Metaoptics are thin, planar surfaces consisting of many subwavelength optical resonators that can be designed to simultaneously control the amplitude, phase, and polarization to arbitrarily shape an optical wavefront much in the same manner as a traditional lens but with a much smaller form factor. Incorporation of metaoptics into a conventional optical system spans multiple length scales between that of the individual metaoptic elements (<λ) and that of the entire size of the optic (≫λ), making computational techniques that accurately simulate the optical response of metaoptics computationally intractable, while more efficient techniques utilizing various approximations suffer from inaccuracies in their prediction of the optical response. To overcome the trade-off between speed and accuracy, we implement a transformer-based neural network solver to calculate the optical response of metaoptics and combine it with commercial ray optics software incorporating Fourier propagation methods to simulate an entire optical system. We demonstrate that this neural net method is over three orders of magnitude faster than a traditional finite-difference time-domain method, with only a 0.47% deviation in total irradiance when compared with a full wave simulation, which is nearly two orders of magnitude more accurate than standard approximation methods for metaoptics. The ability to accurately and efficiently predict the optical response of a metaoptic could enable their optimization, further accelerating and facilitating their application.

From higher computational efficiency to enabling the discovery of novel and complex structures, deep learning has emerged as a powerful framework for the design and optimization of nanophotonic circuits and components. However, both data-driven and exploration-based machine learning strategies have limitations in their effectiveness for nanophotonic inverse design. Supervised machine learning approaches require large quantities of training data to produce high-performance models and have difficulty generalizing beyond training data given the complexity of the design space. Unsupervised and reinforcement learning-based approaches on the other hand can have very lengthy training or optimization times associated with them. Here we demonstrate a hybrid supervised learning and reinforcement learning approach to the inverse design of nanophotonic structures and show this approach can reduce training data dependence, improve the generalizability of model predictions, and significantly shorten exploratory training times. The presented strategy thus addresses several contemporary deep learning-based challenges, while opening the door for new design methodologies that leverage multiple classes of machine learning algorithms to produce more effective and practical solutions for photonic design.

In recent years, hybrid design strategies combining machine learning (ML) with electromagnetic optimization algorithms have emerged as a new paradigm for the inverse design of photonic structures and devices. While a trained, data-driven neural network can rapidly identify solutions near the global optimum with a given dataset's design space, an iterative optimization algorithm can further refine the solution and overcome dataset limitations. Furthermore, such hybrid ML-optimization methodologies can reduce computational costs and expedite the discovery of novel electromagnetic components. However, existing hybrid ML-optimization methods have yet to optimize across both materials and geometries in a single integrated and user-friendly environment. In addition, due to the challenge of acquiring large datasets for ML, as well as the exponential growth of isolated models being trained for photonics design, there is a need to standardize the ML-optimization workflow while making the pre-trained models easily accessible. Motivated by these challenges, here we introduce DeepAdjoint, a general-purpose, open-source, and multi-objective "all-in-one" global photonics inverse design application framework which integrates pre-trained deep generative networks with state-of-the-art electromagnetic optimization algorithms such as the adjoint variables method. DeepAdjoint allows a designer to specify an arbitrary optical design target, then obtain a photonic structure that is robust to fabrication tolerances and possesses the desired optical properties - all within a single user-guided application interface. Our framework thus paves a path towards the systematic unification of ML and optimization algorithms for photonic inverse design.

A fundamental challenge in the design of photonic devices, and electromagnetic structures more generally, is the optimization of their overall architecture to achieve a desired response. To this end, topology or shape optimizers based on the adjoint variables method have been widely adopted due to their high computational efficiency and ability to create complex freeform geometries. However, the functional understanding of such freeform structures remains a black box. Moreover, unless a design space of high-performance devices is known in advance, such gradient-based optimizers can get trapped in local minima valleys or saddle points, which limits performance achievable through this inverse design process. To elucidate the relationships between device performance and nanoscale structuring while mitigating the effects of local minima trapping, we present an inverse design framework that combines adjoint optimization, automated machine learning (AutoML), and explainable artificial intelligence (XAI). Integrated with a numerical electromagnetics simulation method, our framework reveals structural contributions towards a figure-of-merit (FOM) of interest. Through an explanation-based reoptimization process, this information is then leveraged to minimize the FOM further than that obtained through adjoint optimization alone, thus overcoming the optimization's local minima. We demonstrate our framework in the context of waveguide design and achieve between 39% and 74% increases in device performance relative to state-of-the-art adjoint optimization-based inverse design across a range of telecom wavelengths. Results of this work therefore highlight machine learning strategies that can substantially extend and enhance the capabilities of a conventional, optimization-based inverse design algorithm while revealing deeper insights into the algorithm's designs.

