Neuromorphic Photonics

Sharma A., Pande M., Katti A.

Rojas Yanez L., Hu H., Ciracì C., Palomba S.

Interest and excitement in nanophotonics—the study and control of light-matter interactions at the nanoscale—are driven by the ability to confine light to volumes well below a cubic wavelength, and, thereby, achieve extremely high intensities. This leads to light-matter interactions of unprecedented localization and strength. Such extreme behavior—both in terms of field enhancement and localization—can be achieved using plasmonic nanostructures, which concentrate light in regions much smaller than the wavelength of light, reducing the excitation power and, under certain conditions, removing phase-matching requirements in the nonlinear regime. In this study, we theoretically show that metal–dielectric–metal (MDM) slot waveguides (WGs), consisting of a thin dielectric layer sandwiched between metal films, provide the strongest confinement. We also demonstrate that integrating epsilon-near-zero (ENZ) materials within the MDM slot significantly improves the nonlinear conversion efficiency of these structures. The results show that the degenerate four-wave mixing conversion efficiency of these ENZ-MDM structures surpasses that of regular plasmonic structures and their dielectric counterparts, even under low pump power conditions, and remains robust despite higher losses in the ENZ material.

Flodgren V., Das A., Sestoft J.E., Alcer D., Jensen T.K., Jeddi H., Pettersson H., Nygård J., Borgström M.T., Linke H., Mikkelsen A.

Alcer D., Zaiats N., Jensen T.K., Philip A.M., Gkanias E., Ceberg N., Das A., Flodgren V., Heinze S., Borgström M.T., Webb B., Laursen B.W., Mikkelsen A.

Abstract Photonic solutions are potentially highly competitive for energy-efficient neuromorphic computing. However, a combination of specialized nanostructures is needed to implement all neuro-biological functionality. Here, we show that donor-acceptor Stenhouse adduct dyes integrated with III-V semiconductor nano-optoelectronics have combined excellent functionality for bio-inspired neural networks. The dye acts as synaptic weights in the optical interconnects, while the nano-optoelectronics provide neuron reception, interpretation and emission of light signals. These dyes can reversibly switch from absorbing to non-absorbing states, using specific wavelength ranges. Together, they show robust and predictable switching, low energy thermal reset and a memory dynamic range from days to sub-seconds that allows both short- and long-term memory operation at natural timescales. Furthermore, as the dyes do not need electrical connections, on-chip integration is simple. We illustrate the functionality using individual nanowire photodiodes as well as arrays. Based on the experimental performance metrics, our on-chip solution is capable of operating an anatomically validated model of the insect brain navigation complex.

Ahmadi R., Ahmadnejad A., Koohi S.

Neuromorphic engineering has emerged as a promising avenue for developing brain-inspired computational systems. However, conventional electronic AI-based processors often encounter challenges related to processing speed and thermal dissipation. As an alternative, optical implementations of such processors have been proposed, capitalizing on the intrinsic information-processing capabilities of light. Among the various Optical Neural Networks (ONNs) explored within the realm of optical neuromorphic engineering, Spiking Neural Networks (SNNs) have exhibited notable success in emulating the computational principles of the human brain. The event-based spiking nature of optical SNNs offers capabilities in low-power operation, speed, temporal processing, analog computing, and hardware efficiency that are difficult or impossible to match with other ONN types. In this work, we introduce the pioneering Free-space Optical Deep Spiking Convolutional Neural Network (OSCNN), a novel approach inspired by the computational model of the human eye. Our OSCNN leverages free-space optics to enhance power efficiency and processing speed while maintaining high accuracy in pattern detection. Specifically, our model employs Gabor filters in the initial layer for effective feature extraction, and utilizes optical components such as Intensity-to-Delay conversion and a synchronizer, designed using readily available optical components. The OSCNN was rigorously tested on benchmark datasets, including MNIST, ETH80, and Caltech, demonstrating competitive classification accuracy. Our comparative analysis reveals that the OSCNN consumes only 1.6 W of power with a processing speed of 2.44 ms, significantly outperforming conventional electronic CNNs on GPUs, which typically consume 150-300 W with processing speeds of 1-5 ms, and competing favorably with other free-space ONNs. Our contributions include addressing several key challenges in optical neural network implementation. To ensure nanometer-scale precision in component alignment, we propose advanced micro-positioning systems and active feedback control mechanisms. To enhance signal integrity, we employ high-quality optical components, error correction algorithms, adaptive optics, and noise-resistant coding schemes. The integration of optical and electronic components is optimized through the design of high-speed opto-electronic converters, custom integrated circuits, and advanced packaging techniques. Moreover, we utilize highly efficient, compact semiconductor laser diodes and develop novel cooling strategies to minimize power consumption and footprint.

