Chameleon Dataset Overview

Dataset Introduction

The Chameleon dataset is designed for obfuscating power traces in side‑channel analysis, containing real power traces collected from a 32‑bit RISC‑V SoC that implements four masking countermeasures: Dynamic Frequency Scaling (DFS), Random Delay (RD), Morphing (MRP), and Chaffing (CHF). Each side‑channel trace comprises multiple interleaved encryption operations and general‑purpose applications.

Dataset Structure

The dataset is organized by countermeasure, and each file contains the following structure:

• Data: Power traces, containing 134,217,550 time samples. The DFS, RD, MRP, and CHF datasets contain 256, 512, 512, and 1,024 traces respectively. The traces capture interleaved AES encryption and general‑purpose application execution on the SoC.
• Metadata: Divided into three groups:
  • Encryption: Contains AES inputs such as keys and plaintexts.
  • Locations: Contains start and end time‑sample indices for each AES execution within each trace file.
  • Frequency: Provides frequency‑variation labels for each power trace, available only in the DFS dataset.

Dataset Format

The dataset consists of four sub‑datasets, one per countermeasure, stored in separate folders. To reduce individual file size, each sub‑dataset is split into 16 files according to the encryption key; the key is a 16‑byte array where only the first byte varies, the remaining 15 are fixed.

Dataset Creation

The Chameleon dataset addresses the lack of real‑world complexity in existing side‑channel analysis datasets by providing realistic masking methods and segmented encryption operations. Data were collected from a genuine hardware‑software infrastructure using a Picoscope 5244d digital sampling oscilloscope and a NewAE CW305 board; the FPGA implements a SoC containing a 1.5 Mps UART interface, a compute platform, and a localization unit.

Societal Impact

The Chameleon dataset aims to enhance side‑channel security by identifying and mitigating potential weaknesses in cryptographic implementations, fostering the development of stronger countermeasures, and ultimately contributing to a safer digital world.

Dataset Download

• Full dataset:

from datasets import load_dataset
dataset = load_dataset("hardware-fab/Chameleon")

• Select sub‑dataset:

from datasets import load_dataset
sub_dataset = load_dataset("hardware-fab/Chameleon", <sub_dataset>)

Replace <sub_dataset> with DFS, RD, MRP, or CHF.

Dataset Size

Dataset size ranges from 1K < n < 10K.

Dataset Configuration

• DFS: DFS/*.h5
• RD: RD/*.h5
• MRP: MRP/*.h5
• CHF: CHF/*.h5

Dataset Tags

• AES
• RISC‑V
• Random‑Delay
• Dynamic‑Frequency‑Scaling
• Chaffing
• Morphing
• Side‑Channel‑Analysis

Dataset License

Open Data Commons License cc-by-4.0