Combining planar optics such as metalenses or metacorrectors with conventional lenses can drastically improve the optical performance of imaging systems with additional benefits such as cost, size and weight improvements. However, incorporating metacorrectors into conventional lens design requires multiscale simulations to account for the different length scale interactions. Namely, full wave scattering and geometric optics analysis is needed for the metacorrector and hybrid lens design, respectively. Multiscale inverse optimization using Sandia National Laboratories’ MIRaGE along with different wave propagation techniques and commercial-off-the-shelf GO tools are considered to accurately predict hybrid design optical performance.

Incorporating planar optics such as metalenses or metacorrectors into optical designs can drastically improve the performance of imaging systems with additional benefits such as cost, size and weight improvements. However, modeling of such hybrid lenses is challenging because of the multi-scale nature of the simulation. We demonstrate that one can combine ray optic simulations with full wave electromagnetic simulations and Fourier optics approaches to model a whole compound/hybrid lens considering all metasurface unit cell interactions and to study the effect of possible fabrication errors.

We develop an array-level inverse design framework that accounts for the nonideal optical response to optimize the array configuration of reconfigurable metasurfaces. As an example, we validate this approach experimentally and demonstrate high-directivity beam steering with nonideal antenna components exhibiting a phase shift of ~220°.

Sub-wavelength periodic arrays exhibit narrow near-unity reflection bands that arise from guided mode resonances. These resonances have extremely high quality factor (i.e., narrow band features) and are ideal for filtering applications. A high quality factor requires many periods, causing large lateral footprints that limit an imaging system's spatial resolution. We present a 1D ultra-thin (<100 nm) compact finite design of seven periods of amorphous Si slabs with subwavelength periodicity surrounded by Al mirrors, which allow the finite array to approximate an infinite array and enabling a small footprint (∼5 μm), for near-infrared applications (λ = 800–2000 nm). We demonstrate spectral tunability (amplitude, bandwidth, and peak location) via geometric parameter variation and demonstrate the performance of these filters both in experiment and in simulation. This work miniaturizes guided-mode resonance filters, previously limited by extremely large footprints, while being relatively cheap and simple to fabricate compared to many existing designs.

We report an array-level inverse design approach to optimize the beam steering performance of active metasurfaces, thus overcoming the limitations posed by nonideal metasurface phase and amplitude tuning. In contrast to device-level topology optimization of passive metasurfaces, the outlined system-level optimization framework relies on the electrical tunability of geometrically identical nanoantennas, enabling the design of active antenna arrays with variable spatial phase and amplitude profiles. Based on this method, we demonstrate high-directivity, continuous beam steering up to 70° for phased arrays with realistic tunable antenna designs, despite nonidealities such as strong covariation of scattered light amplitude with phase. Nonintuitive array phase and amplitude profiles further facilitate beam steering with a phase modulation range as low as 180°. Furthermore, we use the device geometries presented in this work for experimental validation of the system-level inverse design approach of active beam steering metasurfaces. The proposed method offers a framework to optimize nanophotonic structures at the array level that is potentially applicable to a wide variety of objective functions and actively tunable metasurface antenna array platforms.

We designed mirrored plasmonic filters using an advanced active machine learning algorithm that efficiently explores multiple physical models with different approximation fidelities and costs. This method is applicable to a variety of nanophotonics optimization problems.

Long-term space flights and cis-lunar research platforms require a sustainable and light life-support hardware which can be reliably employed outside the Earth's atmosphere. So-called 'solar fuel' devices, currently developed for terrestrial applications in the quest for realizing a sustainable energy economy on Earth, provide promising alternative systems to existing air-revitalization units employed on the International Space Station (ISS) through photoelectrochemical water-splitting and hydrogen production. One obstacle for water (photo-) electrolysis in reduced gravity environments is the absence of buoyancy and the consequential, hindered gas bubble release from the electrode surface. This causes the formation of gas bubble froth layers in proximity to the electrode surface, leading to an increase in ohmic resistance and cell-efficiency loss due to reduced mass transfer of substrates and products to and from the electrode. Recently, we have demonstrated efficient solar hydrogen production in microgravity environment, using an integrated semiconductor-electrocatalyst system with p-type indium phosphide as the light-absorber and a rhodium electrocatalyst. By nanostructuring the electrocatalyst using shadow nanosphere lithography and thereby creating catalytic 'hot spots' on the photoelectrode surface, we could overcome gas bubble coalescence and mass transfer limitations and demonstrated efficient hydrogen production at high current densities in reduced gravitation. Here, the experimental details are described for the preparations of these nanostructured devices and further on, the procedure for their testing in microgravity environment, realized at the Bremen Drop Tower during 9.3 s of free fall.

