Inverse Design Approach Research Articles

Design, tuning, and blackbox optimization of laser systems

Chirped pulse amplification (CPA) and subsequent nonlinear optical (NLO) systems constitute the backbone of myriad advancements in semiconductor manufacturing, communications, biology, defense, and beyond. Accurately and efficiently modeling CPA+NLO-based laser systems is challenging because of the complex coupled processes and diverse simulation frameworks. Our modular start-to-end model unlocks the potential for exciting new optimization and inverse design approaches reliant on data-driven machine learning methods, providing a means to create tailored CPA+NLO systems unattainable with current models. To demonstrate this new, to our knowledge, technical capability, we present a study on the LCLS-II photo-injector laser, representative of a high-power and spectro-temporally non-trivial CPA+NLO system.

A machine learning-based method for co-design and optimization of microwave-absorbing/load-bearing multifunctional structures

Simultaneously considering the absorption performance and load-bearing capability is a trend in the design of multifunctional structures. Nevertheless, the collaborative design and optimization involved in this process present a challenging problem. Herein, guided by multifunctionality, a lightweight microwave-absorbing/load-bearing multifunctional structure is intelligently inversely designed based on machine learning. A co-design scheme is developed to address the contradiction between the absorption performance and load-bearing performance. An approach for rapid inverse design of metamaterial absorbers containing multilayered frequency-selective surfaces is proposed. The simulation results obtained using multi-objective optimization based on surrogate models indicate that the optimized multifunctional structure achieves more than 90% absorption in the frequency range of 2.5 GHz–18.0 GHz and simultaneously exhibits superior load-bearing performance with an out-of-plane Young’s modulus of 334.8 MPa and an out-of-plane compressive strength of 4.95 MPa, demonstrating the effectiveness of the co-design scheme. Finally, the experimental results are analysed. This study provides a reference for co-design and multi-objective optimization of similar multifunctional structures.

Optimizing Polarization Selective Unidirectional Photoluminescence from Phased‐Array Metasurfaces

AbstractMetasurface‐based optical elements offer a wide design space for miniature and lightweight optical applications. Typically, metasurface optical elements transform an incident light beam into a desired output waveform. Recent demonstrations of light‐emitting metasurfaces highlight the potential for directly producing desired output waveforms via metasurface‐mediated spontaneous emission. In this work, reciprocal finite‐difference time‐domain (FDTD) simulations and machine learning are used to enable the inverse design of highly unidirectional photoluminescent III‐Nitride quantum well metasurfaces capable of directive p‐, s‐, or combined p‐ and s‐ polarized emission at arbitrary angles. In comparison with previous intuition‐guided designs using the same quantum well architectures, the inverse design approach enables new polarization capabilities and experimentally demonstrated improvements in directivity of 54%. An analysis of ways in which the inverse design both validates and contradicts previous intuition‐guided design heuristics is presented. Ultimately, the combination of reciprocal simulations and efficient global optimization (EGO) grants remarkable improvements in emission directivity and results in full control over the polarization and momentum of emitted light, including simultaneous directional emission of s‐ and p‐polarized light.

Deep-learning-based inverse design of colloidal quantum dots

Because optical properties of colloidal quantum dots (QDs) greatly depend on their layer thicknesses at the sub-nanometer scale, accurately predicting optical properties based on these structural parameters can help streamline the development process. However, most of traditional inverse design approaches require time-consuming numerical simulations and optimization processes. In this simulation-only paper, we propose a deep-learning-based inverse design model for colloidal InP/ZnSe/ZnS QDs based on the structure parameters of layer thicknesses and optical properties of the emission and absorption spectrum. Four different deep neural network (DNN) structures are implemented and compared. A tandem DNN structure with the simultaneous use of both emission and absorption spectra as input data shows the best accuracy due to its ability to alleviate the non-uniqueness problem in the inverse design of colloidal QDs. Our inverse design model enables the prediction of layer thicknesses for QDs that exhibit on-demand optical emission/absorption spectra, which are particularly enhanced at the specific target wavelength region. Moreover, the weights and biases of the well-trained tandem DNN for InP/ZnSe/ZnS QDs can be transferred and tuned to implement a transfer-learning-assisted inverse design model for CdSe/ZnSe/ZnS QDs, where the amount of training data to achieve the average prediction accuracy of 99.3 % is reduced by a maximum of 71.4 %.

