Symposium Organizers
Carson Meredith, Georgia Inst of Technology
Sergei Kalinin, Oak Ridge National Laboratory
Momoji Kubo, Tohoku University
Artem Oganov, Skolkovo Institute of Science and Technology
Symposium Support
Applied Materials, Inc.
The Dow Chemical Company
Georgia Institute of Technology
CM7.1: Advances in Data Mining for Materials Development I
Session Chairs
Carson Meredith
Artem Oganov
Tuesday PM, April 18, 2017
PCC North, 100 Level, Room 124 B
2:30 PM - *CM7.1.01
Data Analytics for Mining Process-Structure-Property Linkages for Hierarchical Materials
Surya Kalidindi 1
1 , Georgia Institute of Technology, Atlanta, Georgia, United States
Show AbstractA majority of the materials employed in advanced technologies exhibit hierarchical internal structures with rich details at multiple length and/or structure scales (spanning from atomic to macroscale). Collectively, these features of the material internal structure are here simply referred to as the material structure or just structure, and constitute the central consideration in the development of new/improved hierarchical materials. Indeed, the existence of a causal relationship between the material structure and its properties is the central tenet in the field of materials science and engineering. It should be noted that the word structure is used very broadly in these statements to include and refer to any of the details of the material internal structure (spanning all relevant length or structure scales involved). Although the core connections between the material’s structure, its evolution through various manufacturing processes, and its macroscale properties (or performance characteristics) in service are widely acknowledged to exist, establishing this fundamental knowledge base has proven effort-intensive, slow, and very expensive for most material systems being explored for advanced technology applications. It is anticipated that the multi-functional performance characteristics of a material are likely to be controlled by a relatively small number of salient features in its hierarchical internal structure. However, cost-effective validated protocols do not yet exist for fast identification of these salient features and establishment of the desired core knowledge needed for the accelerated design, manufacture and deployment of new materials in advanced technologies. The main impediment arises from lack of a broadly accepted framework for a rigorous quantification of the material’s structure, and objective (automated) identification of the salient features that control the properties of interest. This presentation focuses on the development of data science algorithms and computationally efficient protocols capable of mining the essential linkages from large ensembles of materials datasets (both experimental and modeling), and building robust knowledge systems that can be readily accessed, searched, and shared by the broader community. The methods employed in this novel framework are based on digital representation of material’s hierarchical internal structure, rigorous quantification of the material structure using n-point spatial correlations, objective (data-driven) dimensionality reduction of the material structure representation using data science approaches (e.g., principal component analyses), and formulation of reliable and robust process-structure-property linkages using various regression techniques. This new framework is illustrated through a number of case studies.
3:00 PM - *CM7.1.02
Multi-Way Hyperspectral Image Analysis Based on Scanning Transmission Electron Microscopy and Associated Spectroscopic Methods
Shunsuke Muto 1 , Motoki Shiga 2 5 , Jakob Spiegelberg 3 , Masahiro Ohtsuka 4 , Jan Rusz 3
1 Institute of Materials and Systems for Sustainability, Nagoya University, Nagoya Japan, 2 Faculty of Engineering, Gifu University, Gifu Japan, 5 PRESTO, Japan Science and Technology Agency, Kawaguchi Japan, 3 Department of Physics and Astronomy, Uppsala University, Uppsala Sweden, 4 Graduate School of Engineering, Nagoya University, Nagoya Japan
Show AbstractThe digital technologies have changed current scientific measurements to automated operations, which has improved their efficiency and accuracy particularly in many repeated measurement processes, thereby inevitably enlarging the data size obtained. Such ‘big data’ obtained without arbitrary choices of specific areas of interest may contain richer information than those obtained with specific purposes/expectations. It is now sophisticated methods called as ‘data mining’ that are available for mining embedded information from the datasets based on information/statistics theories.
Spectrum imaging (SI) techniques by scanning a small probe of electron/x-ray/light on a sample area of interest can provide such a dataset as the two-dimensional data array, each row corresponding to the spectrum at a specific position. A non-negativity matrix factorization (NMF) technique [1] [2] can then apply to the dataset, which decomposes the set of spectra into a product of the constituent pure spectral components and their corresponding relative composition (weight) matrices without any reference spectra. This method allows us to provide a two dimensional spatial distribution map of different chemical states incorporated even when the multiply overlapped spectra.
In the present talk we review the recent progress in our NMF technique, particularly applied to datasets obtained by scanning transmission electron microscopy (STEM) and electron energy-loss spectroscopy (EELS), followed by discussing its intrinsic difficulties and our attempts to solve them. Finally, we discuss the future prospects of the field to extend the present bilinear model to multi-way analysis for more robust modeling of low signal-to-noise ratio data, where, for instance, EELS and EDX spectra are concurrently recorded at each spatial sampling point. In these cases the data structure is better described in the form of a set of matrices with coupled component matrices, where spectral components are unique to each data matrix, whereas spatial profiles are shared between them. Several successful application examples (of such coupled matrix decompositions) are shown.
[1] S. Muto, T. Yoshida and K. Tatsumi, Mater. Trans. 50 (2009) 964-969
[2] M. Shiga, K. Tatsumi, S. Muto, K. Tsuda, T. Mori, T. Tanji, Ultramicroscopy, 170 (2016) 43-59.
3:30 PM - CM7.1.03
A Priori Knowledge vs Computation Cost—Comparison of Different Segmentation Approaches of Large X-Ray Diffraction Datasets
Fang Ren 1 , Travis Williams 2 , Tri Duong 1 , Logan Ward 3 , Jason Hattrick-Simpers 2 , Christopher Wolverton 3 , Apurva Mehta 1
1 , SLAC National Accelerator Center, Menlo Park, California, United States, 2 , University of South Carolina, Columbia, South Carolina, United States, 3 , Northwestern University, Evanston, Illinois, United States
Show AbstractOver the past 3 decades, investment in brighter sources and development of larger and faster multi-pixel detectors has resulted in explosive rise in the amount of data collected. The exponential rise in data has transformed material science. It has enabled move away from investigation of pure materials to monitoring the operation devices composed of hierarchically complex arrangement of materials under realistic conditions. It has also inspired a new mode of material discovery based on high throughput experimentation.
The rate of new discoveries, however, has not kept pace with rate of data collection, mostly because data is still curated and analyzed by humans. Though humans are superb at utilizing prior knowledge for extracting information (as trends and patterns) from low contrast data, they are too slow to keep pace with accelerated pace of HiTp data. New automated/unsupervised methods are needed to extract trends and patterns, in nearly real time.
The challenge here is that many of the material science datasets are noisy and sparsely distributed in high dimensions, and consequently hungry for computational resources. Strategies are needed to incorporate knowledge about external physical constraints to reduce the computation costs. In here, we compare three classes of automated data analysis strategies to identify potential metallic glass formers in Co-Fe-Zr ternary system from HiTp XRD datasets. In the first strategy, we extract an attribute with direct physical linkage to the classification of interest, namely, the width and intensity of diffraction peaks to the degree of crystallinity. This strategy works on one diffraction spectrum at a time and scales linearly with number of points. In the second strategy we assume that closer the data points are to each other in space and time more likely are they to be similar to each other. This strategy works on two pairs of spectra, but still scales linearly with size of the dataset. In the final strategy we use the full dataset, and apply several well-established unsupervised data classification methods. This strategy scales approximately to the factorial of the size of the dataset.
3:45 PM - CM7.1.04
Automatic Segmentation and Fingerprint Matching for Atomic Resolution TEM Images
Eric Schwenker 1 2 , Fatih Sen 2 , Christopher Wolverton 1 , Maria Chan 2
1 , Northwestern University, Evanston, Illinois, United States, 2 , Argonne National Laboratory , Lemont, Illinois, United States
Show AbstractWith the emergence of materials research databases and advanced data acquisition frameworks, there is a growing gap between data generation capabilities and full data comprehension. In order to better link the high volume of datapoints with scientific knowledge, the interpretation process must be adapted for automation. However, interpretation of noisy experimental images (from transmission electron microscopy (TEM) and diffraction imaging) is in general, very difficult to achieve by purely automatic means. This is in large part due to the challenge of image segmentation and effectively representing the data. To this end, we have developed an approach for the automatic segmentation and fingerprinting of atomic resolution TEM images. We will discuss how segmentation and fingerprinting is used to create a flexible similarity metric for comparing microscopy images across datasets and over time, and show how this similarity metric compares simulated and real TEM images in a framework for constructing models of disordered interfaces. Overall, similarity judgements play a critical role in image retrieval tasks and are essential for facilitating atomic-level interpretation of large volumes of microscopy data.
4:30 PM - CM7.1.05
Data Mining for New Two-Dimensional Materials
Gowoon Cheon 1 , Karel-Alexander Duerloo 2 , Austin Sendek 1 , Chase Porter 3 , Yuan Chen 1 , Evan Reed 2
1 Department of Applied Physics, Stanford University, Stanford, California, United States, 2 Department of Materials Science and Engineering, Stanford University, Stanford, California, United States, 3 Department of Mechanical Engineering, Stanford University, Stanford, California, United States
Show AbstractTwo-dimensional materials such as graphene are very attractive for both technological applications and fundamental physics. However, despite the profuse amount of research efforts on two-dimensional materials, only a few dozen materials have been subject to considerable research focus. We employ data-mining techniques to identify 1173 potential 2D materials and 429 atomically thin materials with a variety of forms of piezoelectric tensor. Data on the families of materials, band gaps and point groups for the materials identified in this work are presented. This work significantly extends the scope of potential two-dimensional materials to be investigated.
