README.md

---
license: cc-by-4.0
tags:
- AES
- RISC-V
- Random-Delay
- Dynamic-Frequency-Scaling
- Chaffing
- Morphing
- Side-Channel-Analysis
pretty_name: Chameleon
size_categories:
- 1K<n<10K
configs:
- config_name: BASE
  data_files: "BASE/*.h5" 
- config_name: DFS
  data_files: "DFS/*.h5"
- config_name: RD
  data_files: "RD/*.h5"
- config_name: MRP
  data_files: "MRP/*.h5"
- config_name: CHF
  data_files: "CHF/*.h5"
---

# Chameleon

`Chameleon` is a dataset designed for side-channel analysis of obfuscated power traces. 
It contains real-world power traces collected from a 32-bit RISC-V System-on-Chip implementing four hiding countermeasures: 
Dynamic Frequency Scaling (DFS), Random Delay (RD), Morphing (MRP), and Chaffing (CHF).
The dataset also includes side-channel power traces without any active countermeasure (BASE).
Each side-channel trace includes multiple cryptographic operations 
interleaved with general-purpose applications.

- **Curated by:** hardware-fab
- **License:** Open Data Commons License [cc-by-4.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/cc-by-4.0.md)
- **Paper**: [Chameleon: A Dataset for Segmenting and Attacking Obfuscated Power Traces in Side-Channel Analysis](https://doi.org/10.46586/tches.v2025.i3.389-412)

<div align="center">
   <img src="https://github.com/hardware-fab/chameleon/blob/main/images/chameleon_logo.png?raw=true" alt="Chameleon Logo" width="150">
</div>

The dataset is designed to aid research in:

- Segmentation methods (locate and isolate cryptographic operations)
- Side-channel analysis methods (attacking devices with hiding countermeasures)

## How to Download
Full dataset:  
⚠ **WARNING**: Full dataset requires more than 600 GB of space.
```python 
from huggingface_hub import snapshot_download

snapshot_download(repo_id="hardware-fab/Chameleon", repo_type="dataset", local_dir="<download_path>")
```

One sub-dataset of choice:
```python 
from huggingface_hub import snapshot_download

snapshot_download(repo_id="hardware-fab/Chameleon", repo_type="dataset", local_dir="<download_path>", allow_patterns="<sub_dataset>/*")
```
Replace `<sub_dataset>` with `BASE`, `DFS`, `RD`, `MRP`, `CHF`.

## Dataset Structure

The dataset is divided per hiding countermeasure. Each file has the following structure:
* **Data:** The data are power traces of 134,217,550 time samples.
 BASE, DFS, RD, MRP, and CHF sub-dataset
 contain 256, 256, 512, 512, and 1024 data respectively.
 The traces capture the SoC execution of AES encryptions interleaved with general-purpose applications.
* **Metadata:** The metadata are divided into three groups:
  * **Ciphers:** This group contains the AES inputs:
    * `key`: The secret key used for AES encryption.
    * `plaintexts`: The plaintext used for the AES encryption.
  * **Pinpoints:** This group contains the start and end time samples of each AES execution in each trace file.
    * `start`: The starting sample of the AES encryption. It takes values ranging from 0 to 134,217,550.
    * `end`: The ending sample of the AES encryption. It takes values ranging from 0 to 134,217,550.
  * **Frequencies:** This group provides labels for each power trace, indicating the frequency changes during the measurement.
      _Notably, this metadata is only available for the sub-datasets with DFS enabled_. Each metadata has two fields:
    * `samples`: This field denotes the time sample at which a frequency change happens, with integer values ranging from 0 to 134,217,550.
    * `frequencies`: This field specifies the new operating frequency starting from the corresponding sample. It can take floating values from 5MHz to 100MHz.

