BLADEI v3.0.0

This release introduces dual-head classification, enabling both Trojan detection (Benign vs Malicious) and hardware family identification across six categories (CRYPTO, COMMS, MCU/CPU, BUS/DISPLAY, ITC99, ISCAS89).

Key updates include:

• Pre-trained models built from the sample dataset
• Expanded 278-dimensional feature vector (256 byte histogram + 10 statistical + 12 structural features)
• GridSearchCV hyperparameter optimization for improved model performance
• Optional TSVD and SMOTE via command-line flags (--tsvd, --smote)
• Updated deployment script with confidence scores and automatic quarantine

The sample dataset is now hosted separately as a password-protected archive due to file size constraints. Access is available upon request: ryes@email.sc.edu

BLADEI v2.0.2

This release eliminates unused directories and files from the repository. It also transitions the training and deployment Jupyter notebooks into Python scripts for improved usability.

These updates bring the project to its finalized prototype for Version 2. As of right now, BLADEI operates solely on a ML model framework for generating predictions.

BLADEI v2.0.1

This release introduces a license and includes minor updates to the documentation. Namely, the acknowledgements section has been revised, and the future work section adjusted to reflect the project’s current direction.

BLADEI v2.0.0

This updated release extends BLADEI to support a wider range of PYNQ and ARM-based FPGA platforms. It allows on-device deployment on ARMv7 boards (e.g., PYNQ-Z1/Z2) and both training and deployment on ARMv8 boards (e.g., Zynq UltraScale+ MPSoC, RFSoC, Kria), while maintaining the same lightweight, byte-level classification approach.

BLADEI v1.0.0

This initial release provides a lightweight, byte-level malware detection system designed specifically for the PYNQ-Z1/ Z2 (Zynq-7000 SoC) board. It enables on-device inference without requiring reverse engineering of source code or netlists. However, training the model must be performed externally due to limited computational resources.