Deep learning with coherent nanophotonic circuits

Abstract

These authors contributed equally to this work.Artificial Neural Networks are computational network models inspired by signal processing in the brain. These models have dramatically improved the performance of many learning tasks, including speech and object recognition. However, today's computing hardware is inefficient at implementing neural networks, in large part because much of it was designed for von Neumann computing schemes. Significant effort has been made to develop electronic architectures tuned to implement artificial neural networks that improve upon both computational speed and energy efficiency. Here, we propose a new architecture for a fully-optical neural network that, using unique advantages of optics, promises a computational speed enhancement of at least two orders of magnitude over the state-of-the-art and three orders of magnitude in power efficiency for conventional learning tasks. We experimentally demonstrate essential parts of our architecture using a programmable nanophotonic processor.Modern computers based on the von Neumann architecture are far more power-hungry and less effective than their biological counterparts -central nervous systemsfor a wide range of tasks including perception, communication, learning, and decision making. With the increasing data volume associated with processing big data, developing computers that learn, combine, and analyze vast amounts of information quickly and efficiently is becoming increasingly important. For example, speech recognition software (e.g., Apple's Siri) is typically executed in the cloud since these computations are too taxing for mobile hardware; realtime image processing is an even more demanding task [1]. To address the shortcomings of von Neumann computing architectures for neural networks, much recent work has focused on increasing artificial neural network computing speed and power efficiency by developing electronic architectures (such as ASIC and FPGA chips) specifically tailored to a task [2][3][4][5]. Recent demonstrations of electronic neuromorphic hardware architectures have reported improved computational performance [6]. Hybrid optical-electronic systems that implement spike processing [7][8][9] and reservoir computing [10,11] have also been investigated recently. However, the computational speed and power efficiency achieved with these hardware architectures are still limited by electronic clock rates and ohmic losses.Fully-optical neural networks offer a promising alternative approach to microelectronic and hybrid optical-electronic implementations. Linear transformations (and certain nonlinear transformations) can be performed at the speed of light and detected at rates exceeding 100 GHz [12] in photonic networks, and in some cases, with minimal power consumption [13]. For example, it is well known that a common lens performs Fourier transform without any power consumption, and that certain matrix operations can also be performed optically without consuming power. However, implementing such transformations with bulk opti...

On-chip silicon photonic signaling and processing: a review

“…The future trends of on-chip silicon photonic signaling and processing are the large-scale integration of the whole communication system on a chip and the hybrid integeration of silicon photonics and silicon nanoelectronics on a chip [25,287]. Beyond photonic signaling and processing, numerous other novel applications of photonics, including supercontinuum generation [288][289][290], real-time optical measurements [291][292][293], optomechanics [38,294], sensors [44], time lens [295], optical neural network [296,297], deep learning [298], microwave photonics [299], quantum optics [300][301][302], and mid-infrared photonics [49,50,[303][304][305] come into being. Small footprint, low cost, and high efficiency are highly desired in these applications.…”

“…For example, octave-spanning supercontinuum generation at telecommunication wavelengths has been demonstrated recently using a specificly designed silicon waveguide [289]. Very recently, silicon photonics has been introduced to develope large scale neural network for deep learning [298]. Such nanophotonic chip is able to give equivalent learning performance while potentially achieve three orders of magnitude faster speed than conventional electronic counterparts.…”

Universal programmable photonic architecture for quantum information processing

“…Such devices have a wide range of applications in classical information processing [4,[6][7][8][9][10], and integrated universal photonic circuits provides an especially promising hardware platform for high-throughput, energy-efficient machine learning. [11][12][13][14] These devices also have promising applications in quantum information processing: recent demonstrations of boson sampling [15], quantum transport dynamics [16], photonic quantum walks [17], counterfactual communication [18], and probabilistic two-photon gates [19] have all been performed on this type of programmable photonic hardware. Photonic systems offer a range of unique advantages over other substrates for quantum information processing: optical quantum states have long coherence times and can be maintained at room temperature, since they interact very weakly with their environment; photonic qubits are optimal information carriers for distant nodes within quantum networks; and MZIs provide simple, high-fidelity implementations of single-qubit operations which can be integrated into a photonic chip.…”

Canceling Thermal Cross-Talk Effects in Photonic Integrated Circuits

“…hotonic integrated circuits (PICs) are evolving towards onchip reconfigurable architectures and general purpose programmable photonic processors, enabling the implementation of many different functionalities on-demand [1,2,3,4]. These schemes rely on the use of a large number of optical interferometers, such as Mach-Zehnder interferometers (MZI) and microring resonators (MRRs), whose individual working point is inherently related to the phase delay between the interfering optical beams.…”