Reverse engineering circuits using behavioral pattern mining

TL;DR: A novel formalism and method to systematically derive the highlevel function of an unknown circuit component given its gate-level netlist, based on mining interesting behavioral patterns from the simulation traces of a gate- level netlist and representing them as a pattern graph.

Abstract: Systems are increasingly being constructed from off-the-shelf components acquired through a globally distributed, untrusted supply chain. The lack of trust in these components necessitates additional validation of the components before use. Additionally, hardware trojans are becoming a pressing concern. In this paper, we present a novel formalism and method to systematically derive the highlevel function of an unknown circuit component given its gate-level netlist. We define the highlevel description of a circuit as an interconnection of instantiations of abstract library components characterized using logical specifications. The proposed approach is based on mining interesting behavioral patterns from the simulation traces of a gate-level netlist, and representing them as a pattern graph. A similar pattern graph is also generated for library components. Our method first computes input-output signal correspondences via subgraph isomorphism on the pattern graphs. The general function of the unknown circuit is then determined by finding the closest match in the component library, by model checking the unknown circuit against each logical specification. We demonstrate the effectiveness of our approach on publicly-available circuits.