Abstract Random telegraph noise (RTN) is usually regarded as a hallmark of nanoscale conduction channels, arising from individual trapping events in semiconductors and oxide dielectrics. Here we show that optical excitation can induce “giant” RTN in macroscopically large-area devices based on CuScP 2 S 6 /MoS 2 heterostructures, revealing a mesoscopic regime in which a sparse set of photo-activated defects in an insulating thiophosphate controls the conductance of an extended channel. Under optical illumination, the device conductance exhibits stochastic two-level fluctuations whose amplitudes are nearly independent of illumination strength, whereas the characteristic trapping-detrapping time constants are strongly governed by the incident light intensity. This behavior implies that photons are absorbed in effectively small packets that modulate a sparse ensemble of active traps, giving rise to bimodal noise statistics and illumination-tunable switching kinetics. We further exploit this controllable stochasticity in a proof-of-concept optical encoder that converts image pixels into RTN-driven spike trains, enhancing the robustness of a spiking neural network (SNN) to noise-corrupted MNIST inputs. Our results identify CuScP 2 S 6 as a model platform in which light-tunable RTN connects microscopic defect dynamics to macroscopic conductance fluctuations, opening opportunities to engineer noise itself as a functional degree of freedom in photonic and neuromorphic hardware.
Ghosh et al. (Thu,) studied this question.