neuro-inspired information processing scheme
in an optical network of nano-structures


    General information about NeuroQNet


    The main objective (OI) of NeuroQNet is to implement Reservoir Computing (RC), a neuro-inspired information processing scheme, in an optical network of nano-structures. Quantum Dot (QD) micropillar arrays (QDMPA) (OII) will be connected in an all-optical networks including 100s of such emitters (OIII). The project is based upon an interdisciplinary approach: the Neuromorphic computation is based on Reservoir Computing, QDMPAs provide the neural hardware and diffractive coupling will form the network to implement the neuro-inspired scheme (see Fig. 1).

    The project NeuroQNet was funded under the Integration of Molecular Components in Functional Macroscopic Systems initiative of the Volkswagen Stiftung (Volkswagen Foundation). The project duration is three years, being funded from 01.01.2016 until the 31.12.2018.




    The Volkswagen Foundation


    The Volkswagen Foundation is an independent foundation incorporated under private law with registered office in Hanover. Its overall funding volume of around 150 million euros per year makes it Germany‘s largest private research funding foundation, and indeed one of the country’s largest foundations altogether. The Foundation provides funds only to academic institutions. Since it was founded more than 50 years ago, the Volkswagen Foundation has allocated more than 4.2 billion euros in support of over 30,000 projects. It is thereby one of the largest common benefit foundations under private law in all of Germany.



    The consortium


    The chair of “Optoelectronics and Quantum Devices” at the TU Berlin focuses on the development and experimental study of nanophotonics structures and devices. Led by Prof. S. Reitzenstein the research covers a broad range of topics including the epitaxial growth of semiconductor heterostructures, the deterministic processing of quantum devices, and the quantum optical study of key building blocks for quantum information technology. Furthermore and relevant to the present project, the group has also profound expertise in the external quantum control of nanophotonics systems.

    The FEMTO-ST Institute is a joint research institution, with a broad field of expertise, including nonlinear optics and complex photonics, nano science and robotics. It is at the forefront of Neuromorphic Computation implemented in electro-optical systems.

    The Institute for Cross-Disciplinary Physics and Complex Systems (IFISC) is a joint research Institute of the University of the Balearic Islands (UIB) and the Spanish National Research Council (CSIC).  IFISC’s research is focused on the study of generic phenomena in nonlinear and complex systems. Specific research projects cover a range of applications in different fields with recent emphasis on photonic systems and its applications to information processing. This partner, represented by Prof. Ingo Fischer, is head of the Nonlinear Photonics Lab and the group on Experimental Physics of Complex Systems at IFISC. This lab and group have been built up since 2009 and have been pioneering the photonic implementation of unconventional computing concepts, as well as the realization of reconfigurable coupled laser arrays. Therefore, this partner's competences perfectly matches the profile required to address the tasks in the project related to implementing reconfigurable Quantum Dot micropillar arrays, and characterizing them for their use in photonic information processing systems.


    The Collaboration


    The highly interdisciplinary nature of NEuroQNet makes a collaborative project is therefore essential. Each group contributes according to their specific expertise:

    • Nanophotonics devices (TUB)
    • Optical characterization (IFISC)
    • RC implementation (FEMTO-ST)


    Due to the wide range of expertise required in the implementation of the project, the cooperation between the TUB and FEMTO-ST and the IFISC is essential. From the epitaxial growth of the semiconductor substrate to implementing the final all-optical RC, the consortium encompasses the entire research expertise and research chain.

Project Objectives


    Large scale quantum dot micro-pillar arrays (QDMPA)


    The nanophotonic hardware of NeuroQNet is based on dense arrays of high-quality quantum dot microlasers. Arranged in regular arrays with a pitch down to about 10 µm these structures are optimized to allow for an efficient refractive coupling of a large number (several 100s) microlasers within the envisioned all-optical neural network. The mircolasers are electrically or optically driven and will be optimized in all their parameters to precisely fit the requirements of the external optics. Most importantly is the spectral homogeneity of the microlasers within a given array, which is the key requirement for the refractive coupling of multiple of these lasers. We will apply specific growth and processing schemes, and if necessary also post-processing methods for the required spectral homogeneity of our photonic hardware and in order to ensure optimum performance in the neural-network.



    All-optical neural networks in QDMPA


    Generally, Neural Networks consist of two essential ingredients: nonlinear nodes acting as neurons, and a network providing intra-neuron connectivity. The QDMPs provided by the TU-Berlin are photonic devices which react highly nonlinearly to optical stimulus (phase-amplitude coupling, saturation effects, etc.). We will employ the QDMPs as photonic neurons with ultra-high energy efficiency and bandwidth.

    The missing but crucial step on the way to an all-optical Neural Network are then the inter-neuronal connections, corresponding to axons and dendrites in neuroscience. We are employing holographic techniques to create bi-directional coupling between individual QDMP neurons, as depicted in Fig. 2. As such, all flow of information within the Neural Network, from information injection, over neuronal coupling until the computational results are implemented optically.


