Black box side-channel leakage verification using statistical approach
summary
A practical black‑box verification workflow emerges: run paired randomized blocking experiments, hide test classes from the measurement harness, and use bootstrap resampling to get confidence intervals; apply nonparametric sign/Friedman/Skillings‑Mack tests rather than raw t/Welch/TVLA. The open‑source toolkit (part of TLS‑Puzzle) runs over network or locally, is CI‑friendly, and has validated RSA (Marvin), ECDSA (Minerva) and ML‑KEM cases; with billions of traces and a p‑value cutoff ~1e‑5 it finds no false positives and can reach CI below one CPU clock cycle (≈0.2 ns at 5 GHz).
AI-written summary · about
chapters (13)
- 0:00Side-channel leakage verification
- 1:09Statistical independence in measurements
- 3:39Sample correlation risks in tests
- 7:27Double-blind experiment design
- 12:28Sample size with bootstrapping
- 15:19Flaws in side-channel tests
- 17:53Tests for multiple input classes
- 24:12Reliable negative results
- 25:59Randomized double-blind best practices
- 27:03Tools and resources
- 28:14Python implementation Q&A
- 30:00Observer blindness Q&A
- 31:47Closing remarks
related talks
01h 09mCrypto + ML 1
01h 02mCryptographic Side-Channels on Embedded Devices
14mCollapsing Towers for Side-Channel Security
50mTime Trial: Racing Towards Practical Timing Attacks
41mFault Attacks & Side Channels
41mFormally Verifying Everybody's Cryptography
01h 04mSoftware Verification meets Real-World Cryptography
20mPantomime: Constructive Leakage Proofs via Simulation