Georgiou P., Tselios C., Alexandropoulos D.

We evaluate optically injected Quantum Dot spin Vertical Cavity Surface Emitting Lasers (QD spin-VCSELs) as a platform for realizing high-speed photonic spiking neurons. We study numerically the generation and inhibition of spikes of dual state optically pumped QD spin-VCSELs under the influence of pump ellipticity. The principles and working mechanism of the neuron are explored, emphasizing the role of pump ellipticity on the existence of an activation threshold for spiking. By appropriately configuring these levels, we reveal firing rate up to 5 GHz based on a temporal characteristics analysis on spiking activity. Our results highlight the potential of QD spin-VCSELs for advanced photonic spike processing functionalities.

Anufriev G., Furniss D., Farries M.C., Seddon A.B., Phang S.

AbstractA chemical discrimination system based on photonic reservoir computing is demonstrated experimentally for the first time. The system is inspired by the way humans perceive and process visual sensory information. The electro-optical reservoir computing system is a photonic analogue of the human nervous system with the read-out layer acting as the ‘brain’, and the sensor that of the human eye. A task-specific optimisation of the system is implemented, and the performance of the system for the discrimination between three chemicals is presented. The results are compared to the previously published numerical simulation (Anufriev et al. in Opt Mater Express 12:1767–1783, 2022, 10.1364/OME.449036). This publication provides a feasibility assessment and a demonstration of a practical realisation of photonic reservoir computing for a new neuromorphic sensing system - the next generation sensor with a built-in ‘intelligence’ which can be trained to ‘understand’ and to make a real time sensing decision based on the training data.

Sarantoglou G., Bogris A., Mesaritakis C.

Xia F., Kim K., Eliezer Y., Han S., Shaughnessy L., Gigan S., Cao H.

AbstractOptical information processing and computing can potentially offer enhanced performance, scalability and energy efficiency. However, achieving nonlinearity—a critical component of computation—remains challenging in the optical domain. Here we introduce a design that leverages a multiple-scattering cavity to passively induce optical nonlinear random mapping with a continuous-wave laser at a low power. Each scattering event effectively mixes information from different areas of a spatial light modulator, resulting in a highly nonlinear mapping between the input data and output pattern. We demonstrate that our design retains vital information even when the readout dimensionality is reduced, thereby enabling optical data compression. This capability allows our optical platforms to offer efficient optical information processing solutions across applications. We demonstrate our design’s efficacy across tasks, including classification, image reconstruction, keypoint detection and object detection, all of which are achieved through optical data compression combined with a digital decoder. In particular, high performance at extreme compression ratios is observed in real-time pedestrian detection. Our findings open pathways for novel algorithms and unconventional architectural designs for optical computing.

Zhang X., Mu P., Liu G., Wang Y., Li X.

