Electromagnetically Unclonable Function with Immunity to Machine Learning Attacks

Over the past decade, non-Hermitian physical systems and their abnormal singularities, known as exceptional points (EPs), have deepen our understanding of how to engineer spectral and spatial properties of waves, leading to remarkable advances in signal synthesis, routing, and modulation across the spectrum. In particular, spectral sensitivity near the EP-singularity in non-Hermitian physical systems has recently been extensively explored for making the next-generation photonic, acoustic, and electronic sensors with unprecedented sensitivity and resolvability [1]–[5]. However, it is still debated whether this new class of sensors can surpass traditional sensors in terms of reliability and signal-to-noise ratio, since mesoscopic fluctuations and noises could also be amplified at these singular points [6–8].