Support Vector Machines for Improved IP Detection with Soft Physical Hash Functions

Side-channel analysis is a powerful tool to extract secret information from microelectronic devices. Its most frequently considered application is destructive, i.e. key recovery attacks against cryptographic implementations. More recently, it has also been considered constructively, in the context of intellectual property protection/detection, e.g. through the use of side-channel based watermarks or soft physical hash functions. The latter solution is interesting from the application pointof-view, because it does not require any modification of the designs to protect (hence it implies no performance losses). Previous works in this direction have exploited simple (correlation-based) statistical tools in different (more or less challenging) scenarios. In this paper, we investigate the use of support vector machines for this purpose. We first argue that their single-class extension is naturally suited to the problem of intellectual property detection. We then show experimentally that they allow dealing with more complex scenarios than previously published, hence extending the relevance and applicability of soft physical hash functions.