Inverse Design Approach Research Articles

Deep-learning-based isogeometric inverse design for tetra-chiral auxetics

Auxetic materials with the counter-intuitive effect of negative Poisson’s ratio (NPR) have potentials for diverse applications. Typical shape optimization designs of auxetic structures involve complicated sensitivity analysis and a time-consuming iterative process, which is not beneficial for designing functionally-graded structures where the auxetics at different locations need to be inversely designed. To improve the efficiency of the inverse design and simplify the sensitivity analysis, we propose a deep-learning-based inverse shape design approach for tetra-chiral auxetics. First, a non-uniform rational basis spline (NURBS)-based parameterization of tetra-chiral structures is developed to create design samples and computational homogenization based on isogeometric analysis is used in these samples to generate a database consisting of mechanical properties and geometric parameters. Then, the database is utilized to train deep neural networks (DNN) to generate a surrogate model that represents the effective mechanical properties as a function of geometric parameters. Finally, the surrogate model is directly used in the inverse design framework where sensitivity analysis can be calculated analytically. Numerical examples with verifications are presented to demonstrate the efficiency and accuracy of the proposed design methodology.

A deep learning approach for inverse design of the metasurface for dual-polarized waves

Compared to the conventional metasurface design, machine learning-based methods have recently created an inspiring platform for an inverse realization of the metasurfaces. Here, we have used the Deep Neural Network (DNN) for the generation of desired output unit cell structures in an ultra-wide working frequency band for both TE and TM polarized waves. To automatically generate metasurfaces in a wide range of working frequencies from 4 to 45 GHz, we deliberately design an 8 ring-shaped pattern in such a way that the unit-cells generated in the dataset can produce single or multiple notches in the desired working frequency band. Compared to the general approach, whereby the final metasurface structure may be formed by any randomly distributed "0" and "1", we propose here a restricted output structure. By restricting the output, the number of calculations will be reduced and the learning speed will be increased. Moreover, we have shown that the accuracy of the network reaches 91\%. Obtaining the final unit cell directly without any time-consuming optimization algorithms for both TE and TM polarized waves, and high average accuracy, promises an effective strategy for the metasurface design; thus, the designer is required only to focus on the design goal.

Ultra-coherent nanomechanical resonators based on inverse design

Engineered micro- and nanomechanical resonators with ultra-low dissipation constitute a promising platform for various quantum technologies and foundational research. Traditionally, the improvement of the resonator’s performance through nanomechanical structural engineering has been driven by human intuition and insight. Such an approach is inefficient and leaves aside a plethora of unexplored mechanical designs that potentially achieve better performance. Here, we use a computer-aided inverse design approach known as topology optimization to structurally design mechanical resonators with optimized performance of the fundamental mechanical mode. Using the outcomes of this approach, we fabricate and characterize ultra-coherent nanomechanical resonators with, to the best of our knowledge, record-high Q ⋅ f products for their fundamental mode (where Q is the quality factor and f is the frequency). The proposed approach - which can also be used to improve phononic crystals and coupled-mode resonators - opens up a new paradigm for designing ultra-coherent micro- and nanomechanical resonators, enabling e.g. novel experiments in fundamental physics and extreme sensing.

Deep learning based multistage method for inverse design of supercritical airfoil

In the preliminary aerodynamic design phase of transonic wing, inverse design of supercritical airfoils plays an important role in acquiring practical results. However, traditional inverse design approach in transonic regime highly dependents on the manual design experience and physical model simulation, which is time consuming and poses a challenge for its applications in engineering problems. In order to relieve these drawbacks, a novel method is proposed for inverse design of supercritical airfoils based on Wasserstein generative adversarial network (WGAN), genetic algorithm (GA) and deep convolutional neural network (DCNN). This method consists of three stages. First, design characteristics of transonic pressure coefficient (CP) distribution in physics are well captured and imitated by WGAN. Then, GA is used to select task-oriented CP via adjusting controllable latent variables in generator of WGAN. Next, physical-existing CP selected by GA is recovered to corresponding airfoil by DCNN with high efficiency and accuracy. Cases using the dataset sampled based on RAE2822 airfoil are carried out to show effectiveness of submodules involved. Finally, design performance of proposed multistage method is further verified by specific design goals. Test cases indicate that selected CP by GA according to the aerodynamic design criteria can be accurately reflected on inverse designed airfoils, which are confirmed by well agreements between high-fidelity simulation result and selected CP. The proposed method is a promising design approach that can quickly achieve optimal airfoils without significant loss of accuracy.

