Diffractive optoelectronic-reservoir computing with intrinsic spatiotemporal dynamics.

Optical reservoir computing (RC) is a hardware-efficient paradigm that leverages intrinsic optical properties to process information. This paper proposes a diffractive optoelectronic-reservoir computing (DOE-RC) architecture. The architecture utilizes an optoelectronic feedback loop to combine the spatial computing capabilities of diffractive optics with the temporal integration response of CMOS, providing rich dynamical features while maintaining macroscopic controllability. We numerically studied the dynamics of the DOE-RC with varying hyperparameters, including feedback gain, feedback delay, and integration time constant. The results show that the diffractive phase modulation redistributes spatial energy, suppressing local saturation and oscillation to facilitate spatiotemporal feature filtering and image classification. This study provides a guideline for designing high-throughput, energy-efficient optical RC for artificial intelligence.