Biblio

Logic locking is a well-explored defense mechanism against various types of hardware security attacks. Recent approaches to logic locking replace portions of a circuit with reconfigurable blocks such as look-up tables (LUTs) and switch boxes (SBs) to primarily achieve logic and routing obfuscation, respectively. However, these techniques may incur significant design overhead, and methods that can mitigate the implementation cost for a given security level are desirable. In this paper, we address this challenge by proposing an algorithm for deciding the location and inputs of the LUTs in LUT-based obfuscation to enhance security and reduce design overhead. We then introduce a locking method that combines LUTs with SBs to further robustify LUT-based obfuscation, largely independently of the specific LUT locations. We illustrate the effectiveness of the proposed approaches on a set of ISCAS benchmark circuits.

The proliferation of artificial intelligence based systems in all walks of life raises concerns about their safety and robustness, especially for cyber-physical systems including multiple machine learning components. In this paper, we introduce robustness contracts as a framework for compositional specification and reasoning about the robustness of cyber-physical systems based on neural network (NN) components. Robustness contracts can encompass and generalize a variety of notions of robustness which were previously proposed in the literature. They can seamlessly apply to NN-based perception as well as deep reinforcement learning (RL)-enabled control applications. We present a sound and complete algorithm that can efficiently verify the satisfaction of a class of robustness contracts on NNs by leveraging notions from Lagrangian duality to identify system configurations that violate the contracts. We illustrate the effectiveness of our approach on the verification of NN-based perception systems and deep RL-based control systems.

We introduce a scalable observer architecture to estimate the states of a discrete-time linear-time-invariant (LTI) system whose sensors can be manipulated by an attacker. Given the maximum number of attacked sensors, we build on previous results on necessary and sufficient conditions for state estimation, and propose a novel multi-modal Luenberger (MML) observer based on efficient Satisfiability Modulo Theory (SMT) solving. We present two techniques to reduce the complexity of the estimation problem. As a first strategy, instead of a bank of distinct observers, we use a family of filters sharing a single dynamical equation for the states, but different output equations, to generate estimates corresponding to different subsets of sensors. Such an architecture can reduce the memory usage of the observer from an exponential to a linear function of the number of sensors. We then develop an efficient SMT-based decision procedure that is able to reason about the estimates of the MML observer to detect at runtime which sets of sensors are attack-free, and use them to obtain a correct state estimate. We provide proofs of convergence for our algorithm and report simulation results to compare its runtime performance with alternative techniques. Our algorithm scales well for large systems (including up to 5000 sensors) for which many previously proposed algorithms are not implementable due to excessive memory and time requirements. Finally, we illustrate the effectiveness of our algorithm on the design of resilient power distribution systems.