4:45 PM - CM7.1.06
Computational Design for Stimuli Responsive MOFs
Charles Manion 1 , Laura de Sousa Oliveira 2 , Matthew Campbell 1 , Alex Greaney 2
1 , Oregon State University, Corvallis, Oregon, United States, 2 , University of California, Riverside, Riverside, California, United States
Show AbstractMetal-organic–Frameworks(MOFs) that experience stimuli induced structural transformation could enable a whole new class of materials with remarkable properties, such as externally tunable stiffness, variable porosity, or tunable catalytic properties. Here we will present a novel approach based on using graph grammars that makes the problem of discovering MOFs for specific applications amenable to powerful artificial intelligence tree search algorithms. Our long term goal is to enable invention of MOFs that undergo reversible stimuli induced structural transformation. We will show how this approach can be used to find candidate MOFs with reversible pressure induced changes in porosity. In addition we will also demonstrate how the same approach may be used to find candidate MOFs that experience light induced structural transformation via the action of photoisomerizing moieties in the linker.
5:00 PM - *CM7.1.07
Facilitating the Development of a “High Throughput Experimental Materials Science Virtual Laboratory”
Martin Green 1
1 , National Institute of Standards and Technology, Gaithersburg, Maryland, United States
Show AbstractHigh-throughput experimental (HTE) methods are uniquely suited to rapidly generate the large volumes of high-quality data necessary to inform and validate the materials design process. Therefore, HTE methods play a very significant role in the Materials Genome Initiative (MGI), which seeks to reduce the cost and time of development of new materials. Although some HTE programs exist, they lack effective coordination, resulting in underdeveloped opportunities and capabilities in data collection, curation and analysis. Due to the large cost of even one “brick and mortar” HTE facility, and the need to have multiple facilities dedicated to different materials classes (e.g., catalysts, photovoltaics, lightweight structural materials), we propose a “High Throughput Experimental Materials Science Virtual Laboratory” (HTEMSVL) to generate the aforementioned data. The virtual laboratory would consist of an integrated network of high-throughput synthesis and characterization tools, computational tools, and a best-in-class, configurable data curation system that would enable data to be discoverable, accessible and interoperable. Ultimately, users such as researchers at national labs, universities, and start-ups could leverage the HTEMSVL to facilitate the rapid development and commercialization of novel materials and products.
5:30 PM - CM7.1.08
Order and Disorder in Ternary and Quaternary Atomic Laminates (MAX phases) from DFT Predictions and Material Synthesis
Martin Dahlqvist 1 , Rahele Meshkian 1 , Johanna Rosen 1
1 , The Department of Physics, Chemistry and Biology (IFM), Linköping University, Linköping Sweden
Show AbstractMore than 50 years ago a family of atomically laminated compounds were discovered, being comprised of a transition metal M, an A-group element A, and carbon and/or nitrogen X, and therefore referred to as MAX phases. The possibility to form solid solutions on the M-, A or X-sites have attracted attention for a long time, as an approach for attaining novel MAX phase compositions and allow for property tuning. However, attaining an exact a priori decided composition for a solid solution might be challenging. This is opposed to chemically ordered system where the composition may be governed by, e.g., the underlying symmetry. Here we present theoretical predictions of phase stability of quaternary MAX phase alloys based on density functional theory calculations, assuming both disorder and out-of-plane chemical order on the M-sites. Examples are given of both verifying and predictive simulations of known and hypothetical compounds, with a particular focus on Ti-based materials. Furthermore, synthesis of chemically ordered Mo2ScAlC2 is presented, motivated by theoretical results, and previous synthesis of ordered Mo- and Cr-based quaternary laminates is discussed in light of the theoretical analysis. Materials design including composition as well as chemical order expands the concept of property tailoring in atomic laminates.
5:45 PM - CM7.1.09
Discovery of New Metallic Glass Former in a Ternary Composition Space through a Genomic Approach
Fang Ren 1 , Travis Williams 2 , Logan Ward 3 , Kevin Laws 4 , Jason Hattrick-Simpers 2 , Christopher Wolverton 3 , Apurva Mehta 1
1 , SLAC National Accelerator Center, Menlo Park, California, United States, 2 , University of South Carolina, Columbia, South Carolina, United States, 3 , Northwestern University, Evanston, Illinois, United States, 4 , UNSW, Sydney, New South Wales, Australia
Show AbstractMetallic glasses (MGs) are a class of novel material that combines properties of metals and amorphous ceramics. More recently, thin-film MGs have attracted many research efforts due to their potential applications in biomedical and electrical fields. While efforts over the last two decades have managed to fabricate several MGs, the fabrication of new MGs from earth abundant metals and in an industrially scalable manner is still challenging. The fabrication of MGs over the years has relied on only empirical rules, and thus it still requires extensive experimental efforts to search MGs within multi-element compositions. In this work, we combined computational predictions with high throughput (HiTp) experimentation to discover new MGs in a more efficient way through a joint effort by SSRL, the Hattrick-Simpers group at The University of South Carolina, and the Wolverton group at Northwestern University. Our method demonstrated a new methodology for material discovery based on intimate collaboration between experiments, machine learning tools, and theory. The Wolverton group have recently developed a supervised machine learning tool to identify promising MG alloy based on existing experimental data. From the several thousand new MG formers predicted by the tool, we limited our search to elements that are suitable for co-deposition and excluded expensive elements. In the next step, we exploited capabilities developed by the Hattrick-Simpers group to vary the metastability of deposited films by controlling the energetics of sputtering process. After that, X-ray diffraction was used for screening metallic glasses. We have located some amorphous region in Co-V-Zr ternary that is broadly consistent with the ML predictions. The results were imported into machine learning tool as additional training data and compared with the state-of-art thermodynamic, geometric and structural theories.
Symposium Organizers
Carson Meredith, Georgia Inst of Technology
Sergei Kalinin, Oak Ridge National Laboratory
Momoji Kubo, Tohoku University
Artem Oganov, Skolkovo Institute of Science and Technology
Symposium Support
Applied Materials, Inc.
The Dow Chemical Company
Georgia Institute of Technology
CM7.2/CM3.3: Joint Session: Accelerating Materials Discovery and Design with Computing
Session Chairs
Wednesday AM, April 19, 2017
PCC North, 100 Level, Room 124 B
9:00 AM - *CM7.2.01/CM3.3.01
NIST—The Materials Genome Initiative, and Computation
James Warren 1
1 , National Institute of Standards and Technology, Gaithersburg, Maryland, United States
Show AbstractIn this talk I will present an overview of the US Materials Genome Initiative, and then focus on NIST’s efforts in support of the MGI. After an overview where I will provide insight into community-led activities, I will discuss our attempts at NIST to address some of the challenges to creating the materials innovation infrastructure that lies at the heart of the Materials Genome Initiative. In particular NIST is now devoting considerable effort, in concert with its partners in industry, academia and government, to develop the tools, standards and techniques for (i) establishing model and data exchange infrastructure (ii) establishing best practices and new methods for ensuring data and model quality and (iii) developing the data analytics to enable "data driven" materials science. Given the focus of the conference, I will tie these efforts into the essential role of the MGI data infrastructure in enabling multiscale materials simulation, including specifc platforms targeted at soft materials.
9:30 AM - CM7.2.02/CM3.3.02
Intelligently Navigating Parameter Space with Machine Learning
Matthew Spellings 1 , Sharon Glotzer 1
1 , University of Michigan, Ann Arbor, Michigan, United States
Show AbstractAdvances in hardware and algorithms have increased the space of computationally-feasible simulations available to scientists by orders of magnitude. With processing power amplified by supercomputers, the most difficult component of modern computational science is sometimes the act of deciding the best experimental conditions to test. The most common response to this problem is a uniform grid in parameter space, but for categorical data we often care more about where the transition between two types of behavior in parameter space occurs instead of what happens on the interior of uniform regions. When we design a grid of input parameters and then perform experiments, how much information is lost “in the cracks” between points? Here we discuss approaches to incorporate machine learning into live simulations to maximize the variety of obtained results (in our example, the self-assembled crystal structures within a phase diagram) while minimizing computational time spent re-sampling behaviors that have already been seen.
9:45 AM - CM7.2.03/CM3.3.03
Materials Data Management with Signac
Carl Simon Adorf 1 , Paul Dodd 1 , Sharon Glotzer 1
1 , University of Michigan, Ann Arbor, Michigan, United States
Show AbstractResearchers in computational materials science are regularly posed with the challenge of managing large and heterogeneous data spaces. The amount of data increases in lockstep with computational efficiency multiplied by the amount of available computational resources, which shifts the bottleneck within the scientific process from data acquisition to data post-processing and analysis. We present a framework designed to aid in the integration of various specialized formats, tools and workflows. The signac framework provides all basic components required to create a well-defined and thus collectively accessible data space, simplifying data access and modification through a homogeneous data interface, largely agnostic of the data source, i.e., computation or experiment. The framework's data model is designed not to require absolute commitment to the presented implementation, simplifying adaption into existing data sets and workflows. This approach not only increases the efficiency for the production of scientific results, but also significantly lowers barriers for collaborations requiring shared data access.
