Deep Learning Side Channel Attacks and Randomness Beacons

Vocabulary flashcards covering deep learning-based side channel attacks on smart card hardware and the mechanisms of centralized and decentralized public randomness beacons.

Side Channel Attacks

Attacks that exploit physical leaks such as power usage or electromagnetic emissions from hardware to steal secret keys, rather than breaking the cryptographic algorithm itself.

Profiled Attack

A deep learning attack where the attacker trains a model on a cloned device first to learn leakage patterns before extracting keys from the real target.

Non-profiled Attack

An attack where the attacker goes straight to the target device using statistical methods to retrieve secret keys.

Masking

A countermeasure that splits data into fragments to hide the secret, which deep learning models can now break by learning higher order patterns.

Desynchronization

A defense that shifts operations to random places so leakage does not appear consistently, though CNNs can still find leakage regardless of position.

Secure Logic

A countermeasure designed to keep power consumption constant regardless of the data being processed.

Noise Injection

A signal protection method that floods the signal with fake noise; it can be bypassed by U-Net models that reconstruct the real signal.

TFADs Technology

A highly resistant hardware technology that draws so little power that traditional methods cannot distinguish the correct key from the wrong key.

Layer 1 (Standardized Framework)

The fixed environment setup layer defining trace length, preprocessing, and training settings to ensure tests occur under the same conditions.

Layer 2 (Evaluation Metrics)

The layer using standardized metrics for a pass or fail verdict, currently defined for AES (Advanced Encryption Standard).

Layer 3 (Verification Layer)

The layer responsible for multi-device testing and repeated runs to check if a result is trustworthy across multiple chip copies.

Randomness Beacon

A potential solution for providing a regular output of public randomness that aims to be reliable, verifiable, unpredictable, and unbiasable.

Centralized Randomness Beacon

A beacon relying on a single authority, such as the NIST randomness beacon (using quantum mechanics) or random.org (using atmospheric noise).

Last Reveal Attack

An attack where an adversary waits to reveal their commitment last to compute the output and decide whether or not to publish it.

RANDAU

A protocol in Ethereum’s proof of stake that uses economic punishment, requiring nodes to pledge capital such as $32$ Ethereum to discourage adversarial behavior.

DRAM

A protocol used by the League of Entropy that utilizes threshold cryptography, distributed key generation, and BLS signatures.

BLS Signatures

Deterministic signatures used in DRAM; if a threshold of nodes is corrupted, the output becomes predictable but remains unbiasable.

Verifiable Delay Function (VDF)

A cryptographic primitive that allows a prover to show a specific amount of time was spent on a computation, designed to handle cases with only one honest node.

Threshold Cryptography

A system where a minimum number (threshold) of participating nodes is required to broadcast partial signatures to recover a full random value.