Feasibility and Infeasibility of Secure Computation with Malicious PUFs

A recent line of work has explored the use of

physically uncloneable functions (PUFs)

for secure computation, with the goals of (1) achieving universal composability without (additional) setup, and/or (2) obtaining unconditional security (i.e., avoiding complexity-theoretic assumptions). Initial work assumed that all PUFs, even those created by an attacker, are honestly generated. Subsequently, researchers have investigated models in which an adversary can create

malicious

PUFs with arbitrary behavior. Researchers have considered both malicious PUFs that might be stateful, as well as malicious PUFs that can have arbitrary behavior but are guaranteed to be stateless.

We settle the main open questions regarding secure computation in the malicious-PUF model:

We prove that unconditionally secure oblivious transfer is impossible, even in the stand-alone setting, if the adversary can construct (malicious)

stateful

PUFs.

We show that universally composable two-party computation is possible if the attacker is limited to creating (malicious)

stateless

PUFs. Our protocols are simple and efficient, and do not require any cryptographic assumptions.