### Dataset Format

The dataset is divided into five sub-datasets, one for each hiding countermeasure, 
stored in different folders. To alleviate the size of the individual files, 
we partitioned each sub-dataset into 16 files based on the cryptographic key. 
Keys are 16-byte arrays, we vary only the first byte in each trace, 
keeping the remaining 15 fixed.

| Chunk          | First key byte values | Disk size (GB) | # Data  |
|----------------|----------------|----------------|----------------|
| base_chunk_1.h5  | [0x00-0x0f] | 4.3  | 16 |  
| base_chunk_2.h5  | [0x10-0x0f] | 4.3  | 16 |  
| base_chunk_3.h5  | [0x20-0x2f] | 4.3  | 16 |  
| base_chunk_4.h5  | [0x30-0x3f] | 4.3  | 16 |  
| base_chunk_5.h5  | [0x40-0x4f] | 4.3  | 16 |  
| base_chunk_6.h5  | [0x50-0x5f] | 4.3  | 16 |  
| base_chunk_7.h5  | [0x60-0x6f] | 4.3  | 16 |  
| base_chunk_8.h5  | [0x70-0x7f] | 4.3  | 16 |  
| base_chunk_9.h5  | [0x80-0x8f] | 4.3  | 16 |  
| base_chunk_10.h5 | [0x90-0x9f] | 4.3  | 16 |  
| base_chunk_11.h5 | [0xa0-0xaf] | 4.3  | 16 |  
| base_chunk_12.h5 | [0xb0-0xbf] | 4.3  | 16 |  
| base_chunk_13.h5 | [0xc0-0xcf] | 4.3  | 16 |  
| base_chunk_14.h5 | [0xd0-0xdf] | 4.3  | 16 |  
| base_chunk_15.h5 | [0xe0-0xef] | 4.3  | 16 |  
| base_chunk_16.h5 | [0xf0-0xff] | 4.3  | 16 |
| dfs_chunk_1.h5  | [0x00-0x0f] | 4.3  | 16 |  
| dfs_chunk_2.h5  | [0x10-0x0f] | 4.3  | 16 |  
| dfs_chunk_3.h5  | [0x20-0x2f] | 4.3  | 16 |  
| dfs_chunk_4.h5  | [0x30-0x3f] | 4.3  | 16 |  
| dfs_chunk_5.h5  | [0x40-0x4f] | 4.3  | 16 |  
| dfs_chunk_6.h5  | [0x50-0x5f] | 4.3  | 16 |  
| dfs_chunk_7.h5  | [0x60-0x6f] | 4.3  | 16 |  
| dfs_chunk_8.h5  | [0x70-0x7f] | 4.3  | 16 |  
| dfs_chunk_9.h5  | [0x80-0x8f] | 4.3  | 16 |  
| dfs_chunk_10.h5 | [0x90-0x9f] | 4.3  | 16 |  
| dfs_chunk_11.h5 | [0xa0-0xaf] | 4.3  | 16 |  
| dfs_chunk_12.h5 | [0xb0-0xbf] | 4.3  | 16 |  
| dfs_chunk_13.h5 | [0xc0-0xcf] | 4.3  | 16 |  
| dfs_chunk_14.h5 | [0xd0-0xdf] | 4.3  | 16 |  
| dfs_chunk_15.h5 | [0xe0-0xef] | 4.3  | 16 |  
| dfs_chunk_16.h5 | [0xf0-0xff] | 4.3  | 16 |  
| rd_chunk_1.h5   | [0x00-0x0f] | 8.6  | 32 |  
| rd_chunk_2.h5   | [0x10-0x0f] | 8.6  | 32 |  
| rd_chunk_3.h5   | [0x20-0x2f] | 8.6  | 32 |  
| rd_chunk_4.h5   | [0x30-0x3f] | 8.6  | 32 |  
| rd_chunk_5.h5   | [0x40-0x4f] | 8.6  | 32 |  
| rd_chunk_6.h5   | [0x50-0x5f] | 8.6  | 32 |  
| rd_chunk_7.h5   | [0x60-0x6f] | 8.6  | 32 |  
| rd_chunk_8.h5   | [0x70-0x7f] | 8.6  | 32 |  
| rd_chunk_9.h5   | [0x80-0x8f] | 8.6  | 32 |  
| rd_chunk_10.h5  | [0x90-0x9f] | 8.6  | 32 |  
| rd_chunk_11.h5  | [0xa0-0xaf] | 8.6  | 32 |  
| rd_chunk_12.h5  | [0xb0-0xbf] | 8.6  | 32 |  
| rd_chunk_13.h5  | [0xc0-0xcf] | 8.6  | 32 |  
| rd_chunk_14.h5  | [0xd0-0xdf] | 8.6  | 32 |  
| rd_chunk_15.h5  | [0xe0-0xef] | 8.6  | 32 |  
| rd_chunk_16.h5  | [0xf0-0xff] | 8.6  | 32 |  
| mrp_chunk_1.h5  | [0x00-0x0f] | 8.6  | 32 |  
| mrp_chunk_2.h5  | [0x10-0x0f] | 8.6  | 32 |  
| mrp_chunk_3.h5  | [0x20-0x2f] | 8.6  | 32 |  
| mrp_chunk_4.h5  | [0x30-0x3f] | 8.6  | 32 |  
| mrp_chunk_5.h5  | [0x40-0x4f] | 8.6  | 32 |  
| mrp_chunk_6.h5  | [0x50-0x5f] | 8.6  | 32 |  
| mrp_chunk_7.h5  | [0x60-0x6f] | 8.6  | 32 |  
| mrp_chunk_8.h5  | [0x70-0x7f] | 8.6  | 32 |  
| mrp_chunk_9.h5  | [0x80-0x8f] | 8.6  | 32 |  
| mrp_chunk_10.h5 | [0x90-0x9f] | 8.6  | 32 |  
| mrp_chunk_11.h5 | [0xa0-0xaf] | 8.6  | 32 |  
| mrp_chunk_12.h5 | [0xb0-0xbf] | 8.6  | 32 |  
| mrp_chunk_13.h5 | [0xc0-0xcf] | 8.6  | 32 |  
| mrp_chunk_14.h5 | [0xd0-0xdf] | 8.6  | 32 |  
| mrp_chunk_15.h5 | [0xe0-0xef] | 8.6  | 32 |  
| mrp_chunk_16.h5 | [0xf0-0xff] | 8.6  | 32 |  
| chf_chunk_1.h5  | [0x00-0x0f] | 17.2 | 64 |  
| chf_chunk_2.h5  | [0x10-0x0f] | 17.2 | 64 |  
| chf_chunk_3.h5  | [0x20-0x2f] | 17.2 | 64 |  
| chf_chunk_4.h5  | [0x30-0x3f] | 17.2 | 64 |  
| chf_chunk_5.h5  | [0x40-0x4f] | 17.2 | 64 |  
| chf_chunk_6.h5  | [0x50-0x5f] | 17.2 | 64 |  
| chf_chunk_7.h5  | [0x60-0x6f] | 17.2 | 64 |  
| chf_chunk_8.h5  | [0x70-0x7f] | 17.2 | 64 |  
| chf_chunk_9.h5  | [0x80-0x8f] | 17.2 | 64 |  
| chf_chunk_10.h5 | [0x90-0x9f] | 17.2 | 64 |  
| chf_chunk_11.h5 | [0xa0-0xaf] | 17.2 | 64 |  
| chf_chunk_12.h5 | [0xb0-0xbf] | 17.2 | 64 |  
| chf_chunk_13.h5 | [0xc0-0xcf] | 17.2 | 64 |  
| chf_chunk_14.h5 | [0xd0-0xdf] | 17.2 | 64 |  
| chf_chunk_15.h5 | [0xe0-0xef] | 17.2 | 64 |  
| chf_chunk_16.h5 | [0xf0-0xff] | 17.2 | 64 |

