ML-Based Side Channel Vulnerability Detection

ABSTRACT: Side-channel vulnerabilities remain a persistent threat to secure systems, often bypassing traditional verification and mitigation methods. In this talk, we introduce two machine learning-based frameworks—Channelizer and DOME—designed to validate the effectiveness of hardware and software side-channel mitigations. Channelizer uses supervised ML to detect leakage in decision programs by predicting secrets from performance counters, offering a quantitative measure of system security. DOME extends this validation to arbitrary programs using unsupervised clustering to discover distinct execution patterns that correlate with secret inputs. Together, these tools provide a powerful and practical methodology for identifying, explaining, and fixing side-channel flaws—even in hardened systems.

BIO: Todd Austin is the S. Jack Hu Collegiate Professor of Electrical Engineering and Computer Science at the University of Michigan and Director of the Computer Engineering Lab. His research spans computer architecture, secure system design, verification, and performance analysis. Previously, he directed C-FAR, a multi-university SRC/DARPA-funded computer engineering research center. Before academia, he was a Senior Computer Architect at Intel. He created the SimpleScalar Tool Set and co-authored Structured Computer Architecture, 6th Ed. He also co-founded Agita Labs and InTempo Design. He is an IEEE Fellow, and he has received the ACM Maurice Wilkes and IEEE Ramakrishna Rau awards. He earned his PhD from the University of Wisconsin.