Search for: All records

The explosive growth in computation and energy cost of artificial intelligence has spurred interest in alternative computing modalities to conventional electronic processors. Photonic processors, which use photons instead of electrons, promise optical neural networks with ultralow latency and power consumption. However, existing optical neural networks, limited by their designs, have not achieved the recognition accuracy of modern electronic neural networks. In this work, we bridge this gap by embedding parallelized optical computation into flat camera optics that perform neural network computations during capture, before recording on the sensor. We leverage large kernels and propose a spatially varying convolutional network learned through a low-dimensional reparameterization. We instantiate this network inside the camera lens with a nanophotonic array with angle-dependent responses. Combined with a lightweight electronic back-end of about 2K parameters, our reconfigurable nanophotonic neural network achieves 72.76% accuracy on CIFAR-10, surpassing AlexNet (72.64%), and advancing optical neural networks into the deep learning era. 

Abstract A new release of the Monte Carlo event generator (version 7.3) has been launched. This iteration encompasses several enhancements over its predecessor, version 7.2. Noteworthy upgrades include: the implementation of a process-independent electroweak angular-ordered parton shower integrated with QCD and QED radiation; a new recoil scheme for initial-state radiation improving the behaviour of the angular-ordered parton shower; the incorporation of the heavy quark effective theory to refine the hadronization and decay of excited heavy mesons and heavy baryons; a dynamic strategy to regulate the kinematic threshold of cluster splittings within the cluster hadronization model; several improvements to the structure of the cluster hadronization model allowing for refined models; the possibility to extract event-by-event hadronization corrections in a well-defined way; the possibility of using the string model, with a dedicated tune. Additionally, a new tuning of the parton shower and hadronization parameters has been executed. This article discusses the novel features introduced in version 7.3.0. 

Extended depth of focus (EDOF) optics can enable lower complexity optical imaging systems when compared to active focusing solutions. With existing EDOF optics, however, it is difficult to achieve high resolution and high collection efficiency simultaneously. The subwavelength spacing of scatterers in a meta-optic enables the engineering of very steep phase gradients; thus, meta-optics can achieve both a large physical aperture and a high numerical aperture. Here, we demonstrate a fast $(f/1.75)$EDOF meta-optic operating at visible wavelengths, with an aperture of 2 mm and focal range from 3.5 mm to 14.5 mm (286 diopters to 69 diopters), which is a $250\times$elongation of the depth of focus relative to a standard lens. Depth-independent performance is shown by imaging at a range of finite conjugates, with a minimum spatial resolution of $\sim 9.84\mathrm{\mu m}$(50.8 cycles/mm). We also demonstrate operation of a directly integrated EDOF meta-optic camera module to evaluate imaging at multiple object distances, a functionality which would otherwise require a varifocal lens. 

Abstract Nano-optic imagers that modulate light at sub-wavelength scales could enable new applications in diverse domains ranging from robotics to medicine. Although metasurface optics offer a path to such ultra-small imagers, existing methods have achieved image quality far worse than bulky refractive alternatives, fundamentally limited by aberrations at large apertures and low f-numbers. In this work, we close this performance gap by introducing a neural nano-optics imager. We devise a fully differentiable learning framework that learns a metasurface physical structure in conjunction with a neural feature-based image reconstruction algorithm. Experimentally validating the proposed method, we achieve an order of magnitude lower reconstruction error than existing approaches. As such, we present a high-quality, nano-optic imager that combines the widest field-of-view for full-color metasurface operation while simultaneously achieving the largest demonstrated aperture of 0.5 mm at an f-number of 2. 

Establishing a coherent interaction between a material resonance and an optical cavity is a necessary first step to study semiconductor quantum optics. Here we report on the signature of a coherent interaction between a two-dimensional excitonic transition in monolayer MoSe₂and a zero-dimensional, ultra-low mode volume (V_m ∼ 2(λ/n)³) on-chip photonic crystal nanocavity. This coherent interaction manifests as a dispersive shift of the cavity transmission spectrum, when the exciton-cavity detuning is decreased via temperature tuning. The exciton-cavity coupling is estimated to be ≈6.5 meV, with a cooperativity of ≈4.0 at 80 K, showing our material system is on the verge of strong coupling. The small mode-volume of the resonator is instrumental in reaching the strongly nonlinear regime, while on-chip cavities will help create a scalable quantum photonic platform. 

Controlling the propagation of optical fields in three dimensions using arrays of discrete dielectric scatterers is an active area of research. These arrays can create optical elements with functionalities unrealizable in conventional optics. Here, we present an inverse design method based on the inverse Mie scattering problem for producing three-dimensional optical field patterns. Using this method, we demonstrate a device that focuses 1.55-μm light into a depth-variant discrete helical pattern. The reported device is fabricated using two-photon lithography and has a footprint of 144 μm by 144 μm, the largest of any inverse-designed photonic structure to date. This inverse design method constitutes an important step toward designer free-space optics, where unique optical elements are produced for user-specified functionalities.