Abstract

A compact all-optical network of silicon microring resonators, exploiting nonlinearities from free carriers and thermo-optic effects, produces diverse representations of the same input and behaves as a nonlinear recurrent system suited for reservoir-computing approaches. While task-based evaluations, including our previous works on time-series classification, show the computational potential of the platform, they do not reveal the dynamical properties that determine performance. Here we experimentally study the microring resonator network using metrics adapted from digital reservoir computing, focusing on memory capacity. We show that the richness of the memory regimes increases when moving from a single ring to a matrix of coupled rings. By combining output representations obtained at multiple wavelengths, we also exploit the device’s inherent parallelism and observe how the coexistence of distinct dynamical responses improves the overall performance of the network. This approach provides rapid, task-independent characterization and clearer insight into the reservoir’s behavior.

Related Research Topics

• Photonic Reservoir Computing

Related Projects

• Horizon Europe: Neuropuls (NEUROmorphic energy‑efficient secure accelerators based on Phase change materials aUgmented siLicon photonicS)