Side-channel analysis of cryptographic implementations represents a critical issue for the design of secure hardware architectures. When observing physical properties of the hardware platform during executions, the side-channel attacker is able to leak informations about the secret used by the cryptosystem. Most recent line of works have showed how the machine learning revolution greatly enhance the attacking capabilities of the adversarial. In particular, artificial neural networks and deep learning models have been showed to play an important role for side-channel leakage modelling. Some of the most effective countermeasures to classical attacks relies on the introduction of process variability and desynchronizations in the computing platform. These protections have been proven to be very effective when applied for CMOS devices to counter most common attacks, like the well-known bayesian template attack. Deep learning models have instead showed to be very effective against a wide variety of hardware and software countermeasures. The objective of the thesis is to describe the role of machine learning and deep learning methods for modern side-channel analysis. We will present different techniques to exploit artificial neural networks for side-channel modelling, focusing on the capabilities of deep learning models to effectively counter protected implementations of the cryptosystem.
L’analisi del canale laterale delle implementazioni crittografiche rappresenta un problema critico per la progettazione di architetture hardware sicure. Osservando le proprietà fisiche della piattaforma hardware durante varie esecuzioni, l’attaccante del canale laterale è in grado di estrarre informazioni sulla chiave segreta utilizzata dal crittosistema. Gli studi più recenti hanno mostrato come la rivoluzione dell’apprendimento automatico migliori notevolmente le capacità di attacco dell’avversario. In particolare, è stato mostrato come le reti neurali artificiali e i modelli di apprendimento profondo ricoprano un ruolo importante per la modellazione dell’informazione contenuta nel canale laterale. Alcune delle contromisure più efficaci agli attacchi classici si basano sull’introduzione di variabilità di processo e di desincronizzazioni nella piattaforma di calcolo. Queste protezioni si sono rivelate molto efficaci quando applicate nei dispositivi CMOS per contrastare gli attacchi più comuni, come il noto attacco bayesiano a template. I modelli di apprendimento profondo hanno invece dimostrato di essere molto efficaci contro un’ampia varietà di contromisure hardware e software. L’obiettivo della tesi è quello di mostrare il ruolo dei metodi di apprendimento automatico e apprendimento profondo nella moderna analisi del canale laterale. Presenteremo diverse tecniche per sfruttare le reti neurali artificiali in questo contesto, concentrando la nostra ricerca sulle capacità dei modelli di apprendimento profondo di contrastare efficacemente le implementazioni protette del crittosistema.
Deep learning techniques for side-channel cipher attacks
ROCCAMENA, MASSIMILIANO
2021/2022
Abstract
Side-channel analysis of cryptographic implementations represents a critical issue for the design of secure hardware architectures. When observing physical properties of the hardware platform during executions, the side-channel attacker is able to leak informations about the secret used by the cryptosystem. Most recent line of works have showed how the machine learning revolution greatly enhance the attacking capabilities of the adversarial. In particular, artificial neural networks and deep learning models have been showed to play an important role for side-channel leakage modelling. Some of the most effective countermeasures to classical attacks relies on the introduction of process variability and desynchronizations in the computing platform. These protections have been proven to be very effective when applied for CMOS devices to counter most common attacks, like the well-known bayesian template attack. Deep learning models have instead showed to be very effective against a wide variety of hardware and software countermeasures. The objective of the thesis is to describe the role of machine learning and deep learning methods for modern side-channel analysis. We will present different techniques to exploit artificial neural networks for side-channel modelling, focusing on the capabilities of deep learning models to effectively counter protected implementations of the cryptosystem.| File | Dimensione | Formato | |
|---|---|---|---|
|
Deep Learning Techniques for Side-Channel Cipher Attacks.pdf
accessibile in internet per tutti
Dimensione
2.38 MB
Formato
Adobe PDF
|
2.38 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/186925