Machine Learning Attacks-Resistant Security by Mixed-Assembled Layers-Inserted Graphene Physically Unclonable Function.

Abstract

Mixed layers of octadecyltrichlorosilane (ODTS) and 1H,1H,2H,2H-perfluorooctyltriethoxysilane (FOTS) on an active layer of graphene are used to induce a disordered doping state and form a robust defense system against machine-learning attacks (ML attacks). The resulting security key is formed from a 12×12 array of currents produced at a low voltage of 100mV. The uniformity and inter-Hamming distance (HD) of the security key are 50.0 ± 12.3% and 45.5 ± 16.7%, respectively, indicating higher security performance than other graphene-based security keys. Raman spectroscopy confirmed the uniqueness of the 10,000 points, with the degree of shift of the G peak distinguishing the number of carriers. The resulting defense system has a 10.33% ML attack accuracy, while a FOTS-inserted graphene device is easily predictable with a 44.81%ML attack accuracy.