Photoresponsive MoS2-Based Artificial Synaptic Transistor for Wavelength-Dependent Physical Reservoir Computing

초록

Generating rich reservoir states is essential for physical reservoir computing (PRC), yet photonic-synapse reservoirs typically operate at a single illumination wavelength and hence with a single characteristic relaxation timescale, which limits the temporal feature diversity of the reservoir. Here, we report a back-gated MoS2 photonic synaptic transistor in which the illumination wavelength itself enriches the reservoir dynamics. Under red (660 nm), green (520 nm), and blue (440 nm) illumination, the device exhibits persistent photoconductivity with systematically distinct decay dynamics. Temperature-dependent transient analysis shows that the effective activation energy of the relaxation changes with the illumination wavelength, reflecting different relative contributions from trap states at different energies. The three wavelengths therefore provide three different fading-memory timescales in a single device. These spectral channels are combined into one reservoir without any change in the device structure. In a second-order nonlinear time-series prediction task, the combined reservoir achieves a lower normalized mean squared error than any single-wavelength reservoir. By converting the wavelength dependence of the photoresponse into multi-timescale reservoir dynamics, this work provides a compact route toward energy-efficient photonic PRC hardware.

키워드