Advancing materials science through next-generation machine learning.

For over a decade, machine learning (ML) models have been making strides in computer vision and natural language processing (NLP), demonstrating high proficiency in specialized tasks. The emergence of large-scale language and generative image models, such as ChatGPT and Stable Diffusion, has significantly broadened the accessibility and application scope of these technologies. Traditional predictive models are typically constrained to mapping input data to numerical values or predefined categories, limiting their usefulness beyond their designated tasks. In contrast, contemporary models employ representation learning and generative modeling, enabling them to extract and encode key insights from a wide variety of data sources and decode them to create novel responses for desired goals. They can interpret queries phrased in natural language to deduce the intended output. In parallel, the application of ML techniques in materials science has advanced considerably, particularly in areas like inverse design, material prediction, and atomic modeling. Despite these advancements, the current models are overly specialized, hindering their potential to supplant established industrial processes. Materials science, therefore, necessitates the creation of a comprehensive, versatile model capable of interpreting human-readable inputs, intuiting a wide range of possible search directions, and delivering precise solutions. To realize such a model, the field must adopt cutting-edge representation, generative, and foundation model techniques tailored to materials science. A pivotal component in this endeavor is the establishment of an extensive, centralized dataset encompassing a broad spectrum of research topics. This dataset could be assembled by crowdsourcing global research contributions and developing models to extract data from existing literature and represent them in a homogenous format. A massive dataset can be used to train a central model that learns the underlying physics of the target areas, which can then be connected to a variety of specialized downstream tasks. Ultimately, the envisioned model would empower users to intuitively pose queries for a wide array of desired outcomes. It would facilitate the search for existing data that closely matches the sought-after solutions and leverage its understanding of physics and material-behavior relationships to innovate new solutions when pre-existing ones fall short.

Challenges in nanofabrication for efficient optical metasurfaces.

Optical metasurfaces have raised immense expectations as cheaper and lighter alternatives to bulk optical components. In recent years, novel components combining multiple optical functions have been proposed pushing further the level of requirement on the manufacturing precision of these objects. In this work, we study in details the influence of the most common fabrication errors on the optical response of a metasurface and quantitatively assess the tolerance to fabrication errors based on extensive numerical simulations. We illustrate these results with the design, fabrication and characterization of a silicon nanoresonator-based metasurface that operates as a beam deflector in the near-infrared range.

Deep Learning Meets Nanophotonics: A Generalized Accurate Predictor for Near Fields and Far Fields of Arbitrary 3D Nanostructures.

Deep artificial neural networks are powerful tools with many possible applications in nanophotonics. Here, we demonstrate how a deep neural network can be used as a fast, general purpose predictor of the full near-field and far-field response of plasmonic and dielectric nanostructures. A trained neural network is shown to infer the internal fields of arbitrary three-dimensional nanostructures many orders of magnitude faster compared to conventional numerical simulations. Secondary physical quantities are derived from the deep learning predictions and faithfully reproduce a wide variety of physical effects without requiring specific training. We discuss the strengths and limitations of the neural network approach using a number of model studies of single particles and their near-field interactions. Our approach paves the way for fast, yet universal, methods for design and analysis of nanophotonic systems.

Design of plasmonic directional antennas via evolutionary optimization.

We demonstrate inverse design of plasmonic nanoantennas for directional light scattering. Our method is based on a combination of full-field electrodynamical simulations via the Green dyadic method and evolutionary optimization (EO). Without any initial bias, we find that the geometries reproducibly found by EO work on the same principles as radio-frequency antennas. We demonstrate the versatility of our approach by designing various directional optical antennas for different scattering problems. EO-based nanoantenna design has tremendous potential for a multitude of applications like nano-scale information routing and processing or single-molecule spectroscopy. Furthermore, EO can help to derive general design rules and to identify inherent physical limitations for photonic nanoparticles and metasurfaces.

Deep learning enabled real time speckle recognition and hyperspectral imaging using a multimode fiber array.

