High-performance optical reservoir computing based on spatially extended systems

The present day combination of the von Neumann architecture, digital encoding, and its microelectronic integration might never be challenged in terms of heavy calculation-based procedures. Nevertheless, many tasks such as facial and speech recognition, which the human brain can handle perfectly in an instant, are computationally very demanding and they require large, costly and energy consuming computer resources. Therefore, neuron-inspired or neuromorphic computing techniques have attracted renewed interest. At the same time, photonic manufacturing is booming, driven by the demand for high-capacity optical communication links. The marriage of photonics and neuromorphic computation has reinvigorated the field of photonic based computing. We target efficient (high speed and low power) analogue photonic computing systems based on the concept of reservoir computing (RC). This computing paradigm offers a framework to exploit the transient dynamics of a nonlinear dynamical system for performing useful computations. In this project, we will implement an RC architecture taking as reservoir a spatially extended non-linear photonic system with diffusive coupling between the different regions or neurons. Such an approach makes it possible to use the massive parallelism and high bandwidth offered by optics, while the system architecture remains simple and is easily scalable. This research will thus pave the way to implement optic RC systems for a variety of complex computational tasks.