Switch language

DEMO3: Neuromorphic processor

 Photonic Nonlinear Transient Computing


DEMO3 manager: Laurent LARGER, FEMTO-ST

Co-manager: Jérôme PLAIN, LNIO


Design a neuromorphic chip capable to process extremely complex tasks, according to Reservoir  Computing (RC) principles, and making use of the complexity properties developed in multiple delay nonlinear dynamics. RC is a new brain inspired computational paradigm, that has emerged from neural network computing concepts, as well as from Echo State Network and Liquid State  Machines ones. Physical integration of these complex delay dynamics is envisioned through the utilization of advanced integrated optics technology such as photonic crystals structures and surface plasmon devices.

The research will have to deal with:

  • theoretical studies  of the novel concepts of delay dynamics based RC, trying to understand  deeper the fundamental principle of operation and the  underlying information processing, all this with respect to the physical parameters involved in the practical photonic implementation ;
  • operation optimization,  through the resolution, by our photonic neuromorphic processor, of benchmark tasks and other complex machine learning practical problems, such as classification and prediction; real world practical problems will be also addressed within the project, such as EEG signals analysis, Fuel Cell devices diagnosis and failure detection, real-time laser beam control in complex femtosecond laser nano-machining, as well as econometry problems and related  partly determinsistic time series predictions ;
  • propose and design solutions for the physical implementation of the processing concepts into a photonic integrated chip, with which high speed processing and computing capabilities can be achieved, together with a low power consumption, thanks to the use of  broadband fiber optics telecommunication concepts ;
  • suggest and implement the suited environment for the chip, so that neuromorphic processing can be autonomously operated, e.g. due to a smart interfacing of the complex photonic nonlinear dynamical chip  together with a standard digital processor sequentializing and distributing in real-time the various tasks required for the    neuromorphic computation (input information formatting and injecting and, learning and testing for output readout)    ;
  • and finally propose photonic integration solutions to implement on chip the neuromorphic architecture, e.g. with our know-how in LiNbO3 photonic crystal devices, and plasmonic devices (LiNbO3).



Main outcome:

A universal neuromorphic computer integrated on a photonic chip.

Main features:

  • Computational power emulated by the complexity of accurately controlled nonlinear delay dynamics; 
  •  Integration of the main functionalities on photonic crystal structures (waveguides, modulators, couplers, delay lines, etc.)  and with other similar nano-structuration of optically active materials,
  •  Programmable computational function, through the interfacing with standards digital processors aimed at sequencializing the tasks necessary to perform the computation, as well as the implementation of plasticity functionalities, and autonomous learning
  • Ultra-fast real time information and data processing capabilities through the use of chips with standard optical telecom  bandwidth.


photonoc crystal neuromorphicdiscrete demonstrator neuromorphicdescription neurmorphic processor

Concept of photonic crystal reservoir computing

Discrete device demonstrator

Description of the demonstrator        


Experience of the consortium:
  • Theory and analysis of the complexity developed by nonlinear delay dynamics
  • Experimental demonstration in photonics and optoelectronics of high dimensional nonlinear delay dynamics, theoretical and experimental investigation of their properties
  • Harnessing dynamical properties of nonlinear optoelectronic delay oscillators for various advanced applications, such as:
    (i)  High speed chaos cryptography involved at the physical layer of optical fiber communication systems; this  involves controlled and synchronized chaotic motion of laser light beams;
    (ii) generation  of high spectral purity microwave oscillations typ. for Radar applications;
    (iii) first demonstration of a photonic neuromorphic computer based on Reservoir Computing concepts
  • Quantum cryptography, related quantum key distribution systems for secure fiber optic links
  • Design of nano-photonic devices devices with LiNbO3 photonic crystal structures and plasmonic devices
  • R&D experience from 2004-2009 with SmartQuantum co. for Plug & Play QKD communication systems; collaboration with the start-up Aurea Technology created in  2010, on advanced photon counting systems @ 1,5µm.


Applications and socio-economic challenges:

The photonic neuromorphic chip is expected to provide a high speed high efficiency computational power for many practical complex problems that can not be addressed with conventional digital computers:

  • Classification and medical image processing,
  • Analysis of high dimensional phyisical problems,
  • Forecasting or anticipation of catastrophes (earthquake, tsunami...), financial bubbles
  • Support for industrial strategies, time series prediction, etc.



Keywords: neuromorphic computing, nonlinear transient computing, complex systems, nonlinear delay dynamics, smart components and systems, lithium niobate, metamaterials, photonic nano-components, plasmonics,
computing with Complex Dynamical Systems, Complexity in Nonlinear Delay Differential Equations, on-chip integration via nano photonics and photonic crystals...