Conceptual Understanding through Efficient Automated Design of Quantum Optical Experiments

One crucial question within artificial intelligence research is how this technology can be used to discover new scientific concepts and ideas. We present Theseus, an explainable AI algorithm that can contribute to science at a conceptual level. This work entails four significant contributions. (i) We introduce an interpretable representation of quantum optical experiments amenable to algorithmic use. (ii) We develop an inverse-design approach for new quantum experiments, which is orders of magnitudes faster than the best previous methods. (iii) We solve several crucial open questions in quantum optics, which is expected to advance photonic technology. Finally, and most importantly, (iv) the interpretable representation and drastic speedup produce solutions that a human scientist can interpret outright to discover new scientific concepts. We anticipate that Theseus will become an essential tool in quantum optics and photonic hardware, with potential applicability to other quantum physical disciplines.

Inverse design couplers for the excitation of odd plasmonic pairs in thin semiconducting films

A set of grating couplers that converts plane waves into odd pairs of plasmons supported by extremely thin layers of doped indium tin oxide is designed. The inverse design approach is implemented to improve the efficiency of ordinary periodical couplers significantly. The optimization code based on the gradient descent method and direct Maxwell solver is designed in Matlab. Three models with different optimization levels are compared and discussed. The results of calculations are prepared for subsequent experimental verification. Considered odd plasmonic pairs represent an example of high quality modes tunable using a field effect. They are potentially applicable for the implementation in various active electro-optical devices and lead the way to fully semiconductor based plasmonics.

Neural evolution structure generation: High entropy alloys.

We propose a neural evolution structure (NES) generation methodology combining artificial neural networks and evolutionary algorithms to generate high entropy alloy structures. Our inverse design approach is based on pair distribution functions and atomic properties and allows one to train a model on smaller unit cells and then generate a larger cell. With a speed-up factor of ∼1000 with respect to the special quasi-random structures (SQSs), the NESs dramatically reduce computational costs and time, making possible the generation of very large structures (over 40 000 atoms) in few hours. Additionally, unlike the SQSs, the same model can be used to generate multiple structures with the same fractional composition.

Active-Learning-Based Generative Design for the Discovery of Wide-Band-Gap Materials

Active learning has been increasingly applied to screening functional materials from existing materials databases with desired properties. However, the number of known materials deposited in the popular materials databases such as ICSD and Materials Project is extremely limited and consists of just a tiny portion of the vast chemical design space. Herein we present an active generative inverse design method that combines active learning with a deep variational autoencoder neural network and a generative adversarial deep neural network model to discover new materials with a target property in the whole chemical design space. The application of this method has allowed us to discover new thermodynamically stable materials with high band gap (SrYF$_5$) and semiconductors with specified band gap ranges (SrClF$_3$, CaClF$_5$, YCl$_3$, SrC$_2$F$_3$, AlSCl, As$_2$O$_3$), all of which are verified by the first principle DFT calculations. Our experiments show that while active learning itself may sample chemically infeasible candidates, these samples help to train effective screening models for filtering out materials with desired properties from the hypothetical materials created by the generative model. The experiments show the effectiveness of our active generative inverse design approach.

Inverse Design of Magnetic Fields Using Deep Learning

We here present a novel deep learning (DL) approach for designing structures of permanent magnets. The challenge for the DL method in this kind of problem is to learn the mapping from a desired magnetic field to a simple magnetic structure, i.e., an inverse design approach. We demonstrate this approach by training six different standard convolutional neural network (CNN) structures previously used in the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) to inversely predict the properties of a single hard magnet (magnetization, size, and location) from a given 2-D magnetic field. We show that the best network, ResNeXt-50, can perform this prediction with an error of 0.22% in the properties of the magnet.

