On-chip optical memristors based on ferroelectric-doped graphene

Chip-integrated optical memristors, modulating light in a nonvolatile and semicontinuous manner, are attractive to revolutionize on-chip optical signal processing via the constructions of nonvolatile reconfigurable photonic circuits, in-memory computing, brain-inspired architectures, etc. Mechanisms, including phase-change, filamentation, and ferroelectricity, have been attempted to implement on-chip optical memristors, though their intricate tradeoffs between fabrication compatibility, modulation depth, power consumption, retention time, and cyclability make it desired to pursue new architectures. Here, we demonstrate graphene-based on-chip optical amplitude and phase memristors by electrostatically doping the graphene integrated on a silicon nitride waveguide with a ferroelectric film. Benefiting from graphene’s significant dependence of complex refractive index on its carrier density and the ferroelectric remnant doping, semicontinuous nonvolatile modulation with a maximum depth of ∼32.5 dB is realized with a low programming energy of ∼1.86 pJ/µm², exhibiting good cyclability (fluctuation ratio <0.9%) and long retention time (over 10 years). By integrating the graphene-based optical memristor with cascaded microring resonators, in-memory computings with multiple wavelength channels are demonstrated by analogue matrix-vector multiplication and digital logic gate operations. Combining these merits with CMOS-compatible on-chip graphene integration, the demonstrated graphene-based optical memristor has proven to be a competitive candidate for high-bandwidth neuromorphic computing, convolutional processing, and artificial intelligence on photonic integrated circuits.