  • Ultra fast, photonic and parallel neuromorphic computing


    Once the optical network of QDMPs is established and characterized, we will explore its computational ability following the concept of reservoir computing. To that end, we will determine how many memory steps are stored in the responses of the recurrent network and how diverse the responses are. Three conditions need to be fulfilled in the system to allow for efficient reservoir computing: (i) similar inputs need to result in similar outputs (consistency property), (ii) sufficiently diverse inputs need to give rise to different responses and (iii) for context-sensitive processing the responses need to contain fading memory of previous inputs.

    Once these properties are fulfilled, we will employ the optical network of QDMPs for computation. The advantage of the chosen approach relies in the fact that a network of many QD emitters can be realized, addressed and read out in parallel, with the individual emitter exhibiting a characteristic response timescale of sub-nanoseconds. Therefore, ultra-fast, parallel and efficient neuromorphic computing come into reach.

Project technology


    Reservoir Computing


    Reservoir computing (RC) is a bioinspired, machine-learning paradigm that exhibits state-of-the art performance for processing empirical data. Tasks, which are deemed computationally hard, such as chaotic time series prediction, or speech recognition, among others, can be successfully performed. The main inspiration underlying RC is the insight that the brain processes information generating patterns of transient neuronal activity excited by input sensory signals. Therefore, RC is mimicking neuronal networks.

    RC implementations are generally composed of three distinct parts: an input layer, the reservoir and an output layer. The input layer feeds the input signals to the reservoir via random connections. The reservoir usually consists of a large number of randomly interconnected nonlinear nodes, constituting a recurrent network, i.e., a network that has internal feedback loops. Under the influence of input signals, the network exhibits transient responses. These transient responses are read out at the output layer, usually via a linear weighted sum of the individual node states. The objective of RC is to implement a specific nonlinear transformation of the input signal or to classify the inputs.

    In recent years a number of different photonic implementations of reservoir computing have proven to be ultrafast, showing excellent performance and being hardware and energy-efficient.



    Quantum dot micro-pillar arrays


    Quantum dot micropillar arrays will be realized by state-of-the-art nanotechnology tools of the Center of Nanophotonics at the TU Berlin. This includes the epitaxial growth of high-quality semiconductor heterostructures with self-assembled quantum dots as active medium. Based on these heterostrutures dense arrays of quantum dot micropillars will be patterned by means of high-resolution electron beam lithography and plasma etching. The required spectral homogeneity within the micropillar-arrays will be ensured by the growth of distributed Bragg reflector (DBR) microcavities with highly homogenous layer thicknesses, diameter tuning of the micropillars and electro-optical resonance tuning.


    more info

    Diffractive coupling


    Diffractive coupling, schematically illustrated in Fig. 4, is a holographic technique recently developed by members of the consortium. The goal is to couple individual elements within an array of regularly spaced QDMPs. A Diffractive Optical element (DOE) is used to create multiple copies of the imaged object in the image plane. A mirror reflects the image of the object and its copies back onto the object plane. Back in the object plane, diffractive copies overlap with elements of the original array for correct diffraction angles.


    more info


Oral Conferences and Seminar presentations

  • D. Brunner, "All-Optical Neuromorphic Computing in Optical Networks of Semiconductor Lasers," invited seminar presentation, Institut für Kognitionswissenschaft, 09.11.2016, Osnabrück, Germany.
  • D. Brunner, S. Reitzenstein, I. Fischer, "All-Optical Neuromorphic Computing in Optical Networks of Semiconductor Lasers," oral presentation, IEEE ICRC 2016, 17-19.10.2016, San Diego, USA.
  • D. Brunner, “Towards the Creation of Cognitive Devices and the Role of Photonics,” oral presentation, Max Planck Institute of Microstructure Physics, 21.06.2016, Halle, Germany.
  • I. Fischer, J. Bueno, D. Brunner, M. Soriano, C. Mirasso, "Photonic Reservoir Computing for Ultra-Fast Information Processing Using Semiconductor Lasers," invited oral presentation, CLEO-FOCUS Meeting on Optical Information Processing, ECOC 2016, 18-22.09.2016, Düsseldorf, Germany.
  • J. Bueno, N. Oliver, D. Brunner, M. Soriano, I. Fischer, "Photonic information processing using semiconductor lasers," invited oral presentation, International Tandem Workshop: Pattern Dynamics in Nonlinear Optical Cavities (PDNOC 2016), 15-19.08.2016, Dresden, Germany.
  • I. Fischer, S. Reitzenstein, D. Brunner, “Neuromorphic Computing using networks of quantum dot emitters,” invited oral presentation, Beyond von Neumann Workshop, 18-21.05.2016, Berlin, Germany.