High-Throughput Quantum-Mechanics computation of material properties by ab-initio methods has become the foundation of an effective approach to materials design, discovery and characterization. One of the major challenges in mapping the materials genome is in developing efficient computational tools that are cost-effective and accurate at the same time. In this talk, we will discuss the recently developed Agapito-Curtarolo-Boungiorno-Nardelli (ACBN0) pseudo-hybrid Hubbard density functional where the Hubbard energy within the DFT + U formulation is calculated self consistently by direct evaluation of the local Coulomb integrals. The U depends on the electron density and depends both on the geometry and chemical environment of the system. We show that ACBN0 improves the description of both the structural and electronic properties in a range of complex materials from the Zn and Cd based chalcogenides to the transition metal oxides. The magnetic moments and the magnetic ordering energies are better described compared to the LDA/GGA functionals. We will also discuss the application of the ACBN0 approach to surfaces, doped and multi-valent systems where it is possible to evaluate U for different sites and different chemical bonding. For all the complex materials studied here, we #64257;nd that the electronic properties are signi#64257;cantly improved over the DFT values and the accuracy is at par with the HSE values at a fraction of the computational cost.

Machine learning (ML) methods were used to solve the equations for two basic models systems of quantum many-body condensed-matter physics: the Anderson impurity model and dynamical mean-field theory (DMFT) for the Hubbard model. The key technical issues are defining the functional mapping of an input function to an output function, and distinguishing metallic from Mott insulating solutions. The resulting formalism can be used either as an impurity solver or to replace the entire DMFT self-consistency loop. In addition, it is general enough to be applied to other functions learning problems. The validity of the machine learning scheme is assessed by comparing predictions of full correlation functions, of quasi-particle weight and particle density all compared to values directly obtained. We also discuss how ML can be used as a reverse engineering tool for correlated materials. The results indicate that with modest further development, machine learning approach may be an attractive computational efficient option for real materials predictions for strongly correlated systems. Our scheme being very general, complications arising from Hunds coupling, multi-bands etc present in real materials should fit into the formalism.

Growth and phase transformations of metastable polymorphs at nano-scale, especially ones that are very close in energy are of immense scientific interest. Growths of metastable compounds at nano-scale are studied using high-resolution transmission electron microscope (HRTEM) to elucidate local crystal structure and atomic arrangement. Density functional theory based modeling can help explain observed crystal structures and growth mechanism.
In this work we integrate first-principles electronic structure modeling and high-resolution transmission electron microscopy to explain the growth of metastable zincblende AlN in Al/AlN/TiN trilayer and its transformation to wurtzite AlN under uniaxial stress. An unusual growth mechanism of metastable zincblende AlN thin film by diffusion of nitrogen atoms into Al lattice is established. Using first-principles density functional theory, we studied the possibility of thermodynamic stability of AlN as zincblende phase due to epitaxial strains and interface effect, which fails to explain the formation of zincblende AlN. We then compared the formation energetics of rocksalt and zincblende AlN in fcc Al through direct diffusion of nitrogen atoms to Al octahedral and tetrahedral interstitials. The formation of zincblende AlN thin film is determined to be a kinetically driven process than a thermodynamically driven process.

The electron band structure serves as a primary indicator of the suitability of a material for a given application, e.g., as an electronic or catalytic material. Computed band structures, however, are subject to a host of approximations, some of which are more obvious (e.g., the treatment of the exchange-correlation of self-energy) and others less obvious (e.g., the treatment of core, semicore, or valence electrons, handling of relativistic effects, or the accuracy of the underlying basis set used). One important contribution is the handling of spin-orbit coupling, an increasingly important material design target. Another is the employed approximation to exchange-correlation itself; here, hybrid density functionals show improved performance for indicators such as band gaps compared to earlier, cheaper semi-local functionals. However, reliable benchmark data for spin-orbit coupled band structures, especially for hybrid functionals, is relatively sparse. We here provide a set of accurate Kohn-Sham band structure benchmarks, using the numeric atom-centered all-electron electronic structure code FHI-aims combined with the “traditional” PBE functional and the hybrid HSE functional, to calculate the occupied and low-lying unoccupied bands of a set of 2D and 3D materials, validated against WIEN2k for PBE. For perturbatively applied spin-orbit coupling, we highlight the importance of a qualitatively sound band structure as a starting point, the functional dependence, and the reach.

The Materials Genome Initiative aims to accelerate the process of materials discovery and reduce the time to commercialization of advanced materials. In the field of catalysis the development and optimization of new catalyst formulations involve the systematic investigation of a multi-dimensional synthesis-structure-property space in plug-flow reactors mimicking those used in industry. Over the past ten years high-throughput methodologies have been increasingly used to search for new catalyst formulations and more recently interactions between model catalyst systems, theory, and data science have begun creating integrated accelerated materials innovation workflows. In the current presentation I will highlight recent work from our group in the development of novel catalysts for the formation and reformation of hydrocarbon feedstocks. The first part of the talk will focus on breaking complex hydrocarbons down into low carbon (C₁-C₄) fuels that can be used as fuel for solid oxide fuel. In this portion of the work theory, primary screening on model systems, and powder screening studies were combined to optimize multiple figures of merit resulting in the identification of a catalyst formulation that exceeded nominal specifications in less than 9 months. The second portion of the presentation will discuss recent work in our lab centered around the use of design of experiments to systematically screen the role of catalyst morphology on activity. Particular emphasis will be placed on a recent “hit” which resulted in the formation of catalysts that could spontaneously reduce during the production of hydrocarbons from CO and CO₂.

The High Throughput Experimentation (HTE) project of the Joint Center for Artificial Photosynthesis (JCAP, http://solarfuelshub.org/) performs accelerated discovery of new earth-abundant photoabsorbers and electrocatalysts. The materials discovery framework integrates high throughput experimentation with materials theory and machine learning, and an overview of the research program will be provided through specific examples of material discoveries. Light absorbers with excellent band energetics for solar fuels applications, for example Mn₂V₂O₇, have been discovered through a combination of detailed electronic structure calculations and high throughput synthesis and screening experiments. In this effort, the portfolio of theoretical and experimental techniques has been designed to provide a comprehensive understanding of the light absorber materials through establishing composition-structure-property relationships. In the area of electrocatalysis, where the relationship between bulk electronic structure and catalytic activity is rather tenuous, different modes of experiment-computation integration are employed. Particular attention will be given to the development of a genetic programming algorithm for the automated identification of composition-property trends in high order composition spaces. The algorithm produces clusters of catalyst compositions with unique composition-activity trends and has identified a composition region whose unique performance has been corroborated though the experimental identification of a multi-phase nanostructure.
This material is based upon work performed by the Joint Center for Artificial Photosynthesis, a DOE Energy Innovation Hub, supported through the Office of Science of the U.S. Department of Energy under Award Number DE-SC0004993.

A major challenge in the development of novel catalytic materials is the quick and effective exploration of the material parameter space. This process can be expedited by synthesizing, characterizing, and testing multiple materials in parallel using high-throughput experimentation. However, the efficacy of high-throughput experimentation is limited by our ability to screen and optimize materials in an intelligent manner. In particular, variability in the outcome of a nanoparticle synthesis arises from a large number of inextricable factors, ranging from the type of synthesis method used to the particular heating regimen and set of chemical constituents. While research has been somewhat successful in identifying factor-response relationships using trial and error efforts, the complexities of the full design space are not yet well understood. To this end, statistical design of experiments was used to elucidate the role of different design factors in one pot metal salt reduction, a widely used nanoparticle synthesis method. Using cobalt as a case study, the full spectrum of possible numeric and categorical design factors were screened systematically using factorial design of experiments. The as-synthesized particles were characterized for their crystal structure, particle size, and size distribution using X-ray Diffraction (XRD) and Transmission Electron Microscopy (TEM). Minitab Statistical Software was used to discriminate between significant and insignificant synthesis outcomes, facilitating the development of a reproducible response surface. Following oxidation of the nanoparticles to Co3O4, the unsupported cobalt oxides were tested for their CO oxidation activity in a 16-channel fixed bed reactor with a parallel infrared imaging system. Subsequently, catalytic turn over frequency (TOF) was calculated to build quantitative structure-activity relationships in order to direct further design space exploration.

Formulating low-dimensional descriptions of complex, dynamical systems is an area of major scientific and technological interest. These low-dimensional descriptions are used to develop computationally tractable models of complex systems and, thus, enable engineering analyses and precise control strategies over these systems.One application for such a reduced model is in the descriptionof self and directed assembly of clusters of colloidal particles. Of particular interest is the assembly of clusters of colloidal particles into highly ordered crystalline structures.Diffusion mapping is a machine learning technique that can provide dynamically rigorous, low-dimensional descriptions for such complex systems. This technique has been applied successfully to describe a broad range of physical phenomena from protein folding to chemical reaction kinetics.
In this presentation, weuse diffusion mapping to generate low-dimensional descriptions of the directed assembly of a quasi-2D collection of colloidal particles under the action of an externally applied electric field.The dynamic data set for implementation of diffusion mapping for this system has been generated by Brownian-dynamics simulations of the colloidal assembly process.This material dynamical system exhibits complex behavior, including order-to-disorder transitions, as well as formation of various crystalline defects in ordered assemblies of colloidal particles. This variety in the structure and physical behavior of the colloidal assembly presents a unique challenge in the application of diffusion mapping, as the implementation of this machine learning technique requires a distance metric that is able to consistently distinguish between different configurations characterized by this broad range of structural features. This is captured successfully by a novel local-environment-basedmetric for this system, which is superior to other standard distance metrics considered in describing structural diversity. The application of this coarse-grained description of directed colloidal assembly, properly obtained through diffusion mapping, allows for a deeper understanding of the physical processes and governing thermodynamics and kinetics in these colloidal particle systems.

Determining the full three-dimensional (3D) atomistic structure of metastable nanoparticles is critical to understanding and controlling their properties. Here we develop an integrated genetic algorithm (GA) [1] optimization tool called StructOpt [2] that can determine the 3D structure of a nanoparticle by matching forward modeling to experimental scanning transmission electron microscopy (STEM) [3,4] data and simultaneously minimizing the energy. This tool integrates the power of GAs for complex optimization and utilizes both available experimental and energetics data simultaneously. We validate the accuracy of finding the target structure and promising orientation on stable and metastable 309-atom Au nanoclusters, demonstrating excellent agreement with model STEM data. We then use the tool to determine the 3D structure of an experimentally observed ~6000 atom metastable Au nanoparticle [5]. Our tool will enable practical prediction of 3D chemical and topological structure of nanoscale materials from intrinsically limited experimental data, and will provide critical, experimentally validated starting structures for simulation study of materials&’ properties. The StructOpt code developed in this study is available under open source licensing as part of the MAterials Simulation Toolkit (MAST) and can be obtained from https://pypi.python.org/pypi/MAST.
[1] Johnston, R. L.; Evolving Better Nanoparticles: Genetic Algorithms for Optimising Cluster Geometries. Dalton Trans.2003, 4193.
[2] Kaczmarowski, A.; Yang, S.; Szlufarska, I.; Morgan, D. Genetic Algorithm Optimization of Defect Clusters in Crystalline Materials. Comput. Mater. Sci.2015, 98, 234.
[3] Kirkland, E. J. Advanced Computing in Electron Microscopy. (Plenum Press, New York/London) (1998).
[4] Yankovich, A. B.; Berkels, B.; Dahmen, W.; Binev, P.; Sanchez, S. I.; Bradley, S. A.; Li, A.; Szlufarska, I.; Voyles, P. M. Picometre-Precision Analysis of Scanning Transmission Electron Microscopy Images of Platinum Nanocatalysts. Nat. Commun.2014, 5, 4155.
[5] Yankovich, A. B.; Berkels, B.; Dahmen, W.; Binev, P.; Voyles, P. M. High-Precision Scanning Transmission Electron Microscopy at Coarse Pixel Sampling for Reduced Electron Dose. Adv. Struct. Chem. Imaging2015, 1, 2.

