A critical view on the real-world security of logic locking

With continuously shrinking feature sizes of integrated circuits, the vast majority of semiconductor companies have become fabless, outsourcing to foundries across the globe. This exposes the design industry to a number of threats, including piracy via IP-theft or unauthorized overproduction and subsequent reselling on the black market. One alleged solution for this problem is logic locking, where the genuine functionality of a chip is “locked” using a key only known to the designer. Solely with a correct key, the design works as intended. Since unlocking is handled by the designer only after production, an adversary in the supply chain should not be able to unlock overproduced chips. In this work, we focus on logic locking against the threat of overproduction. First, we survey existing locking schemes and characterize them by their handling of keys, before extracting similarities and differences in the employed attacker models. We then compare said models to the real-world capabilities of the primary adversary in overproduction—a malicious foundry. This comparison allows us to identify pitfalls in existing models and derive a more realistic attacker model. Then, we discuss how existing schemes hold up against the new attacker model. Our discussion highlights that several attacks beyond the usually employed SAT-based approaches are viable. Crucially, these attacks stem from the underlying structure of current logic locking approaches, which has never changed since its introduction in 2008. We conclude that logic locking, while being a promising approach, needs a fundamental rethinking to achieve real-world protection against overproduction.