10:00 AM - CM7.2.04/CM3.3.04
Digital Alchemy—An Inverse Approach to Mesoscale Soft Materials Design
Greg van Anders 1 , Paul Dodd 1 , Yina Geng 1 , Sharon Glotzer 1
1 , University of Michigan, Ann Arbor, Michigan, United States
Show AbstractTheory-led approaches have the potential to revolutionize the design of soft materials, but developing them is a challenge. We present recent advances on a mesoscale method for inverse materials design we term “digital alchemy.” Digital alchemy is a first-principles, statistical mechanics technique that can be used to compute thermodynamically optimal attributes of colloidal particles to self-assemble target materials. We present results from several recent investigations that demonstrate the use of digital alchemy to design colloid attributes such as particle shape for target materials, even in cases where collective behavior and competing interactions complicate the underlying materials physics. We demonstrate applications to high-dimensional design-parameter spaces that are beyond the reach of high throughput computing techniques.
10:15 AM - CM7.2.05/CM3.3.05
Pressure-Induced Phase Transitions in Shape Space
Rose Cersonsky 1 , Greg van Anders 1 , Paul Dodd 1 , Sharon Glotzer 1
1 , University of Michigan, Ann Arbor, Michigan, United States
Show AbstractIn designing new materials for synthesis, the inverse materials design approach posits that, given a structure, we can predict a building block optimized for self-assembly. How does that building block change as pressure is varied to maintain the same crystal structure? We address this question for entropically stabilized colloidal crystals by working in a generalized statistical thermodynamic ensemblewhere an “alchemical potential” variable is fixed and its conjugate variable -- particle shape – is allowed to fluctuate. We show that there are multiple regions of shape behavior and phase transitions in shape space between these regions. Furthermore, while past literature has looked towards packing arguments for proposing shape-filling candidate building blocks for structure formation, we show that even at very high pressures, a structure will attain lowest free energy by modifying these space-filling shapes.
10:30 AM - CM7.2.06/CM3.3.06
Determining Molecular Orientation via Physics Based Polymer Models with Polarized X-Ray Scattering
Adam Hannon 1 2 , Daniel Sunday 1 , Donald Windover 1 , Christopher Liman 1 , Alec Bowen 3 , Gurdaman Khaira 4 , Juan de Pablo 3 , Dean DeLongchamp 1 , R. Kline 1
1 , NIST, Gaithersburg, Maryland, United States, 2 , Georgetown University, Washington, District of Columbia, United States, 3 , University of Chicago, Chicago, Illinois, United States, 4 , Mentor Graphics, Wilsonville, Oregon, United States
Show AbstractFlexible electronics and photovoltaics, composites, and stimuli-responsive materials all require better methods to characterize molecular orientation at the nanoscale. The optical, mechanical, electronic, and transport properties of the devices made from such materials are determined from how the molecules orient in space. Methods such as scanning transmission electron microscopy (STEM) can obtain atomic resolution in inorganic materials, but severely damage organic polymer systems. Resonant soft X-ray scattering (RSoXS) has been used to determine the morphology of organic systems such as block copolymers and disordered organic semiconductors. This technique is non-destructive when performed below the absorption edge used to obtain resonance. Molecular orientation information can be obtained in addition to the morphology in anisotropic samples by varying the polarization of the incident X-ray beam (P-RSoXS).
Because RSoXS does not directly measure the real space structure, we develop inverse search methods to find the structure that best fits the measured scattering. We have incorporated physics based models such as self-consistent field theory (SCFT) and theoretically informed coarse-grained Monte Carlo (TI-CGMC) simulations into our RSoXS inverse search algorithm to fit scattering profiles for isotropic thin film block copolymer systems. These models allow for the measurement of thermodynamic properties in addition to the morphological shape while limiting the number of model parameters for such morphological detail rich systems. We have extended these models to obtain the molecular orientation in anisotropic systems from P-RSoXS measurements by considering polarization and molecular orientation effects explicitly. In this presentation, we first show examples of how the methodology has been successful for isotropic block copolymer samples. We then show the implementation of an SCFT model using a wormlike chain partition function to model a rigid-rod block in a rod-coil block copolymer. The density profiles and orientation profiles of the rod block is used to simulate the theoretical scattering profile for a P-RSoXS experiment. These simulated profiles are used with an inverse fitting evolutionary strategy algorithm that shows the model can be used in P-RSoXS experiments to find the average molecular orientation. Based on the results, experimental systems to explore with the technique are suggested.
10:45 AM - CM7.2/CM3.3
BREAK
11:15 AM - *CM7.2.07/CM3.3.07
Evolutionary Structure Prediction from Complex Crystals to Defects
Qiang Zhu 1
1 , University of Nevada, Las Vegas, Las Vegas, Nevada, United States
Show AbstractNowadays, the urgent demand for new technologies has greatly exceeds the capabilities of materials research. Understanding the atomic structure of a material is the first step in materials design. There have been tremendous progresses in the accurate prediction of crystal structures from first principles based on a variety of global optimization methods combing density functional theory (DFT) calculations. However, there remain many challenges on predicting complex systems such as organic crystals. Furthermore, recent experiments have revealed highly complex interface structures in different solids. The understanding of the atomic arrangements in the interfaces is crucial for the engineering control of materials properties on an upper level. In this talk, I will discuss the recent progresses in applying the evolutionary algorithm to study the organic crystal polymorphism and the structural phase transformations in metallic grain boundaries. The encouraging results so far suggest a major role of this approach in the prediction and design future functional and structural materials.
11:45 AM - CM7.2.08/CM3.3.08
Large-Scale Molecular Dynamics Simulation on Fracture Properties of Ni Anode for Highly Durable Solid Oxide Fuel Cell
Jingxiang Xu 1 , Yuji Higuchi 1 , Nobuki Ozawa 1 , Momoji Kubo 1
1 , Institute for Materials Research, Tohoku University, Sendai Japan
Show AbstractSolid oxide fuel cell (SOFC) is used as a highly efficient electronic-energy conversion device without environmental pollution and greenhouse gases. Currently, Ni-based anode is widely used as an anode for the SOFC; however it possesses a fracture problem due to the heat cycle operation of the SOFC. Moreover, the fracture of Ni-based anode is affected by the content of the water vapor in the fuel. Thus, an understanding of fracture properties of the Ni-based anode in water vapor is necessary for improving the durability of the SOFC and many experimental results have been reported so far. For the design of the durable anode, we need not only experimental but also atomic-scale theoretical studies. However, atomic-scale theoretical studies on fracture properties in the water vapor based on molecular dynamics (MD) simulation are not carried out widely. In this study, we investigated the fracture properties of polycrystalline Ni substrate in the water vapor by using our developed large-scale MD simulator [1]. In the tensile test simulation in the presence of the water vapor, a stacking fault firstly generates in the surface area of the polycrystalline Ni substrate when the strain is 0.033. In the tensile test simulation in the absent of the water vapor, the polycrystalline Ni substrate shows little change when the strain is 0.033. A large compressive stress is observed in the surface area of the polycrystalline Ni substrate in the presence of the water vapor by calculating the atomic stress distribution, whereas it does not be observed in the absent of the water vapor. Then, we investigate the component of stress and find that the coulomb interaction induced by the charge transfer between Ni and water molecules contributes to the large compressive stress. Thus, our large-scale MD simulation reveals that the water vapor accelerates the generation of stacking fault in the surface of the polycrystalline Ni substrate. Next, the effect of the grain size in polycrystalline Ni substrate in the water vapor are also discussed. Our study can provide a database for the development of the new anode material. [1] J. Xu et al., J. Mater. Chem. A 3 (2015) 21518.
12:00 PM - CM7.2.09/CM3.3.09
Integrated Imaging and Simulation to Investigate Lattice Deformations in Externally Stimulated Nanocrystals
Kiran Sasikumar 1 , Mathew Cherukara 1 , Thomas Peterka 1 , Ross Harder 1 , Subramanian Sankaranarayanan 1
1 , Argonne National Laboratory, Lemont, Illinois, United States
Show AbstractDespite the increasing role of nanomaterials in technology, their mechanical and dynamical properties under external stimulation are not well understood. One such problem is the pulsed laser excitation of a diverse class of nanomaterials, such as ZnO nanorods, WSe2 nanopillars, and Au-Al core-shell bimetallic nanocrystals. Another class of problems is the investigation of lattice deformations in nanostructured catalysts during multi-electron transfer processes. These constitute an important class of materials systems for catalysis, biomedical and energy applications. Understanding the temporal behavior of such nanomaterials under conditions of external stimulation is, thus, crucially important for energy research. In addition, characterizing lattice distortions can provide key insights into the behavior of nanomaterials and nanoscale interfaces.
Recently, experimental techniques have evolved to conduct time-dependent lattice dynamics measurements in nanomaterials. In particular, Bragg Coherent Diffraction Imaging (BCDI) has been used to directly image ultrafast lattice distortions in laser-heated nanocrystals. BCDI measurements have also been successfully used to observe reversible lattice distortions in metallic nanocrystals facilitating chemical reactions at low-coordination corner and edge sites. Suitable simulation models prove to be an ideal foil to explore the underlying mechanisms behind the observed lattice deformations. With the convergence of time and length scales accessible by both experiments and simulations, we are now able to integrate experimental observations with classical molecular dynamics (MD) simulations and continuum finite element calculations to enhance the fundamental understanding of materials behavior under external stimulation.
