XOR-PUF Modeling with Custom SVM Solvers

Introduction

This project focuses on enhancing cybersecurity in Operational Technology (OT) systems by analyzing and predicting XOR-based Physical Unclonable Functions (PUFs). By understanding XOR-PUF behavior, we aim to improve the security of OT systems against potential threats, specifically modeling these non-linear physical systems to prevent cloning and tampering.## Key Technical Achievements

1. Custom SVM Implementation

Unlike standard library implementations, this project features a custom Support Vector Machine (SVM) solver built from scratch.

• Methodology: Solved a complex binary classification challenge using Primal Gradient Descent.
• Optimization: Tailored the objective function to handle the specific constraints of cryptographic hardware data.

2. Advanced Feature Mapping

To accurately capture the complexities of XOR-PUFs, we implemented advanced feature mapping techniques.

• Non-Linear Modeling: These techniques map raw challenge-response data into higher dimensions to model the behavior of the non-linear physical system effectively.
• Complexity Handling: Successfully transforms the physical intricacies of the PUF into a format solvable by the linear SVM.

3. Performance & Results

The model was rigorously tested to ensure robustness against adversarial attacks.

• Accuracy: Achieved 99% accuracy on a 20,000 sample test set.
• Impact: Demonstrated a robust method for predicting XOR-PUF responses, validating the system's ability to mitigate tampering and cloning risks in critical OT infrastructure.

Objectives

• Develop SVM-based Solvers: Create solvers from scratch to model XOR-PUF behavior without reliance on "black box" libraries.
• Cybersecurity Focus: Enhance security in OT systems by validating hardware authenticity.
• Performance Evaluation: Validate the model's predictive power using extensive datasets.