Memlumor: a luminescent memory device for photonic neuromorphic computing

Abstract

Neuromorphic computing promises to transform the current paradigm of traditional computing towards Non-Von Neumann dynamic energy-efficient problem solving. Thus, dynamic memory devices capable of simultaneously performing nonlinear operations (volatile) similar to neurons and also storing information (non-volatile) alike brain synapses are in the great demand. To satisfy these demands, a neuromorphic platform has to possess intrinsic complexity reflected in the built-in diversity of its physical operation mechanisms. Herein, we propose and demonstrate the novel concept of a memlumor - an all-optical device combining memory and luminophore, and being mathematically a full equivalence of the electrically-driven memristor. By utilizing metal halide perovskites as a memlumor material platform, we demonstrate the synergetic coexistence of both volatile and non-volatile memory effects within a broad timescale from ns to days. We elucidate the origin of such complex response to be related to the phenomena of photodoping and photochemistry activated by a tunable light input and explore several possible realizations of memlumor computing. Leveraging on the existence of a history-dependent photoluminescent quantum yield in various material platforms, the memlumor device concept will trigger multiple new research directions in both material science and optoelectronics. We anticipate that the memlumor, as a new optical dynamic computing element, will add a new dimension to existing optical technologies enabling their transition into application in photonic neuromorphic computing.