Here, we demonstrate the workflow(s) to integrate BCDI measurements with large-scale atomistic molecular dynamics simulations and finite element models to investigate lattice dynamics in externally stimulated nanocrystals. We will demonstrate the suitability of the workflow(s) as applied to a diverse class of materials systems and external stimulus. We show that direct comparisons between experiments and simulations are possible by using the appropriate level of theory or a combination of simulation techniques. In addition, the integrated experiment-informed simulation approach yields new insight into deformation mechanisms of nanomaterials that cannot be obtained and validated by either approach alone.
12:15 PM - CM7.2.10/CM3.3.10
DFT Applied to Transition Metals and Binaries—Developing the V/DM-17 Test Set
Elizabeth Decolvenaere 1 , Ann Mattsson 2
1 , University of California, Santa Barbara, Santa Barbara, California, United States, 2 Multiscale Computational Materials Methods, Sandia National Laboratories, Albuquerque, New Mexico, United States
Show AbstractDensity functional theory (DFT) is undergoing a shift from a descriptive to a predictive tool in the field of solid state physics, with undertakings like the Materials Project, OQMD, and AFLOW leading the way in utilizing high-throughput data to predict and seek novel materials properties. However, methods to rigorously evaluate the validity and accuracy of these studies is lacking in both the availability and utilization of techniques. The natural disconnect between simulated and experimental length-scales and temperatures, combined with this lack of validation, raises serious questions when simulation and experiment disagree. In response, we have developed the V-DM/17 test set, designed to evaluate the experimental accuracy of DFT’s various implementations for periodic transition metal solids. Our test set evaluates 26 transition metal elements and 80 transition metal alloys across three physical observables: lattice constants, elastic coefficients, and formation energy of alloys. Whether or not a functional can accurately evaluate the formation energy offers key insights into whether the relevant physics are being captured in a simulation, an especially important question in transition metals where active d-electrons can thwart the accuracy of an otherwise well-performing functional. Our test set captures a wide variety of cases where the unique physics present in transition metal binaries can undermine the effectiveness of “traditional” functionals. By application of the V/DM-17 test set, we aim to better characterize the performance of existing functionals on transition metals, and to offer a new tool to rigorously evaluate the performance of new functionals in the future.
Sandia National Laboratories is a multi-mission laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energys National Nuclear Security Administration under contract DE-AC04-94AL85000.
12:30 PM - CM7.2.11/CM3.3.11
Development of Crystal Structure Prediction Method for Magnet Materials
Tomoki Yamashita 1 2 , Hiori Kino 1 , Takashi Miyake 3 1 , Koji Tsuda 4 1 , Tamio Oguchi 2 1
1 , National Institute for Materials Science (NIMS), Tsukuba, Ibaraki, Japan, 2 , Osaka University, Ibaraki, Osaka, Japan, 3 , National Institute of Advanced Industrial Science and Technology (AIST), Tsukuba, Ibaraki, Japan, 4 , The University of Tokyo, Kashiwa, Chiba, Japan
Show AbstractSm-Co and Nd-Fe-B intermetallic compounds are known for high-performance rare-earth permanent magnets which are one of the key materials in magnetic and energy conversion devices. For material search of permanent magnets, developments of basic methods with data-driven approaches have been highly desired because of the rapid growth of supercomputer performances. One of the difficulties in finding new permanent magnet materials is how to predict the complex crystal structures. For example, the best magnet, Nd2Fe14B, contains 68 atoms in the unit cell. This system is quite difficult to predict the crystal structure because number of configuration explosively increases as atoms increase. In this study, we have developed and investigated methods of crystal structure predictions to overcome those difficulties. First, random search algorithm in combination with structure optimization technique using first-principles calculations was employed. Furthermore, Bayesian optimization method was added to accelerate crystal searches. The usability of crystal structure predictions for finding new rare-earth magnet materials is discussed.
We started with simple systems to test the random search algorithm. Crystal structure prediction simulations were carried out for RCo5 and R2Co17 (R means rare-earth element such as Y and Sm) which are important compositions of Sm-Co magnets. Our predicted structures were in complete agreement with structures in experiments. The most stable structures of YCo5 and Y2Co17 were obtained with probabilities of 8 and 3%, respectively. These results show that the random search algorithm is highly efficient for relatively small unit cell (less than 20 atoms). For a large unit cell including 4 Y and 34 Co atoms, however, we could not obtain stable structures within trials of 300 structures. We further try a Bayesian optimization method to search the vast space of Y4Co34 efficiently, and demonstrate the usability of Bayesian optimization for large systems. Several descriptors of crystal structures are discussed.
CM7.3: Advances in Data Mining for Materials Development II
Session Chairs
Wednesday PM, April 19, 2017
PCC North, 100 Level, Room 124 B
2:30 PM - CM7.3.01
Automated First-Principles Calculations for Building Dopant Property Database of Doped-ZnO
Kanghoon Yim 1 , Joohee Lee 1 , Ho-Hyun Nahm 2 , Seungwu Han 1
1 , Seoul National University, Seoul Korea (the Republic of), 2 , Institute for Basic Science, Seoul Korea (the Republic of)
Show AbstractOver the past decades, zinc oxide (ZnO) has been receiving a great attention for applications in electronic, optoelectronic, catalystic and magnetic applications. Among the wide-direct gap oxides, ZnO has promising properties for next transparent conducting oxide (TCO) such as high mobility, abundance and fabrication compatibility. Another important merit of ZnO is a wide range of external doping, which can overcome the limitation of carriers by intrinsic defects and tunes the optical, electrical, chemical, and magnetic properties. Therefore, extensive studies have conducted for doped-ZnO that cover most elements in the Periodic Table as dopants according to specific target applications. However, the detailed mechanism underlying the property tuning is not easily revealed by experimental analysis alone because defects or dopants are difficult to investigate with spectroscopic tools. In this respect, the first-principles method is a powerful tool to reveal the atomistic origin of dopant-related properties. While many theoretical studies successfully describe the behaviors of dopant in ZnO, systematic and complete studies have been limited to only a few elements. In this wok, we systematically calculate the single-element doping property on ZnO for 61 elements in the Periodic Table. Using an in-house automation code, we calculate thousands of dopant configurations and screened the configurations with low-lying formation energies. The automation code successfully and efficiently yields formation energy profiles with high accuracy by combining conventional density functional theory and hybrid functional method. The results of well-known dopants such as Li, Al, and N show good agreements with former studies. On the other hand, we also identified several interesting dopant configurations that have not been reported as far as we are aware. To examine the utility of dopant for electronic applications, we identified major dopant configurations and computed net carrier concentrations at 300 K. In addition, the localized magnetic moments of major dopant configurations are analyzed for spintronic applications. From the result, we suggest several new candidate elements for dopants for electrode and magnetic applications.
2:45 PM - CM7.3.02
Machine Learned Approximations to Density Functional Theory Hamiltonians—Towards High-Throughput Screening of Electronic Structure and Transport in Materials
Ganesh Hegde 1 , R. Chris Bowen 1
1 Advanced Logic Lab, Samsung Semiconductor Inc, Austin, Texas, United States
Show AbstractDensity Functional Theory (DFT)-based electronic structure calculations are an indispensable tool in computational material science.
Industrial applications often require high-throughput screening of the electronic structure of material systems for variations in system size, boundary conditions, geometry and strain. It is well known that DFT electronic structure and transport calculations scale poorly with system size. Even the so-called Order(N) DFT approaches do not scale to the extent required for time-sensitive electronic structure screening applications.
We present results from our recent work on direct machine learning of DFT Hamiltonians. In showing that approximating DFT Hamiltonians accurately by direct learning is feasible, we also make the case that this approach is preferable to existing semi-empirical approaches to the problem.
The technique we have proposed requires little manual intervention or arbitrary model parameters and can be applied to any material system or geometry for quick-yet-accurate predictions of DFT Hamiltonians. We also discuss open questions and challenges that need to be addressed to integrate this technique into existing genomic approaches to materials discovery.
3:00 PM - CM7.3.03
High-Throughput Detection and Data Mining of Coordination Environments in Oxides
David Waroquiers 1 , Xavier Gonze 1 , Gian-Marco Rignanese 1 , Cathrin Welker-Nieuwoudt 2 , Frank Rosowski 3 , Michael Goebel 2 , Stephan Schenk 2 , Peter Degelmann 2 , Rute Andre 2 , Robert Glaum 4 , Geoffroy Hautier 1
1 , Université catholique de Louvain, Louvain-la-Neuve Belgium, 2 , BASF, Ludwigshafen Germany, 3 , BasCat (UniCat BASF JointLab), Berlin Germany, 4 , University of Bonn, Bonn Germany
Show AbstractCoordination environments (e.g, tetrahedra and octahedra) are powerful descriptors of the structure of a solid. An automatic and robust detection of these environment is an important step towards the data mining of large materials databases (experimental or theoretical) currently available to chemists and materials scientists.
