Close

Presentation

SEALing Neural Network Models in Encrypted Deep Learning Accelerators
Time
Location
Event Type
Research Manuscript
Virtual Programs
Hosted in Virtual Platform
Keywords
Hardware Security: Primitives, Architecture, Design & Test
Topics
Security
DescriptionDeep learning (DL) accelerators suffer from a new security problem, i.e., being vulnerable to physical-access-based attacks. Memory encryption becomes important for DL accelerators to improve the security but causes significant performance degradation. To address this problem, our paper proposes SEAL to enhance the performance of encrypted DL accelerators. SEAL leverages a criticality-aware smart encryption scheme that identifies partial data having no impact on the security of NN models and allows them to bypass the encryption engine, thus reducing the amount of data to be encrypted without affecting security. Experimental results demonstrate SEAL improves the performance by 1.34-1.4 times.