Material scientists have often focused on the properties of previously reported compounds, but neglect numerous unreported but chemically plausible compounds that could have interesting properties. For example, the 18-valence electron ABX family of compounds features examples of topological insulators, thermoelectrics and piezoelectrics, but only 83 out of 483 of these possible compounds have been made. Using first-principles thermodynamics we examined the theoretical stability of the 400 unreported members and predict that 54 should be stable. Of those previously unreported ‘missing&’ materials now predicted to be stable, 15 were grown in this study; X-ray studies agreed with the predicted crystal structure in all 15 cases. Among the predicted and characterized properties of the missing compounds are potential transparent conductors, thermoelectric materials and topological semimetals. This integrated process—prediction of functionality in unreported compounds followed by laboratory synthesis and characterization—could be a route to the systematic discovery of hitherto missing, realizable functional materials .See R. Gautier, X. Zhang, L. Hu, L. Yu, Y. Lin, T. O. L. Sunde, D. Chon, K. R. Poeppelmeier, A. Zunger Nature Chemistry 7, 308-316 (2015). Functionality-directed theoretical search discovered unusual transparent conductors in this group see F. Yan, X. Zhang, Y. Yu, L. Yu, A. Nagaraja, T.O. Mason, and Alex Zunger , Nature Communication 6, 7308 (2015). Focusing on the above noted ABX we search theoretically for electronic structures that simultaneously lead to optical transparency while accommodating intrinsic defect structures that produce uncompensated free holes. This leads to the prediction of a stable, never before synthesized TaIrGe compound made of all-metal heavy atom compound. Laboratory synthesis then found it to be stable in the predicted crystal structure and p-type transparent conductor remarkably high hole mobility of 2,730 cm2 Vminus;1 sminus;1 at room temperature. This methodology opens the way to future searches of transparent conductors in unexpected chemical groups.

The Materials Genome Initiative has the objective to accelerate materials discovery and development through computational modeling and data-driven new approaches. The Materials Project (www.materialsproject.org) and other initiatives have demonstrated the large-scale delivery of predicted materials properties. In the Materials Project experimental data is used not only for benchmarking and calibration of computational methods, but also to broaden their impact into areas where computational predictions are still difficult. I will show examples involving aqueous stability of materials, and in assessing the accuracy of density functional theory methods for predicting the structure of materials.
Through high-throughput computing centers such as the Materials Project, one can envision a future where many of the properties of compounds are known well before they are created, making it likely that materials synthesis will become the more limiting factor. I will discuss some recent approaches showing how data-driven science, combined with computing can be used to predict how both stable and metastable materials can be made.

Our research group have hitherto developed the high-throughput preparation and evaluation system “M-ist Combi” for powder, film and liquid library. Basic concept of the preparation process in the “M-ist Combi” system is the electrostatic spray deposition (ESD) method which is one of solution process. And, the applied voltage to starting materials solution is contributed to library geometry. In addition, the “M-ist Combi” system can easily exchange from the ESD apparatus to various type of evaluation probe. Therefore, efficient materials exploration can be performed by using a small-sized accessory device together in limited time and space. In candidate cathode materials exploration, we found a number of knowledge about avantage effect of multielement substitution using the “M-ist Combi” system.
In case of layered rock salt type Li(Ni,Co)O₂, Ti element substitution to the transition metal site attributed to miniaturization of crystallite size. And, Li(Ni,Co)_0.9Ti_0.1O₂ library showed better cycle performance than the other layered rock salt type Li(Ni,Co,Ti)O₂.
Recently, we have also established reaction phase diagrams of pseudo-ternary LiFePO₄-LiMnPO₄-LiCoPO₄ for finding newly olivine-type compounds using the “M-ist Combi” system because a part of LiFe_1-x-yMn_xCo_yPO₄ have also studied in order to improve the energy density. All powder library heat-treated at 500-7000C for 2 hours under argon (97%)/hydrogen (3%) atmosphere indexed as olivine-type structures (space group: Pnma). However, a part of Cobalt-rich compounds included Co₂P as by-product. On the other hands, we found that LiFe_1-xMn_xCo_0.1PO₄ library showed higher discharge capacity than the other powder library.
From these results, multielement substitution to well-known materials is promising improvement of function.

We theoretically explored new two-dimensional materials near the ionic instability (three-dimensional structures are favored), with covalent bonded systems (graphene) sitting at the opposite end of the spectrum. Accordingly, monolayer alkaline earth and transition metal halides, many of their bulk forms being layered structures, were investigated with density functional calculations. We thus predicted a new class of two-dimensional materials by performing structure relaxations, cohesive/formation energies and full phonon dispersion calculations. These materials exhibit strong ionic bonding characters, as revealed by significant charge transfers. The superior charge donating/accepting abilities and the large specific area make these new materials promising for adsorption and catalytic reactions. We demonstrated adsorptions and diffusions of Li on these materials, which are relevant for Li ion battery electrodes and hydrogen storage. Also the new materials with varied charge donating abilities and their nanostructures can enhance and tune catalytic reactions, such as Ziegler-Natta catalysts. Moreover, they exhibit diverse electronic properties that can be of great application interests, ranging from insulators, metals, and even spin-polarized semiconductors.

Combined approaches in materials simulations and materials informatics are now increasingly becoming popular for use in the rapid evaluation of various material properties for vast number of materials. These techniques could offer an indispensable tool in the battery field to aid in the search of, for example, high-performance solid electrolytes, as only a limited number of potential materials are still known up to date and with most of them still falling short in terms of Li diffusivity. New solid electrolytes are indeed still continually sought in the hope of realizing the next major breakthrough in the field of all-solid-state lithium ion batteries.
Recently, olivine-type (LiFePO₄) and tavorite-type (LiFeSO₄F) materials have been determined as battery cathodes with high Li ion insertion rates. Taking this inspiration, we evaluated the two structure types for solid electrolyte use by using Li ion migration energy (ME) as the screening criterion. We surveyed the olivine LiMXO₄ (M³⁺minus;X⁴⁺ and M²⁺minus;X⁵⁺) and tavorite LiMTO₄F (M³⁺minus;T⁵⁺ and M²⁺minus;T⁶⁺) compositions, where the ionically-substituted M cation is a nontransition metal and X/T cations are typical tetrahedral cations. In order to efficiently predict ME, we built a machine learning model of it based from descriptors extracted from DFT-optimized structure data. We determined a number of compositions with enormously low ME values (le; 0.30 eV), especially for the olivine-type compounds. Based from the generated ME model, we found out that with ionic substitutions at M and X/T sites, competing effects among Li pathway bottleneck size, polyanion covalency, and local lattice distortion are crucial for controlling ME.

Harvesting and storing solar energy via photoisomerization is a prospective usage of photoresponsive materials originally proposed almost forty years ago. However, design of an efficient and cost-effective photoresponsive material for this application, known as solar thermal fuels (STF), has been a great challenge for decades. Recently, we have developed a high-throughput ab initio simulations approach to tackle this challenge¹. Using our approach, we have not only discovered materials as high-energy density STF candidates, but also physical principles for STF materials design. One of the high performance candidate materials predicted from our high-throughput simulations has been synthesized and characterized in our group. Building on this method and the materials database we obtained from this work, we develop an adaptive high-throughput ab initio simulation method, which integrates prior knowledge into the design of materials. The capability of our adaptive high-throughput method is demonstrated with the new high performance STF candidates predicted.
1. Y. Liu and J. C. Grossman, Nano Lett. 14 (12), 7046-7050 (2014).

Rational, data-driven materials discovery would be an immense boon for research and development, making these efforts far faster and cheaper. In such a paradigm, computer models trained to find patterns in massive chemical datasets would rapidly scan compositions and systematically identify attractive candidates for technological applications, such as new batteries, lightweight alloys, solar cells, and so on. Here, we present several examples of our work on developing machine learning (ML) methods capable of creating predictive models using a diverse range of materials data. As input training data, we demonstrate ML on both large computational datasets of density functional theory calculations, as implemented in the Open Quantum Materials Database (oqmd.org), and also experimental databases of materials properties. We construct ML models using a large and chemically diverse list of attributes, which we demonstrate are suitable for describing a range of properties, and a novel method for partitioning the data set into groups of similar materials in order to boost the predictive accuracy of the ML models. Our results demonstrate how ML can be used as an effective tool to automatically learn intuitive design rules, predict diverse properties of crystalline and amorphous materials, such as formation energy, specific volume, band gap energy, and glass-forming ability, and accelerate combinatorial searches.

Solar energy harvesting is an attractive field due to the vast abundance of solar radiation. Recently, solar thermal fuels (STFs) have seen remarkable breakthroughs where solar energy harvesting and storage are accomplished in a single photoactive molecule. The key to this recent success was the combination of computational design and experiments. Recently reported materials involving azoebenzene molecules templated onto carbon nanotubes led to energy storage densities comparable to lead batteries, with potential energy densities comparable to commercial Lithium-ion batteries¹. Here we take this a step further and make use of computation to engineer solar thermal fuels on a molecular level and experimentally synthesize and characterize the molecule. We are able to achieve unprecedentedly high energy densities for single molecules (approaching 1 eV per molecule), without the need for relatively expensive and complicated templating.
Computation designs via ab-initio simulations have been experimentally verified thus enabling a path towards new solar thermal fuel design. Using ab-initio simulations, high throughput screening was done on a large number of molecules to determine high energy density candidate solar thermal fuels by considering formation of metastable structures and corresponding isomerization enthalphies². Out of the top solar thermal fuel candidates, azobenzene functionalized with bis(trityl alcohol) was selected for experimental characterization.
The isomerization enthalphy of this solar thermal fuel candidate was calculated to be on the order of 1eV/molecule. We synthesized the molecule and characterized the heat release to be about 0.9 eV/molecule, which is in good agreement with the computational values. The key to this increase in isomerization enthalpy from 0.49eV/molecule of azobenzene is the presence of the hydroxy group and bulky phenol groups. The half-life of the molecule at room temperature is fourteen days compared to 4.2days for azobenzene.
With the use of ab-initio simulations to screen solar thermal fuel molecules, it was possible to screen out potential candidates for solar thermal fuels. Experiments showed that the screening method was effective and stable photoswitchable molecules were synthesized with high energy storage and longer storage lifetime. With the combination of computation and experiements, new classes of materials can be found by applying robust design criteria, increasing the energy stored as well as the storage lifetime.
Reference
1. Kucharski, T. J. et al.Nat. Chem.advance online publication, (2014).
2. Liu, Y. & Grossman, J. C. Nano Lett.14, 7046-7050 (2014).