We present here a tool to automatically identify coordination environments in a given crystal structure solely based on geometrical considerations. Distortions are taken into account and our approach allows the description of an atom's neighborhood as a mixture of several local environments.
After outlining our algorithm, we illustrate the approach by presenting a statistical analysis of coordination environments for about 8,000 oxides from the Inorganic Crystal Structure Database. We present the local environment distribution for the most common cations in oxides and discuss their chemical origin. Our work complements the often empirical knowledge of the solid state chemists and materials scientists with a statistically sound large scale analysis.
3:15 PM - CM7.3.04
Making Materials Data Discoverable, Accessible, Interoperable and Reusable—A Focus on High-Throughput Experimental Materials Science
Zachary Trautt 1 , Nam Nguyen 1 , Martin Green 1
1 , National Institute of Standards and Technology, Gaithersburg, Maryland, United States
Show AbstractThere is increasing recognition of the need to improve the discovery, access, interoperability, and reusability of scholarly data, which is also a central aspect of the Materials Genome Initiative (MGI). Tremendous progress has been made in the five years since the launch of the MGI, specifically in the area of data discovery and access through the creation of materials resource registries and materials data repositories. However, significant challenges remain in improving the interoperability and reusability of materials data. We will give a brief overview of NIST efforts in creating materials resource registries and materials data repositories, followed by a detailed overview of our efforts to facilitate interoperable data curation, exchange, and dissemination within high-throughput experimental materials science.
4:30 PM - *CM7.3.05
A Data Platform for Materials Research by Information Integration
Yibin Xu 1
1 , National Institute for Materials Science, Tsukuba Japan
Show AbstractIn addition to experiment, theory and computation, data-driven research is becoming be the fourth method of materials science. The new approach of materials development starts from analyzing a large, comprehensive and systematic data set, by statistical or machine learning methods, discover the relationships between materials process, structure and properties, and then design and optimize the material component and structure in order to obtain the properties required. Materials data and databases are the fundamental of this new approach. During the past tens of years, National Institute for Materials Science has developed a federal materials database system MatNavi, which contains more than 10 materials databases covering the basic and engineering properties of inorganic materials, polymers and structure materials. We have successfully set up a process of data collection and verification from literatures, calculation and experiments, and developed a series of database techniques to fit various types of materials data.
In 2015, a Japanese national project Materials Research by Information Integration Initiative (MI2I) has been launched. As a basis of this project, a new data platform integrating materials data, data analysis tools and HPC environment has been constructed. This system supports high throughput computation to generate data, provides API access to MatNavi databases and allows users to bring their own data and merge with MatNavi data for data mining. In the presentation, the MI2I data platform and some research results on thermal management materials design by machine learning will be introduced.
5:00 PM - *CM7.3.06
Data-Driven Routes to New Insight into Materials Properties
Claudia Draxl 1 2
1 Physics, Humboldt University of Berlin, Berlin Germany, 2 Theory, Fritz Haber Institute of the Max Planck Society, Berlin Germany
Show AbstractAb initio computational high-throughput studies are producing data with an exponential growth rate. Typically, only a small fraction of its content is finally published, while most of the results on the quantum-mechanical many-body problem are thrown away. Keeping the data may be considered an unnecessary big-data problem. However, it could also be considered a chance – the chance to learn about physical properties and processes.
How to exploit the wealth of information, inherently inside the materials data, hopefully extracting unprecedented insight? On the one hand, new tools need to be developed for exploring similarities among materials and their properties, and finding out trends and anomalies. These tools comprise approaches of data-analytics, like machine-learning, compressed sensing, and alike. But there are other factors that need to be addressed for this new branch of materials research to be successful. This concerns issues of accuracy, error bars, and comparability of data.
The NOMAD (Novel Materials Discovery) Laboratory – a European Center of Excellence [1] tackles all these questions. It starts from the NoMaD Repository [2], which promotes the idea of open access and sharing of materials data. At present, this repository contains input and output files of more than 3 Mio. calculations. It currently supports about 30 different electronic-structure and force-field codes. This large collection of materials data opens an avenue towards explorations and discoveries. To do so, the first step is to convert the data to make them unbiased with respect to the underlying code. Also critical is to estimate their error bars [3,4]. Based on this, the NOMAD Laboratory creates a Materials Encyclopedia and develops big-data analytics tools for materials science.
I will demonstrate examples how data analytics combined with domain-specific knowledge can lead to new scientific insight and discuss the question whether we can find hitherto unknown correlations and new equations based on big data. These concern classification problems [5], but also the prediction of materials properties, like the structural data of binary and ternary alloys [6].
[1] NOMAD Center of Excellence, funded by the EU within HORIZON2020: http://nomad-CoE.eu
[2] The Novel Materials Discovery (NoMaD) Repository: https://nomad-repository.eu
[3] K. Lejaeghere et al., Reproducibility in density-functional theory calculations of solids, Science 351, aad3000 (2016).
[4] C. Carbogno, K. Thygesen, et al., to be published.
[5] L. Ghiringhelli, J. Vybiral, S. V. Levchenko, C. Draxl, and M. Scheffler, Big Data of Materials Science - Critical Role of the Descriptor, Phys. Rev. Lett. 114, 105503 (2015).
[6] B. Hoock, S. Rigamonti, L. Ghiringhelli, M. Scheffler, and C. Draxl, preprint.
5:30 PM - CM7.3.07
Multi-Fidelity Information Fusion for Materials Informatics
Ghanshyam Pilania 1 , James Gubernatis 1 , Turab Lookman 1
1 , Los Alamos National Laboratory, Los Alamos, New Mexico, United States
Show AbstractGiven that accurate first principles computations or measurements of a wide range of materials’ properties are time consuming and resource intensive, use of statistical learning methods to analyze accurate chemical trends at a lower computational cost and provide insights into the fundamental structure-property relationships of materials is of great interest. In this talk, we will discuss our recent work demonstrating the use of a multi-fidelity information fusion method for cost-effective and reliable property predictions in high throughput chemical space explorations.
Taking electronic bandgaps of 600 elpasolite compounds as an example target property dataset, we will show that the variable-fidelity fusion method can efficiently combines many inexpensive lower accuracy computations of bandgaps (i.e., carried out using local- or semi-local density functionals) with fewer expensive higher accuracy computations (i.e., self-interaction corrected bandgaps) to predict bandgaps whose accuracies are comparable to those produced by the higher accuracy calculations alone. We also note that the presented statistical learning method is not restricted to bandgaps or electronic structure methods and extends the utility of high throughput high-fidelity property predictions in a significant way.
References:
[1] G. Pilania, J. E. Gubernatis, T. Lookman, Submitted to Phys. Rev. X, under review, (2016).
5:45 PM - CM7.3.08
Materials Synthesis Insights Guided by Data Mining and First-Principles Calculations
Edward Kim 1 , Kevin Huang 1 , Adam Saunders 2 , Andrew McCallum 2 , Gerbrand Ceder 3 , Elsa Olivetti 1
1 , Massachusetts Institute of Technology, Cambridge, Massachusetts, United States, 2 , University of Massachusetts Amherst, Amherst, Massachusetts, United States, 3 , University of California, Berkeley, Berkeley, California, United States
Show AbstractA materials synthesis database, constructed from text-mining a corpus of 500,000+ journal articles, is used to reveal key linkages between synthesis parameters and resulting materials properties. By combining machine learning algorithms with first-principles computed thermochemical properties (e.g., phase stability), we present quantitative boundaries and correlations regarding the determination of end-product material properties. This study focuses on understanding synthesis-property relations across several transition metal oxide systems (e.g., driving synthesis conditions for titania nanotube formation) and linking these relations to known synthesis mechanisms.
CM7.4: Poster Session: High-Throughput Screening in Materials Design
Session Chairs
Thursday AM, April 20, 2017
Sheraton, Third Level, Phoenix Ballroom
9:00 PM - CM7.4.01
Data Mining in Ti-6Al-4V Alloy Design
Yibin Xu 1 , Lei Fang 1 , Satoshi Emura 1 , Donald Shih 2
1 , National Institute for Materials Science, Tsukuba, Ibaraki, Japan, 2 , Boeing Research & Technology, St. Louis, Montana, United States
Show AbstractTi-6Al-4V is an important and common engineering titanium alloy in aerospace. But it is difficult to predict its mechanical properties from microstructures or vice versa due to the complex nonlinear relationship between them. Data mining is a useful tool to extract meaningful information hidden in data. As a data mining methodology, decision tree was used to explore the relationship between property and microstructure of Ti-6Al-4V alloy in this study.
There are three main microstructural morphologies for Ti-6Al-4V alloy: bimodal, equiaxed, and lamellar. Based on the knowledge of Ti-6Al-4V alloy research, primary alpha grain size, volume fraction of primary alpha phase, prior beta grain size, alpha colony size and alpha lath width were selected as microstructure features, and tensile yield strength, ultimate tensile strength, elongation, reduction of area and high cycle fatigue strength as mechanical properties. Microstructure features and properties were grouped to input and output data sets accordingly to predict outputs. The prediction was conducted from microstructure to property or from property to microstructure.