Broadband and large-aperture metasurface edge encoders for incoherent infrared radiation.

The prevalence of computer vision systems necessitates hardware-based approaches to relieve the high computational demand of deep neural networks in resource-limited applications. One solution would be to off-load low-level image feature extraction, such as edge detection, from the digital network to the analog imaging system. To that end, this work demonstrates incoherent, broadband, low-noise optical edge detection of real-world scenes by combining the wavefront shaping of a 24-mm aperture metasurface with a refractive lens. An inverse design approach is used to optimize the metasurface for Laplacian-based edge detection across the 7.5- to 13.5-μm LWIR imaging band, allowing for facile integration with uncooled microbolometer-based LWIR imagers to encode edge information. A polarization multiplexed approach leveraging a birefringent metasurface is also demonstrated as a single-aperture implementation. This work could be applied to improve computer vision capabilities of resource-constrained systems by leveraging optical preprocessing to alleviate the computational requirements for high-accuracy image segmentation and classification.

Novel machine learning insights into the QM7b and QM9 quantum mechanics datasets.

This paper (i) explores the internal structure of two quantum mechanics datasets (QM7b, QM9), composed of several thousands of organic molecules and described in terms of electronic properties, and (ii) further explores an inverse design approach to molecular design consisting of using machine learning methods to approximate the atomic composition of molecules, using QM9 data. Understanding the structure and characteristics of this kind of data is important when predicting the atomic composition from physical-chemical properties in inverse molecular designs. Intrinsic dimension analysis, clustering, and outlier detection methods were used in the study. They revealed that for both datasets the intrinsic dimensionality is several times smaller than the descriptive dimensions. The QM7b data is composed of well-defined clusters related to atomic composition. The QM9 data consists of an outer region predominantly composed of outliers, and an inner, core region that concentrates clustered inliner objects. A significant relationship exists between the number of atoms in the molecule and its outlier/inliner nature. The spatial structure exhibits a relationship with molecular weight. Despite the structural differences between the two datasets, the predictability of variables of interest for inverse molecular design is high. This is exemplified by models estimating the number of atoms of the molecule from both the original properties and from lower dimensional embedding spaces. In the generative approach the input is given by a set of desired properties of the molecule and the output is an approximation of the atomic composition in terms of its constituent chemical elements. This could serve as the starting region for further search in the huge space determined by the set of possible chemical compounds. The quantum mechanic's dataset QM9 is used in the study, composed of 133,885 small organic molecules and 19 electronic properties. Different multi-target regression approaches were considered for predicting the atomic composition from the properties, including feature engineering techniques in an auto-machine learning framework. High-quality models were found that predict the atomic composition of the molecules from their electronic properties, as well as from a subset of only 52.6% size. Feature selection worked better than feature generation. The results validate the generative approach to inverse molecular design.

Validation and characterization of algorithms and software for photonics inverse design

In this work, we present a reproducible suite of test problems for large-scale optimization (“inverse design” and “topology optimization”) in photonics, where the prevalence of irregular, non-intuitive geometries can otherwise make it challenging to be confident that new algorithms and software are functioning as claimed. We include test problems that exercise a wide array of physical and mathematical features—far-field metalenses, 2d and 3d mode converters, resonant emission and focusing, and dispersion/eigenvalue engineering—and introduce an a posteriori lengthscale metric for comparing designs produced by disparate algorithms. For each problem, we incorporate cross-checks against multiple independent software packages and algorithms, and reproducible designs and their validations scripts are included. We believe that this suite should make it much easier to develop, validate, and gain trust in future inverse-design approaches and software.