Inverse-design of non-Hermitian potentials for on-demand asymmetric reflectivity.

We propose a genetic algorithm-assisted inverse design approach to achieve 'on- demand' light transport in periodic and non-periodic planar structures containing dielectric and gain-loss layers. The optimization algorithm efficiently produces non-Hermitian potentials from any arbitrarily given real (or imaginary) permittivity distribution for the desired frequency selective and broadband asymmetric reflectivity. Indeed, we show that the asymmetric response is directly related to the area occupied by the obtained permittivity distribution in the complex plane. In particular, unidirectional light reflection can be designed in such a way that it switches from left to right (or vice versa) depending on the operating frequency. Moreover, such controllable unidirectional reflectivity is realized using a stack of dielectric layers while keeping the refractive index and gain-loss within realistic values. We believe this proposal will benefit the integrated photonics with frequency selective one-way communication.

Inverse Design for Silicon Photonics: From Iterative Optimization Algorithms to Deep Neural Networks

Silicon photonics is a low-cost and versatile platform for various applications. For design of silicon photonic devices, the light-material interaction within its complex subwavelength geometry is difficult to investigate analytically and therefore numerical simulations are majorly adopted. To make the design process more time-efficient and to improve the device performance to its physical limits, various methods have been proposed over the past few years to manipulate the geometries of silicon platform for specific applications. In this review paper, we summarize the design methodologies for silicon photonics including iterative optimization algorithms and deep neural networks. In case of iterative optimization methods, we discuss them in different scenarios in the sequence of increased degrees of freedom: empirical structure, QR-code like structure and irregular structure. We also review inverse design approaches assisted by deep neural networks, which generate multiple devices with similar structure much faster than iterative optimization methods and are thus suitable in situations where piles of optical components are needed. Finally, the applications of inverse design methodology in optical neural networks are also discussed. This review intends to provide the readers with the suggestion for the most suitable design methodology for a specific scenario.

Deep learning in nano-photonics: inverse design and beyond

Deep learning in the context of nano-photonics is mostly discussed in terms of its potential for inverse design of photonic devices or nano-structures. Many of the recent works on machine-learning inverse design are highly specific, and the drawbacks of the respective approaches are often not immediately clear. In this review we want therefore to provide a critical review on the capabilities of deep learning for inverse design and the progress which has been made so far. We classify the different deep-learning-based inverse design approaches at a higher level as well as by the context of their respective applications and critically discuss their strengths and weaknesses. While a significant part of the community’s attention lies on nano-photonic inverse design, deep learning has evolved as a tool for a large variety of applications. The second part of the review will focus therefore on machine learning research in nano-photonics “beyond inverse design.” This spans from physics-informed neural networks for tremendous acceleration of photonics simulations, over sparse data reconstruction, imaging and “knowledge discovery” to experimental applications.

Crystallography companion agent for high-throughput materials discovery.

The discovery of new structural and functional materials is driven by phase identification, often using X-ray diffraction (XRD). Automation has accelerated the rate of XRD measurements, greatly outpacing XRD analysis techniques that remain manual, time-consuming, error-prone and impossible to scale. With the advent of autonomous robotic scientists or self-driving laboratories, contemporary techniques prohibit the integration of XRD. Here, we describe a computer program for the autonomous characterization of XRD data, driven by artificial intelligence (AI), for the discovery of new materials. Starting from structural databases, we train an ensemble model using a physically accurate synthetic dataset, which outputs probabilistic classifications-rather than absolutes-to overcome the overconfidence in traditional neural networks. This AI agent behaves as a companion to the researcher, improving accuracy and offering substantial time savings. It is demonstrated on a diverse set of organic and inorganic materials characterization challenges. This method is directly applicable to inverse design approaches and robotic discovery systems, and can be immediately considered for other forms of characterization such as spectroscopy and the pair distribution function.