Three decision trees were studied in Ti-6Al-4V alloy design, which are simple regression tree, ensemble regression tree and compact ensemble regression tree. The collected raw data were standardized to exclude outliers. To avoid too complex regression tree, pruning was performed to treat tree depth and data population in each leaf. By comparing prediction scatter and calculating cross-validation error, it was found that the compact ensemble regression tree models have the best performance, the simple decision trees obtain the smallest model size, and some ensemble decision tree models present better results than the simple tree models. In addition, both ensemble regression tree and compact ensemble regression tree methods are insensitive to outliers and deviations, and they are harder to produce overfitting than the simple regression tree method.
9:00 PM - CM7.4.02
Molecular Dynamics Simulations of Substitutional Diffusion
Xiaowang Zhou 1 , Reese E. Jones 1 , Jacob Gruber 2
1 , Sandia National Labs, Livermore, California, United States, 2 , Drexel University, Philadelphia, Pennsylvania, United States
Show AbstractIn atomistic simulations, diffusion energy barriers are usually calculated for each atomic jump path using a nudged elastic band method. Practical materials often involve thousands of distinct atomic jump paths that are not known a priori. Hence, it is often preferred to determine an overall diffusion energy barrier and an overall pre-exponential factor from the Arrhenius equation constructed through molecular dynamics simulations of mean square displacement of the diffusion species at different temperatures. This approach has been well established for interstitial diffusion, but not for substitutional diffusion at the same confidence. Using In0.1Ga0.9N as an example, we have identified conditions where molecular dynamics simulations can be used to calculate highly converged Arrhenius plots for substitutional alloys. This may enable many complex diffusion problems to be easily and reliably studied in the future using molecular dynamics, provided that moderate computing resources are available.
Acknowledgement: Sandia National Laboratories is a multi-mission laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the US Department of Energy's National Nuclear Security Administration under contract DE-AC04-94AL85000. The project is supported by a laboratory directed research and development (LDRD) project (project #180899).
9:00 PM - CM7.4.03
Genetically Tunable M13 Phage Films Utilizing Evaporating Sessile Droplet Techniques
Erik Alberts 1 , Chris Warner 1 , Eftihia Barnes 1 , Kevin Pilkiewicz 1 , Edward Perkins 1 , Aimee Poda 1
1 , U.S. Army Corps Engineering Research and Development Center ERDC, Vicksburg, Mississippi, United States
Show AbstractNature provides strategies for materials synthesis which rely on evolutionary approaches for directed assembly. Building blocks can be assembled for materials synthesis from the information stored within the genetic code. Bacteriophage has great potential as a building block for functional materials due to the tunability of the viral particle via genetic engineering. The unique self-assembly properties of phage provide opportunities for controlled material synthesis at the nano- and micro- scales. In this work, we have characterized the self-assembling properties of two strains of genetically modified M13 phage which exhibit distinct pattering and film structure. We examined the use of evaporating sessile-drops in fabricating phage films which have the potential to bypass scale-up, reproducibility, and film heterogeneity challenges. The coffee-ring effect, where suspended particles in an evaporating droplet accumulate at the pinned contact line, provides a unique opportunity to form biomaterials with genetically-tunable functional properties. Films were fabricated utilizing modified M13 phage, each which express unique surface chemistries. The formation of coffee-ring in drop-cast films was examined as well as self-assembly of phage within the film across a range of temperatures and concentrations. Suppression of the coffee ring effect was observed in both strains of M13, likely as a result of an increase in evaporation rate of phage droplets at elevated temperatures. We also observe repeatable and dramatically different internal morphologies between genetic constructs, with charged phage assembling into ordered structures and the more neutral forming a crosslinked mat. Fabrication of biomaterials using genetically tunable bacteriophage in controllable sessile droplets provide a platform approach for accelerated material development.
9:00 PM - CM7.4.04
Combinatorial Synthesis of New High-Dielectric Constant Film Materials for Nanoelectronics
Takahiro Nagata 1 2 , Sung-Gi Ri 3 1 , Kenichiro Takahashi 3 1 , Yoshifumi Tsunekawa 3 1 , Setsu Suzuki 3 1 , Toyohiro Chikyow 1
1 , NIMS, Tsukuba Japan, 2 , JST-PRESTO, Kawaguchi Japan, 3 , COMET Inc., Tsukuba Japan
Show AbstractThe development of high-dielectric constant thin film materials is essential for future active and passive nanoelectronics devices. Recently we developed high-dielectric constant thin film materials for thin film capacitor and gate dielectrics by combinatorial synthesis.
In the case of the passive nanoelectronics devices, we focused on the power device application. Compound semiconducter based power devices such as SiC require high-temperature operational passive devices such as a capacitor. SiC based active devices are capable of operating over 250°C. For module and system level advancement, the high-temperature operational passive should be developed urgently. In the casse of capacitors, current available capacitors can efficiently function up to 175°C only. Therefore, a thin-film capacitor that can be operated at high-temperatures and integrated monolithically in the proximity of active device should be developed. The BaTiO3 based relaxor ferroelectrics wiht a high dielectric constant (> 200) and free of hazardous elements, are promising candidates. Among the BaTiO3 based relaxor ferroelectrics, we have chosen x[BaTiO3]-(1-x)[Bi(Mg2/3Nb1/3)O3] (BT-BMN) due to its high dielectric constant and temperature stability in the bulk form. However, a thin-film process of this ceramic is not available yet. Moreover, the control of Bi composition, which affects on dielectric constant and ferroelectricity strongly, is challenging due to high volatility of Bi. In this context, high throughput combinatorial synthesis method is effective in the fast optimization of the Bi composition. In this paper, we have developed a thin-film processing technology for BT-BMN films by employing combinatorial pulse laser deposition method.
The composition spread film with a linear variation of Bi from stoichiometric to 10 wt% Bi excess regions was formed by ablating stoichiometric and 10 wt% rich BT-BMN targets alternatively and employing a moving mask to form alternative thickness gradient layers. Scanning nonlinear dielectric microscopy revealed that the 7 wt% Bi excess region showed the disappearance of ferroelectricity and the high dielectric constant. The dielectric constant is close to 250 and the stability is <13% from room temperature to 400°C, which are very promising as a high-temperature dielectric medium.
In the presentation, we also briefly introduce the combinatorial synthesis of high-k gate materials for a Ge channel as the active nanoelectronics devices, which has been attracting a lot of attention as a replacement for the Si channel used in current Si-based metal-oxide-semiconductor (CMOS) devices. This is because the Ge channel has high electron and hole mobility, which lead to a higher drive current, and Ge has a narrower band gap than Si thus allowing supply voltage scaling. For these application, we have proposed the direct growth of (110) rutile TiO2 or non-oxide materials on (100) Ge substrates, and performed the combinatorial synthesis.
9:00 PM - CM7.4.05
Graph Theoretical Representation, Analysis and Synthesis of Amorphous Metal Oxide Networks
. Divya 1
1 Materials Science and Engineering, Indian Institute of Technology, Kanpur, Kanpur India
Show AbstractWith the advent of amorphous oxide semiconductors (AOS) like amorphous indium gallium zinc oxide (a-IGZO), the analysis and prediction of amorphous structures has regained importance, more so, since first principles based studies are being increasingly employed to explain device behavior. Negative bias illumination stress in a-IGZO thin film transistors is one such example. However, the amorphous atomic structure is complex and defect or dopant studies require each site to be modeled independently and this leads to significant computational costs. Therefore, a simplification in the representation of the amorphous oxide network is effected so that it may lead to identifying similar atomic sites. The amorphous network is visualized as a network of polyhedra. The polyhedra has at its center a cation with the bonded oxygen atoms at the vertices and it comprises the short range interactions characterized by bond lengths and bond angles. Based on a first principles study of 10 a-IGZO models containing 36 cations each, it was found that the 360 polyhedra of the a-IGZO models can actually be described with only ten polyhedral motifs. These polyhedra are then connected to each other via a shared vertex or an edge; face-sharing was not observed in these models. Graph theory is used to map this network using either a graph of cationic polyhedra as the nodes or a bipartite graph (composed of cations and anions as individual nodes), each of which is described using the respective adjacency matrix. The second nearest interactions are characterized by the degree of each vertex and each atomic site is now characterized by a polyhedron and network metrics; and hence, can be compared with same-element sites. The changes in network itself, are quantified as the composition changes, when varying the ratio of In:Ga:Zn in a-IGZO. For example, the average vertex connectivity of a pair of indium sites reflects on the continuity of overlap between the In-5s orbitals which compose the conduction band minimum in a-IGZO, which, in turn, affects the transport properties of the semiconductor. Thus, the long range interactions of the physical amorphous network are described by the graph metrics. Moreover, evolutionary algorithm in conjunction with this graph theoretic representation can be used to generate new amorphous models. Two parent graphs are chosen, split into two and then spliced. The new graph is then reverse-engineered to form an amorphous model which then undergoes ionic and volume relaxation in the framework of density functional theory. The resulting child is a new amorphous model, with its energy as the fitness criterion.
