Deep learning with coherent nanophotonic circuits
01 Jul 2017-Vol. 11, Iss: 7, pp 441-446
TL;DR: A new architecture for a fully optical neural network is demonstrated that enables a computational speed enhancement of at least two orders of magnitude and three order of magnitude in power efficiency over state-of-the-art electronics.
Abstract: Artificial Neural Networks have dramatically improved performance for many machine learning tasks. We demonstrate a new architecture for a fully optical neural network that enables a computational speed enhancement of at least two orders of magnitude and three orders of magnitude in power efficiency over state-of-the-art electronics.
Citations
More filters
TL;DR: This article reviews in a selective way the recent research on the interface between machine learning and the physical sciences, including conceptual developments in ML motivated by physical insights, applications of machine learning techniques to several domains in physics, and cross fertilization between the two fields.
Abstract: Machine learning (ML) encompasses a broad range of algorithms and modeling tools used for a vast array of data processing tasks, which has entered most scientific disciplines in recent years. This article reviews in a selective way the recent research on the interface between machine learning and the physical sciences. This includes conceptual developments in ML motivated by physical insights, applications of machine learning techniques to several domains in physics, and cross fertilization between the two fields. After giving a basic notion of machine learning methods and principles, examples are described of how statistical physics is used to understand methods in ML. This review then describes applications of ML methods in particle physics and cosmology, quantum many-body physics, quantum computing, and chemical and material physics. Research and development into novel computing architectures aimed at accelerating ML are also highlighted. Each of the sections describe recent successes as well as domain-specific methodology and challenges.
1,504 citations
TL;DR: Monolithically integrated lithium niobate electro-optic modulators that feature a CMOS-compatible drive voltage, support data rates up to 210 gigabits per second and show an on-chip optical loss of less than 0.5 decibels are demonstrated.
Abstract: Electro-optic modulators translate high-speed electronic signals into the optical domain and are critical components in modern telecommunication networks1,2 and microwave-photonic systems3,4. They are also expected to be building blocks for emerging applications such as quantum photonics5,6 and non-reciprocal optics7,8. All of these applications require chip-scale electro-optic modulators that operate at voltages compatible with complementary metal–oxide–semiconductor (CMOS) technology, have ultra-high electro-optic bandwidths and feature very low optical losses. Integrated modulator platforms based on materials such as silicon, indium phosphide or polymers have not yet been able to meet these requirements simultaneously because of the intrinsic limitations of the materials used. On the other hand, lithium niobate electro-optic modulators, the workhorse of the optoelectronic industry for decades9, have been challenging to integrate on-chip because of difficulties in microstructuring lithium niobate. The current generation of lithium niobate modulators are bulky, expensive, limited in bandwidth and require high drive voltages, and thus are unable to reach the full potential of the material. Here we overcome these limitations and demonstrate monolithically integrated lithium niobate electro-optic modulators that feature a CMOS-compatible drive voltage, support data rates up to 210 gigabits per second and show an on-chip optical loss of less than 0.5 decibels. We achieve this by engineering the microwave and photonic circuits to achieve high electro-optical efficiencies, ultra-low optical losses and group-velocity matching simultaneously. Our scalable modulator devices could provide cost-effective, low-power and ultra-high-speed solutions for next-generation optical communication networks and microwave photonic systems. Furthermore, our approach could lead to large-scale ultra-low-loss photonic circuits that are reconfigurable on a picosecond timescale, enabling a wide range of quantum and classical applications5,10,11 including feed-forward photonic quantum computation. Chip-scale lithium niobate electro-optic modulators that rapidly convert electrical to optical signals and use CMOS-compatible voltages could prove useful in optical communication networks, microwave photonic systems and photonic computation.
1,358 citations
TL;DR: 3D-printed D2NNs are created that implement classification of images of handwritten digits and fashion products, as well as the function of an imaging lens at a terahertz spectrum.
Abstract: Deep learning has been transforming our ability to execute advanced inference tasks using computers. Here we introduce a physical mechanism to perform machine learning by demonstrating an all-optical diffractive deep neural network (D2NN) architecture that can implement various functions following the deep learning-based design of passive diffractive layers that work collectively. We created 3D-printed D2NNs that implement classification of images of handwritten digits and fashion products, as well as the function of an imaging lens at a terahertz spectrum. Our all-optical deep learning framework can perform, at the speed of light, various complex functions that computer-based neural networks can execute; will find applications in all-optical image analysis, feature detection, and object classification; and will also enable new camera designs and optical components that perform distinctive tasks using D2NNs.
1,145 citations
TL;DR: An optical version of a brain-inspired neurosynaptic system, using wavelength division multiplexing techniques, is presented that is capable of supervised and unsupervised learning.
