Exploring Integrated Device Implementation for FRET-Based Optical Reservoir Computing

Abstract

AbstractThis chapter explores a reservoir computing (RC) device based on fluorescence resonance energy transfer (FRET) between quantum dots (QDs). We propose a compact structure in which optical input/output and quantum dots are adjacently placed without lenses or delay lines. The proposed structure exploits the QD-based optical reservoir as an intermediate layer and adopts memory to enable recurrent inputs. We evaluate the feasibility of the proposed structure by applying tasks that require nonlinearity. Simulation-based experimental results show that the proposed device can perform logistic map, time-series XOR, and NARMA10. A proof-of-concept implementation with a commercial image sensor demonstrates that the proposed structure can solve XOR and MNIST tasks. Also, we discuss the energy advantage over conventional digital circuit implementations.