With growing interest in the Materials Genome Initiative, there is a need for engineered features and representations to study materials from a data-scientific perspective. We use representation learning, in which machine learning is used to extract useful representations of data, to investigate the thermodynamic and kinetic properties of atomistic systems. Using this approach, we study silicon and silicon-lithium alloys in various phases. The learned representation is used to investigate phase transitions between the ordered phases, as well as the relaxation of the disordered phases, directly from atomistic structure. Since neither phase information nor dynamics were included in the training of the neural networks, we conclude that their resulting representations have broader applications than the specific supervised learning task they were trained for. This approach paves the way for data-driven materials design guided by deep learning.

The materials design process requires the identification of specific materials that meet a desired application or property need. The traditional routes adopted thus far to meet such design goals involve the determination of the relevant properties of a large number of potential candidates, via high-throughput experiments or computations. While powerful and successful ^[1,2], this strategy suffers from two primary drawbacks. (1) Consideration of each material in a case-by-case manner is laborious and time-consuming, especially if one were to ignore the availability of past data on the same or “similar” candidate materials. (2) Furthermore, the prevalent strategy addresses the materials design problem in an “inverted” manner, i.e., instead of approaching the “desired properties --> suitable materials” design problem, the “materials --> properties” problem is tackled, and the former design aspect is addressed indirectly through enumeration.
In the present contribution, we demonstrate the feasibility of a directed design scheme for the example of polymer dielectrics. Pathways to surmount both the above hurdles are proposed. Polymers-- quasi 1-dimensional systems composed of well-defined blocks-- offer a perfect testing ground for this demonstration. Building upon past work ^[1,3], we used crystal structure prediction and density functional theory calculations to generate a database for the relevant properties of a number of polymers generated in a combinatorial manner starting with specified building units. We then used a regression technique on this data to train property prediction models; this is performed via an intermediate ‘fingerprinting&’ step, where every polymer is converted to a series of meaningful numbers ^[2] based on its constituent combination of motifs. The robust, rapid prediction models thus obtained were then used to estimate the properties of a large number of new polymers not previously considered in generating the data, thereby significantly expanding on the number of cases in the chemical space we possess knowledge about. In order to directly address the “materials by design” problem, we adopted a genetic algorithm approach, where we start with a random population of polymers and systematically move towards ones that display desired target properties. With this development, we have the means of tackling the ‘desired properties --> suitable materials&’ aspect. The models we developed enable us to trivially determine properties of any polymers in the chemical space as well as recommend polymers suitable for specific applications for experimental synthesis and characterization. ^[4]
References
[1] V. Sharma et al., Nature Communications. 5, 4845 (2014)
[2] G. Pilania, C.C. Wang, X. Jiang, S. Rajasekaran, R. Ramprasad, Scientific Reports 3 (2013)
[3] A. F. Baldwin et al., Adv. Mater. 27, 346 (2015)
[4] A. Mannodi-Kanakkithodi et al., (manuscript under preparation)

Molecularly imprinted polymers are advanced recognition materials selectively rebinding a target molecule present during the synthesis of the polymer matrix [1]. It is commonly understood that the templating process is based on embedding the complex formed between the template and functional monomers into a co-polymer matrix while maintaining their spatial arrangement, thus resulting in a molecular imprint of the template. Template removal then ideally leads to a distribution of synthetic recognition sites within the polymer matrix that are complementary in functionality, size, and shape to the target. The template may then rebind following the lock-and-key principle similar to molecular recognition processes associated with enzymes or antibodies.
We introduce an innovative theoretical concept using fully atomistic molecular dynamics simulations for developing the first theoretical demonstration for molecular templating processes leading to virtually imprinted polymers (VIPs). A VIP for 17-β-estradiol was successfully created demonstrating selective binding via virtual chromatography experiments, thereby not only providing fundamental insight on the nature of molecularly imprinted polymers, but theoretically confirming the existence of a molecular imprinting effect.
1. L. Ye and K. Mosbach, K. Molecular imprinting: synthetic materials as substitutes for biological antibodies and receptorsdagger;. Chem. Mater.20, 859-868 (2008).
2. S. A. Piletsky, et al. Substitution of antibodies and receptors with molecularly imprinted polymers in enzyme-linked and fluorescent assays. Biosens. Bioelectron.16, 701-707 (2001).
3 S. Zink, F. A. Moura, P. A. S. Autreto, D. S. Galvao, and B. Mizaikoff, Virtually Imprinted Polymers (VIPs): Understanding Molecularly Templated Materials via Advanced Molecular Dynamics Simulations - to be published.

DNA-stabilized silver clusters (Ag-DNAs) are novel fluorophores that are finding numerous applications in nanophotonics, chemical sensing, and bioimaging. The fluorescence colors of Ag-DNAs can be tuned from blue-green into the near-infrared by selecting the sequence of the single-stranded DNA that templates the cluster. Recent work has shown that Ag-DNA color correlates to silver cluster size. However, the connection between cluster size and DNA sequence remains elusive, and the immense space of possible DNA template sequences limits researchers&’ ability to design Ag-DNAs for specific applications. By applying bioinformatics and machine learning tools to large experimental data sets, we gain understanding of how DNA selects silver cluster size and color and develop predictive design tools for multi-base templates for specific Ag-DNAs.
Using robotic pipetting, Ag-DNAs were synthesized on ~700 different 10-base DNA template strands with randomly selected sequences, and the fluorescence spectrum associated with each strand was measured. We find that certain discriminative multi-base motifs correlate to templates for brightly fluorescent clusters, and the lengths of these base motifs agree with previous Ag-DNA models that suggest elongated cluster structures. Using these motifs to parameterize DNA templates, we develop a machine learning-based tool to design DNA templates that stabilize brightly fluorescent Ag-DNAs. Our framework increases the probability of selecting DNA templates that stabilize bright Ag-DNAs by a factor of > 3, as compared to random guessing [1]. We then extend these methods to design DNA templates that select specific Ag-DNA colors. Our results not only provide much-needed design rules for researchers using Ag-DNAs but also illustrate that machine learning methods can enable new discoveries about the underlying physics and chemistry of materials.
[1] S. M. Copp, P. Bogdanov, M. Debord, A. Singh, E. Gwinn. Base Motif Recognition and Design of DNA Templates for Fluorescent Silver Clusters by Machine Learning. Adv. Mater. 2014, 26, 5839.

Biomechanical imaging techniques with acoustic radiation force (ARF) have been developed for characterization of biological soft tissue and detection of tumors. The response induced by ARF excitation is measured to estimate the mechanical properties of soft tissue. The challenge to quantitatively solve this inverse problem arises due to the unknown stress distribution and local boundary conditions in the region of excitation (ROE). In the presence of the above uncertain information, an optimization procedure that fits the parameters to a deterministic finite element (FE) model of soft tissue cannot take full advantage of the prior information. To solve the above problem, a statistical inverse FE characterization method based on Bayesian approach is undertaken in this study. The uncertain prior information that is used to build the FE model is formulated as probability distribution, and a posterior distribution of the estimated parameter is calculated with Markov Chain Monte Carlo (MCMC) method. In order to make the MCMC simulation computationally tractable, a machine learning technique, Gaussian process (GP) metamodeling, is applied as a statistical approximation of the FE model in the Bayesian estimation process. In addition, sparsity of the GP is ensured by selecting the data set that is used in the Bayesian inference with a gradient optimization. The inverse algorithm is then applied to estimate the time constant that represents the ratio of viscosity to elasticity in ARF induced creep imaging, and its performance is evaluated with synthetic data generated from FE models with different parameter configurations. Simulation results indicate that the inverse procedure is computationally efficient, and provides a comprehensive and accurate statistical interpretation of the estimated parameter in the presence of model parameter uncertainty and measurement data noise.

A bottle neck that hinders valorisation studies of nanoparticles and micro-capsules in functional materials is very often the limited capacity of research groups to synthesize sufficient amounts of nanoparticles that allow preliminary formulation or coating tests. Typical lab scale synthesis yields milligrams to grams of nanoparticle product. However, for first ‘tests&’ in coating formulations, usually up to 10 to 100 g of material is required.
Additionally, material industries are struggling with a continuous need to reduce the time to market for new or improved products. High-throughput technologies can be applied for the accelerated discovery of new compounds, the optimization of synthesis conditions up to the extraction of synthesis knowhow.
Flamac has responded to these needs with the development of unique automated parallel synthesis platforms for nanocrystals and micro-capsules. Via these platforms it is now possible to synthesize tailor made nanoparticles and micro-capsules at a reasonable scale to allow a first set of experiments to assess the potential of these building blocks in functional materials.
This presentation will highlight the successful development and use of these unique platforms in areas such as nanocrystals for printed photovoltaic applications, micro-capsules for self-healing composite materials, and many more.

The ability to design and refine materials has long accentuated the development of civilization. The ever-increasing spectrum of functionalities required for developing and optimizing materials fundamental to modern civilization requires efficient paradigms for materials discovery and design going beyond current serendipitous discoveries and classical synthesis-characterization-theory approaches. To bridge these complex issues will require integrated and direct feedback from multi-scale functional measurements to theory and must allow real-time and archival experimental data to be incorporated effectively. Although computational approaches have recently allowed screening bulk properties of materials in existing structures in the inorganic (or organic) crystal structure databases, these efforts often lack a concerted effort to understand how a particular functionality comes about in a compound (or a set of compounds) and to use this knowledge to discover new materials. In this presentation, I will delineate pathways for atomic level probing of structure-property relationships on atomic level as exemplified by single-atom electrochemical and superconductive properties. I will further discuss how the information contained in experimental data can provide input into computational methods to predict and understand new materials. Finally, text analytics and process monitoring tools can establish and integrate the knowledge base for experimental materials fabrication.
This research is supported by and performed at the Center for Nanophase Materials Sciences, which is sponsored at Oak Ridge National Laboratory by the Scientific User Facilities Division, BES DOE.

Materials discovery from an experimental point of view necessitates seamless integration of composition library fabrication with different types of testing methods. For structural materials, mechanical properties are invariably important. Nanoindentation has emerged as the most convenient and versatile technique for probing mechanical properties at the micro/nano scale. However, high throughput nanomechanical testing of thin film composition libraries is not trivial and straight forward. As a rule of thumb, indentation depths should be less than 10% of film thickness to avoid substrate elastic effects[1]. However, at very shallow depths, additional problems arise due to enhanced effects of calibration errors (e.g. tip shape function), increased noise due to the indenter pushed to its resolution limits and large scatter in experimental data. In such a scenario, statistical data analysis methodologies and informatics [2] that can handle the noise, scatter and calibration errors in nanoindentation data becomes vital. In addition to this level of complexity, often mechanical properties like yield strength, ductility, strain rate sensitivity, etc. are also a part of the initial check-list in the screening process which cannot be extracted from standard nanoindentation tests alone.
This talk will address these relevant issues in high throughput experimentation and materials discovery for structural applications by presenting the case study of bulk metallic glass compositions based on binary CuZr and ternary CuZrTi system. For metallic glasses, good modulus, high strength and reasonably attainable ductility to avoid catastrophic failure are desirable mechanical properties. Extensive nanoindentation was performed on these composition libraries to correlate the hardness and modulus with their chemical compositions, determined using X-ray fluorescence (XRF) and Energy Dispersive X-Ray spectroscopy (EDS). Large scatter and noise in nanoindentation data was handled by using dimensionality reduction techniques to select the most representative load-displacement curves for particular compositions and to improve signal-to-noise ratio of the experimental data. The most promising compositions based on these criteria (modulus, hardness) were selected. Micropillar compression of these selected compositions was carried out to assess their yield strength and compressive ductility. The issues surrounding such an endeavor and the integration of various kinds of experimental data generated using this combinatorial approach will be discussed. It is hoped that this study will provide a rational basis for high throughput experimentation technique for accelerated structural materials discovery.
[1] R. Saha, W.D. Nix, Acta materialia, 50 (2002) 23-38.
[2] K. Rajan, Materials Today, 8 (2005) 38-45.

