Deep physical neural networks trained with backpropagation.

Deep-learning models have become pervasive tools in science and engineering. However, their energy requirements now increasingly limit their scalability1. Deep-learning accelerators2-9 aim to perform deep learning energy-efficiently, usually targeting the inference phase and often by exploiting physical substrates beyond conventional electronics. Approaches so far10-22 have been unable to apply the backpropagation algorithm to train unconventional novel hardware in situ. The advantages of backpropagation have made it the de facto training method for large-scale neural networks, so this deficiency constitutes a major impediment. Here we introduce a hybrid in situ-in silico algorithm, called physics-aware training, that applies backpropagation to train controllable physical systems. Just as deep learning realizes computations with deep neural networks made from layers of mathematical functions, our approach allows us to train deep physical neural networks made from layers of controllable physical systems, even when the physical layers lack any mathematical isomorphism to conventional artificial neural network layers. To demonstrate the universality of our approach, we train diverse physical neural networks based on optics, mechanics and electronics to experimentally perform audio and image classification tasks. Physics-aware training combines the scalability of backpropagation with the automatic mitigation of imperfections and noise achievable with in situ algorithms. Physical neural networks have the potential to perform machine learning faster and more energy-efficiently than conventional electronic processors and, more broadly, can endow physical systems with automatically designed physical functionalities, for example, for robotics23-26, materials27-29 and smart sensors30-32.

Engineering a Kerr-Based Deterministic Cubic Phase Gate via Gaussian Operations.

We propose a deterministic, measurement-free implementation of a cubic phase gate for continuous-variable quantum information processing. In our scheme, the applications of displacement and squeezing operations allow us to engineer the effective evolution of the quantum state propagating through an optical Kerr nonlinearity. Under appropriate conditions, we show that the input state evolves according to a cubic phase Hamiltonian, and we find that the cubic phase gate error decreases inverse quartically with the amount of quadrature squeezing, even in the presence of linear loss. We also show how our scheme can be adapted to deterministically generate a nonclassical approximate cubic phase state with high fidelity using a ratio of native nonlinearity to linear loss of only 10^{-4}, indicating that our approach may be experimentally viable in the near term even on all-optical platforms, e.g., using quantum solitons in pulsed nonlinear nanophotonics.

Several new directions for ultrafast fiber lasers [Invited].

Ultrafast fiber lasers have the potential to make applications of ultrashort pulses widespread - techniques not only for scientists, but also for doctors, manufacturing engineers, and more. Today, this potential is only realized in refractive surgery and some femtosecond micromachining. The existing market for ultrafast lasers remains dominated by solid-state lasers, primarily Ti:sapphire, due to their superior performance. Recent advances show routes to ultrafast fiber sources that provide performance and capabilities equal to, and in some cases beyond, those of Ti:sapphire, in compact, versatile, low-cost devices. In this paper, we discuss the prospects for future ultrafast fiber lasers built on new kinds of pulse generation that capitalize on nonlinear dynamics. We focus primarily on three promising directions: mode-locked oscillators that use nonlinearity to enhance performance; systems that use nonlinear pulse propagation to achieve ultrashort pulses without a mode-locked oscillator; and multimode fiber lasers that exploit nonlinearities in space and time to obtain unparalleled control over an electric field.

Self-seeded, multi-megawatt, Mamyshev oscillator.

We demonstrate a fiber oscillator that achieves 3 MW peak power, is easily started, and is environmentally stable. The Mamyshev oscillator delivers 190-nJ pulses that can be compressed externally to 35 fs duration. Accurate numerical modeling of the gain medium provides insight into the behavior and performance of the device.

Observation of multimode solitons in few-mode fiber.

We experimentally isolate and directly observe multimode solitons in few-mode graded-index fiber. We rely on Raman frequency shifts to spectrally isolate these multimode solitons. By varying the input energy and modal composition of the launched pulse, we observe a continuous variation of multimode solitons with different spatiotemporal properties. They exhibit an energy-volume relation that is distinct from those of single-mode and fully spatiotemporal solitons.