Optical filters made from random metasurfaces using Bayesian optimization

Abstract We theoretically investigate the ability to design optical filters from a single material and a single layer of randomly dispersed resonant dielectric particles, defining a random metasurface. Using a Bayesian and generalized Mie inverse-design approach, we design particle radii distributions that give rise to longpass, shortpass, bandpass, and bandstop spectral bands in the infrared. The optical response is shown to be directly related to electric and magnetic multipole scattering of the constituent particles and their near field coupling. We discuss the effect of the particle size distribution and particle–particle coupling interactions on filter design in random systems lacking long-range order.

Perfect Excitation of Topological States by Supersymmetric Waveguides.

Topological photonic states provide intriguing strategies for robust light manipulations, however, it remains challenging to perfectly excite these topological eigenstates due to their complicated mode profiles. In this work, we propose to realize the exact eigenmode of the topological edge states by supersymmetric (SUSY) structures. By adiabatically transforming the SUSY partner to its main topological structure, the edge modes can be perfectly excited with simple single-site input. We experimentally verify our strategy in integrated silicon waveguides in telecommunication wavelength, showing a broad working bandwidth. Moreover, a shortcut-to-adiabaticity strategy is further applied to speed up the adiabatic pump process by inverse-design approaches, thus enabling fast mode evolutions and leading to reduced device size. Our method is universal and beneficial to the topology-based or complex eigenmodes systems, ranging from photonics and microwaves to cold atoms and acoustics.

A Neural Network-Empowered Inverse FSS Design and Synthesis Approach for 5G Shielding Applications

A Neural Network-Empowered Inverse FSS Design and Synthesis Approach for 5G Shielding Applications

Efficient inverse design of optical multilayer nano-thin films using neural network principles: backpropagation and gradient descent.

Optical multilayer thin films have a wide range of applications due to their ability to manipulate transmissive or reflective wavelengths by adjusting the thickness of composed layers, enabling diverse uses. Although their light weight, flexible nature and ease of fabrication position them as promising components for future devices, determining their optimal layer thickness for the desired functionality demands extensive simulations, leading to inefficient utilization of computational resources and time. To overcome these challenges, inverse design methods, leveraging machine learning and deep learning, are being explored. However, these methods necessitate learning processes, despite the presence of well-established formulas that elucidate these phenomena. Furthermore, deriving accurate answers for conditions not included in the learning process proves to be challenging. This paper introduces an innovative inverse design approach that utilizes the backpropagation of a networked transfer matrix, effectively explaining the characteristics of optical multilayer thin films. By exploiting the chain rule of the network, this method calculates gradients to discern how each layer thickness influences the outcomes. Consequently, the optimal thickness is determined without the need for an additional learning process. Mathematical elucidation of the operational principle of this approach is precisely described. Optimization of computing resource utilization through network configuration reduces the calculation time compared to conventional methods. The efficacy of this method is demonstrated through its application in the inverse design of transmissive and reflective films, verifying its potential for enhancing efficiency and accuracy in optical multilayer thin-film design and manufacturing processes.

Inverse design of skull osteoinductive implants with multi-level pore structures through machine learning.

How to accurately design a personalized matching implant that can induce skull regeneration is the focus of current research. However, the design space for the porous structure of implants is extensive, and the mapping relationships between these structures and their mechanical and osteogenic properties are complex. At present, the forward design of skull implants mainly relies on expert experience, leading to high cost and a lengthy process, while the existing inverse design approaches face challenges due to data dependence and manufacturing process errors. This study presents an efficient inverse design method for personalized multilevel structures of skull implants using a machine learning pipeline composed of a finite element method, topological optimization, and neural networks. Based on the mechanical response of the human body falls, this method can tailor multi-level structures for implants in various defect positions. The results show that the proposed method establishes a bidirectional relationship between topological parameters and mechanical properties, enabling the customization of mechanical behavior at low computational cost while accounting for manufacturing errors in the 3D printing process. Additionally, the design results are also mutually consistent with analytical relationships between lattice parameters and the elastic modulus obtained from experiments and finite element simulations. Thus, this study provides a general and practical approach to rapidly design skull osteoinductive implants.

