Optical neuromorphic processing with soliton crystal Kerr microcombs: Scaling the network in size and speed

Convolutional neural networks (CNNs), inspired by biological visual cortex systems, are a powerful category of artificial neural networks that can extract the hierarchical features of raw data to greatly reduce the network parametric complexity and enhance the predicting accuracy. They are of significant interest for machine learning tasks such as computer vision, speech recognition, playing board games and medical diagnosis [1-7]. Optical neural networks offer the promise of dramatically accelerating computing speed to overcome the inherent bandwidth bottleneck of electronics. Here, we demonstrate a universal optical vector convolutional accelerator operating beyond 10 Tera-OPS (TOPS - operations per second), generating convolutions of images of 250,000 pixels with 8-bit resolution for 10 kernels simultaneously — enough for facial image recognition. We then use the same hardware to sequentially form a deep optical CNN with ten output neurons, achieving successful recognition of full 10 digits with 900 pixel handwritten digit images with 88% accuracy. Our results are based on simultaneously interleaving temporal, wavelength and spatial dimensions enabled by an integrated microcomb source. We show that this approach is scalable and trainable to much more complex networks for demanding applications such as unmanned vehicle and real-time video recognition.Keywords: Optical neural networks, neuromorphic processor, microcomb, convolutional accelerator

An optical neural network using less than 1 photon per multiplication

Deep learning has become a widespread tool in both science and industry. However, continued progress is hampered by the rapid growth in energy costs of ever-larger deep neural networks. Optical neural networks provide a potential means to solve the energy-cost problem faced by deep learning. Here, we experimentally demonstrate an optical neural network based on optical dot products that achieves 99% accuracy on handwritten-digit classification using ~3.1 detected photons per weight multiplication and ~90% accuracy using ~0.66 photons (~2.5 × 10−19 J of optical energy) per weight multiplication. The fundamental principle enabling our sub-photon-per-multiplication demonstration—noise reduction from the accumulation of scalar multiplications in dot-product sums—is applicable to many different optical-neural-network architectures. Our work shows that optical neural networks can achieve accurate results using extremely low optical energies.

Deep Learning for Photonic Design and Analysis: Principles and Applications

Innovative techniques play important roles in photonic structure design and complex optical data analysis. As a branch of machine learning, deep learning can automatically reveal the inherent connections behind the data by using hierarchically structured layers, which has found broad applications in photonics. In this paper, we review the recent advances of deep learning for the photonic structure design and optical data analysis, which is based on the two major learning paradigms of supervised learning and unsupervised learning. In addition, the optical neural networks with high parallelism and low energy consuming are also highlighted as novel computing architectures. The challenges and perspectives of this flourishing research field are discussed.

Intelligent systems based on photonics

This paper discusses the prospects of photonics, shows the relevance and applicability of photonics research. The poten-tial of photonics technologies to answer the socio-economic challenges of the digital transformation age is revealed. Opportunities that emerge with the introduction of photonic devices to various technical systems designed for environ-mental protection and quality of life improvement are demonstrated. Concrete photonics structures and devices for such key applications as spectroscopy, analog optical calculations, and optical neural networks are closely examined. Possi-ble applications for photonic sensors and new type spectrometers are outlined, their competitive advantages explored. Various geometries of extra fine compact photonic spectrometers are presented: based on digital planar diagrams, inte-grated into the photonic waveguides, metasurfaces, diffraction gratings with varying parameters. The benefits of analog optical computations against conventional electronic devices are discussed. Various nanophotonic structures designed for differential and integral operators are studied, solutions for edge detection are proposed. The concept for artificial intelligence implementation on the photonics platform using optical neural networks is analyzed. Various solutions are examined: containing sequences of diffraction elements and based on Huygens–Fresnel principle, as well as planar structures comprised of waveguides that interact as Mach–Zehnder interferometer. SPIE estimation of the international photonics market proposes that the peak of interest for this field is yet to be achieved and photonics will claim its place in the future technological landscape.

A scheme for simulating multi-level phase change photonics materials

Chalcogenide phase change materials (PCMs) have been extensively applied in data storage, and they are now being proposed for high resolution displays, holographic displays, reprogrammable photonics, and all-optical neural networks. These wide-ranging applications all exploit the radical property contrast between the PCMs’ different structural phases, extremely fast switching speed, long-term stability, and low energy consumption. Designing PCM photonic devices requires an accurate model to predict the response of the device during phase transitions. Here, we describe an approach that accurately predicts the microstructure and optical response of phase change materials during laser induced heating. The framework couples the Gillespie Cellular Automata approach for modelling phase transitions with effective medium theory and Fresnel equations. The accuracy of the approach is verified by comparing the PCM’s optical response and microstructure evolution with the results of nanosecond laser switching experiments. We anticipate that this approach to simulating the switching response of PCMs will become an important component for designing and simulating programmable photonics devices. The method is particularly important for predicting the multi-level optical response of PCMs, which is important for all-optical neural networks and PCM-programmable perceptrons.