Related Experiment Video
Updated: Apr 28, 2026
11:18
Closed-loop Neuro-robotic Experiments to Test Computational Properties of Neuronal Networks
Published on: March 2, 2015
10.2K
Clock Glitch Fault Attacks on Deep Neural Networks and Their Countermeasures.
Sangwon Lee1, Suhyung Kim1, Seongwoo Hong2
1Department of Information Security, Hoseo University, Asan-Si 31499, Republic of Korea.
Sensors (Basel, Switzerland)
|May 14, 2025
Summary
Clock glitch fault attacks can cause deep neural networks (DNNs) on edge devices to misclassify data with high probability. This research proposes algorithmic countermeasures to effectively prevent such security vulnerabilities in DNN models.
Area of Science:
- Computer Science
- Artificial Intelligence
- Cybersecurity
Background:
- Deep neural networks (DNNs) are increasingly deployed on edge devices for real-time processing in resource-constrained environments.
- Deployment on edge devices makes DNNs vulnerable to physical attacks, including fault injection, which can compromise their integrity.
- Clock glitch fault attacks are a specific threat that can manipulate the execution of DNN models on embedded systems.
Purpose of the Study:
- To investigate the effectiveness of clock glitch fault attacks in inducing misclassifications in DNN models executed on microcontrollers.
- To develop and evaluate algorithmic countermeasures against clock glitch attacks targeting Softmax and Sigmoid activation functions.
Main Methods:
- Implementation of DNN models for image classification (MNIST, CIFAR-10, Kaggle datasets) on a microcontroller.
- Application of clock glitch fault attacks during DNN model execution to observe misclassification patterns.
- Development of algorithmic defenses at the Softmax and Sigmoid function level to mitigate fault attack impacts.
Main Results:
- Clock glitch fault attacks were demonstrated to cause a high probability of serious misclassifications in DNNs for both multi-class and binary image classification tasks.
- Experimental results confirmed the vulnerability of DNNs executed on microcontrollers to these specific fault injection attacks.
- The proposed algorithmic countermeasures successfully prevented misclassification incidents caused by clock glitch attacks.
Conclusions:
- Clock glitch fault attacks pose a significant security threat to DNNs deployed on edge devices, leading to potentially critical misclassifications.
- Algorithmic-level countermeasures targeting Softmax and Sigmoid functions are effective in defending against these attacks.
- The findings highlight the need for robust security measures in embedded AI systems to ensure reliable and secure operation.