Structure prediction from spectra amidst dynamical heterogeneity in melanin.

Melanin is a biopolymer pigment that plays a central role in skin photoprotection. Its extensive chemical and dynamical heterogeneity imparts this property through a broad featureless ultraviolet/visible absorption spectrum. Conventionally, the rational design of synthetic photoprotective pigments revolves around establishing the structure-spectra correlation and developing biomimetic materials with desired optical properties. This approach fails to explain the mechanistic details of melanin's absorption spectrum because it arises from an ensemble of structures rather than a local minimum on the potential energy surface. Here, we propose an inverse design approach to elucidate the contributions of dominant chromophoric units in various wavelength domains of the melanin spectrum.

Single-step-etched ultra-compact metamaterial grating coupler enabled by a hierarchical inverse design approach

Single-step-etched ultra-compact metamaterial grating coupler enabled by a hierarchical inverse design approach

Recent advances in metamaterial integrated photonics

Since the invention of the silicon subwavelength grating waveguide in 2006, subwavelength metamaterial engineering has become an essential design tool in silicon photonics. Employing well-established nanometer-scale semiconductor manufacturing techniques to create metamaterials in optical waveguides has allowed unprecedented control of the flow of light in photonic chips. This is achieved through fine-tuning of fundamental optical properties such as modal confinement, effective index, dispersion, and anisotropy, directly by lithographic imprinting of a specific subwavelength grating structure onto a nanophotonic waveguide. In parallel, low-loss mode propagation is readily obtained over a broad spectral range since the subwavelength periodicity effectively avoids losses due to spurious resonances and bandgap effects. In this review we present recent advances achieved in the surging field of metamaterial integrated photonics. After briefly introducing the fundamental concepts governing the propagation of light in periodic waveguides via Floquet–Bloch modes, we review progress in the main application areas of subwavelength nanostructures in silicon photonics, presenting the most representative devices. We specifically focus on off-chip coupling interfaces, polarization management and anisotropy engineering, spectral filtering and wavelength multiplexing, evanescent field biochemical sensing, mid-infrared photonics, and nonlinear waveguide optics and optomechanics. We also introduce a nascent research area of resonant integrated photonics leveraging Mie resonances in dielectrics for on-chip guiding of optical waves, with the first Huygens’ metawaveguide recently demonstrated. Finally, we provide a brief overview of inverse design approaches and machine-learning algorithms for on-chip optical metamaterials. In our conclusions, we summarize the key developments while highlighting the challenges and future prospects.

Modulated high-order Hermite-Gaussian beams with uniform intensity distribution

Hermite-Gaussian beams have been studied since the early years of laser modes. With the recent development of structured light, different modulation methods have been demonstrated to control the energy, polarization and propagation behavior of Hermite-Gaussian beams. In this work we introduce an inverse design approach to modulate the energy distribution among different lobes of Hermite-Gaussian beams. In this approach, the desired intensity profiles of Hermite-Gaussian beams are the starting point. The required spatial phase masks are calculated through inverse Fourier transform, then projected on a spatial light modulator to convert conventional Gaussian beams into modulated Hermite-Gaussian beams. The experimental modulated high-order (m = 10, n = 10) Hermite-Gaussian beams show much improved energy distribution among different lobes over standard Hermite-Gaussian beams. These modulated Hermite-Gaussian beams also have improved beam quality factor M2 after modulation. These results demonstrated the feasibility to fine-tune the intensity profiles of Hermite-Gaussian beams, which have applications in structured illumination, particle manipulation and optical communications.

Application of Inverse Design Approaches to the Discovery of Nonlinear Optical Switches.

