Chameleon: A Dataset for Segmenting and Attacking Obfuscated Power Traces in Side-Channel Analysis

Davide Galli, Giuseppe Chiari, Davide ZoniIACR Trans. Cryptogr. Hardw. Embed. Syst 2025 · OpenReview

Abstract

Side-channel attacks exploit unintended information leakage emitted by cryptographic devices to extract sensitive data. Hiding techniques are a cost-effective countermeasure designed to obfuscate the side-channel leakage and hinder these attacks. Available open datasets rely on artificial models to simulate hiding effects, preventing a realistic assessment of these countermeasures and, thus, leaving a pressing need for datasets offering real-world, obfuscated side-channel measurements. Chameleon introduces the first comprehensive dataset of real-world, obfuscated power traces collected from a RISC-V-based System-on-Chip. The traces are obfuscated using four state-of-the-art hiding techniques: dynamic frequency scaling, random delay, morphing, and chaffing. Chameleon captures real leakage deformations introduced by actual hardware implementations, making it a realistic and valuable tool for evaluating side-channel countermeasures. A key feature of Chameleon is its dual focus on the segmentation and attack stages of the side-channel analysis process. It is the first dataset designed to facilitate the challenging task of segmenting cryptographic operations from obfuscated traces, offering precise metadata that pinpoints the start and end of each operation. The high-quality metadata enables systematic research into segmentation techniques, a critical step often overlooked in previous datasets.

Introduction

In the realm of hardware security, Side-Channel Analysis (SCA) represents a formidable threat. By measuring physical outputs like power consumption, attackers can infer secret cryptographic keys. To defend against this, engineers implement hiding techniques—countermeasures that desynchronize or blur the signal. However, much existing research utilizes datasets where this obfuscation is added mathematically post-acquisition, failing to account for the complex physical deformations inherent in real hardware. Furthermore, the critical task of segmentation—identifying the exact temporal location of a cryptographic operation within a noisy trace—is frequently bypassed in research datasets, despite being one of the most challenging aspects of a real-world attack.

Methodology and Infrastructure

The researchers developed a robust acquisition infrastructure using a RISC-V CPU implemented on an FPGA. This platform natively supports four major hiding techniques:

  1. Dynamic Frequency Scaling (DFS): Randomly varying the CPU clock speed during execution.
  2. Random Delay (RD): Inserting dummy instructions to desynchronize the trace.
  3. Morphing (MRP): Utilizing mathematically equivalent instructions with distinct power profiles.
  4. Chaffing (CHF): Executing decoy encryption threads concurrently with the target operation.
Hardware acquisition infrastructure featuring the FPGA board, pinpointing unit, and host system.

A core innovation is the hardware-based Pinpointing Unit, which records the exact start and end times of every cryptographic operation. This provides the ground-truth metadata necessary for systematic research into both segmentation and attack methodologies.

The Chameleon Dataset

The Chameleon dataset consists of over 600 GB of data, representing more than 60 hours of acquisition time. It is organized into five sub-datasets:

  • BASE: Unprotected traces serving as a control group.
  • DFS, RD, MRP, CHF: Traces protected by each respective hiding technique.

Each entry includes raw power traces, plaintexts, keys, and precise pinpoint metadata, making it a comprehensive resource for the SCA community.

Segmentation and Attack Analysis

Chameleon is the first dataset to explicitly support the evaluation of the two-stage SCA process: segmentation and attack.

The two-stage analysis process: (a) segmentation of operations from raw traces and (b) the subsequent attack stage to recover secret keys.

Key Results

  • Segmentation Success: Using a Convolutional Neural Network (CNN) with a sliding window approach, the authors achieved an Intersection over Union (IoU) of 84% to 91% across all obfuscation types. By applying custom heuristics to filter the CNN output, they achieved perfect (0% False Negative/Positive) localization of encryptions.
  • Attack Resistance:
    • Recovering keys from the BASE dataset required only 14 traces using a CNN attack.
  • Chaffing (CHF) was the most effective defense, requiring 104 traces for a deep-learning attack, while traditional Correlation Power Analysis (CPA) failed completely.
  • DFS also proved highly resilient, with a CNN attack success rate of only 0.7%, highlighting the difficulty of attacking frequency-modulated signals.
  • Physical Reality: The data confirms that real hardware noise, such as transitions between clock frequencies, is significantly more complex than the Gaussian noise typically used in simulations.

Conclusion

Chameleon addresses a critical gap in side-channel research by providing a large-scale, real-world dataset of obfuscated traces. It demonstrates that while modern AI can effectively segment even heavily distorted signals, certain hiding techniques like Chaffing and DFS remain highly effective at protecting secret data. Future work will focus on more granular sample-level classification and the inclusion of masking countermeasures to further enhance the realism and utility of the dataset.

Summary updated

Abstract

Side-channel attacks exploit unintended information leakage emitted by cryptographic devices to extract sensitive data. Hiding techniques are a cost-effective countermeasure designed to obfuscate the side-channel leakage and hinder these attacks. Available open datasets rely on artificial models to simulate hiding effects, preventing a realistic assessment of these countermeasures and, thus, leaving a pressing need for datasets offering real-world, obfuscated side-channel measurements. Chameleon introduces the first comprehensive dataset of real-world, obfuscated power traces collected from a RISC-V-based System-on-Chip. The traces are obfuscated using four state-of-the-art hiding techniques: dynamic frequency scaling, random delay, morphing, and chaffing. Chameleon captures real leakage deformations introduced by actual hardware implementations, making it a realistic and valuable tool for evaluating side-channel countermeasures. A key feature of Chameleon is its dual focus on the segmentation and attack stages of the side-channel analysis process. It is the first dataset designed to facilitate the challenging task of segmenting cryptographic operations from obfuscated traces, offering precise metadata that pinpoints the start and end of each operation. The high-quality metadata enables systematic research into segmentation techniques, a critical step often overlooked in previous datasets. Chameleon provides an essential platform for researchers to develop and test new side-channel attacks, highlighting the vulnerabilities of current hiding techniques. By offering a more realistic assessment of countermeasure effectiveness, Chameleon is an invaluable tool for advancing the state-of-the-art in the side-channel evaluation.
BibTeX
@article{galli2025chameleon,
  author = {Davide Galli and Giuseppe Chiari and Davide Zoni},
  title = {{Chameleon: A Dataset for Segmenting and Attacking Obfuscated Power Traces in Side-Channel Analysis}},
  journal = {IACR Trans. Cryptogr. Hardw. Embed. Syst.},
  year = {2025},
  url = {https://openreview.net/forum?id=Enr0WFOxgT},
}
Figure