Recently, much attention has been paid to layered perovskite oxides exhibiting noncentrosymmetry due to oxygen octahedral rotations (OORs) towards the development of new series of ferroelectrics and multiferroelectrics.¹ We have reported the OOR-induced noncentrosymmetry of A-site-ordered n=1 Ruddlesden-Popper phase NaA&’TiOshy;₄ (A&’=rare earth),² in which inversion symmetry is broken by OORs represented by (Phi;00)(0Phi;0) in Aleksandrov notation.³ The Na ions can be replaced by different monovalent cations such as H⁺, Li⁺, K⁺, and Ag⁺, possibly leading to the discovery of large piezoelectric family. One would expect that smaller A-site cations induce larger OOR in the framework of concept of tolerance factor. Here we report experimental and theoretical work showing, however, that the K substitutes also exhibit the OOR-induced noncentrosymmetry and, surprisingly, the OOR instability is enhanced with substitution of larger K ions for smaller Na ions.
Synchrotron x-ray diffraction (SXRD) and second harmonic generation (SHG) measurements for polycrystalline KA&’TiO₄ (A&’=Sm, Eu) samples revealed that KA&’TiO₄ (A&’=Sm, Eu) belongs to a noncentrosymmetric P-42₁m space group with a (Phi;00)(0Phi;0)-type OOR at room temperature as well as the Na substitutes.² The variable-temperature SXRD and SHG measurements showed that KA&’TiO₄ has higher structural phase transition temperatures relevant to the OOR than the Na substitutes, while the transition temperatures are higher for smaller rare-earth ions.² It was found from first-principles calculations that the rare-earth and alkaline ions play different roles in the OOR although both the ions occupy the same sites in a crystallographic viewpoint: the rare-earth ions strongly attract O^2- ions to optimize their oxygen coordination due to the trivalent positive charge, while the alkaline oxide layers impose in-plane strain on the other layers so as to tune the OOR instability. Our results suggest the importance of considering not only the tolerance factors but also the chemical feature of A-site cations for predicting the OOR instability, in particular, if multiple A-site cations are involved.
HA and ASG were supported by the National Science foundation DMR-1420620.
References
[1] N. A. Benedek et al, Dolton. Trans.44, 10543 (2015).
[2] H. Akamatsu et al., Phys. Rev. Lett.112, 187602 (2014).
[3] K. S. Aleksandrov, Crystallogr. Rep.40, 251 (1995).

As a rapidly growing area of materials science, high-throughput (HT) computational materials design is playing a crucial role in developing new materials. In this presentation, I will talk about our recent research work on the HT computational design of the perovskite-based two-dimensional electron gas (2DEG) systems. The 2DEG formed on the perovskite oxide heterostructure (HS) has potential applications in next-generation nanoelectronic devices. In order to achieve practical implementation of the 2DEG in the device design, desired physical properties such as high charge carrier density and mobility are necessary. Here we show that using HT electronic structure calculation methods and introducing a series of combinatorial descriptors, we have successfully identified a series of candidate 2DEG systems on the basis of perovskite oxides. This work provides another exemplar of applying HT computational screening and design strategy for the discovery of advanced functional materials.

Transparent conducting oxides (TCOs) are large band gap materials (to favor transparency) doped with electrons (n-type) or holes (p-type). TCOs are essential to many technologies from solar cell to transparent electronics and strong efforts are directed towards the discovery of new TCOs. High mobility p-type TCOs are especially challenging to develop due to the low hole effective mass and flat valence bands present in most oxides. The identification of new p-type TCOs can be tremendously accelerated through high-throughput computing. In this talk, we will report on a new double-perovskite oxide : Ba₂BiTaO₆ identified among thousands of materials candidates through high-throughput computing. Our talk will present our computational screening and how this material emerged as an exceptionally low hole effective mass, high band gap oxide. Next to the computational results, we will show experimental data on Ba₂BiTaO₆ (including thin film synthesis and characterization) confirming the computational prediction and the very high mobility of this novel perovskite. Finally, we will rationalize the chemical and structural factors leading to such a high mobility and propose avenues for the further optimization of this promising material.

We will discuss our latest efforts in actively incorporating various data driven approaches to facilitate the combinatorial strategy for accelerated exploration of different functional materials. We are constantly implementing new machine learning techniques to rapidly analyze a large number of diffraction and other spectral and image data taken on thin film composition spread samples which typically span large fractions of multinary compositional phase diagrams. The current goal in this effort is focused on achieving accuracy and integrity in mapping distribution of structural phases across phase diagrams, as well as carrying out semi-automated Rietveld analysis and peak indexing of diffraction patterns from composition spread samples. We will also discuss developing techniques to perform data mining and visualization of existing databases compiled from the literature. Such practices are useful and important in trying to unveil hidden correlations between different physical parameters. We are also actively addressing the standardization issues of data from combinatorial libraries with the ultimate goal of building databases out of reliable data from combinatorial experimentation. The materials topics of current interest are transparent conductors, superconductors, and magnetic materials. This work is carried out in collaboration with A. G. Kusne, S. Curtarolo, M. Nardelli, M. Fornari, A. Mehta, M. Green, S. Baron, and T. Chikyow. This work is funded by NIST, ONR, AFOSR, and NSF.

Thermoelastic cooling has high efficiency (~84% Carnot) and minimum environmental impact. In 2014, U.S. Department of Energy ranked it as the most promising new HVAC technology to replace vapor compression. Currently, the only viable material for thermoelastic cooling application is the NiTi based shape memory alloy. While NiTi has high latent heat (cooling power) and long fatigue life under compression, its prohibitively high cost makes it difficult for wide spread industrial and consumer applications. CuAlNi is a candidate thermoelastic cooling material. It is cost effective, and has a high efficiency (COP~11) comparable to that of the NiTi alloy. Unfortunately, CuAlNi is brittle. It cannot be deformed million times without fracture. To improve CuAlNi&’s ductility, an alloying strategy aimed at grain-refinement was development. Such strategy requires large quantity of bulk specimens covering a wide range of compositional space. Here, we present our efforts on development high throughput arc melting synthesis method to rapidly prepare, process and characterize a large number of bulk CuAlNi based samples. Our results show a small addition of Ti will refine the grain size, which leads to enhancement of ductility. The obtained CuAlNi-Ti alloys were examined for its phase transformation, latent heat and mechanical properties.

We have developed a first-of-its-kind Autonomous REsearch System (ARES) capable of designing, executing, and analyzing its own experiments autonomously. The closed loop, iterative method enables ARES to design new experiments based on prior results dynamically, after each experiment; a first for materials research.
We are applying this method to understand and control the synthesis of single wall carbon nanotubes, in this case optimizing growth rate in (7) - dimensional parameter space. We use automated in situ Raman spectroscopy characterization of growth rate for CVD synthesis of carbon nanotubes as a metric for a target objective used by our NMD-M3 artificial intelligence planner. NMD-M3 uses a random forest learning approach which models experimental results as a function of experimental inputs, and a genetic algorithm planner to propose new experiments expected to achieve the targeted growth rate.
We expect ARES to be a disruptive advance in the near future, combining advances in robotics, artificial intelligence, data sciences and in operando methods to enable us to attack high dimensional research problems that were previously intractable by current research processes. Already we are applying the ARES method to multiple problems, including Additive Manufacturing and defect engineering in graphene.

For high-throughput discovery of new materials, unsupervised and accurate calculations of surface energies are often of crucial importance. For polar materials, i.e. materials where the top and the bottom surfaces of the crystal structure are different, the commonly used approaches involve detailed case-by-case considerations to find approximation or reorganization of atomic positions to form equivalent structure with identical, known surfaces. Although accurate, such strategies are cumbersome and unsuitable when the target of screening is a large material database. In this talk we propose a new method for the calculation of interface energies with density-functional theory that elevates this problem. The basic idea is to cap any dangling bond of the bottom surface with properly chosen atoms or molecules such that the bottom surface is essentially bulk-like. We have a developed set of rules where these atoms or molecules are determined uniquely based on the structure of the periodic table. Our approach allows for the systematic treatment of any surface, is computationally inexpensive, and can easily be turned into an algorithm and implemented in any software. We demonstrate this method for a number of group IV and III-V semiconductors and show that the error is around (or below) 2% in all cases.

Understanding and controlling the Bi incorporation and defect formation are critical to the successful application of GaAsBi to IR laser diodes with improved energy efficiency. Using density functional theory (DFT) calculations, we investigated the growth mechanisms during the molecular beam epitaxy growth of GaAsBi films. We studied the surface structures, adsorption/desorption processes, surface diffusion barriers, and surface reactions relevant to the Bi species. One major result we found is that As₂ can easily exchange with a (2×1)-Bi surface both thermodynamically and kinetically, which potentially explains the low Bi incorporation even when a large number of Bi atoms are present on the surface. Based on this discovery, we proposed to increase the Bi incorporation under higher growth temperature (such as >350 °C) either by cutting off the As₂ supply for the As₂/Bi exchange process through sequential deposition method or by increasing the barrier of the exchange process through applying asymmetric in-plane strain to the substrate.
Due to the current use of low growth temperatures to incorporate Bi, GaAsBi films usually possess a large number of defects, which dramatically reduce the carrier lifetime and PL intensity [1]. Using DFT with hybrid functional to correct the band-gap errors and the MAST software to automate defect calculations [2], we studied the formation energies and energy levels of various defects in GaAsBi. Under the usual As-rich growth condition, we found that the thermodynamically dominant point defects are As_Ga, Bi_Ga, V_Ga, and Bi_As, while the dominant pair defects are As_Ga+Bi_As and V_Ga+Bi_As. The predicted defect energy levels are in good agreement with available experiments and may help explain previous deep-level transient spectroscopy measurements of GaAsBi [3]. To understand the formation of Bi-rich clusters after thermal annealing, we investigated the formation of two types of clusters, nBi_As+V_Ga and nBi_As+V_As (n = 1, 2, hellip;), extending a previous work [4], and found very strong driving force for Bi_As to concentrate around either V_As or V_Ga. Thus it seems promising to reduce the Bi-rich clusters by reducing the contents of V_As and V_Ga.
Acknowledgements: This research was supported by the NSF-funded University of Wisconsin Materials Research Science and Engineering Center (DMR-1121288) and also benefitted from the use of the Extreme Science and Engineering Discovery Environment (XSEDE) computing resources, which is supported by NSF grant OCI-1053575.
[1] V. Pacebutas, et al., J. Mater. Sci.-Mater. Electron. 20, 363 (2009); V. Pa#269;ebutas, et al., Thin Solid Films 520, 6415 (2012).
[2] T. Angsten, et al., New J. Phys. 16, 015018 (2014); T. Mayeshiba, et al., MAterials Simulation Toolkit (MAST), https://pypi.python.org/pypi/MAST/
[3] P. M. Mooney, et al., J. Appl. Phys. 113, 133708 (2013).
[4] M. P. J. Punkkinen, et al., Semicond. Sci. Technol. 29, 115007 (2014).

