ELPHOR ENERGY-EFFICIENT LEARNING WITH PHOTONIC RESERVOIR COMPUTING
ELPHOR aims to advance fundamental understanding and broaden the research team's expertise in photonic reservoir computing. The project will investigate innovative approaches for performing computation using photonic hardware that leverages nonlinear optical dynamics. While the research to be implemented is still at the prototype stage, it goes beyond curiosity-driven exploration. It aims to advance the capabilities that photonic reservoir computing (RC) with SLs has shown in initial implementations, and to propose and demonstrate energy-efficient devices and architectures that are expected to have significant long-term impact in the photonic computing field. The outcomes may contribute to the future development of low-power nonlinear photonic processors in integrated devices and sustainable high-speed information-processing solutions. The project will address the following questions from a physicist’s perspective, but with an extension to consider as well engineering and technological limitations and possibilities of the proposed topologies, emphasizing:
- How can edge-emitting SLs lasers with ultra-low threshold and high-slope optical power efficiency be engaged as nonlinear elements in energy-efficient photonic time delay RC (TDRC).
- What are the benefits of these devices in terms of their energetic footprint in the photonic TDRC architectures and what are the advantages they offer in terms of dynamics and computational performance, when compared to the available schemes with standard SLs.
- What improvements in terms of performance will deep neural network approaches bring by incorporating additional nonlinear layers (multiple SL nonlinearities), intermediate amplification stages with saturation nonlinearities, and addition of multiple external optical paths for linear/nonlinear memory enhancement.
- What are the strategic designs facilitating integrating such architectures and active components in photonic integrated circuits, using hybrid integration concepts.
Investigadors
Apostolos Argyris
Miguel C. Soriano
Ingo Fischer
Sandor Battaglini
Mohammadreza Mokhtari
Adhila Nazarudeen
Muhammad Riaz