Following the structure of the dataset, below are HDF5 fields used and their atomic type: 

```
.  
├── data  
│    └── traces  
│       ├── trace_0 [int16]  
│       ├── ...  
│       └── trace_n [int16]  
└── metadata  
    ├── ciphers  
    │   ├── ciphers_0  
    │   │   ├── key [('k', uint8, (16,))]  
    │   │   └── plaintexts [('p', uint8, (16,))]  
    │   ├── ...  
    │   └── ciphers_n  
    │       ├── key [('k', uint8, (16,))]
    │       └── plaintexts [('p', uint8, (16,))] 
    ├── pinpoints  
    │   ├── pinpoints_0 [('start', int32), ('end', int32)]  
    │   ├── ...  
    │   └── pinpoints_n [('start', int32), ('end', int32)]  
    └── frequencies  
        ├── frequencies_0  [('sample', int32), ('frequency', float32)]  
        ├── ...  
        └── frequencies_n  [('sample', int32), ('frequency', float32)]  
```

## Dataset Creation

Existing datasets for side-channel analysis lack real-world complexity. 
Chameleon addresses this by providing the first dataset of:

- **Real-world hiding methods**: Traces are collected from a real system implementing four actual obfuscation countermeasures (DFS, RD, MRP, CHF).
- **Segmentable cryptographic operations**: Chameleon includes multiple operations interleaved with real-world applications,
  mimicking real-world use and necessitating segmentation techniques for attack.