Point defects such as oxygen vacancies in metal oxides are extensively used to manipulate vital material properties. While methods to predict such defect formation energies have advanced significantly, high throughput calculations of defective systems remain a challenge largely due to prohibitive computational costs. We demonstrate an approach to move beyond the current high throughput calculations of perfect materials to rapid predictions of defect systems, a key for enabling materials genome inspired design and discovery of materials. We use first principles calculations to study the connection between intrinsic (bulk) material properties and the energy to form a single, charge neutral oxygen vacancy (E_v). We investigate 45 binary and ternary oxides and find that a simple model which combines (i) the oxide enthalpy of formation, (ii) the mid-gap energy relative to the O-2p band center, and (iii) atomic electronegativities reproduces calculated E_v within ~0.2 eV. This result provides both a direct method to predict E_v and valuable insights into the key properties influencing E_v. We predict the E_v of ~1800 oxides and validate the predictive nature of our approach against direct defect calculations for a subset of 18 randomly selected materials. We advocate use of this direct and simple method to predict E_v at significantly reduced computational cost in order to identify candidate materials for applications where oxygen vacancies (or lack thereof) are of critical significance.

Metastable metal oxide polymorphs potentially offer opportunities for novel functionalities, such as enhanced optical, electrical and photocatalitic properties. Key to any of these applications is the ability to selectively grow a targeted polymorph in competition to surrounding metastable phases. It is therefore first necessary to establish a set of heuristics and experimental methods that can be used to stabilize energetically unfavorable polymorphs. This work focuses on the growth and characterization of specific VO₂ polymorphs stimulated by the high number of polymorphic variants and the interesting correlated electron affects observed in this material system. First, the deposition conditions were determined for phase pure, ground state (P2₁/c) VO₂ films deposited by pulsed laser deposition onto an amorphous substrate. The same processing parameters were then used in an inverse combinatorial experiment, where an array of substrates of different compositions and crystallographic orientations were used simultaneously for deposition while the processing parameters were kept constant. A coincidence lattice model was developed and paired with strain calculations to downselect substrates for the inverse combinatorial experiments. The boundary conditions include two-dimensional strain of less than 9% and a coincident lattice area of less than 400 Å. Anatase (001) TiO₂ thin film substrates stabilized a phase mixture of A and B phase VO₂ which is consistent with the predictions of the coincidence lattice model. Brookite TiO₂ substrates, however, were predicted to form Brookite VO₂ but stabilized the high temperature Rutile VO₂ phase instead. The temperature dependence of the conductivity of the VO₂ films on various substrates was also measured to determine the impacts of epitaxial stabilization on the metal-insulator transition and the stability of the phase under temperature cycling. We will discuss the potential limits of epitaxy templating for selected polymorphs.

High-throughput computational materials science provides researchers with the opportunity to rapidly generate large databases of materials properties. To rapidly add thermal properties to the AFLOWLIB consortium [1, 2, 3, 4] and Materials Project repositories [5], we have implemented an automated quasi-harmonic Debye model, the Automatic GIBBS Library (AGL) [6, 7]. This enables us to screen thousands of materials for thermal conductivity, bulk modulus, thermal expansion and related properties. The search and sort functions of the online database can then be used to identify suitable materials for more in-depth study using more precise computational or experimental techniques. AFLOW-AGL source code is public domain and will soon be released within the GNU-GPL license.
[1] S. Curtarolo et al., Comp. Mat. Sci. 58, 218 (2012).
[2] S. Curtarolo et al., Comp. Mat. Sci, 58, 227 (2012).
[3] www.aflowlib.org
[4] R. H. Taylor, F. Rose, C. Toher, O. Levy, K. Yang, M. Buongiorno Nardelli and S. Curtarolo, Comp. Mat. Sci. 93, 178 (2014).
[5] A. Jain et al., APL Mater. 1, 011002 (2013).
[6] C. Toher, J. J. Plata, O. Levy, M. de Jong, M. Asta, M. Buongiorno Nardelli and S. Curtarolo, Phys. Rev. B 90, 174107 (2014).
[7] M. A. Blanco et al., Comput. Phys. Comm. 158, 57 (2004).

The need to effectively develop new technological playgrounds is diffusing the traditional separation between theory and experiment. As a result, innovative assemblies are being developed, where theory is an integral part of the research effort. The concurrent development of next generation detectors grants the acquisition on high throughput modes of synchrotron spectroscopy data, featuring extremely high energy and spatial resolution, as well as large surface analysis.
Indeed, last generation detectors enabling large area imaging capabilities, while maintaining excellent surface sensitivity, are finally opening the door to mesoscopic characterization at device interfaces, capable of systematically assessing growth and processing. In this scheme, acquired data effectively results in knowledge to inform synthesis and processing. This knowledge subsequently affects experimental/theoretical design; conducive to scaling of highly anticipated graphene-based electronics at wafer-level. Further, adoption of an in house theory effort, allows for quick turn around in this affection.
In this work, we present our approach to the study of property measurement and nanointerface behavior of transferred CVD-grown graphene in an effort that combines experiment, first-principle theory and last generation large-area synchrotron detectors.

High-throughput structure determination for combinatorial materials discovery leads to unique computational challenges. I will discuss our work on solving the phase map diagram problem for composition libraries. While standard statistical and machine learning methods are important to address this challenge, they fail to incorporate relationships arising from the physics of the underlying materials. I will discuss how an effective approach requires a tight integration of statistical machine learning methods, to deal with noise and uncertainty in the measurement data, and optimization and inference techniques, to incorporate the rich set of constraints arising from the underlying physics. Moreover, certain parts of the interpretation process can still benefit from subtle human insight and intuition by incorporating a human-computation component.

Since 2002, National Institute for Materials Science (NIMS) has been carried out the development and management of an Internet materials database system MatNavi. Today, MatNavi contains more than 10 materials databases covering the basic and engineering properties of inorganic materials, polymers and structure materials. For the past years, we have successfully set up a process of data collection and verification from literatures, calculation and experiments. We have designed a general materials identification system and a database structure, which can be applied to any type of materials and properties. The MatNavi integrated search system enables data search crossing organizations and nations.
In recent years, materials exploration and design has become an important purpose of materials database, in addition to reference data and materials selection. The new approach of materials development starts from analyzing a large, comprehensive and systematic data set, by statistical or machine learning methods, in order to discover the relationships between materials process, structure and properties. The big data concept is changing traditional materials database in various aspects including data collection, data storage and data service. A new concept of NIMS data platform will be introduced.

In this talk, we present an under-the-hood technical overview of Citrination, a cloud-based software platform that performs three broad tasks: (1) automatically extracting and ingesting materials research data from various sources, including papers, patents, data sheets, and raw experimental output; (2) structuring these data within a unified and open-source JSON-based data standard; and (3) applying machine learning algorithms to the resulting large-scale data in order to uncover novel property-structure-processing linkages. Beyond this general overview, we will also focus on: bridging the gap between computation-only and experiment-only repositories; the major problem of lack of incentives for data sharing; and broadening the appeal of materials informatics to the entirety of the materials science community.

The ability to probe chemistry and structure at the atomic scale has not only allowed a deeper understanding of fundamental processes in materials, but also promised to open knowledge-driven pathways to search the Materials Genome for new materials and devices needed to address many of the technological and environmental challenges we face today. With this vision in mind, there has been continuing investments in brighter sources and development of faster detectors. Consequently the rate of data generation is perpetually accelerating, however, the pace of new discoveries and deepening scientific understanding hasn&’t seen an equally dramatic increase. The major bottleneck in the workflow for converting data into knowledge is our ability to efficiently distill large complex and noisy data sets into a reduced set of characteristics and scientifically interesting features and trends.
The traditional path of extracting knowledge from data relies on the superb ability of the human eye and brain to detect subtle changes and seeing patterns buried in noise. But as the speed of data generation increases reliance on humans to curate and analyze the data becomes a bottleneck. Efficient and reliable computational methods are therefore urgently needed for extracting hidden features, with minimal human intervention, from large datasets. Advances in image processing, statistical machine learning, and Probabilistic model approaches have opened new pathways for data mining, and feature and trend extraction. But these approaches are still not completely successful, especially when the object of the search is unknown and/or unexpected but scientifically most interesting structural phases that are hidden in these datasets are minority components occurring in a small region of the dataset.
In here, we will illustrate an approach that uses eigenvector analysis and clustering approaches to extract structural phase information from large x-ray spectroscopy datasets containing over half a million spectra, which was acquired in less than an hour at a synchrotron based spectro-microscopy facility. We will also discuss a proposal to iteratively improve upon the results of initial extraction by use of sophisticated error analysis. We believe that integrating these approaches in a search engine operating in nearly real-time will have large impact in accelerating the rate of knowledge discovery from high throughput data sources.

A database with band-structure and optical properties for semiconductors and insulators is of great value for a wide range of energy applications of electronic materials, and reliable band gap predictions from first principles would be highly desirable. While GW quasi-particle energy calculations provide fairly accurate band gap predictions for conventional main group compounds, a clear picture has yet to emerge for electronically complex materials such as transition metal oxides. The GW approach of Ref. [1] was designed as a computationally efficient compromise allowing the calculation of hundreds to thousands of materials with sufficient accuracy to generate useful data of real value. By placing the 3d orbitals and the sp bands in an integrated picture, this approach provides a consistent description of conventional semiconductors with itinerant band states and Mott insulators with localized orbitals, as well as materials that lie between those limits [2]. The currently available data for ~300 materials is part of the publicly accessible database [3], serving as a resource for materials design and discovery. As an example, the computational predictions for a hypothetical tetrahedral phase of MnO [4] have led to the design and realization of novel wurtzite structure Mn_1-xZn_xO alloys with x < 0.5, with a proof-of-concept demonstration of solar water splitting [5].
[1] Band-structure calculations for the 3d transition metal oxides in GW, S. Lany, Phys. Rev. B 87, 085112 (2013).
[2] Semiconducting Transition metal oxides, S. Lany, J. Phys.: Cond. Matter. (topical review) (in press, 2015).
[3] http://materials.nrel.gov
[4] Semiconducting transition metal oxides based on d⁵ cations: Theory for MnO and Fe₂O₃, H. Peng, S. Lany, Phys. Rev. B 85, 201202(R) (2012).
[5] Design of semiconducting tetrahedral Mn_1-xZn_xO alloys and their application to solar water splitting, H. Peng, P. Ndione, D.S. Ginley, A. Zakutayev, and S. Lany, Phys. Rev. X 5, 021016 (2015).

Recent studies have highlighted the potential for materials databases to serve as resources for creating predictive models. These methods generally work by using machine learning to infer a functional relationship between a material property and attributes designed to capture the chemical effects using a machine learning algorithm. As a step to making this capability available to the materials community on a wide scale, we have a created a general-purpose framework for creating predictive models and a software toolkit for using and sharing these models. Our method relies on a diverse set of attributes and partitioning a database into groups of chemically-similar groups before training separate models on each partition. In this work, we explore the capability of this framework to predict a wide variety of material properties by constructing models for properties as diverse as band gap energy, formation energy, and glass forming ability.