Kerr self-cleaning of femtosecond-pulsed beams in graded-index multimode fiber.

We observe a nonlinear spatial self-cleaning process for femtosecond pulses in graded-index (GRIN) multimode fiber (MMF). Pulses with ∼80 fs duration at 1030 nm are launched into GRIN MMF with 62.5 µm core. The near-field beam profile at the output end of the fiber evolves from a speckled pattern to a centered, bell-shaped transverse structure with increasing pulse energy. The experimental observations agree well with numerical simulations, which show that the Kerr nonlinearity underlies the process. This self-cleaning process may find applications in ultrafast pulse generation and beam-combining.

Ultrafast fiber lasers based on self-similar pulse evolution: a review of current progress.

Self-similar fiber oscillators are a relatively new class of mode-locked lasers. In these lasers, the self-similar evolution of a chirped parabolic pulse in normally-dispersive passive, active, or dispersion-decreasing fiber (DDF) is critical. In active (gain) fiber and DDF, the novel role of local nonlinear attraction makes the oscillators fundamentally different from any mode-locked lasers considered previously. In order to reconcile the spectral and temporal expansion of a pulse in the self-similar segment with the self-consistency required by a laser cavity's periodic boundary condition, several techniques have been applied. The result is a diverse range of fiber oscillators which demonstrate the exciting new design possibilities based on the self-similar model. Here, we review recent progress on self-similar oscillators both in passive and active fiber, and extensions of self-similar evolution for surpassing the limits of rare-earth gain media. We discuss some key remaining research questions and important future directions. Self-similar oscillators are capable of exceptional performance among ultrashort pulsed fiber lasers, and may be of key interest in the development of future ultrashort pulsed fiber lasers for medical imaging applications, as well as for low-noise fiber-based frequency combs. Their uniqueness among mode-locked lasers motivates study into their properties and behaviors and raises questions about how to understand mode-locked lasers more generally.

Spatiotemporal dynamics of multimode optical solitons.

As optical fiber communications and fiber lasers approach fundamental limits there is considerable interest in multimode fibers. In nonlinear science, they represent an exciting environment for complex nonlinear waves. As in single-mode fiber, solitons may be particularly important. Multimode solitons consist of synchronized, non-dispersive pulses in multiple spatial modes, which interact via the Kerr nonlinearity of the fiber. They are expected to exhibit novel spatiotemporal characteristics, dynamics and, like single-mode solitons, may provide a convenient intuitive tool for understanding more complex nonlinear phenomena in multimode fibers. Here we explore experimentally and numerically basic properties and spatiotemporal behaviors of these solitons: their formation, fission, and Raman dynamics.

Ultrabroadband Dispersive Radiation by Spatiotemporal Oscillation of Multimode Waves.

In nonlinear dynamical systems, qualitatively distinct phenomena occur depending continuously on the size of the bounded domain containing the system. For nonlinear waves, a multimode waveguide is a bounded three-dimensional domain, allowing observation of dynamics impossible in open settings. Here we study radiation emitted by bounded nonlinear waves: the spatiotemporal oscillations of solitons in multimode fiber generate multimode dispersive waves over an ultrabroadband spectral range. This work suggests routes to sources of coherent electromagnetic waves with unprecedented spectral range.

Divided-pulse lasers.

We demonstrate the use of coherent division and recombination of the pulse within an ultrafast laser cavity to manage the nonlinear phase accumulation and scale the output pulse energy. We implement the divided-pulse technique in an ytterbium-doped fiber laser and achieve 16 times scaling of the pulse energy, to generate 6 nJ and 1.4 ps solitons in single-mode fiber. Potential extensions of this concept are discussed.

The hardware is the software.

Human brains and bodies are not hardware running software: the hardware is the software. We reason that because the physics of artificial intelligence hardware and of human biological "hardware" is distinct, neuromorphic engineers need to be selective in the inspiration we take from neuroscience.

Deep nonlinear ablation of silicon with a quasi-continuous wave fiber laser at 1070 nm.