Abstract: Software implementations of brain-inspired computing underlie many important computational tasks, from image processing to speech recognition, artificial intelligence and deep learning applications. Yet, unlike real neural tissue, traditional computing architectures physically separate the core computing functions of memory and processing, making fast, efficient and low-energy computing difficult to achieve. To overcome such limitations, an attractive alternative is to design hardware that mimics neurons and synapses. Such hardware, when connected in networks or neuromorphic systems, processes information in a way more analogous to brains. Here we present an all-optical version of such a neurosynaptic system, capable of supervised and unsupervised learning. We exploit wavelength division multiplexing techniques to implement a scalable circuit architecture for photonic neural networks, successfully demonstrating pattern recognition directly in the optical domain. Such photonic neurosynaptic networks promise access to the high speed and high bandwidth inherent to optical systems, thus enabling the direct processing of optical telecommunication and visual data. An optical version of a brain-inspired neurosynaptic system, using wavelength division multiplexing techniques, is presented that is capable of supervised and unsupervised learning.
862 citations
TL;DR: A tandem neural network architecture is demonstrated that tolerates inconsistent training instances in inverse design of nanophotonic devices and provides a way to train large neural networks for the inverseDesign of complex photonic structures.
Abstract: Data inconsistency leads to a slow training process when deep neural networks are used for the inverse design of photonic devices, an issue that arises from the fundamental property of nonuniqueness in all inverse scattering problems. Here we show that by combining forward modeling and inverse design in a tandem architecture, one can overcome this fundamental issue, allowing deep neural networks to be effectively trained by data sets that contain nonunique electromagnetic scattering instances. This paves the way for using deep neural networks to design complex photonic structures that require large training data sets.
619 citations
References
More filters
Proceedings Article•
19 Jun 2016TL;DR: This work constructs an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned, and demonstrates the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.
Abstract: Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very longterm dependencies.
630 citations
"Deep learning with coherent nanopho..." refers background in this paper
...Furthermore, with this on-chip training scheme, one can readily parametrize and train unitary matrices–an approach known to be particularly useful for deep neural networks [42]....
[...]
TL;DR: This work reports an optoelectronic implementation of reservoir computing based on a recently proposed architecture consisting of a single non linear node and a delay line that is sufficiently fast for real time information processing.
Abstract: Reservoir computing is a recently introduced, highly efficient bio-inspired approach for processing time dependent data. The basic scheme of reservoir computing consists of a non linear recurrent dynamical system coupled to a single input layer and a single output layer. Within these constraints many implementations are possible. Here we report an optoelectronic implementation of reservoir computing based on a recently proposed architecture consisting of a single non linear node and a delay line. Our implementation is sufficiently fast for real time information processing. We illustrate its performance on tasks of practical importance such as nonlinear channel equalization and speech recognition, and obtain results comparable to state of the art digital implementations.
606 citations
TL;DR: Numerical and experimental results presented show that the approach is capable of introducing accuracy and robustness to optical processing while maintaining the traditional advantages of optics, namely, parallelism and massive interconnection capability.
Abstract: Optical implementation of content addressable associative memory based on the Hopfield model for neural networks and on the addition of nonlinear iterative feedback to a vector–matrix multiplier is described. Numerical and experimental results presented show that the approach is capable of introducing accuracy and robustness to optical processing while maintaining the traditional advantages of optics, namely, parallelism and massive interconnection capability. Moreover a potentially useful link between neural processing and optics that can be of interest in pattern recognition and machine vision is established.
584 citations
TL;DR: A very high optical bandwidth, estimated up to 120GHz, was evidenced in 10 µm long Ge photodetectors selectively grown at the end of silicon waveguides using three kinds of experimental set-ups.
Abstract: We report on lateral pin germanium photodetectors selectively grown at the end of silicon waveguides. A very high optical bandwidth, estimated up to 120GHz, was evidenced in 10 µm long Ge photodetectors using three kinds of experimental set-ups. In addition, a responsivity of 0.8 A/W at 1550 nm was measured. An open eye diagrams at 40Gb/s were demonstrated under zero-bias at a wavelength of 1.55 µm.
417 citations
TL;DR: In this article, the intrinsic properties of monolayer graphene allow it to act as a more effective saturable absorber for mode-locking fiber lasers when compared to multilayer graphene.
Abstract: We demonstrate that the intrinsic properties of monolayer graphene allow it to act as a more effective saturable absorber for mode-locking fiber lasers when compared to multilayer graphene. The absorption of monolayer graphene can be saturated at lower excitation intensity compared to multilayer graphene, graphene with wrinkle-like defects, or functionalized graphene. Monolayer graphene has a remarkably large modulation depth of 65.9%, whereas the modulation depth of multilayer graphene is greatly reduced due to nonsaturable absorption and scattering loss. Picosecond ultrafast laser pulses (1.23 ps) can be generated using monolayer graphene as a saturable absorber. Due to the ultrafast relaxation time, larger modulation depth and lower scattering loss of monolayer graphene, it performs better than multilayer graphene in terms of pulse shaping ability, pulse stability, and output energy.
406 citations