Compressive sensing (CS) is a recently developed mathematical theory that allows high resolution images to be obtained for sampling rates lower than the Nyquist rate. In scanning transmission electron microscopy (STEM), CS has been shown to give electron dose reductions of up to 93%. In this case, the degree of undersampling depends on the requirements of the image being acquired—both the speed and overall spatial resolution that is needed. Despite these improvements, there are many materials for which the dose still exceeds the stability of the material, making them impossible to study by STEM methods—such as the identification of active sites in catalysts and high resolution protein structure identification.
By going a step further and combining CS with machine learning (ML) it should be possible to realize the full potential of CS and apply it to many more materials systems. Moreover, ML models can be customized for specific materials science observations rather than splitting reconstruction and statistical analysis into separate steps. In this work we primarily focus on dose minimization as a way to achieve structure identification under the minimum possible dose.
Within a Bayesian framework we apply active learning to a convolutional factor analysis model to determine the most informative sensing locations (e.g. pixels). Once the locations are found they are measured and the process repeats until a certain dose budget has been met. For example, if there was a budget of 20% of the pixels, we could measure a random 2% of the pixels, then use active learning in 9 more 2% batches. The convolutional model can help identify common structure within a specimen, thereby further reducing dose and generally producing superior reconstructions relative to a completely random approach. The potential for these techniques to study a wide range of beam sensitive materials as well as to improve the sensitivity/resolution of in situ/operando methods will also be discussed.

Carrier lifetime is one of the few semiconductor parameters that not only is used to monitor the material&’s defect density, but also to characterize the device properties. The behavior of a large number of semiconductor devices, ranging from imaging sensors to power MOSFETs, depends on the lifetime parameter. There are various methods that have been developed over the years for measuring the lifetime. The pulsed MOS technique is one of these techniques that has gained popularity due to its easy and practical implementation in industrial environments. Pulse MOS is primarily used for measuring the generation lifetime. This technique is especially useful when characterizing thin epitaxial layers. Although a lot has been done in improving the speed and accuracy of the technique, little study has been done on the role that the test structure plays in yielding accurate measurements. In this work we investigate the effect that the test structure design have on the accuracy of the results obtained from pulsed MOS method. Two important factors that requires attention are the MOS gate geometry and feature size. These are controversial parameters among engineers when it comes to fabricating the appropriate test structures. Although it is easier to fabricate square gates on product wafers (from mask point of view), it will be shown that more accurate results would be obtained when the gates are chosen circular. There is also an optimum gate area that yields accurate lifetime values. Using the physics behind the measurement and employing 3-D TCAD simulations it will be demonstrate that how varying design parameters can affect the measured lifetime values and causing them to deviate from their actual values.

A solid cylindrical object placed horizontally melting into a liquid-solid two phase system involves two processes, heat conduction and buoyancy lifting. The heat conduction process with phase change in a cylindrical system could be considered as an axisymmertric Stefan problem . On the other hand, when liquid appears in the system, the density difference and viscosity break the symmetry of the system by moving the center of the solid cylinder upward or downward. But the relationship between heat conduction, buoyancy and viscosity is not well understood. To study the effect of buoyancy and viscosity on the overall melting process, we propose a dimensionless constant involving the ratio of the characteristic time of melting process obtained from Stefan boundary condition, and the time scale for buoyancy lifting process obtained by balancing viscous drag and buoyancy lifting forces. At the same time, the cylindrical structure melting process without the presence of gravity was simulated. The results show that in the absence of buoyancy and viscosity, the ice rod melting time is fairly close to the characteristic time of melting process. However, under the influence of buoyancy and viscosity, the melting time was shortened. These results have implications in the melting of a cylinder in a constrained space, such as a crystalline semiconductor core in a viscous glass cladding during the drawing of a mid-IR optical fiber.

In this work we explore how misfit dislocations nucleate, propagate, and interact through characterization and simulation of growth and annealing of lattice-mismatched heteroepitaxial semiconductor thin films. Dislocations can greatly affect the optoelectronic and electronic properties of semiconductor devices. In particular, III-nitride (hetero)epitaxy generally involves dislocation densities above 10⁸ cm^-2 for growth on non-native substrates. While great progress has been made on III-nitride (opto)electronic devices despite these high defect densities, there is much evidence correlating dislocations in III-nitrides to decreased device performance and lifespan.
We have previously created a “process simulator” for the Ge_xSi_(1-x)/Si(100) system, which predicts the nucleation, growth, and interaction of misfit dislocations, and thus the total dislocation density at any stage during film growth or annealing. This simulator was refined using many in-situ TEM measurements of dislocation nucleation and velocity. We have recently extended the architecture of this simulator to the Ge_xSi_(1-x)/Si(110) system - which is clearly closely related to the Ge_xSi_(1-x)/Si(100) system, but which has distinct dislocation geometries and structures - as an exercise to see how efficiently this simulator can be extended to a new system, through comparison of its predictions to a limited experimental data set. Specifically we show how by comparing the simulator predictions to a set of about two dozen experimental measurements of dislocation spacing in different samples, we are able to parameterize a dislocation nucleation model in the (110) system such that the overall simulator robustly predicts the experimental data.
These concepts are now being employed to extend the simulator into the III-nitride system. Initial work is focusing on non-polar (a-plane and m-plane) heterointerfaces, such that there is a resolved misfit stress onto the operative glide planes, in contrast to conventional polar c-plane epitaxy. Our overall goal is to integrate experiment, simulation and data refinement methods to develop robust predictive simulators of the development of dislocation arrays during growth and/or annealing of heteroepitaxial systems, and to determine how to most efficiently extend this concept into new materials systems.
We would like to acknowledge the support of the NSF through grant DMR-1309535;

Alloying has been a common way to tune and manipulate material properties. Up to now, studies of alloys have primarily focused on homogeneous alloys, resulting in materials which are typically stable or have low enthalpies of mixing. However, heterogeneous alloys, often highly metastable with high enthalpies of mixing, offer the additional contribution of a structural change, at some composition, to alter the properties. The study of heterogeneous alloys allows for better understanding of metastability and the benefits obtainable from metastable states. This type of alloying can be applied to anisotropic materials with correspondingly anisotropic properties like orthorhombic SnS, widely considered for applications in PV devices due to high absorption and suitable carrier concentration. SnS has carrier mobilities which are excellent in-plane, but are quite low out-of-plane, and thus lower inter-layer current, rendering it unsuitable for solar cell application. By alloying with an isotropic material, the properties would become isotropic, including the mobility.
Here we demonstrate analysis of alloying of heterogeneous materials: orthorhombic SnS and cubic CaS. The entire range of compositions of Sn_shy;1-xshy;Ca_xS from 0<x<1 was deposited as thin films by pulsed layer deposition. The optical properties (bandgap and absorption) and structural properties (lattice constants, crystal structure, and coordination number) were measured and analyzed as a function of composition. Property trends were then compared to theoretical predictions to demonstrate the usefulness of theoretically driven or assisted studies and to show the interesting trends attainable by hetero-structural alloying. The compositional region of structural transition was investigated with particular interest as properties were predicted to change much more dramatically. With the transition from orthorhombic to cubic, the bandgap was expected to drop significantly and the mobility to experience a dramatic increase. Additionally, explanations of differences between experiment and theory are presented.

The approach “Materials by Design” has strongly facilitated the discovery of new materials and the deliberate optimization of compounds in the last 3 years. However, one of the big challenges is the experimental realization of theoretically predicted materials. In this presentation, we focus on the experimental realization in the field of new earth abundant tin oxide and nitride compounds that are attractive due to their chemical stability. The nitrides are especially interesting, because they might exhibit even better semiconducting properties in comparison to the oxides due to the more covalent bonding.
We deposited (Sn,Zn)O and Sn₃N₄ alloys by reactive magnetron sputtering from metallic targets for high-throughput investigations. The spatial separation of the targets together with a unique substrate holder geometry allowed for the preparation of crossed gradients of the temperature and the composition over 2 x 2 inch substrates. This approach enables us to screen the phase space much faster than it is possible with a non-combinatorial approach. All films were then analyzed spatially resolved by X-ray fluorescence (XRF) for the composition and by X-ray diffraction (XRD) for the phase. More in depth characterization of the structural, electrical, and optical properties was performed for selected samples.
In this poster, we will present experimental work on the process conditions needed to reach the desired metastable phases and we will report their properties in comparison with theoretical predictions [1,2]. First results show that the Sn oxidation state in (Sn,Zn)O changes from Sn⁰ via Sn²⁺ to Sn⁴⁺ with increasing Zn content in the films. It is therefore necessary to precisely tune the oxygen content in the sputtering atmosphere to maintain the SnO structural phase while incorporating the Zn. Initial experiments on Sn₃N₄ alloys showed that we are able to reach chemical potentials of nitrogen in the order of Δµ_N = 1 eV with our atomic nitrogen source, which should suffice to deposit the desired phases. However, the oxygen background pressure needs to be further reduced to suppress the formation of mixed anion O-N phases.
[1] H. Peng et al., arXiv:1504.01168 (2015)
[2] C. Caskey et al., J. Mater. Chem. C 3, 1389 (2015)

Multi-principle element high entropy alloys (HEA) are relatively new material systems with near-equiatomic proportions of five or more constituent elements in a single alloy phase. These alloys are of interest in high temperature applications due to their creep resistance, diffusion resistance, and high tensile strength. However, the stabilization mechanism that causes the preferential formation of the alloy phase as opposed to the intermetallic phase is not well understood. The initially proposed mechanism is that the large entropies of mixing for these materials stabilize the alloy phase. These large entropies of mixing lead to a higher free energy of mixing compared to the free energy of secondary phase formation. An opposing theory states that the stabilization mechanism relies on the mutual solubility of the constituent elements within a single crystal lattice. The miscibility of each binary pair of elements in the system drives growth toward a stable alloy phase.
We utilize high-throughput experiments as a method to synthesize and screen compositional variations of an (Al_1-x-yCu_xMo_y)FeNiTiVZr HEA system for potential secondary phase formation. Al, Cu, and Mo were selected to be varied in the system because different alloy phases were observed in the binary phase diagrams of each pair. A thin film compositional spread was deposited using magnetron sputtering on three inch Si substrates. In-situ synchrotron x-ray studies were performed to observe phase transformation across the combinatorial library on the sample and SEM images were taken to observe grain growth at different sputtering conditions. We will discuss how these techniques were used to determine which mechanism is dominant in the growth and stabilization of the HEA system.

Our research group have hitherto established the combinatorial system for high-throughput exploration of functional oxides films or fine particles. Furthermore, we have recently developed the Seebeck coefficient (S) evaluation probe for high-throughput thermoelectric materials exploration and the tool for library preparation under high pressure (~ 200 MPa).
In case of thermoelectric oxides exploration using our combinatorial system, the electrical conductivity (σ) is lower than the anticipated results because the density of obtained powder library is very low. Therefore, it was difficult for calculation of the power factor value PF (σS²).
In this study, we tried the improvement of the electrical conductivity evaluation by adding the high-pressure process in powder library preparation. As an example, in case of CaMnO_3-δ type isomorphic compounds exploration, powder library before sintering process was densified using the above high-pressure preparation tool. In case of 5% Bi substituted CaMnO_3-δ, σ of the densified powder showed better value (63 S#65381;cm^-1) than non-high pressure treatment (13 S#65381;cm^-1).