### Data Collection

The data are collected from a real-world hardware-software infrastructure, available [online](https://doi.org/10.48550/arXiv.2407.17432). 
The setup comprises a host PC, 
a [Picoscope 5244d](https://www.picotech.com/download/datasheets/picoscope-5000d-series-data-sheet.pdf) 
digital sampling oscilloscope (DSO), and 
a [NewAE CW305](https://rtfm.newae.com/Targets/CW305%20Artix%20FPGA/)
board which hosts an [AMD Artix-7 FPGA](https://docs.amd.com/v/u/en-US/ds180_7Series_Overview). 
The board is specifically designed to facilitate the deployment 
of digital designs targeting FPGAs and studying their side-channel behavior. 
The sampling rate of the DSO is set to 125Msample/s 
with a resolution of 12 bits for the entire dataset.

The FPGA implements a system-on-chip consisting of a 1.5Mps UART 
interface to communicate with the host, a computing platform 
to execute the user applications, and a pinpointing unit 
to record the beginning and end time sample for each cryptographic operation in the traces. 
The computing platform implements an in-order 32-bit RISC-V CPU 
that has been modified to implement the analyzed hiding methods. In particular,
we implement random delay and dynamic frequency scaling in hardware,
while morphing and chaffing are software-implemented. Notably, 
the CPU is clocked at 50MHz for all acquisition campaigns, 
except for the DFS ones, for which the DFS actuator 
is instructed to change the operating frequency of
the computing platform randomly at its maximum speed.

As the cryptographic operation of choice, we selected 
the [OpenSSL AES implementation](https://github.com/openssl/openssl), 
representing the standard for symmetric cryptography.

## Social Impact of Dataset

Chameleon has been developed to enhance side-channel security.
Notably, the side-channel analysis represents a standard procedure 
for evaluating novel countermeasures. Indeed, 
the [NIST FIPS-140v3](https://doi.org/10.6028/NIST.FIPS.140-3) 
standard enforces side-channel security as a mandatory step 
in the security validation of any novel software- and 
hardware-implemented cryptographic device. To this end, 
Chameleon is a valuable asset in strengthening real-world security 
by enabling researchers to identify and address potential 
weaknesses in cryptographic implementations. 
By promoting the creation of robust countermeasures, 
this dataset ultimately contributes to a more secure digital world.

As creating a high-quality training dataset is a fundamental requirement, the quality of Chameleon
sits on the time-consuming acquisition process that requires a clean-room acquisition setup and
system-on-chip. Without considering the design time to obtain the implementation of the computing
platform and the working acquisition setup, the time required by the acquisition procedure exceeded
58 hours.

## Citation

**BibTeX:**
```
@article{Galli_Chiari_Zoni_2025,
    title={Chameleon: A Dataset for Segmenting and Attacking Obfuscated Power Traces in Side-Channel Analysis},
    author={Galli, Davide and Chiari, Giuseppe and Zoni, Davide},
    volume={2025}, 
    number={3},
    journal={IACR Transactions on Cryptographic Hardware and Embedded Systems},
    year={2025},
    month={Jun.},
    pages={389–412},
    DOI={10.46586/tches.v2025.i3.389-412},
    url={https://tches.iacr.org/index.php/TCHES/article/view/12221}
}
```

 **APA:**
 
 > Galli, D., Chiari, G., & Zoni, D. (2025). Chameleon: A Dataset for Segmenting and Attacking Obfuscated Power Traces in Side-Channel Analysis. IACR Transactions on Cryptographic Hardware and Embedded Systems, 2025(3), 389-412.


## Note

This repository is protected by copyright and licensed under the 
Open Data Commons License [cc-by-4.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/cc-by-4.0.md) file.

© 2025 hardware-fab