Although fluorescence and lifetimes of nanoscale emitters can be manipulated by plasmonic materials, it is harder to control polarization due to strict requirements on emitter environments. An ability to engineer 3D nanoarchitectures with nanoscale precision is needed for controlled polarization of nanoscale emitters. Here, we show that prescribed 3D heterocluster architectures with polarized emission can be successfully assembled from nanoscale fluorescent emitters and metallic nanoparticles using DNA-based self-assembly methods. An octahedral DNA origami frame serves as a programmable scaffold for heterogeneous nanoparticle assembly into prescribed clusters. Internal space and external connections of the frames are programmed to coordinate spherical quantum dots (QDs) and gold nanoparticles (AuNPs) into heterocluster architectures through site-specific DNA encodings. We demonstrate and characterize assembly of these architectures using in situ and ex situ structural methods. These cluster nanodevices exhibit polarized light emission with a plasmon-induced dipole along the QD-AuNP nanocluster axis, as observed by single-cluster optical probing. Moreover, emittance properties can be tuned via cluster design. Through a systematic study, we analyzed and established the correlation between cluster architecture, cluster orientation, and polarized emission at a single-emitter level. Excellent correspondence between the optical behavior of these clusters and theoretical predictions was observed. This approach provides the basis for rational creation of single-emitter 3D nanodevices with controllable polarization output using a highly customizable DNA assembly platform.

Rapid advances in image sensor technology have generated a mismatch between the small size of image sensor pixels and the achievable filter spectral resolution. This mismatch has prevented the real...

We report the optical properties obtained through experiments, simulation, and theory of ultrathin (<0.1λ), amorphous Si nanopillar arrays embedded in a thin film of SiO_2 designed for narrowband filtering for multi- and hyperspectral imaging in the near-infrared. The fabricated nanopillar arrays are square-packed with subwavelength periodicity, heights of ∼100 nm, and a radius-to-spacing ratio, r/a, of ∼0.2. Specular reflection measurements at normal incidence demonstrate that these arrays behave as narrow stopband filters in the near-infrared (λ = 1300–1700 nm) and attain ∼90% reflectivity in band and a full width at half-maximum as low as 20 nm. Using a combination of full-wave simulations and theory, we demonstrate that these narrowband filtering properties arise from efficient grating coupling of light into guided modes of the array because the nanopillar arrays serve as photonic crystal slabs. This phenomenon is known as a guided mode resonance. We discover that the spectral location of these resonances is passively tunable by modifying array geometry and is most sensitive to nanopillar spacing. Theoretical photonic crystal slab band diagrams accurately predict the spectral locations of the observed resonance and provide physical insights into and support the guided mode resonance formulation. This work demonstrates that these ultrathin all-dielectric nanopillar arrays have advantages over existing hyperspectral filter designs because they are polarization independent, do not suffer from material absorption loss, and have significant implications for minimizing imaging device size.

We describe the fabrication and use of arrays of TiO2 nanocones to yield high optical transmission into semiconductor photoelectrodes covered with high surface loadings of light-absorbing electrocatalysts. Covering over 50% of the surface of a light absorber with an array of high-refractive-index TiO2 nanocones imparted antireflective behavior (<5% reflectance) to the surface and allowed >85% transmission of broadband light to the underlying Si, even when thick metal contacts or opaque catalyst coatings were deposited on areas of the light-facing surface that were not directly beneath a nanocone. Three-dimensional full-field electromagnetic simulations for the 400-1100 nm spectral range showed that incident broadband illumination couples to multiple waveguide modes in the TiO2 nanocones, reducing interactions of the light with the metal layer. A proof-of-concept experimental demonstration of light-driven water oxidation was performed using a p+n-Si photoanode decorated with an array of TiO2 nanocones additionally having a Ni catalyst layer electrodeposited onto the areas of the p+n-Si surface left uncovered by the TiO2 nanocones. This photoanode produced a light-limited photocurrent density of ∼28 mA cm-2 under 100 mW cm-2 of simulated air mass 1.5 illumination, equivalent to the photocurrent density expected for a bare planar Si surface even though 54% of the front surface of the Si was covered by an ∼70 nm thick Ni metal layer.