In the context of Materials by Design (or MGI) approaches to materials development, high-throughput experimental materials synthesis and characterization are used to rapidly create structure-property relation maps to complement or test the results of high-throughput theoretical screening. In this effort, the primary current bottleneck is the accurate determination of crystalline phase content from x-ray diffraction (XRD) data for sample sets with 100&’s or 1000&’s of samples. In our work, the XRD patterns for thin film composition gradient samples, grown by co-sputtering, are measured using a commercial (Bruker D-8 Discover) diffractometer equipped with a 2D detector. Cation composition is determined by x-ray fluorescence mapping. The resultant data files for both are then automatically harvested and made available in a database. Hence, the big challenge, and the topic of this talk, is to identify structure-phase maps for large ensembles in a rapid (nearly real time), semi-automated fashion. We will describe a flexible Python and R based analysis framework that allows the user to interactively visualize and interrogate the data as well as to implement and test a variety of advanced machine learning algorithms. Current work is aimed at the reliable detection of real XRD peaks in datasets, which are often noisy and/or have overlapping peaks. This includes preprocessing to normalize the data and remove the substrate contribution, noise filtering (such as with spline fitting), and peak detection (via using, e.g., continuous wavelet transformations). The resultant extracted XRD peak sets are compared to reference data (pulled from materials.nrel.gov or elsewhere). Two user tools we have developed are “heat maps,” which simultaneously compare peaks sets for multiple samples with multiple references, and “hit maps” which provide a more detailed view of the overlap between the peaks in a single measured spectra and peaks in multiple reference spectra. We will also describe progress towards the application of machine learning algorithms to provide higher levels of analysis automation. In particular, we will examine and compare the effectiveness of methods based on non-negative matrix factorization, spectral and mean shift clustering, sparse regression and neural networks for the purpose of characterizing and classifying samples. This analysis development work will be demonstrated through application to the Cu-Sn-S and Sn-S materials systems currently being explored at NREL for applications as solar cell absorbers.

The recent success of methylammonium lead halide perovskites suggests that the existing approach for computationally identifying promising semiconductors for PV absorbers may be inadequate. Emphasis on optical absorption may miss more critical screening parameters, the minority carrier lifetime and mobility. These properties enable high-performing optoelectronic devices, but it remains difficult to assess them from first principles and to measure them rapidly for a large number of semiconductors. However, we can consider potential electronic structure criteria that could lead to high lifetimes in the presence of defects. We propose that this “defect-tolerance” may be enabled by properties such as a high dielectric constant, disperse band edges due to spin-orbit interactions coupled to crystal structure, and antibonding character to the valence band maximum. [1]
The family of lead halide perovskites satisfies these criteria, but they are not the only semiconductors that do so. In this work, we identify several classes of semiconductors that include partially-oxidized p-block cations, and which share a similar electronic structure to the lead halide perovskites. These are identified through a broad search of the MaterialsProject.org [2] electronic structure database based on the developed design rules.
We then pursue a targeted experimental screening approach informed by the first principles calculations, starting with bismuth triiodide (BiI₃). This consists of two components: first, we optimize the growth and verify the phase, purity, and morphology of the desired semiconductor; second, we use minority carrier lifetime as the response variable to evaluate each candidate semiconductor. We demonstrate a full growth and characterization cycle in less than a day, allowing for fast experimental screening. In doing so, we combine more targeted first principles screening with higher-throughput experimental screening to leverage the best of both approaches, allowing faster identification of promising semiconductors.
[1] R. E. Brandt, V. Stevanovicacute;, D.S. Ginley, T. Buonassisi. MRS Communications, 1-12 (2015). DOI: 10.1557/mrc.2015.26
[2] A. Jain, S.P. Ong, G. Hautier, W. Chen, et al. APL Mater. 1, 011002 (2013). doi:10.1063/1.4812323.

Accelerated materials innovation has made enormous progress over the past several years. Several “Materials by Design” or “Materials Genome” success stories are associated with developing large computational databases for 1000s of compounds. Such databases usually contain materials properties needed for specific functionalities, such as thermoelectrics and batteries. These data enable users to choose 10s of the most promising candidates for further down-selection by experimental studies.
However, there is still a lot of work needed to turn 10s of promising candidates into 1 working device prototype, that then can be fully optimized and hopefully commercialized. Doing this materials research in a traditional “trial and error” mode (1 student = 1 material) is a one way to perform this last step of materials innovation. An alternative approach is to apply the accelerated materials innovation methods to this step.
In this talk, I will discuss how high-throughput materials science methods, such as combinatorial experiments and first-principles simulations, can be adopted to accelerate the last step in materials innovation. The focus of the presentation is how to bridge the gap between the outputs of the high-throughput materials research activities (10s of material candidates) and the traditional applied technology/device optimization (1 device configuration). I will illustrate this on the example from the area of thin film solar cells based on Cu-M-Q (M = Sn, Sb, Q = S, Se) photovoltaic absorber materials.
By combinatorial experiments and first-principles computations we were able to (a) down-select from ~20 Cu-M-Q material with different stoichiometries, to 2 most promising Cu₂SnS₃ and CuSbQ₂ photovoltaic absorbers using simple thermodynamic stability and optical absorption metrics [1,2]; (b) choose the CuSbQ₂ absorber among these two based on defect /doping and electrical transport properties [3,4,5]; and (c) integrate the CuSbQ₂ in a solar cell prototype with standard CdS and Mo contacts and initial 1-3% energy conversion efficiency [6,7]. Currently, combinatorial device development is underway to optimize the CuSbQ₂ absorbers and to re-design the contact layers. The goal is to push the efficiency towards 5-10% level, where more traditional device optimization can begin.
This work was supported by U.S. Department energy. Experimental contributions from Adam Welch, Lauryn Baranowski, Willian Lucas, as well as the theoretical collaborations with Pawel Zawadzki, Haowei Peng and Stephan Lany are gratefully acknowledged
[1] Appl. Phys. Lett., 103, 232106 (2013)
[2] Chem. Mater., 26, 4951 (2014)
[3] Phys. Rev. Appl. 3, 034007 (2015)
[4] Solar Energy Materials and Solar Cells 132, 499 (2015)
[5] arXiv:1504.01327 (2015)
[6] arXiv:1504.01345 (2015)
[7] arXiv:1505.02311 (2015)

A PV device made exclusively from metal oxides (MOs) would be an ideal source of sustainable energy, as MOs are stable, non-toxic, abundant, and can be manufactured by low-cost methods. However, the electronic properties of most known MOs, i.e. short lifetime of excited electronic states and low mobility of charge carriers, prevent their use as active solar cell materials. To bring all-oxide PV to a breakthrough, new materials have to be developed including complex stoichiometries, crystal phases, doping, etc. The prospect of finding these unique materials lies in combinatorial material science which can produce novel MOs consisting of two, three, four or more elements. While most binary MOs are known, the number of unknown compositions is drastically increasing with the number of components.
Our approach involves high throughput synthesis which provides arrays of material compositions on a single substrate, high throughput material characterization by chemical, structural, optical and electronic techniques, and high throughput device testing under PV operating conditions. The acquired information is stored in a dedicated database for standard scientific analysis and for data mining by statistical tools that have been learned from state of the art machine learning and other big data tools. Our preliminary results show that a combination of the logical and statistical analysis with a large set of data points leads to the target MOs. This is best seen in the constant increases of cell performance that are being achieved. The methodology, new photoactive MOs and all oxide photovoltaic cells will be reported.

Visible-light driven photocatalysts have attracted great attention in the context of “green” technology. Recent studies have indicated that a few oxides of lone-pair cations show high visible-light photocatalytic activity; some examples are BiVO₄, SnNb₂O₆, and SnWO₄. Among them, β-SnWO₄ was reported to show good photocatalytic activity for the organic contaminants degradation under simulated daylight. The aim of this present study is rational and systematic exploration of novel visible-light driven photocatalysts within Sn(II) based complex oxides by combining of a large set of density functional theory (DFT) calculations, synthesis and characterization experiments. We focus on the ternary Sn(II) oxides with qA group elements (q=4, 5 and 6) of nd⁰ states, which have valency of +q. Structural models for the calculation are obtained from the International Crystal Structure Database (ICSD). DFT calculations are carried out for the whole range of SnO-MO_q/2 (M : Ti, Zr, and Hf (q=4); V, Nb, and Ta (q=5); Cr, Mo, and W (q=6)) pseudo-binary systems with many different compositions and structures. The VASP code was used for DFT calculations. All prototypical structures registered in ICSD, namely 584 structures for q=4 systems, 201 structures for q=5, and 362 structures for q=6, are chosen in addition to 21 non-prototypical structures of known Sn(II) complex oxides. Formation energies against SnO and MO_q/2, dynamical stabilities, and band gaps of a total of approximately 3500 compounds are calculated and used as the basis for screening. After screening of compounds by DFT calculations at the ordinary GGA-PBE level, additional calculations with HSE06 functional are made for some attractive candidates. In order to validate the results by the screening, we actually synthesized some compounds and examined their photocatalytic activity.

The large-scale materials demand of energy applications has spurred recent research on earth abundant and nontoxic materials. The limited materials space spanned by these requirements motivates the need to exploit both structure-property and composition-structure relationships of promising earth-abundant materials to overcome materials-intrinsic barriers. Low valent tin chalcogenides such as SnS and SnSe have attracted particular attention for optoelectronic and energy conversion applications but their performance suffers in part from their anisotropic crystal structure. The low valent state of Sn(II) in these materials results in a stereo-active lone pair leading to a local relaxation (C₃ coordinated Sn) and orthorhombic crystal structure. In contrast, the alkali earth chalcogenides, e.g. CaS or SrSe, are octahedrally coordinated in the rock-salt structure. The atomic structures of the two materials are compatible, as the C₃ orthorhombic phase of SnS and SnSe can be viewed as a distorted variant of the rock-salt structure. Therefore, alloying of low valent tin chalcogenides with alkali earth metals may be exploited to induce a phase transition to the higher symmetry rock-salt structure and enable more desirable materials properties.
We use ab initio electronic structure methods to calculate the atomic structure and the alloy mixing enthalpies and materials properties for Sn_1-xM_xCh alloys (M = Ca, Sr; Ch = S, Se). We predict the transition from the C₃ orthorhombic structure into the octahedral rocksalt structure for each alloy and determine the respective T-x phase diagrams. The carrier effective masses and band gaps, and the role of the local atomic coordination environments on structure-property relationships are also reported and discussed. The change of the lattice symmetry from the phase transition leads to a discontinuity in the calculated band gaps and electron/hole effective masses. In contrast to solid solution homostructural alloys, the miscibility gap temperatures for the binodal and spinodal lines do not share a coincident critical point, which is a consequence of the asymmetric shape of the mixing enthalpy in heterostructural alloys. Therefore, growth above the spinodal gap, but below the binodal gap temperature may allow the realization of functional metastable alloys with a homogeneous microstructure and more desirable materials properties. A theory driven complementary experimental study is performed over the entire composition range of Sn_shy;1-xshy;Ca_xS. Theoretical predictions are validated against the measured optical and structural properties. The prediction, synthesis, and characterization of Sn_shy;1-xshy;Ca_xS exemplify how the integrated strategy between theory and experiment is used to design and realize functional metastable materials by heterostructural alloying, and demonstrates a potential route to control the structure-property relationships and access the nonequilibrium phase space of other promising earth-abundant materials.

