Abstract:<p>In the field of image recognition, machine learning technologies, especially deep learning, have been rapidly advancing alongside the advances of hardware such as GPUs. In image recognition, in general, large numbers of labeled images to be identified are input to a neural network, and repeatedly learning the images enables the neural network to identify objects with high accuracy. A new profiling side-channel attack method, the deep learning side-channel attack (DL-SCA), utilizes the neural network's high identifying ability to unveil a cryptographic module's secret key from side-channel information. In DL-SCAs, the neural network is trained with power waveforms captured from a target cryptographic module, and the trained network extracts the leaky part that depends on the secret. However, at this stage, the main target of investigation has been software implementation, and studies regarding hardware implementation, such as ASIC, are somewhat lacking. In this paper, we first depict deep learning techniques, profiling side-channel attacks, and leak models to clarify the relation between secret and side channels. Next, we investigate the use of DL-SCA against hardware implementations of AES and discuss the problem derived from the Hamming distance model and ShiftRow operation of AES. To solve the problem, we propose a new network training method called "mixed model dataset based on round-round XORed value." We prove that our proposal solves the problem and gives the attack capability to neural networks. We also compare the attack performance and characteristics of DL-SCA to conventional analysis methods such as correlation power analysis and conventional template attack. In our experiment, a dedicated ASIC chip for side-channel analysis is utilized and the chip is also equipped with a side-channel countermeasure AES. We show how DL-SCA can recover secret keys against the side-channel countermeasure circuit. Our results demonstrate that DL-SCA can be a more powerful option against side-channel countermeasure implementations than conventional SCAs.</p>

Pareto Frequency-Aware Power Side-Channel Countermeasure Exploration on CNN Systolic Array.

RE-Specter: Examining the Architectural Features of Configurable CNN with Power Side-Channel

Dual-Rail Precharge Logic-Based Side-Channel Countermeasure for DNN Systolic Array

Design of Convolutional Neural Networks Architecture for Non-Profiled Side-Channel Attack Detection

Defense against ML-based Power Side-channel Attacks on DNN Accelerators with Adversarial Attacks

A Desynchronization-Based Countermeasure Against Side-Channel Analysis of Neural Networks

Reverse engineering convolutional neural networks through side-channel information leaks

A Novel Power System Side Channel Attack Method Based on Machine Learning CNN-Transformer

Energy-Efficient Side-Channel Attack Countermeasure with Awareness and Hybrid Configuration Based on It

Power Side-Channel Attacks on BNN Accelerators in Remote FPGAs

Research on Side-Channel Analysis Based on Deep Learning with Different Sample Data

I Know What You See: Power Side-Channel Attack on Convolutional Neural Network Accelerators

Shift-Invariance Robustness of Convolutional Neural Networks in Side-Channel Analysis

Power Supply Noise Aware Evaluation Framework for Side Channel Attacks and Countermeasures

An Energy-Efficient Mixed-Signal Parallel Multiply-Accumulate (MAC) Engine Based on Stochastic Computing

PowerGAN: A Machine Learning Approach for Power Side-Channel Attack on Compute-in-Memory Accelerators

Power Side Channels in Security ICs: Hardware Countermeasures

Side-Channel Information Leakage Analysis And Countermeasures In An Embedded Cpu Microarchitecture

Physical Time-Varying Transfer Functions as Generic Low-Overhead Power-SCA Countermeasure

Deep learning side-channel attack against hardware implementations of AES

Energy Efficient Obfuscation of Side-Channel Leakage for Preventing Side-Channel Attacks