirements

• Trace datasets should be accompanied by detailed metadata, including:
  • FPGA implementation specifics
  • Description of applied defense techniques
  • Methodology for input generation
• The metadata description could further follow details in our earlier document.

Examples of existing datasets: Kyber https://eprint.iacr.org/2025/811