We demonstrate the use of deep learning for fast spectral deconstruction of speckle patterns. The artificial neural network can be effectively trained using numerically constructed multispectral datasets taken from a measured spectral transmission matrix. Optimized neural networks trained on these datasets achieve reliable reconstruction of both discrete and continuous spectra from a monochromatic camera image. Deep learning is compared to analytical inversion methods as well as to a compressive sensing algorithm and shows favourable characteristics both in the oversampling and in the sparse undersampling (compressive) regimes. The deep learning approach offers significant advantages in robustness to drift or noise and in reconstruction speed. In a proof-of-principle demonstrator we achieve real time recovery of hyperspectral information using a multi-core, multi-mode fiber array as a random scattering medium.

Enhancement of electric and magnetic dipole transition of rare-earth-doped thin films tailored by high-index dielectric nanostructures.

We propose a simple experimental technique to separately map the emission from electric and magnetic dipole transitions close to single dielectric nanostructures, using a few-nanometer thin film of rare-earth-ion-doped clusters. Rare-earth ions provide electric and magnetic dipole transitions of similar magnitude. By recording the photoluminescence from the deposited layer excited by a focused laser beam, we are able to simultaneously map the electric and magnetic emission enhancement on individual nanostructures. In spite of being a diffraction-limited far-field method with a spatial resolution of a few hundred nanometers, our approach appeals by its simplicity and high signal-to-noise ratio. We demonstrate our technique at the example of single silicon nanorods and dimers, in which we find a significant separation of electric and magnetic near-field contributions. Our method paves the way towards the efficient and rapid characterization of the electric and magnetic optical response of complex photonic nanostructures.

Pushing the limits of optical information storage using deep learning.

Diffraction drastically limits the bit density in optical data storage. To increase the storage density, alternative strategies involving supplementary recording dimensions and robust readout schemes must be explored. Here, we propose to encode multiple bits of information in the geometry of subwavelength dielectric nanostructures. A crucial problem in high-density information storage concepts is the robustness of the information readout with respect to fabrication errors and experimental noise. Using a machine-learning-based approach in which the scattering spectra are analysed by an artificial neural network, we achieve quasi-error-free readout of sequences of up to 9 bits, encoded in top-down fabricated silicon nanostructures. We demonstrate that probing few wavelengths instead of the entire spectrum is sufficient for robust information retrieval and that the readout can be further simplified, exploiting the RGB values from microscopy images. Our work paves the way towards high-density optical information storage using planar silicon nanostructures, compatible with mass-production-ready complementary metal-oxide-semiconductor technology.

Polarization conversion in plasmonic nanoantennas for metasurfaces using structural asymmetry and mode hybridization.

Polarization control using single plasmonic nanoantennas is of interest for subwavelength optical components in nano-optical circuits and metasurfaces. Here, we investigate the role of two mechanisms for polarization conversion by plasmonic antennas: Structural asymmetry and plasmon hybridization through strong coupling. As a model system we investigate L-shaped antennas consisting of two orthogonal nanorods which lengths and coupling strength can be independently controlled. An analytical model based on field susceptibilities is developed to extract key parameters and to address the influence of antenna morphology and excitation wavelength on polarization conversion efficiency and scattering intensities. Optical spectroscopy experiments performed on individual antennas, further supported by electrodynamical simulations based on the Green Dyadic Method, confirm the trends extracted from the analytical model. Mode hybridization and structural asymmetry allow address-ing different input polarizations and wavelengths, providing additional degrees of freedom for agile polarization conversion in nanophotonic devices.

Evolutionary multi-objective optimization of colour pixels based on dielectric nanoantennas.

The rational design of photonic nanostructures consists of anticipating their optical response from systematic variations of simple models. This strategy, however, has limited success when multiple objectives are simultaneously targeted, because it requires demanding computational schemes. To this end, evolutionary algorithms can drive the morphology of a nano-object towards an optimum through several cycles of selection, mutation and cross-over, mimicking the process of natural selection. Here, we present a numerical technique that can allow the design of photonic nanostructures with optical properties optimized along several arbitrary objectives. In particular, we combine evolutionary multi-objective algorithms with frequency-domain electrodynamical simulations to optimize the design of colour pixels based on silicon nanostructures that resonate at two user-defined, polarization-dependent wavelengths. The scattering spectra of optimized pixels fabricated by electron-beam lithography show excellent agreement with the targeted objectives. The method is self-adaptive to arbitrary constraints and therefore particularly apt for the design of complex structures within predefined technological limits.