Cryptanalysis of the Hardware Implementation of GIFT-COFB Cryptographic System
Abstract
In recent years, the increase of devices that are interconnected has led to a growing need for lightweight cryptographic systems in order to preserve data integrity, leading NIST to make a call for algorithms in search of new lightweight cryptographic systems. These systems must be designed for implementation in resource limited environments with one example of such an algorithm being the NIST finalist GIFT-COFB. Lightweight cryptographic ciphers are differentiated from conventional ciphers and may have different vulnerabilities as such. Side-channel attacks have proven to be an effective method of breaking cryptographic systems. Due to recent advances in methodology, there is a need to better understand the efficacy and intricacy of side-channel attacks using deep learning techniques. This project is focused on the study of classical side-channel attack methods, such as correlation power analysis, along with newer attack methods that utilize a deep-learning structure and their efficacy attacking the hardware implemented GIFT-COFB cryptographic system. Specifically, this project focused on studying recently introduced methods of deep learning power analysis and differential deep learning analysis with varying underlying neural net architectures such as convolutional and multilayer perceptron. The efficacy of these attacks was examined. Due to the novelty of these methods, there is a need for understanding the role of different architectures of attack and their impact on success. For example, the network structure, trace sample size, underlying metric, and network hyperparameters play an imperative role in attack performance to successfully attack the GIFT-COFB cryptographic system.
Cryptanalysis of the Hardware Implementation of GIFT-COFB Cryptographic System
In recent years, the increase of devices that are interconnected has led to a growing need for lightweight cryptographic systems in order to preserve data integrity, leading NIST to make a call for algorithms in search of new lightweight cryptographic systems. These systems must be designed for implementation in resource limited environments with one example of such an algorithm being the NIST finalist GIFT-COFB. Lightweight cryptographic ciphers are differentiated from conventional ciphers and may have different vulnerabilities as such. Side-channel attacks have proven to be an effective method of breaking cryptographic systems. Due to recent advances in methodology, there is a need to better understand the efficacy and intricacy of side-channel attacks using deep learning techniques. This project is focused on the study of classical side-channel attack methods, such as correlation power analysis, along with newer attack methods that utilize a deep-learning structure and their efficacy attacking the hardware implemented GIFT-COFB cryptographic system. Specifically, this project focused on studying recently introduced methods of deep learning power analysis and differential deep learning analysis with varying underlying neural net architectures such as convolutional and multilayer perceptron. The efficacy of these attacks was examined. Due to the novelty of these methods, there is a need for understanding the role of different architectures of attack and their impact on success. For example, the network structure, trace sample size, underlying metric, and network hyperparameters play an imperative role in attack performance to successfully attack the GIFT-COFB cryptographic system.