Tamper resistant design of convolutional neural network hardware accelerator
Tamper resistant design of convolutional neural network hardware accelerator
The globalisation of supply chains and manufacturing processes can lead to loss of control over the manufacturing process and exposure to potentially malicious third parties, thus making the security of Convolutional Neural Network hardware accelerators compromised by emerging attacks (e.g., hardware Trojan(HT) insertion attacks and backdoor attacks from third-party dataset providers). In this paper, a new defence mechanism, called Shuffle and Substitution-Based Defence Mechanism(SSDM), is proposed to effectively defend against attacks launched by attackers from the third-party dataset providers and the Fabrication phase. The new countermeasure proposed in this paper can not only effectively suppress the activation of most existing HTs, but also greatly increase the difficulty for adversaries from third-party dataset providers to successfully execute backdoor attacks. The experimental results show that the new defensive countermeasures are effective in preventing HTs from being activated and significantly increasing the difficulty of backdoor attacks.
Accelerator, Backdoor attack, CNN, Countermeasure, Hardware Trojans, Shuffle, Substitution
Yu, Haosen
71ded974-69df-4edd-8c1a-ad192b83a27e
Sun, Peiyao
e517faec-75c2-43e4-a45e-90f47e80d195
Halak, Basel
8221f839-0dfd-4f81-9865-37def5f79f33
Shanthakumar, Karthik
6ff8be75-4a65-4ce6-a3ed-1e7961f6dabf
Kazmierski, Tomasz
a97d7958-40c3-413f-924d-84545216092a
24 January 2024
Yu, Haosen
71ded974-69df-4edd-8c1a-ad192b83a27e
Sun, Peiyao
e517faec-75c2-43e4-a45e-90f47e80d195
Halak, Basel
8221f839-0dfd-4f81-9865-37def5f79f33
Shanthakumar, Karthik
6ff8be75-4a65-4ce6-a3ed-1e7961f6dabf
Kazmierski, Tomasz
a97d7958-40c3-413f-924d-84545216092a
Yu, Haosen, Sun, Peiyao, Halak, Basel, Shanthakumar, Karthik and Kazmierski, Tomasz
(2024)
Tamper resistant design of convolutional neural network hardware accelerator.
In Proceedings of the 2023 Asian Hardware Oriented Security and Trust Symposium (AsianHOST).
IEEE.
5 pp
.
(doi:10.1109/AsianHOST59942.2023.10409319).
Record type:
Conference or Workshop Item
(Paper)
Abstract
The globalisation of supply chains and manufacturing processes can lead to loss of control over the manufacturing process and exposure to potentially malicious third parties, thus making the security of Convolutional Neural Network hardware accelerators compromised by emerging attacks (e.g., hardware Trojan(HT) insertion attacks and backdoor attacks from third-party dataset providers). In this paper, a new defence mechanism, called Shuffle and Substitution-Based Defence Mechanism(SSDM), is proposed to effectively defend against attacks launched by attackers from the third-party dataset providers and the Fabrication phase. The new countermeasure proposed in this paper can not only effectively suppress the activation of most existing HTs, but also greatly increase the difficulty for adversaries from third-party dataset providers to successfully execute backdoor attacks. The experimental results show that the new defensive countermeasures are effective in preventing HTs from being activated and significantly increasing the difficulty of backdoor attacks.
This record has no associated files available for download.
More information
Published date: 24 January 2024
Venue - Dates:
2023 Asian Hardware Oriented Security and Trust Symposium (AsianHOST), Tianjin, China, 2023-12-13 - 2023-12-15
Keywords:
Accelerator, Backdoor attack, CNN, Countermeasure, Hardware Trojans, Shuffle, Substitution
Identifiers
Local EPrints ID: 486706
URI: http://eprints.soton.ac.uk/id/eprint/486706
PURE UUID: 095db779-33df-4983-ad8b-bbf5f0dcf783
Catalogue record
Date deposited: 02 Feb 2024 17:32
Last modified: 18 Mar 2024 04:10
Export record
Altmetrics
Contributors
Author:
Haosen Yu
Author:
Peiyao Sun
Author:
Basel Halak
Author:
Karthik Shanthakumar
Author:
Tomasz Kazmierski
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics