Data-Oblivious ML Accelerators Using Hardware Security Extensions

Hossam Elatali; John Z. Jekel; Lachlan J. Gunn; N. Asokan

doi:10.1109/HOST55342.2024.10545398

Data-Oblivious ML Accelerators Using Hardware Security Extensions

Hossam Elatali, John Z. Jekel, Lachlan J. Gunn, N. Asokan

University of Waterloo

Tutkimustuotos: Artikkeli kirjassa/konferenssijulkaisussa › Conference article in proceedings › Scientific › vertaisarvioitu

6 Lataukset (Pure)

Abstrakti

We present Dolma, which applies dynamic information flow tracking (DIFT) to the Gemmini matrix multiplication accelerator. With the BliMe CPU extensions, it efficiently guarantees client data confidentiality, even in the presence of malicious/vulnerable software and side channel attacks on servers.

Alkuperäiskieli	Englanti
Otsikko	Proceedings of the 2024 IEEE International Symposium on Hardware Oriented Security and Trust, HOST 2024
Kustantaja	IEEE
Sivut	373-377
Sivumäärä	5
ISBN (elektroninen)	979-8-3503-7394-3
DOI - pysyväislinkit	https://doi.org/10.1109/HOST55342.2024.10545398
Tila	Julkaistu - 2024
OKM-julkaisutyyppi	A4 Artikkeli konferenssijulkaisussa
Tapahtuma	IEEE International Symposium on Hardware Oriented Security and Trust - McLean, Yhdysvallat Kesto: 6 toukok. 2024 → 9 toukok. 2024

Julkaisusarja

Nimi	Proceedings of the IEEE International Symposium on Hardware-Oriented Security and Trust
ISSN (elektroninen)	2765-8406

Conference

Conference	IEEE International Symposium on Hardware Oriented Security and Trust
Lyhennettä	HOST
Maa/Alue	Yhdysvallat
Kaupunki	McLean
Ajanjakso	06/05/2024 → 09/05/2024

Pääsy asiakirjaan

10.1109/HOST55342.2024.10545398

SCI_Elatali_etal_HOST_2024Accepted author manuscript, 572 KB

https://arxiv.org/abs/2401.16583

Muut tiedostot ja linkit

Linkki julkaisuun Scopuksessa

Gunn Lachlan: Rigorous security guarantees for run-time integrity
Gunn, L. (Vastuullinen tutkija)
01/09/2021 → 31/08/2024
Projekti: RCF Postdoctoral Researcher

Siteeraa tätä

Elatali, H., Jekel, J. Z., Gunn, L. J., & Asokan, N. (2024). Data-Oblivious ML Accelerators Using Hardware Security Extensions. teoksessa Proceedings of the 2024 IEEE International Symposium on Hardware Oriented Security and Trust, HOST 2024 (Sivut 373-377). (Proceedings of the IEEE International Symposium on Hardware-Oriented Security and Trust). IEEE. https://doi.org/10.1109/HOST55342.2024.10545398

@inproceedings{9255a661df474b988d0196cc81b3abd1,

title = "Data-Oblivious ML Accelerators Using Hardware Security Extensions",

abstract = "We present Dolma, which applies dynamic information flow tracking (DIFT) to the Gemmini matrix multiplication accelerator. With the BliMe CPU extensions, it efficiently guarantees client data confidentiality, even in the presence of malicious/vulnerable software and side channel attacks on servers.",

author = "Hossam Elatali and Jekel, {John Z.} and Gunn, {Lachlan J.} and N. Asokan",

note = "Publisher Copyright: {\textcopyright} 2024 IEEE.; IEEE International Symposium on Hardware Oriented Security and Trust, HOST ; Conference date: 06-05-2024 Through 09-05-2024",

year = "2024",

doi = "10.1109/HOST55342.2024.10545398",

language = "English",

series = "Proceedings of the IEEE International Symposium on Hardware-Oriented Security and Trust",

publisher = "IEEE",

pages = "373--377",

booktitle = "Proceedings of the 2024 IEEE International Symposium on Hardware Oriented Security and Trust, HOST 2024",

address = "United States",

}

Elatali, H, Jekel, JZ, Gunn, LJ & Asokan, N 2024, Data-Oblivious ML Accelerators Using Hardware Security Extensions. julkaisussa Proceedings of the 2024 IEEE International Symposium on Hardware Oriented Security and Trust, HOST 2024. Proceedings of the IEEE International Symposium on Hardware-Oriented Security and Trust, IEEE, Sivut 373-377, IEEE International Symposium on Hardware Oriented Security and Trust, McLean, Yhdysvallat, 06/05/2024. https://doi.org/10.1109/HOST55342.2024.10545398

Data-Oblivious ML Accelerators Using Hardware Security Extensions. / Elatali, Hossam; Jekel, John Z.; Gunn, Lachlan J. et al.
Proceedings of the 2024 IEEE International Symposium on Hardware Oriented Security and Trust, HOST 2024. IEEE, 2024. s. 373-377 (Proceedings of the IEEE International Symposium on Hardware-Oriented Security and Trust).

Tutkimustuotos: Artikkeli kirjassa/konferenssijulkaisussa › Conference article in proceedings › Scientific › vertaisarvioitu

TY - GEN

T1 - Data-Oblivious ML Accelerators Using Hardware Security Extensions

AU - Elatali, Hossam

AU - Jekel, John Z.

AU - Gunn, Lachlan J.

AU - Asokan, N.

PY - 2024

Y1 - 2024

N2 - We present Dolma, which applies dynamic information flow tracking (DIFT) to the Gemmini matrix multiplication accelerator. With the BliMe CPU extensions, it efficiently guarantees client data confidentiality, even in the presence of malicious/vulnerable software and side channel attacks on servers.

AB - We present Dolma, which applies dynamic information flow tracking (DIFT) to the Gemmini matrix multiplication accelerator. With the BliMe CPU extensions, it efficiently guarantees client data confidentiality, even in the presence of malicious/vulnerable software and side channel attacks on servers.

UR - http://www.scopus.com/inward/record.url?scp=85196103995&partnerID=8YFLogxK

U2 - 10.1109/HOST55342.2024.10545398

DO - 10.1109/HOST55342.2024.10545398

M3 - Conference article in proceedings

AN - SCOPUS:85196103995

T3 - Proceedings of the IEEE International Symposium on Hardware-Oriented Security and Trust

SP - 373

EP - 377

BT - Proceedings of the 2024 IEEE International Symposium on Hardware Oriented Security and Trust, HOST 2024

PB - IEEE

T2 - IEEE International Symposium on Hardware Oriented Security and Trust

Y2 - 6 May 2024 through 9 May 2024

ER -

Elatali H, Jekel JZ, Gunn LJ, Asokan N. Data-Oblivious ML Accelerators Using Hardware Security Extensions. julkaisussa Proceedings of the 2024 IEEE International Symposium on Hardware Oriented Security and Trust, HOST 2024. IEEE. 2024. s. 373-377. (Proceedings of the IEEE International Symposium on Hardware-Oriented Security and Trust). doi: 10.1109/HOST55342.2024.10545398

Data-Oblivious ML Accelerators Using Hardware Security Extensions

Abstrakti

Julkaisusarja

Conference

Pääsy asiakirjaan

Muut tiedostot ja linkit

Sormenjälki

Projektit

Gunn Lachlan: Rigorous security guarantees for run-time integrity

Siteeraa tätä