We achieve high aspect-ratio laser ablation of silicon with a strong nonlinear dependence on pulse duration while using a power density 10(6) times less than the threshold for typical multiphoton-mediated ablation. This is especially counter-intuitive as silicon is nominally transparent to the modulated continuous wave Yb:fiber laser used in the experiments. We perform time-domain finite-element simulations of thermal dynamics to investigate thermo-optical coupling and link the observed machining to an intensity-thresholded runaway thermo-optically nonlinear process. This effect, cascaded absorption, is qualitatively different from ablation observed using nanosecond-duration pulses and is general enough to potentially facilitate high-quality, high aspect-ratio, and economical processing of many materials.

Quantum-noise-limited optical neural networks operating at a few quanta per activation.

A practical limit to energy efficiency in computation is ultimately from noise, with quantum noise [1] as the fundamental floor. Analog physical neural networks [2], which hold promise for improved energy efficiency and speed compared to digital electronic neural networks, are nevertheless typically operated in a relatively high-power regime so that the signal-to-noise ratio (SNR) is large (>10). We study optical neural networks [3] operated in the limit where all layers except the last use only a single photon to cause a neuron activation. In this regime, activations are dominated by quantum noise from the fundamentally probabilistic nature of single-photon detection. We show that it is possible to perform accurate machine-learning inference in spite of the extremely high noise (signal-to-noise ratio ~ 1). We experimentally demonstrated MNIST handwritten-digit classification with a test accuracy of 98% using an optical neural network with a hidden layer operating in the single-photon regime; the optical energy used to perform the classification corresponds to 0.008 photons per multiply-accumulate (MAC) operation, which is equivalent to 0.003 attojoules of optical energy per MAC. Our experiment also used >40× fewer photons per inference than previous state-of-the-art low-optical-energy demonstrations [4, 5] to achieve the same accuracy of >90%. Our training approach, which directly models the system's stochastic behavior, might also prove useful with non-optical ultra-low-power hardware.

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.

High-power femtosecond pulses without a modelocked laser.

We demonstrate a fiber system which amplifies and compresses pulses from a gain-switched diode. A Mamyshev regenerator shortens the pulses and improves their coherence, enabling subsequent amplification by parabolic pre-shaping. As a result, we are able to control nonlinear effects and generate nearly transform-limited, 140-fs pulses with 13-MW peak power-an order-of-magnitude improvement over previous gain-switched diode sources. Seeding with a gain-switched diode results in random fluctuations of 2% in the pulse energy, which future work using known techniques may ameliorate. Further development may allow such systems to compete directly with sources based on modelocked oscillators in some applications while enjoying unparalleled robustness and repetition rate control.

Spatiotemporal mode-locking in multimode fiber lasers.

A laser is based on the electromagnetic modes of its resonator, which provides the feedback required for oscillation. Enormous progress has been made toward controlling the interactions of longitudinal modes in lasers with a single transverse mode. For example, the field of ultrafast science has been built on lasers that lock many longitudinal modes together to form ultrashort light pulses. However, coherent superposition of longitudinal and transverse modes in a laser has received little attention. We show that modal and chromatic dispersions in fiber lasers can be counteracted by strong spatial and spectral filtering. This allows locking of multiple transverse and longitudinal modes to create ultrashort pulses with a variety of spatiotemporal profiles. Multimode fiber lasers thus open new directions in studies of nonlinear wave propagation and capabilities for applications.

Megawatt peak power from a Mamyshev oscillator.

We demonstrate a fiber source with the best performance from an ultrafast fiber oscillator to date. The ring-cavity Mamyshev oscillator produces ~50-nJ and ~40-fs pulses. The peak power is an order of magnitude higher than that of previous lasers with similar fiber mode area. This performance is achieved by designing the oscillator to support parabolic pulse formation which enables the management of unprecedented nonlinear phase shifts. Experimental results are limited by available pump power. Numerical simulations reveal key aspects of the pulse evolution, and realistically suggest that (after external compression) peak powers that approach 10 MW are possible from ordinary single-mode fiber. The combination of practical features such as environmental stability, established previously, with the performance described here make the Mamyshev oscillator extremely attractive for applications.