Sub-wavelength arrays have garnered significant interest for many potential optoelectronics applications. We fabricated sub-wavelength silicon nanopillar arrays with a ratio of radius, r and a center-to-center distance, a, of r/a approximate to 0.2 that were fully embedded in SiO2 for narrow stopband filters that are compact and straightforward to fabricate compared to conventional Bragg stack reflectors. These arrays are well-suited for hyperspectral filtering applications in the infrared. They are ultra-thin (< 0.1 lambda), polarization-independent, and attain greater efficiencies enabled by low loss compared to plasmonic-based designs. The choice of Si as the nanopillar material stems from its low cost, high index of refraction, and a band gap of 1.1 eV near the edge of the visible. These arrays exhibit narrow near-unity reflectivity resonances that arise from coupling of an incident wave into a leaky waveguide mode via a grating vector that is subsequently reradiated, also known as guided mode resonances (GMRs). Simulations reveal reflectivities of >99% with full width at half maxima (FWHM) of approximate to 0.01 mu m. We demonstrate a fabrication route for obtaining nanopillar arrays that exhibit these GMRs. We experimentally observed a GMR with an amplitude of similar to 0.8 for filter arrays fabricated on silicon on insulator (SOI) substrates, combined with Fabry-Perot interference that stems from the underlying silicon layer.

Photoelectrochemical (PEC) cells offer the possibility of carbon-neutral solar fuel production through artificial photosynthesis. The pursued design involves technologically advanced III-V semiconductor absorbers coupled via an interfacial film to an electrocatalyst layer. These systems have been prepared by in situ surface transformations in electrochemical environments. High activity nanostructured electrocatalysts are required for an efficiently operating cell, optimized in their optical and electrical properties. We demonstrate that shadow nanosphere lithography (SNL) is an auspicious tool to systematically create three-dimensional electrocatalyst nanostructures on the semiconductor photoelectrode through controlling their morphology and optical properties. First results are demonstrated by means of the photoelectrochemical production of hydrogen on p-type InP photocathodes where hitherto applied photoelectrodeposition and SNL-deposited Rh electrocatalysts are compared based on their J-V and spectroscopic behavior. We show that smaller polystyrene particle masks achieve higher defect nanostructures of rhodium on the photoelectrode which leads to a higher catalytic activity and larger short circuit currents. Structural analyses including HRSEM and the analysis of the photoelectrode surface composition by using photoelectron spectroscopy support and complement the photoelectrochemical observations. The optical performance is further compared to theoretical models of the nanostructured photoelectrodes on light scattering and propagation.

Long-term space missions require extra-terrestrial production of storable, renewable energy. Hydrogen is ascribed a crucial role for transportation, electrical power and oxygen generation. We demonstrate in a series of drop tower experiments that efficient direct hydrogen production can be realized photoelectrochemically in microgravity environment, providing an alternative route to existing life support technologies for space travel. The photoelectrochemical cell consists of an integrated catalyst-functionalized semiconductor system that generates hydrogen with current densities >15 mA/cm(2) in the absence of buoyancy. Conditions are described adverting the resulting formation of ion transport blocking froth layers on the photoelectrodes. The current limiting factors were overcome by controlling the micro-and nanotopography of the Rh electrocatalyst using shadow nanosphere lithography. The behaviour of the applied system in terrestrial and microgravity environment is simulated using a kinetic transport model. Differences observed for varied catalyst topography are elucidated, enabling future photoelectrode designs for use in reduced gravity environments.

Metasurfaces offer significant potential to control far-field light propagation through the engineering of the amplitude, polarization, and phase at an interface. We report here the phase modulation of an electronically reconfigurable metasurface and demonstrate its utility for mid-infrared beam steering. Using a gate-tunable graphene-gold resonator geometry, we demonstrate highly tunable reflected phase at multiple wavelengths and show up to 237° phase modulation range at an operating wavelength of 8.50 μm. We observe a smooth monotonic modulation of phase with applied voltage from 0° to 206° at a wavelength of 8.70 μm. Based on these experimental data, we demonstrate with antenna array calculations an average beam steering efficiency of 23% for reflected light for angles up to 30° for this range of phases, confirming the suitability of this geometry for reconfigurable mid-infrared beam steering devices. By incorporating all nonidealities of the device into the antenna array calculations including absorption losses which could be mitigated, 1% absolute efficiency is achievable up to 30°.