Significant progress has been made in the research of all-optical neural networks in recent years. In this paper, we theoretically explore the properties of a neural system composed of semiconductor ring lasers (SRLs). Our study demonstrates that external optical signals generated by a tunable laser (TL) are injected into the first semiconductor ring laser photonic neuron (SRL1). Subsequently, the responses of SRL1 in the clockwise (CW) and counterclockwise (CCW) directions are unidirectionally injected into the CW and CCW directions of the second semiconductor ring laser photonic neuron (SRL2), respectively, which then exhibits similar spiking inhibition behaviors. Numerical simulations reveal that the spiking inhibition behavior of the SRL response can be precisely controlled by adjusting the perturbation time and intensity of the external injection signal, and this behavior is highly repeatable. Most importantly, we successfully achieve the stable transmission of these responses between the two SRL photonic neurons. These inhibition behaviors are analogous to those of biological neurons, but with a response speed reaching the sub-nanosecond level. Additionally, we indicate that SRL photonic neurons undergo a refractory-period-like phenomenon when subjected to two consecutive perturbations. These findings highlight the immense potential for the design and implementation of future all-optical neural networks, providing critical theoretical foundations and support for them.

Blow E.C., Zhang J., Zhang W., Bilodeau S., Lederman J., Shastri B.J., Prucnal P.R.

Farmakidis N., Dong B., Bhaskaran H.

Using photons in lieu of electrons to process information has been an exciting technological prospect for decades. Optical computing is gaining renewed enthusiasm, owing to the accumulated maturity of photonic integrated circuits and the pressing need for faster processing to cope with data generated by artificial intelligence. In neuromorphic photonics, the bosonic nature of light is exploited for high-speed, densely multiplexed linear operations, whereas the superior computing modalities of biological neurons are imitated to accelerate computations. Here, we provide an overview of recent advances in integrated synaptic optical devices and on-chip photonic neural networks focusing on the location in the architecture at which the optical to electrical conversion takes place. We present challenges associated with electro-optical conversions, implementations of optical nonlinearity, amplification and processing in the time domain, and we identify promising emerging photonic neuromorphic hardware. Neuromorphic photonics is an emerging computing platform that addresses the growing computational demands of modern society. We review advances in integrated neuromorphic photonics and discuss challenges associated with electro-optical conversions, implementations of nonlinearity, amplification and processing in the time domain.

Zheng M., liu W., Shi L., Zi J.

In order to harness diffractive neural networks (DNNs) for tasks that better align with real-world computer vision requirements, the incorporation of gray scale is essential. Currently, DNNs are not powerful enough to accomplish gray-scale image processing tasks due to limitations in their expressive power. In our work, we elucidate the relationship between the improvement in the expressive power of DNNs and the increase in the number of phase modulation layers, as well as the optimization of the Fresnel number, which can describe the diffraction process. To demonstrate this point, we numerically trained a double-layer DNN, addressing the prerequisites for intensity-based gray-scale image processing. Furthermore, we experimentally constructed this double-layer DNN based on digital micromirror devices and spatial light modulators, achieving eight-level intensity-based gray-scale image classification for the MNIST and Fashion-MNIST data sets. This optical system achieved the maximum accuracies of 95.10% and 80.61%, respectively.

Blow E.C., Bilodeau S., Zhang W., Ferreira de Lima T., Lederman J.C., Shastri B., Prucnal P.R.

Broadband analog signal processors utilizing silicon photonics have demonstrated a significant impact in numerous application spaces, offering unprecedented bandwidths, dynamic range, and tunability. In the past decade, microwave photonic techniques have been applied to neuromorphic processing, resulting in the development of novel photonic neural network architectures. Neuromorphic photonic systems can enable machine learning capabilities at extreme bandwidths and speeds. Herein, low‐quality factor microring resonators are implemented to demonstrate broadband optical weighting. In addition, silicon photonic neural network architectures are critically evaluated, simulated, and optimized from a radio‐frequency performance perspective. This analysis highlights the linear front‐end of the photonic neural network, the effects of linear and nonlinear loss within silicon waveguides, and the impact of electrical preamplification.

Morison H., Singh J., Al Kayed N., Aadhi A., Moridsadat M., Tamura M., Tait A.N., Shastri B.J.

Recurrent neural networks based on silicon photonics can take on a wide range of dynamical features at high bandwidth, but experimental demonstrations are being held back by of the lack of a physical-level simulation platform that accounts for parasitic effects. This study uses photonic Verilog-A models to demonstrate characteristic neural dynamics. Simulation reveals that these dynamics exhibit a topological equivalence to the continuous-time recurrent-neural-network model.