Quantum-Dot-Based Photonic Reservoir Computing

Abstract

AbstractReservoir computing is a novel computational framework based on the characteristic behavior of recurrent neural networks. In particular, a recurrent neural network for reservoir computing is defined as a reservoir, which is implemented as a fixed and nonlinear system. Recently, to overcome the limitation of data throughput between processors and storage devices in conventional computer systems during processing, known as the Von Neumann bottleneck, physical implementations of reservoirs have been actively investigated in various research fields. The author’s group has been currently studying a quantum dot reservoir, which consists of coupled structures of randomly dispersed quantum dots, as a physical reservoir. The quantum dot reservoir is driven by sequential signal inputs using radiation with laser pulses, and the characteristic dynamics of the excited energy in the network are exhibited with the corresponding spatiotemporal fluorescence outputs. We have presented the fundamental physics of a quantum dot reservoir. Subsequently, experimental methods have been introduced to prepare a practical quantum dot reservoir. Next, we have presented the experimental input/output properties of our quantum dot reservoir. Here, we experimentally focused on the relaxation of fluorescence outputs, which indicates the characteristics of optical energy dynamics in the reservoir, and qualitatively discussed the usability of quantum dot reservoirs based on their properties. Finally, we have presented experimental reservoir computing based on spatiotemporal fluorescence outputs from a quantum dot reservoir. We consider that the achievements of quantum dot reservoirs can be effectively utilized for advanced reservoir computing.