The damage caused from a loss-of-coolant accident (LOCA) for a nuclear reactor is greatly increased by hydrogen explosions. During a LOCA, high temperature steam is created, and oxidizes the Zircaloy cladding materials. This oxidation depletes the steam of oxygen, creating an explosive hydrogen atmosphere in the reactor that can cause catastrophic damage, as was seen during the recent nuclear disaster at the Fukushima Daiichi Nuclear Power Plant in 2011. To mitigate this risk, high-temperature alloy coatings that can resist oxidation could be applied to the current cladding materials. We have applied a combination of computationally guided materials synthesis, high-throughput structural characterization and data analysis tools to investigate the feasibility of coatings from the Fe-Cr-Al alloy system. The region of the phase diagram investigated was guided by regions identified by previous bulk studies as forming protective oxides. To understand initial oxide and metallurgical phase dynamics, analysis at intermediate temperatures was performed via in situ synchrotron glancing incidence x-ray diffraction at temperatures up to 690 K. This initial screening showed that a composition region with an Al concentration greater than 3.08 at%, and between 20.0 at% to 32.9 at% Cr exhibited high oxidation resistance. A series of compositionally targeted samples were deposited on stubs and their oxidation behavior at high temperature, 1373 K, was investigated to investigate if a passivating oxide would form at these high temperatures. This targeted study showed that the oxidation resistant composition region identified by the initial in situ screening facilitated the growth of a passivating oxide over a period of 6 hours at 1373 K.

The discovery of new high-temperature, oxidation-resistant, alloy coatings is vital to increasing turbine efficiencies and preventing hydrogen explosions during nuclear meltdowns. Unfortunately, the search space for such alloys is immense. Conventional coatings contain 5 - 7 elements built around a single principle alloying element, and new multiple-principle element alloys have been proposed that alloy 10+ element as a solid-solution. Regardless of their complexity, new coatings must be metallurgically stable in aggressive environments, must form a passivating oxide, and must exhibit strong surface adhesion. Here we will discuss our materials genome based method for systematically investigating novel coating materials. Combining physical vapor deposition modeling, in situ high-throughput crystal structure characterization, and machine learning algorithms, we rapidly screen new materials systems to identify promising new materials and elucidate fundamental alloy stability mechanisms. Here I will present our recent work on identifying novel nuclear cladding coatings. I will also provide experimental evidence that mutual solubility is the primary stabilizing force for multi-component alloys.

High throughput experimentation (HTE) is a recent development and has come to represent a methodology that seeks to speed up the discovery of new materials. It encompasses several fields including: design of experimental strategies; fabrication of combinatorial material libraries; automation of existing characterization techniques; development of new, rapid test methods; and intelligent data analysis. It has been implemented successfully in multiple fields including pharmaceutical development, biological research, and functional materials science. Challenges exist in applying high throughput strategies to the discovery of structural alloys, especially high entropy or multi-principal element alloys where the design space is extremely large. In this presentation we will discuss the considerations that are needed when designing HTE for microstructurally sensitive materials and suggest a possible framework for moving forward. We will also touch briefly on some of the recent developments in high throughput techniques that make HTE relevant in current experimental research on structural alloys.

Calorimetry is a powerful technique to study the thermodynamics and kinetics of materials and is often used to determine thermophysical properties that are related to the glass forming ability of metallic glasses. With extremely high sensitivity and a wide range of heating and cooling rates, nanocalorimetry further offers an opportunity to study metallic glasses over a huge dynamic range of cooling and heating.
In this work, we provide a methodology based on combinatorial nanocalorimetry to investigate metallic glasses using composition spreads. Over sixty different metallic glass alloys can be sputter deposited simultaneously onto an array of nanocalorimetry sensors. A combination of DC and AC techniques enables kinetics studies with heating and cooling rates ranging from isothermal to 10⁵ K/s. Various parameters related to the glass forming ability, including glass transition, crystallization, and melting temperatures, heats of crystallization and solidification, as well as the heat capacity can be measured as a function of composition much faster than with conventional techniques. We present results for Zr-based metallic glasses and demonstrate that the effect of composition on glass forming ability is readily investigated.

A strategy for accelerated discovery and exploration of multi-principal element alloys was developed and used to identify new alloys within a design window of desired microstructures and properties. As an example, the strategy was applied to analyze thousands of 3-4-,5- and 6-component alloys at equiatomic compositions of the selected palette of alloying elements. Currently available thermodynamic databases were used to assess equilibrium phase diagrams for these alloys. The validity and reliability of the calculated phase diagrams were estimated based on the extent of experimental binary and ternary data used to build the respective thermodynamic databases. Alloys with specific characteristics, such as alloys with possible use temperatures above 1000 C, were identified using an automated analysis of the calculated phase diagrams. This automated analysis includes the search for 2-phase alloys that may be solutionized, quenched and aged to produce precipitation-strengthened microstructures. The density, elastic moduli and costs of these alloys were estimated using the rule of mixtures of pure elements and were used as additional criteria for alloy selection. This approach allowed rapid, albeit preliminary, screening of many thousands of alloys and identification of promising candidate compositions for more time intensive experimental validations and assessments. The issues and opportunities for rapid searching of multi-component phase diagrams will be discussed.

The MAterials Simulation Toolkit (MAST) is a python toolkit developed at UW-Madison for managing high-throughput ab-initio calculations. MAST allows for the automation of complex defect and diffusion workflows consisting of multi-step interconnected VASP calculations. We demonstrate the utility of MAST with results on more than 200 dilute solute diffusion systems in FCC alloys. We find good agreement with experimental diffusion measurements and see clear diffusion trends across the periodic table. We then use the database and machine learning approaches to determine descriptors for predicting the diffusion coefficients of new systems without requiring further calculations. MAST enables such high-throughput screenings to be managed in an easy and consistent manner.

From glasses and colloidal suspensions to foams and sand, disordered materials are ubiquitous. However, the study of amorphous systems has lagged behind their crystalline counterparts in part because the lack of positional order in these systems obscures the relationship between structure and material properties. We discuss the use of machine learning to investigate these complex systems and build an understanding of their behavior. In particular, we use representation learning to identify a function of local structure - that we call ``softness'' - that correlates strongly with dynamics in these systems. We begin with an investigation of softness and its properties, using Hidden Markov Models to study the time evolution of softness in these systems. We then leverage softness to investigate the role of structure in two important phenomena of glassy systems: aging and failure. Aging manifests itself in dynamics that grow increasingly sluggish as glasses are left to rest below their glass transition temperature. We show that this phenomena is intimately related to the evolution of softness in these systems.

Developing new permanent magnet alloys with decreased critical materials requires combining bulk combinatorial synthesis techniques with rapid assessment methodologies to identify materials with desirable structures and properties. Our approach utilizes laser engineered net shaping (LENS) synthesis of bulk samples (> 3 mm diameter) in libraries over large composition ranges. These bulk samples are critical for rapid assessment of intrinsic magnetic properties (such as the Curie temperature (T_c), saturation magnetization (M_s) and anisotropy) necessary for identifying candidate materials for magnet manufacturing. Here we present results for this approach applied to the Fe₂B-Co₂B-M (where M is minor alloying addition) system. Starting from single crystal measurements, we have confirmed the accuracy of this approach as well as investigated the effects of various alloying effects on the magnetic properties of the Fe₂B-Co₂B system. The application of the approach to other systems is also discussed.

Selective Laser Melting (SLM) is an additive manufacturing technology to produce metal parts layer by layer by melting metallic powders with a high-powered laser. The final material properties of parts made via SLM are extremely sensitive to the powder characteristics (e.g., latent heat of fusion, laser absorptivity, etc.) and laser parameters (e.g., beam size, power, scan rate, etc.). Considering the beneficially large material set available to SLM, it is a significant challenge to identify the optimal operating parameters and in situ control mechanisms to rapidly and reliably produce parts with the desired properties without defects and residual stresses. In this talk, we discuss our efforts to determine the appropriate process parameters to produce smooth, high-density (>99% bulk value) parts from various metal powders. In particular, we use machine learning to develop logistic regression models that map out the threshold of acceptable laser scan rates across the range of laser powers currently used in SLM.

The CALMAR is a fully integrated system in which materials libraries of oxides, metals or nonmetals can be prepared by various PVD methods such as thermal evaporation (THEO), electron beam deposition (SCUBA) and sputtering (CODO). Even complex combinations solely in UHV are possible. Besides the usual physical characterization methods such as XRF, EDX the strength of the system lays in the scanning electrochemical methods namely scanning Kelvin probe (SKP) or scanning droplet cell microscopy (SDCM) coupled to a downstream analytics by inductively coupled plasma mass spectrometry (ICP-MS). Both techniques were extended to allow for the activation of photoelectrochemical processes. A number of examples will demonstrate possible applications such as water splitting photoelectrodes, flexible electronics or corrosion protecting coatings are used.

Mechanical properties of polycrystalline metals such as ductility, strength and fatigue resistance are connected directly to the material microstructure. The microstructure is in turn strongly dependent on the associated thermal processing, thus, control over such processing is critical. In this presentation we will describe new approaches to this dynamic control of the microstructure.
This work is a part of a multi-disciplinary research program targeted at integrating simulation, experimental sensing and control algorithms for dynamically controlling grain size distribution in polycrystalline films by way of controlling the local temperature distribution. Micro-heater arrays were designed and fabricated—using a photolithography based process sequence—for performing controlled annealing of an overlying copper film in-situ in the SEM. Each of these heater arrays had a set of ten independent resistive heaters on-board for creating tunable temperature distributions across macroscopic areas (c. 1mm²) of the copper films. The design of the micro-heater arrays was guided by finite element calculations. After micro-heater fabrication, the resulting chips were wire bonded to chip carriers, such that each individual heater line could be individually controlled. Control algorithms (decentralized feedback augmented by consensus feedback, and fully decentralized feedback) were then applied for generating various resistance profiles (therefore, temperature profiles) across the heater arrays. For example, it was possible to convert initial 1D non-uniform resistance profiles across a heater array into uniform profiles by applying suitable controls. Also, linear, periodic and uniform resistance increment profiles were successfully applied by using appropriate resistance control algorithms.
We are currently integrating these arrays into a scanning electron microscope, to provide in-situ sensing of the evolution of the local microstructure, crystallography and temperature during heating (using secondary electron imaging and electron backscattered diffraction). Coupled with feed forward and feedback control algorithms executed during annealing, this will allow real time control of the evolving materials microstructure. Control will be enabled both through feedback algorithms based on real-time sensing and through feed-forward algorithms guided by extensive grain growth simulations. Although, this work is currently targeted at controlling grain growth in copper films during thermal processing, the methods developed should be extensible to a broader range of materials and processing parameters.
This work was funded by National Science Foundation under NSF-CMMI 1334283 as part of the DMREF program.