Optomechanical Reservoir Computing

Reservoir computing is a promising paradigm of physical intelligence where a nonlinear, dynamical system performs information processing within the system's physics by computing input-output mappings. However, the interplay of nonlinearities between different physics in the system has been less explored and presents an opportunity to yield signal processing capacities that are unique to the multiphysics coupling. We examine an optomechanical reservoir computer consisting of bilinear springs and elastomeric optical fibers, which bend in tandem with the springs to produce additional nonlinearity to the sensed reservoir state. After developing a frequency content analysis, we leveraged this analysis with a novelty search algorithm to show that the springs and fibers add different types of spectral content to the reservoir's response. Tracking these shifts in spectral content helps interpret the effect of each source of nonlinearity on the overall signal processing capacity of the system. Reservoir designs with high frequency content demonstrated good performance in simulation and experiment for computational tasks utilizing those frequencies. This work highlights the benefit of leveraging nonlinearities from different physics for reservoir computing.