Two Birds, One Mask: Side-Channel-Resistant Unified Hardware for ML-DSA and ML-KEM (source code and datasets)

Two Birds, One Mask: Side-Channel-Resistant Unified Hardware for ML-DSA and ML-KEM

Source code

Released for the non-commercial use only. For licensing, please contact [e-mail].

Unified masked hardware architecture of ML-KEM (Kyber) and ML-DSA (Dilithium) compliant with NIST standards FIPS203 and FIPS204. It is based on the unprotected design DiKy.

Dataset

This dataset contains power consumption traces captured from a unified hardware implementation of the FIPS 203 (ML-KEM) and FIPS 204 (ML-DSA) post-quantum cryptography standards. The data is intended for side-channel analysis, specifically for evaluating the leakage of signing and decapsulation operations.

Data Format & Structure

The datasets are provided in HDF5 format for efficient access to large-scale trace data. Each file is organized as follows:

Fixed Input Traces: Located in fixed/traces with corresponding inputs in fixed/input (or fixed/inputs for full phases where public components vary).

Random Input Traces: Located in random/traces with corresponding inputs in random/inputs.

Metadata: The settings group contains oscilloscope configurations and sampling parameters.

Input Data Specification

To assist with leakage assessment and attack verification, the input byte ranges are structured as follows:

Byte Range (ML-DSA-44)
Description
Byte Range (ML-KEM-512)
Description

0–2559
Secret key
0–1631
Secret key

2560
Context
1632–2399
Ciphertext

2561–2588
Message
2400–2431
Shared key

2589–5008
Signature

Note on Sampling: Full-phase datasets have been down-sampled to 10 samples per clock cycle to optimize storage while maintaining signal integrity for analysis.

一掩双用：适配ML-DSA与ML-KEM的抗侧信道统一硬件

源代码
本代码仅可用于非商业用途，授权事宜请联系[e-mail]。

本架构为符合美国国家标准与技术研究院（NIST）FIPS203与FIPS204标准的ML-KEM（Kyber）与ML-DSA（Dilithium）统一掩码硬件架构，其原型基于未做防护的DiKy设计。

数据集
本数据集包含基于FIPS 203（ML-KEM）与FIPS 204（ML-DSA）后量子密码标准的统一硬件实现所采集的功耗轨迹数据，旨在用于侧信道分析，专门用于评估签名与解密封装操作的侧信道泄漏。

数据格式与结构
本数据集采用HDF5格式存储，以实现大规模轨迹数据的高效访问。每个文件的组织形式如下：
1. 固定输入轨迹：存储于fixed/traces路径下，对应输入数据存储于fixed/input路径；若为全阶段场景且公共组件存在差异，则对应输入数据存储于fixed/inputs路径。
2. 随机输入轨迹：存储于random/traces路径下，对应输入数据存储于random/inputs路径。
3. 元数据：settings组包含示波器配置与采样参数。

输入数据规范
为便于泄漏评估与攻击验证，输入字节范围的结构如下：

| ML-DSA-44字节范围 | 对应内容 | ML-KEM-512字节范围 | 对应内容 |
|-------------------|----------|-------------------|----------|
| 0–2559 | 私钥 | 0–1631 | 私钥 |
| 2560 | 上下文 | 1632–2399 | 密文 |
| 2561–2588 | 消息 | 2400–2431 | 共享密钥 |
| 2589–5008 | 签名 | 无 | 无 |

采样说明
全阶段数据集已被下采样至每个时钟周期10个采样点，在优化存储占用的同时，保留了分析所需的信号完整性。