The glass maze: hiding keys in spin glasses

Abstract

Key-binding mechanisms allow to use one's biometric features as a universal digital key without having them ever stored anywhere. Security, ease of use and privacy concerns are addressed in one stroke. The best known proposal for such a mechanism, the Fingerprint Vault, treats biometric data as projections of a polynomial encoding the key. Its security is based on the difficulty of polynomial reconstruction. Here I propose a new key-binding mechanism based on associative pattern recall and making use of a totally different security principle, that of the difficulty of energy optimization of spin glasses. The idea is to exploit the mixed ferromagnetic and spin glass phase of the Hopfield neural network to encode the key as a local minimum configuration of the energy functional, ”lost” amidst the exponentially growing number of valleys and minima representing the spin glass. The correct fingerprint will be able to retrieve the key by dynamical evolution to the nearest attractor. Other fingerprints will be driven far away from the key. Known vulnerabilities of the Fingerprint Vault are eliminated by this new security principle.