9:00 PM - CM7.4.06
The Predictive Semi-gSEM Models of Waterborne Acrylic Coatings under Multi-Factor Accelerated Weathering Test
Donghui Li 2 , Noah Tietsort 2 , Emily Morris 1 , Yiyang Sheng 3 , Adam Joselson 4 , Laura Bruckman 2 , Roger French 2 5
2 Department of Material Science and Engineering, Case Western Reserve University, Cleveland, Ohio, United States, 1 Department of Chemistry, Case Western Reserve University, Cleveland, Ohio, United States, 3 Department of Mathematics, Applied Mathematics and Statistics, Case Western Reserve University, Cleveland, Ohio, United States, 4 Department of Mechanical and Aerospace Engineering, Case Western Reserve University, Cleveland, Ohio, United States, 5 Department of Macromolecular Science and Engineering, Case Western Reserve University, Cleveland, Ohio, United States
Show AbstractDriven by the social and political awareness, coupled with the strict environmental legislation, waterborne coating plays an important role in the global coating market. The service life of waterborne acrylics is of importance to the waterborne exterior coating application. In this study, effect of photoreactivity of titanium dioxide (TiO2) pigments on the durability and lifetime of acrylic waterborne coatings was investigated. And predictive models, semi-gSEM (semi-supervised generalized structural equation modeling) models, were constructed from gloss and colorimetry responses. Two acrylic waterborne coatings with different amount of anatase TiO2 were exposed in outdoor exposure and accelerated exposures at periodic intervals. And the accelerated exposures are artificially-controlled conditions of temperature, humidity and solar irradiance as prescribed by ASTM G154 and G155. The physical and chemical degradation of coating were monitored by Fourier transform infrared spectroscopy (FTIR-ATR), colorimetry and glossmetry to evaluate the degradation of coatings. FTIR-ATR analysis indicated that chain scission is the main mechanism of polymer binder under all exposure conditions. Lastly, Predictive models, ftir as degradation mechanism and gloss\colorimetry as responses, showed that different degradation modes dominate the response change of samples in constant accelerated and cyclic accelerated exposure. The quantitative framework can be utilized for the cross-correlation of accelerated and real world exposures.
9:00 PM - CM7.4.07
Pycroscopy—An Open Source Approach for Analyzing and Storing Material Science Data
Suhas Somnath 1 , Chris Smith 1 , Stephen Jesse 1 , Rama Vasudevan 1 , Nouamane Laanait 1
1 , Oak Ridge National Laboratory, Oak Ridge, Tennessee, United States
Show AbstractMaterials science is undergoing profound changes, driven by experimental data that are rapidly growing in dimensionality and size, increased accessibility to high-performance computing (HPC) resources, and more sophisticated computer algorithms than ever before. These changes are most pronounced in the functional imaging of materials. However, the softwares supplied with the instruments such as microscopes are typically very expensive, do not provide access to advanced or user-defined data analysis routines, and store experimental data in proprietary formats. Furthermore, these proprietary software and data formats not only impede data analysis but also hinder continued research and instrument development, especially in the age of “big data”. Therefore, moving to the forefront of data-intensive materials research requires general and unified data curation and analysis platforms that are HPC-ready and open source.
We have developed a platform called Pycroscopy that uses an open-source approach for analyzing and storing data. Pycroscopy is freely available via popular software repositories, and therefore lifts any financial burden for handling data. Pycroscopy uses an open and hierarchical data format (HDF) that can be interrogated using any programming language and scales well from kilobyte to terabyte sized datasets, and can readily be used in HPC environments unlike proprietary data formats. More crucially, Pycroscopy uses a universal data format that is curation-ready and therefore both meets the guidelines for data sharing issued to federally funded agencies and satisfies the implementation of digital data management as outlined by the United States Department of Energy. This instrument-independent data format has also greatly simplified the correlation of data acquired from multiple instruments, necessary for comprehensive studies of materials. Unlike many other open-source packages that focus on analytical or processing routines specific to an instrument, the general definition of the Pycroscopy data format can be readily adopted for different microscopy techniques. Furthermore, the generality of Pycroscopy provides material scientists access to a vast and growing library of community-driven data processing and analysis routines that far exceed those provided by instrument manufacturers and are desperately needed in the age of big data. In summary, Pycroscopy can greatly accelerate materials research and discovery through the realms of big, deep, and smart data.
This research was conducted at the Center for Nanophase Materials Sciences, which is sponsored at Oak Ridge National Laboratory by the Scientific User Facilities Division, Office of Basic Energy Sciences, U.S. Department of Energy.
9:00 PM - CM7.4.08
High-Throughput Study of Defect Energetics and Proton Transport in Cubic Perovskites
Janakiraman Balachandran 1 , Jilai Ding 2 1 , Lianshan Lin 1 , Yongqiang Cheng 1 , Raymond Unocic 1 , Nazanin Bassiri-Gharb 2 , Gabriel Veith 1 , Craig Bridges 1 , Weiju Ren 1 , Panchapakesan Ganesh 1
1 , Oak Ridge National Laboratory, Oak Ridge, Tennessee, United States, 2 Materials Science and Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States
Show AbstractTargeted design of new materials that can rapidly transport protons can enable the deveopment of proton conducting solid oxide fuel cells (P-SOFC) with higher efficiency and better reliability. However, proton transport in solid oxide materials is a complex phenomena that depends on the formation, migration and interaction energies of various defects (such as acceptor dopants, oxygen vacancies and proton interstitials). Understanding these energetics and their dependency on other variables such as strain, doping and local structure becomes the necessary first step towards targeted design of new materials.
Acceptor doped perovskites such as Y doped BaZrO3 have been previously shown to exhibit high proton conductivity. However, a lack of fundamental understanding on why certain pervoskite/dopant combination are better at conducting protons than others have hampered our ability to modify and improve their transport properties. To address this issue, we develop ab initio models of for Y doped BaZrO3 to study defect formation, migration and interaction energies for various dopant concentrations. These models are systematically benchmarked by a wide range of experimental characterization tools such as inelastic neutron scattering (INS), kelvin probe force microscopy (KPFM), atom probe tomography (APT) and electron microscopy (STEM).
These benchmarked models are then integrated with a massively scalable high throughput workflow to study defect energetics (comprising of acceptor dopants, oxygen vacancies and proton interstitials) on more than hundred cubic perovskite systems with different dopant atoms. The data obtained from these high throughput workflows are analyzed employing machine learning approaches to identify correlations between these defect energetics with local chemical environment, structural distortion and defect transition levels. This analysis enables us to obtain valuable insights and trends that can accelerate the design and development of new materials with improved proton transport properties.
This work was sponsored by Laboratory Directed Research and Development Program of Oak Ridge National Laboratory (ORNL). Computations were performed at NERSC and OLCF supercomputing facilities. Neutron experiments were performed at Spallation Neutron Source (SNS) at ORNL. Transport and microscopy measurements were performed at Center for Nanophase Materials Sciences (CNMS) at ORNL.
9:00 PM - CM7.4.09
Influence of Strain and Doping on Local Material Structure and Proton Transport Properties in Disordered Oxides
Janakiraman Balachandran 1 , Yongqiang Cheng 1 , Niina Jalarvo 1 , Gabriel Veith 1 , Craig Bridges 1 , Panchapakesan Ganesh 1
1 , Oak Ridge National Laboratory, Oak Ridge, Tennessee, United States
Show AbstractSolid materials that selectively transport protons provides are critical to develop high efficiency electrochemical devices. Recently various disordered oxide materials such as La28−xW4+xO54+δ (disordered fluorite) and La1−xBa1+xGaO4− x/2 (orthorhombic structure with disordered GaO4 tetrahedral motifs) that possess different local environments for same stoichiometry have shown to exhibit superior proton transport properties. However a lack of fundamental understanding on the intrinsic correlations between the local chemical/structural environment, proton dynamics and proton transport creates an impediment to improvize upon these materials. In this work, we combine ab intio modeling with inelastic (INS) and quasi-elastic (QENS) neutron scattering experiments to understand how the local chemical and lattice environments influences the vibrational dynamics and transport of protons in these material systems.
In this work, we sample a large number of configurations in a high throughput fashion to identify energetically favorable configurations for these disordered materials. Ab initio based phonon and molecular dynamics simulations performed on these disordered structural configurations enable us to interpret the neutron scattering data and identify the influence of local structure on proton dynamics and transport properties.
The insights obtained this analysis helped us to develop simple descriptors based on lattice and chemical structure that correlates well with the change in vibrational dynamics and in turn proton transport. These descriptor based models are interfaced with highthroughput workflows to explore how doping and strain modifies local environment and in turn the vibrational dynamics and transport of protons in disordered oxide materials.
Acknowledgement
This work was sponsored by Laboratory Directed Research and Development Program of Oak Ridge National Laboratory (ORNL). Computations were performed at NERSC and OLCF supercomputing facilities. Neutron experiments were performed at Spallation Neutron Source (SNS) at ORNL. Transport and microscopy measurements were performed at Center for Nanophase Materials Sciences (CNMS) at ORNL.
9:00 PM - CM7.4.10
Critical Current by Design through Large-Scale Simulations
Andreas Glatz 1 2
1 Materials Science Division, Argonne National Laboratory, Lemont, Illinois, United States, 2 Department of Physics, Northern Illinois University, DeKalb, Illinois, United States
Show AbstractUnderstanding the dynamic behavior of vortex matter in complicated pinning landscapes is a major challenge for both fundamental science and energy applications. In particular, optimizing type, size and density of pinning centers can significantly enhance the critical current. Based on the time-dependent Ginzburg-Landau equation, we developed a numerical approach towards finding these optimal pinning configurations.
