Reservoir Computing is a novel computing paradigm that uses a nonlinear recurrent dynamical system to carry out information processing. Recent electronic and optoelectronic Reservoir Computers based on an architecture with a single nonlinear node and a delay loop have shown performance on standardized tasks comparable to state-of-the-art digital implementations. Here we report an all-optical implementation of a Reservoir Computer, made of off-the-shelf components for optical telecommunications. It uses the saturation of a semiconductor optical amplifier as nonlinearity. The present work shows that, within the Reservoir Computing paradigm, all-optical computing with state-of-the-art performance is possible.

We present a numerical study of a passive integrated photonics reservoir computing platform based on multimodal Y-junctions. We propose a novel design of this junction where the level of adiabaticity is carefully tailored to capture the radiation loss in higher-order modes, while at the same time providing additional mode mixing that increases the richness of the reservoir dynamics. With this design, we report an overall average combination efficiency of 61% compared to the standard 50% for the single-mode case. We demonstrate that with this design, much more power is able to reach the distant nodes of the reservoir, leading to increased scaling prospects. We use the example of a header recognition task to confirm that such a reservoir can be used for bit-level processing tasks. The design itself is CMOS-compatible and can be fabricated through the known standard fabrication procedures.

We propose using a neural network approach in conjunction with digital holographic microscopy in order to rapidly determine relevant parameters such as the core and shell diameter of coated, non-absorbing spheres. We do so without requiring a time-consuming reconstruction of the cell image. In contrast to previous approaches, we are able to obtain a continuous value for parameters such as size, as opposed to binning into a discrete number of categories. Also, we are able to separately determine both core and shell diameter. For simulated particle sizes ranging between 7 and 20 μm, we obtain accuracies of (4.4±0.2)% and (0.74±0.01)% for the core and shell diameter, respectively.

High-throughput cell sorting with flow cytometers is an important tool in modern clinical cell studies. Most cytometers use biomarkers that selectively bind to the cell, but induce significant changes in morphology and inner cell processes leading sometimes to its death. This makes label-based cell sorting schemes unsuitable for further investigation. We propose a label-free technique that uses a digital inline holographic microscopy for cell imaging and an integrated, optical neural network for high-speed classification. The perspective of dense integration makes it attractive to ultrafast, large-scale cell sorting. Network simulations for a ternary classification task (monocytes/granulocytes/lymphocytes) resulted in 89% accuracy.

Abstract-Reservoir Computing (RC) is a computing scheme related to recurrent neural network theory. As a model for neural activity in the brain it attracts a lot of attention, especially because of its very simple training method. However, building a functional, on-chip, photonic implementation of RC remains a challenge. Scaling delay lines down from optical fibre scale to chip scale results in RC systems that compute faster, but at the same time require that the input signals are also scaled up in speed, which might be impractical or expensive. In this paper, we show that this problem can be alleviated by a masked RC system in which the amplitude of the input signal is modulated by a binary-valued mask. For a speech recognition task we demonstrate that the necessary input sample rate can be a factor of 40 smaller than in a conventional RC system. Additionally, we also show that linear discriminant analysis and input matrix optimisation is a well-performing alternative to linear regression for reservoir training.