RESEARCH UNIT ON NEUROMORPHIC COMPUTING AND PHOTONICS (RNCP)
Photonic neuromorphic computing is a rapidly evolving scientific field which attracts the interest worldwide due to its intrinsic scientific value and the niche applications it can enable. Bio-inspired computing can be performed in photonics with the use of pulsed lasers and complex waveguide structures achieving neuron-like behavior in unprecedented speeds and low consumption. The intrinsic parallelism of photonic circuits paves the wave for enhanced connectivity and scalability so as to solve complex problems at the speed of photons. RNCP members have a strong activity in the area as can be found in the publication record and in the participation in EU and national R&D projects.
Optical communication systems suffer from linear and nonlinear effects of the optical channel. Although modern ASICs can handle linear effects such as polarization mode dispersion and chromatic dispersion, the major limitation regarding the maximum capacity that can be achieved comes from the nonlinearities attributed to Kerr effect. RCNP investigates state of the art recursive neural networks and reservoir computing techniques so as to mitigate the nonlinear effects in long-haul WDM transmission systems. The same activity seeks for applying low complexity machine learning algorithms in short-area networks where consumption matters.
One of the key areas that MCP has invested is the envision-design and development of cryptographic devices based on electronic-photonics hardware for cyber-physical applications. In particular MCP has developed multiple designs of optical modules as non-replicable authentication tokens and secure pseudo-random generators able to be integrated in IoT ecosystems and solidifying their resilience against cyber-physical attacks. More importantly MCP is working towards combining its solid background on neuromoprhic engineering and crypto-systems, so as to spawn a new generation of electronic-photonic neuro-cryptographic devices able to offer a twofold advantage. On one hand, provide solid security features, such as data encryption and authentication, based on physical properties and secondly employ the same modules for anomaly detection and edge machine learning
Mimicking the brain (live)
A special live show brought to you from the Berlin Science Week 2022. A recent study in Communications Engineering presents a photonic setup to create a neural network. Tune in to find out more…
Featuring Charis Mesaritakis (University of the Aegean) and Miranda Vinay (Communications Engineering)
Hosted by Cristiano Matricardi (Nature Communications)
Ref: Sozos, K., Bogris, A., Bienstman, P. et al. High-speed photonic neuromorphic computing using recurrent optical spectrum slicing neural networks. Commun Eng 1, 24 (2022). https://doi.org/10.1038/s44172-022-00024-5
On your wavelength is a (physics) podcast where we discuss a paper published recently in a Nature Portfolio journal. We invite the author(s) and the editor of the paper for an interview, and explore the story behind the paper, including the conception of the research and aspects that you do not find written in the publication — essentially, an account of the ‘scientific journey’ of the paper until completion. We also go through the editorial process with the editor to give a different point of view and discuss post-publication content (if any) and future outlooks.
During this episode, we will talk about neuromorphic computing, which can be interpreted as biologically inspired computing facilitated by deep learning algorithms. The paper we will discuss brings the implementation of this new way of computing one step closer to real-world devices. The authors use light and a clever filtering system that could allow high-bandwidth communications and high-speed imaging.
Authors: Adonis Bogris, University of West Attica, Greece & Charis Mesaritakis, University of the Aegean, Greece
Editor: Miranda Vinay, Communications Engineering
Host: Cristiano Matricardi, Nature Communications