Conference Agenda

We review recent works in signal shaping and advanced characterization techniques within the framework of nonlinear fiber optics. Here, we focus on characterization methods based on the dispersive Fourier transform to monitor incoherent spectral broadening processes with enhanced resolution and sensitivity. In this framework, we further discuss recent studies of modulation instability in a noise-driven regime. Paired with suitable optical monitoring techniques, we show that controlled coherent optical seeding can be leveraged using suitable machine learning approaches to tailor and optimize incoherent spectral broadening dynamics.

Photonics has proven to be a promising platform for the implementation of neuromorphic architectures. In this context, we implement an optoacoustic recurrent operator (OREO) based on stimulated Brillouin scattering in a highly nonlinear fiber. We demonstrate how OREO can establish a connection between optical pulses through acoustic waves. We show how the depth of our network benefits from cooling the fiber, as this extends the acoustic lifetime. We use OREO under this concept to perform a pattern classification task with 67% accuracy.

High harmonic generation (HHG) stands as one of the most complex processes in strong-field physics, as it enables the conversion of laser light from the infrared to the extreme-ultraviolet or even the soft x-rays, enabling the synthesis and control of pulses lasting as short as tens of attoseconds. Accurately simulating this nonlinear and non-perturbative phenomena requires the coupling the dynamics of laser-driven electronic wavepackets, described by the three-dimensional time-dependent Schrödinger equation (3D-TDSE), with macroscopic Maxwell’s equations. Such calculations are extremely demanding due to the duality of microscopic and macroscopic nature of the process, thereby requiring the use of approximations. We develop a HHG method assisted by artificial intelligence that facilitates the simulation of macroscopic HHG within the framework of 3D-TDSE. This approach is particularly suited to simulate HHG driven by structured laser pulses. In particular, we demonstrate a self-interference effect in HHG driven by Hermite-Gauss beams. The theoretical and experimental agreement allows us to validate the AI-based model, and to identify a unique signature of the quantum nature of the HHG process.

Disordered, self-assembled media, contain a large amount of information, which can be seen as a huge set of random and uncontrolled memory patterns. In the framework of optics, in which an opaque medium may modelled with the transmission matrix approach, each transmitted mode “contains” a memory element which is embodied by the correspondent transmission vector. Even if the stored amount of information in this system is huge, these random memories cannot be tailored easily. Here we present a new approach to write, read, and classify memory patterns: the photonic emergent learning. The writing paradigm is borrowed form a-physical- mathematical model for the biological memory, the emergent archetype, which we translated to photonics. In our approach the random patterns enclosed in the transmitted electromagnetic modes, are used as prototypes which are summed in constructive fashion in order write our target archetype-memory into our disordered optical memory (DOM). The DOM can work as a content addressable memory, retrieving at the lightning speed which memory in the library is the closest to an optically proposed query pattern. Moreover, the optical memories can be organized into super structures containing memories of the same thus efficiently delivering a classification task.

We describe our use of deep learning to optimize the multi-dimensional parameter space of systems-on-chip as an important step towards the scalable production of photonic solutions and their widespread integration into high-volume applications. The challenges of transitioning between prototype and volume production are highlighted, and the suitability of deep neural networks for navigating the multi-dimensional design space of today’s photonic circuits is discussed. We adopt multi-path neural network architectures to reduce the computational requirements of model training and to mitigate the risk of overfitting. We demonstrate the use of a multi-path neural network to optimize the construction parameters of photonic designs in a high-volume production environment. Lastly, we discuss the advantages of using machine learning not only as a highly capable tool for navigating the multi-dimensional design space of complex systems-on-chip but also as an effective strategy for compensating for fabrication process non-uniformities that are undetectable by standard process metrology instruments.