Deep Learning-based
Side Channel Attack Detection System

Research on next-generation security solutions that analyze physical side effects
to detect and defend vulnerabilities in cryptographic systems

Yonsei University | 2025 Spring Semester

Research Overview

🔒

Side Channel Attack

An attack technique that extracts secret information by analyzing physical side effects such as power consumption, electromagnetic radiation, and processing time

🧠

Deep Learning-based Detection

Utilizing three deep learning models—MLP, CNN1D, and LSTM—to detect and classify side channel attacks

🛡️

Defense System

Developing real-time detection-based defense strategies and hardware-based security solutions

Types of Side Channel Attacks

SPA (Simple Power Analysis)

Visually analyzing power consumption patterns during computation through simple power analysis

📊

DPA (Differential Power Analysis)

Using statistical techniques to analyze differences in power consumption and compare with predicted results

📈

TA (Template Analysis)

Constructing a probabilistic model from a pre-collected device for comparative analysis

🎯

Performance Comparison of Deep Learning Models

MLP

Multi-Layer Perceptron
99.4%
Training Time 12.3 min
Memory Usage 560 MB

LSTM

Long Short-Term Memory
93.1%
Training Time 18.5 min
Memory Usage 690 MB

CNN1D

1D Convolutional Neural Network
89.6%
Training Time 20.8 min
Memory Usage 780 MB

Experimental Data Visualization

Synthetic Data Distribution (PCA)

• Samples: 200,000 • Features: 100 dimensions • Classes: 2

Learning Curve

■ MLP ■ LSTM ■ CNN1D

Research Results

🎯 Key Achievements

Successfully classified side channel attack patterns with over 90% accuracy using deep learning techniques.

Excellence of MLP Model

Despite its relatively simple structure, MLP achieved the best classification performance (99.4%).

Efficiency Validation

MLP was also confirmed to be the most efficient in terms of training time and memory usage.