Applications of Photonic Neural Architectures to Optical Cryptography

Abstract

Rapid growth in the number of Internet connected devices along with coming advances in high-speed wireless networking will require adaptable cryptographic suites that support rapid authentication, key exchange, and encryption without occupying significant on-board resources. With traditional digital architectures starting to fall short of expectations set by Moore's Law, ``more-than-Moore" integrated silicon photonic platforms are well positioned to provide such ultrafast cryptographic functionality, in accordance with novel cross-protocol design philosophies. Previous work has characterized the usefulness of Photonic Neural Networks (PNNs) in replicating deterministic dynamical chaotic attractors by “solving” associated systems of equations, with the goal of using the resulting network as a Pseudo-random Function (PRF) for use in cryptography applications. We continue this work where we had previously left off, using techniques proposed in literature to prove that chaos that we can model with photonic hardware can be partitioned cleverly to create a pseudo-random number generator (PRNG) capable of producing cryptographically pseudo random bit streams. We quantitatively analyze the output of the proposed generator to evaluate whether it displays characteristics of statistically random binary sequence outputs, and consider the efficiency of the generators, in concept and in the simulated PNN setup, as a function of our quantization strategy. We find that partitioned PNN-generated chaotic flows succeed in generating high-quality random information, passing 11/15 tests at speeds of 4.33 Megabits/sec. Further work will be required to confirm that the system passes additional tests needed to verify full readiness for cryptographic applications, as well as to improve the overall speed of the generator and make it operate synchronously without adaptation.