Implementation and Analysis of Side-Channel Attack Mitigation Based on Autoencoder

Abstract

A side-channel attack is a method of stealing confidential data from a computer chip's encryption process. This kind of attack specifically targets a computing device's hardware by exploiting its power consumption and electromagnetic radiation discharge while operational. Attackers may use wired or wireless probes to try and access the chip's power pins or intercept the leaked radiation. After collecting a lot of signals, they analyze the features using statistics and machine learning to figure out the key information of the encryption algorithm. This is a big danger to the device's security. To prevent SCA, they can add noise to the output signal. This makes the attack less effective. Injecting noise is an often used method. Introducing noise into the power pins of a chip can disrupt its original features and use the noise waveform as a defense against attacks. Our research presents a novel approach to improve the efficacy of side-channel attacks. We employ an autoencoder (AE) to preprocess noised signal traces, which is seamlessly integrated into the ChipWhisperer hardware. The AE excels at significantly reducing interference caused by noise, which ultimately enhances the effectiveness of the side-channel attacks. We tested the encryption algorithm on the Arduino Uno board and introduced a strategy to reduce distinctive signals produced by the power supply through a filter capacitor. This approach is a highly effective defense against correlation power analysis (CPA) attacks.