Neural networks are one of the disruptive computing concepts of our time. However, they fundamentally differ from classical, algorithmic computing. These differences result in equally fundamental, severe and relevant challenges for neural network computing using current computing substrates. Neural networks urge for parallelism across the entire processor and for a co-location of memory and arithmetic, i.e. beyond von Neumann architectures. Parallelism in particular made photonics a highly promising platform, yet until now scalable and integratable concepts are scarce. Here, we demonstrate for the first time how a fully parallel and fully implemented photonic neural network can be realized by spatially multiplexing neurons across the complex optical near-field of a semiconductor multimode laser. Discrete spatial sampling defines ∼90 nodes on the surface of a large-area vertical cavity surface emitting laser that is optically injected with the artificial neural networks input information. Importantly, all neural network connections are realized in hardware, and our processor produces results without pre- or post-processing. Input and output weights are realized via the complex transmission matrix of a multimode fiber and a digital micro-mirror array, respectively. We train the readout weights to perform 2-bit header recognition, a 2-bit XOR logical function and 2-bit digital to analog conversion, and obtain < 0.9 × 10 − 3 and 2.9 × 10⁻² error rates for digit recognition and XOR, respectively. Finally, the digital to analog conversion can be realized with a standard deviation of only 5.4 × 10⁻². Crucially, our proof-of-concept system is scalable to much larger sizes and to bandwidths in excess of 20 GHz.

Brain-inspired computing concepts like artificial neural networks have become promising alternatives to classical von Neumann computer architectures. Photonic neural networks target the realizations of neurons, network connections and potentially learning in photonic substrates. Here, we report the development of a nanophotonic hardware platform of fast and energy-efficient photonic neurons via arrays of high-quality vertical cavity surface emitting lasers (VCSELs). The developed 5 × 5 VCSEL arrays provide high optical injection locking efficiency through homogeneous fabrication combined with individual control over the laser wavelengths. Injection locking is crucial for the reliable processing of information in VCSEL-based photonic neurons, and we demonstrate the suitability of the VCSEL arrays by injection locking measurements and current-induced spectral fine-tuning. We find that our investigated array can readily be tuned to the required spectral homogeneity, and as such show that VCSEL arrays based on our technology can act as highly energy efficient and ultra-fast photonic neurons for next generation photonic neural networks. Combined with fully parallel photonic networks our substrates are promising for ultra-fast operation reaching 10 s of GHz bandwidths, and we show that a single non-linear transformation based on our lasers will consume only about 100 fJ per VCSEL, which is highly competitive, compared to other platforms.

Single-emitter vertical-cavity surface-emitting lasers (VCSELs) and multiple-emitter VCSEL arrays designed for emission at 980 nm, and with large oxide aperture diameters (φ) from ~10.5 to 33.5 μm and with φ of ~7.5 μm that are configured in three-emitter and seven-emitter electrically parallel triangular and hexagonal arrays, respectively, have record bandwidths and optical output powers for VCSELs with such large cumulative emitting areas. We demonstrate room temperature (RT) maximum small-signal modulation bandwidths (f _3dBmax ) of 26.6 to 18.6 GHz with corresponding continuous wave (CW) optical output powers (L) of 16.2 and 47 mW for VCSELs with φ ~13.5 and 33.5 μm, respectively. For parallel 980 nm triple and septuple arrays with φ ~7.5 μm for each VCSEL, we demonstrate RT _f3dBmax of 25.5 and 24.8 GHz with corresponding CW L of 22.7 and 50.1 mW, respectively. We perform a comparative analysis of the RT results, which consist of standard static light output power-current-voltage (LIV) characteristics to determine the typical VCSEL figures-of-merit including wall plug efficiency (WPE), LI slope efficiency, spectral emission versus current (I), and small-signal frequency response curves. We examine the tradeoffs in optical output power, small-signal modulation bandwidth, and WPE for the various VCSEL designs, and find that while the combination of f _3dBmax and L are record values for single VCSELs and VCSEL arrays, these parameters fall off as the cumulative VCSEL array area increases to emit higher overall optical power.

Many bosons can occupy a single quantum state without a limit. It is described by the quantum-mechanical Bose–Einstein statistic, which allows Bose–Einstein condensation at low temperatures and high particle densities. Photons, historically the first considered bosonic gas, were late to show this phenomenon, observed in rhodamine-filled microcavities and doped fibre cavities. These findings have raised the question of whether condensation is also common in other laser systems with potential technological applications. Here we show the Bose–Einstein condensation of photons in a broad-area vertical-cavity surface-emitting laser with a slight cavity-gain spectral detuning. We observed a Bose–Einstein condensate in the fundamental transversal optical mode at a critical phase-space density. The experimental results follow the equation of state for a two-dimensional gas of bosons in thermal equilibrium, although the extracted spectral temperatures were lower than the device’s. This is interpreted as originating from the driven-dissipative nature of the photon gas. In contrast, non-equilibrium lasing action is observed in the higher-order modes in more negatively detuned device. Our work opens the way for the potential exploration of superfluid physics of interacting photons mediated by semiconductor optical nonlinearities. It also shows great promise for enabling single-mode high-power emission from a large-aperture device.

Artificial neural networks have become a staple computing technique in many fields. Yet, they present fundamental differences with classical computing hardware in the way they process information. Photonic implementations of neural network architectures potentially offer fundamental advantages over their electronic counterparts in terms of speed, processing parallelism, scalability and energy efficiency. Scalable and high performance photonic neural networks (PNNs) have been demonstrated, yet they remain scarce. In this work, we study the performance of such a scalable, fully parallel and autonomous PNN based on large area vertical-cavity surface-emitting lasers (LA-VCSEL). We show how the performance varies with different physical parameters, namely, injection wavelength, injection power, and bias current. Furthermore, we link these physical parameters to the general computational measures of consistency and dimensionality. We present a general method of gauging dimensionality in high dimensional nonlinear systems subject to noise, which could be applied to many systems in the context of neuromorphic computing. Our work will inform future implementations of spatially multiplexed VCSEL PNNs.