Deep neural network-based automatic metasurface design with a wide frequency range

Beyond the scope of conventional metasurface, which necessitates plenty of computational resources and time, an inverse design approach using machine learning algorithms promises an effective way for metasurface design. In this paper, benefiting from Deep Neural Network (DNN), an inverse design procedure of a metasurface in an ultra-wide working frequency band is presented in which the output unit cell structure can be directly computed by a specified design target. To reach the highest working frequency for training the DNN, we consider 8 ring-shaped patterns to generate resonant notches at a wide range of working frequencies from 4 to 45 GHz. We propose two network architectures. In one architecture, we restrict the output of the DNN, so the network can only generate the metasurface structure from the input of 8 ring-shaped patterns. This approach drastically reduces the computational time, while keeping the network’s accuracy above 91%. We show that our model based on DNN can satisfactorily generate the output metasurface structure with an average accuracy of over 90% in both network architectures. Determination of the metasurface structure directly without time-consuming optimization procedures, an ultra-wide working frequency, and high average accuracy equip an inspiring platform for engineering projects without the need for complex electromagnetic theory.

Inverse design of a single-frequency diffractive biosensor based on the reporter cleavage detection mechanism.

Vertically interrogated porous silicon (PSi) interferometric biosensors have shown high potential for sensing bio-molecules as they combine high detection sensitivity with simplicity of fabrication, functionalization, optical coupling, and interfacing with microfluidic systems. However, most interferometric sensor designs require either broadband or wavelength-tunable light sources as well as wide-angle detection schemes, increasing their complexity and cost for point-of-care biosensing applications. The limit of detection of such sensors is also constrained by the small size and low refractive index of biological molecules, making it hard to detect very low concentrations of pathogens. In this work, we use a large-scale computational "inverse design" technique to demonstrate a single-frequency, fixed-angle PSi-based biosensor, which exploits a recently developed high-contrast reporter cleavage detection (HCCD) technique. The HCCD sensors detect high-index reporter cleavage events instead of low-index target analyte capture events as typical for traditional label-free optical biosensors. We use the inverse design approach to discover an optimal configuration of a PSi biosensor that makes use of the extended achievable range of cleavage-induced PSi effective index variations and can be interrogated at a single frequency and at a fixed angle. The optimized design in the form of a one-dimensional PSi grating exhibits the change in the reflectance up to 55 % at the interrogation angle of 12∘ and wavelength of 600 nm, which is caused by cleavage of Au nanoparticle reporters initially occupying 2% of the sensor surface area. The maximum possible change in reflectance is predicted to be 222 % (for a two-dimensional freeform design not amenable to fabrication). This demonstration may pave the way for developing new or redesigned conventional interferometric and colorimetric point-of-care biosensor systems in combination with the cleavage-based detection schemes.

Inverse design of photonic meta-structure for beam collimation in on-chip sensing

Designed or patterned structured surfaces, metasurfaces, enable the miniaturization of complex arrangements of optical elements on a plane. Most of the existing literature focuses on miniaturizing the optical detection; little attention is directed to on-chip optical excitation. In this work, we design a metasurface to create a planar integrated photonic source beam collimator for use in on-chip optofluidic sensing applications. We use an iterative inverse design approach in order to optimize the metasurface to achieve a target performance using gradient descent method. We then fabricate beam collimators and experimentally compare performance characteristics with conventional uniform binary grating-based photonic beam diffractors. The optimal design enhances the illumination power by a factor of 5. The reinforced beam is more uniform with 3 dB beam spot increased almost ~ 3 times for the same device footprint area. The design approach will be useful in on-chip applications of fluorescence imaging, Raman, and IR spectroscopy and will enable better multiplexing of light sources for high throughput biosensing.

Broadband vectorial ultrathin optics with experimental efficiency up to 99% in the visible region via universal approximators