Molecular switches, in which a stimulus induces a large and reversible change in molecular properties, are of significant interest in the domain of photonics. Due to their commutable redox states with distinct nonlinear optical (NLO) properties, hexaphyrins have emerged as a novel platform for multistate switches in nanoelectronics. In this study, we employ an inverse design algorithm to find functionalized 26R→28R redox switches with maximal βHRS contrast. We focus on the role of core modifications, since a synergistic effect with meso-substitutions was recently found for the 30R-based switch. In contrast to these findings, the inverse design optima and subsequent database analysis of 26R-based switches confirm that core modifications are generally not favored when high NLO contrasts are targeted. Moreover, while push-pull combinations enhance the NLO contrast for both redox switches, they prefer a different arrangement in terms of electron-donating and electron-withdrawing functional groups. Finally, we aim at designing a three-state 26R→28R→ 30R switch with a similar NLO response for both ON states. Even though our best-performing three-state switch follows the design rules of the 30R-based component, our chemical compound space plots show that well-performing three-state switches can be found in regions shared by high-responsive 26R and 30R structures.

Multichannel Quantum Emission with On-Chip Emitter-Coupled Holographic Metasurfaces.

Multichannel quantum emission is in high demand for advanced quantum photonic applications such as quantum communications, quantum computing, and quantum cryptography. However, to date, the most common way for shaping photon emission from quantum emitters (QEs) is to utilize free-standing (external) bulky optical components. Here, we develop the multichannel holography approach for flexibly designing on-chip QE-coupled metasurfaces that make use of nonradiatively QE-excited surface plasmon polaritons for generating far-field quantum emission, which propagates in designed directions carrying specific spin and orbital angular momenta (SAM and OAM, respectively). We further design, fabricate, and characterize on-chip quantum light sources of multichannel quantum emission encoded with different SAMs and OAMs. The holography-based inverse design approach developed and demonstrated on-chip quantum light sources with multiple degrees of freedoms, thereby enabling a powerful platform for quantum nanophotonics, especially relevant for advanced quantum photonic applications, e.g., high-dimensional quantum information processing.

A Continuous Action Space Tree search for INverse desiGn (CASTING) framework for materials discovery

Material properties share an intrinsic relationship with their structural attributes, making inverse design approaches crucial for discovering new materials with desired functionalities. Reinforcement Learning (RL) approaches are emerging as powerful inverse design tools, often functioning in discrete action spaces. This constrains their application in materials design problems, which involve continuous search spaces. Here, we introduce an RL-based framework CASTING (Continuous Action Space Tree Search for inverse design), that employs a decision tree-based Monte Carlo Tree Search (MCTS) algorithm with continuous space adaptation through modified policies and sampling. Using representative examples like Silver (Ag) for metals, Carbon (C) for covalent systems, and multicomponent systems such as graphane, boron nitride, and complex correlated oxides, we showcase its accuracy, convergence speed, and scalability in materials discovery and design. Furthermore, with the inverse design of super-hard Carbon phases, we demonstrate CASTING’s utility in discovering metastable phases tailored to user-defined target properties and preferences.

Rapid inverse design of metamaterials based on prescribed mechanical behavior through machine learning

Designing and printing metamaterials with customizable architectures enables the realization of unprecedented mechanical behaviors that transcend those of their constituent materials. These behaviors are recorded in the form of response curves, with stress-strain curves describing their quasi-static footprint. However, existing inverse design approaches are yet matured to capture the full desired behaviors due to challenges stemmed from multiple design objectives, nonlinear behavior, and process-dependent manufacturing errors. Here, we report a rapid inverse design methodology, leveraging generative machine learning and desktop additive manufacturing, which enables the creation of nearly all possible uniaxial compressive stress‒strain curve cases while accounting for process-dependent errors from printing. Results show that mechanical behavior with full tailorability can be achieved with nearly 90% fidelity between target and experimentally measured results. Our approach represents a starting point to inverse design materials that meet prescribed yet complex behaviors and potentially bypasses iterative design-manufacturing cycles.