Integrating conventional optics into compact nanostructured surfaces is the goal of flat optics. Despite the enormous progress in this technology, there are still critical challenges for real-world applications due to the limited operational efficiency in the visible region, on average lower than 60%, which originates from absorption losses in wavelength-thick (≈ 500 nm) structures. Another issue is the realization of on-demand optical components for controlling vectorial light at visible frequencies simultaneously in both reflection and transmission and with a predetermined wavefront shape. In this work, we developed an inverse design approach that allows the realization of highly efficient (up to 99%) ultrathin (down to 50 nm thick) optics for vectorial light control with broadband input–output responses in the visible and near-IR regions with a desired wavefront shape. The approach leverages suitably engineered semiconductor nanostructures, which behave as a neural network that can approximate a user-defined input–output function. Near-unity performance results from the ultrathin nature of these surfaces, which reduces absorption losses to near-negligible values. Experimentally, we discuss polarizing beam splitters, comparing their performance with the best results obtained from both direct and inverse design techniques, and new flat-optics components represented by dichroic mirrors and the basic unit of a flat-optics display that creates full colours by using only two subpixels, overcoming the limitations of conventional LCD/OLED technologies that require three subpixels for each composite colour. Our devices can be manufactured with a complementary metal-oxide-semiconductor (CMOS)-compatible process, making them scalable for mass production at low cost.

Analysis of Deep Neural Network Models for Inverse Design of Silicon Photonic Grating Coupler

Deep neural networks (DNNs) have been introduced to achieve the rapid design of photonic devices by creating a nonlinear function mapping the geometric structure to the optical response. By building the DNN with a finite-difference time-domain (FDTD) solver, we have demonstrated that both forward and inverse design approaches can be used to design efficiently a silicon photonic grating coupler-one of the fundamental silicon photonic devices with a wavelength-sensitive optical response, respectively. A systematic study on the model parameters including number of hidden layers, number of nodes in each layer, initial learning rate, size of training batches, number of evolution epochs, and dataset size/distribution has been carried out to analyze the relationship between the DNNs and the performances of inverse-designed devices. The study shows that the forward design approach based on an optimal forward-modeling network can achieve a peak coupling efficiency with a prediction accuracy as high as 91.7% for the coupler. And the inverse design approach based on an optimal inverse-prediction network can obtain target optical response spectrum as well as provide possibility to get an alternative design for the device. This work is helpful for the designers to improve the machine learning methods and expedite the design progress towards the creation of novel silicon photonic devices.

Optimization of Resonant PZT MEMS Mirrors by Inverse Design and Electrode Segmentation

This paper presents a systematic approach to transduction optimization of resonant piezo-based MEMS devices. The method is based on inverse design approach using direct piezoelectric effect and exploits mechanical action of desired mode to calculate the polarization map of the piezoelectric ceramics. Optimal transduction is achieved by matching the excited electric field distribution to the inversely retrieved polarization map using the electrode segmentation and topology optimization technique. The proposed method was applied to improve the design of existing state-of-the-art piezoelectric MEMS mirrors. The optimized devices feature 70% increase in the optical scan angle ( $\theta _{\mathrm {opt}}$ ) as well as 47 % reduction in power consumption compared to original design. At the same optical scan angle, the optimized devices required 2.28-fold less driving voltage amplitude. [2020-0342]

Evaluating gantry crane-way pavement performance: An inverse approach

Gantry-crane pavements and foundations are significant assets within intermodal facilities. The performance of these pavements, which are subjected to highly-variable loads, is critical to safe operations. Traffic interruptions and costs associated with maintaining and rehabilitating distressed or failed pavements in associated areas are of particular importance. The purpose of this study was to evaluate structural behavior and improve design procedures for gantry crane way pavement used at intermodal facilities by assessing interactions among pavements, subgrades, and operational loading conditions. The performance of the gantry crane pavements and foundations was assessed using a finite-element (FE) model, while pavement structural response to a crane load was measured using strain gages installed in the field. Measuring material properties using in-situ/laboratory tests was not possible in this work due to the limited resources, restricted access to intermodal facilities, and tight schedule of rail freight transport(i.e. tight schedule of cranes). To verify the materials properties and ensure a consistent behavior of the developed FE model with respect to the field measurements, an inverse design approach was implemented. Because of the significant cost of FE simulation, the gradient-based solver used in this study is based on a trust region algorithm. The verified model was used to predict the critical responses of Portland cement concrete (PCC) layer, base course, and subgrade soil. These parameters were then used to conduct a pavement fatigue damage analysis, parametric analyses of material strength and slab geometry were carried out based on Model Code, and resulting fatigue-life recommendations for improved designs were made. The developed FE model along with the inverse approach create a framework that can be used in further health monitoring problems when a specific parameter cannot be directly observed.