Performing large-scale simulations of the vortex dynamics [1,2], we analyzed a number of different inclusion types and found optimal pinning configurations corresponding to the largest critical current in the geometries under consideration [3]. In particularly, we studied the interplay between vortex-vortex and vortex-inclusion interactions in a system including nanorod and columnar defects. This system represents a superconducting tape irradiated by heavy ions at an angle. Our simulation results agree with several experimental results and predict how the observed critical current could be further increased [4,5].
References:
1. I. A. Sadovskyy, A. E. Koshelev, C. L. Phillips, D. A. Karpeev, A. Glatz, J. of Comp. Phys. 294, 639 (2015).
2. Carolyn L. Phillips, Tom Peterka, Dmitry Karpeyev, and Andreas Glatz, Phys. Rev. E 91, 023311 (2015).
3. A. E. Koshelev, I. A. Sadovskyy, C. L. Phillips, A. Glatz, Phys. Rev. B 93, 060508(R) (2016).
4. Ivan A. Sadovskyy, Ying Jia, Maxime Leroux, Jihwan Kwon, Hefei Hu, Lei Fang, Carlos Chaparro, Shaofei Zhu, Ulrich Welp, Jianmin Zuo, Venkat Selvamanickam, George W. Crabtree, Alexei E. Koshelev, Andreas Glatz, and Wai-Kwong Kwok, Advanced Materials 28, 4593 (2016).
5. I. A. Sadovskyy, A. E. Koshelev, A. Glatz, V. Ortalan, M. W. Rupich, M. Leroux, Phys. Rev. Applied 5, 014011 (2016).
Symposium Organizers
Carson Meredith, Georgia Inst of Technology
Sergei Kalinin, Oak Ridge National Laboratory
Momoji Kubo, Tohoku University
Artem Oganov, Skolkovo Institute of Science and Technology
Symposium Support
Applied Materials, Inc.
The Dow Chemical Company
Georgia Institute of Technology
CM7.5: Machine Learning in Materials Discovery and Development
Session Chairs
Anastassia Alexandrova
Artem Oganov
Alexander Tropsha
Thursday AM, April 20, 2017
PCC North, 100 Level, Room 124 B
9:00 AM - *CM7.5.01
Machine Learning for Materials Discovery: Low-LTC Compounds, Grain Boundaries, Superlattices and RNAs
Koji Tsuda 1
1 , The University of Tokyo, Kashiwa Japan
Show AbstractThe scientific process of discovering new knowledge is often
characterized as search through a space of candidate hypotheses.
Machine learning can accelerate the search by properly modeling the
data and determining which candidate on which to apply an experiment.
In many cases, experiments can be substituted by first principles
calculations. Automated search algorithms such as Bayesian
optimization and Monte Carlo tree search can be used to increase
efficiency in designing molecules and materials. I report successful
case studies in discovery of low LTC compounds, grain boundary
optimization, automated design of Si-Ge superlattices and RNA inverse
folding.
9:30 AM - CM7.5.02
A Semi-Supervised Feature Learning Mechanism for Microstructure Reconstruction and Optimal Design
Ruijin Cang 1 , Yi Ren 1
1 Mechanical and Aerospace Engineering, Arizona State University, Tempe, Arizona, United States
Show AbstractComputational material design often requires finding a solution in the structural space for optimal material properties of interest, a task that can be cast as a high-dimensional non-convex optimization problem. To enable tractable computation of this inverse design problem, one would need to identify a mapping between the structural space and a lower dimensional feature space, where optimal solutions can be searched more effectively.
Existing low-dimensional representations of material systems, such as geometric and statistical features, are often constructed based on the designer's intuition and knowledge about the material systems. Such handmade representations may not be available for the design of more complex material systems where multi-scale microstructural features exists. In this oral presentation, we propose a semi-supervised feature learning mechanism that automatically (1) identifies multi-scale features that are correlated with the properties of interest, and (2) derives a continuous low-dimensional feature space from where new material designs with smooth microstructural changes can be sampled and searched. We also test the reconstruction performance of the proposed method under four material systems, including Ti-6Al-4V alloy, Pb63-Sn37 alloy, Fontainebleau sandstone, and spherical colloids, to show that the proposed method preserves material features both image- and property-wise.
9:45 AM - CM7.5.03
Transforming Data into Knowledge using Machine Learning Applied to Experimental Data
John Perkins 1 , Andriy Zakutayev 1 , Caleb Phillips 1 , Jacob Hinkle 1 , Robert White 1 , Kristin Munch 1 , Marcus Schwarting 1
1 , NREL, Golden, Colorado, United States
Show AbstractTo effectively use high-throughput experiments (HTE) as envisioned in the Materials Genome Initiative (MGI) requires not only rapid experiments but also the ability to efficiently transform the resultant large sets into usable knowledge. To address this issue, we are test-driving machine learning on the specific challenge of rapid analysis of large amounts of x-ray diffraction (XRD) and composition data to create structure phase maps. Purpose specific data visualizations to compare measured XRD patterns with reference patterns in one to many or many to many views have been developed. In addition, clustering analysis is used to group measured XRD patterns with spectral clustering being our currently preferred algorithm due, in part, to the ability it provides to selectively balance xrd pattern similarity with compositional proximity. This analysis will be demonstrated by application to the Co-Zn-Ni-O material system. These composition-gradient thin-film libraries were grown on glass substrates using off-axis co-sputtering. XRD patterns were measured using a commercial (Bruker D-8 Discover) diffractometer equipped with a 2D detector and the cation composition was measured using x-ray fluorescence mapping. In the course of this work, we have also created a prototype project-specific analytics database, the NREL High Throughput Experiments for Materials database (HTEM DB). Project relevant experimental data and metadata harvested by the NREL Laboratory Information Management System (LIMS) is automatically extracted, pre-processed when appropriate, and then expressed in the HTEM DB which currently contains x-ray diffraction, compositional, electrical, optical and synthesis data for ~ 50,000 effectively different samples. This data can be accessed via an interactive web interface, an applications programming interface (API) or a structured query language (SQL) interface. Select data visualization and machine learning based analytics tools have been implemented in unmix_xrd, a custom Python package, which is built upon the substantial open source machine learning resources available for Python in scikit-learn (scikit-learn.org). This analytics package can be run from the command line, from a Python notebook or from within any data analysis software capable of issuing a system command. Using this last approach, we have implemented a custom user-friendly analysis tool in Igor Pro, a commercial analysis program widely used at NREL. The overall result is a scientist-friendly extensible analysis environment with project specific machine learning and visualizations, which is now being put into practice within the MGI-oriented Center for Next Generation Materials by Design (www.cngmd-efrc.org), a DOE Energy Frontier Research Center.
10:00 AM - CM7.5.04
Connecting Structure and Kinetics—Using Machine Learning Representations to Model Disordered Materials
Ekin Cubuk 1 , Samuel Schoenholz 2 , Andrea J. Liu 3 , Efthimios Kaxiras 4
1 , Stanford University, Stanford, California, United States, 2 Google Brain, Google, Mountain View, California, United States, 3 Physics, University of Pennsylvania, Philadelphia, Pennsylvania, United States, 4 Physics, Harvard University, Cambridge, Massachusetts, United States
Show AbstractRealistic models of solids have complex atomistic configurations. Previous attempts at studying the influence of structure on dynamical properties have failed to be predictive, relying on simple descriptors like coordination number and bond orientational order. By using machine learning methods and a more comprehensive set of descriptors, we show that structural information can be used to predict dynamics a priori, even in disordered solids. We then use machine-learned representations as physical variables to study model systems for metallic glasses and network glass formers. We show that non-equilibrium dynamics and fragility can be modelled and predicted using only instantaneous structural information.
10:15 AM - CM7.5.05
A Computational Graph-Based Approach for Stochastic Reconstruction of Microstructures Using a Deep Learning Framework
Xiaolin Li 1 , Yichi Zhang 1 , He Zhao 1 , Zijiang Yang 1 , Yixing Wang 1 , L. Catherine Brinson 1 , Wei Chen 1
1 , Northwestern University, Evanston, Illinois, United States
Show AbstractIn developing advanced materials, design of microstructure has been recognized as an important intermediate component that bridges the gap between processing conditions and material properties. Instead of representing these microstructures in a deterministic way such as using 2D and 3D images, it is preferred to describe them in statistical means to capture the random nature of microstructures. With variation of nature in different material systems, researchers have developed different characterization approaches to statistically reveal the intrinsic information hidden in the microstructure. Among these approaches, correlation functions, especially two-point correlation function, has been regarded as one of the most complete ways to describe microstructures. While characterizing microstructure with two-point correlation functions is very efficient, the corresponding statistical reconstruction process is computationally costly. The stochastic optimization approaches, such as simulated annealing (SA) and Generic Algorithm (GA) for reconstruction requires a large amount of iterations, which makes it very difficult to scale up to large 2D and 3D realizations.
As the emerging deep learning technique keeps drawing more attentions within the last decade, the computer science and engineering community has dedicated efforts to developing hardware and software that facilitate the training of complex deep neural networks. Specifically, the training of deep neural networks requires efficient gradient calculations, as most of the solvers for deep learning are gradient-based. To fulfill this need, a novel network representation approach, computational graphs, and the associated GPU-accelerated frameworks (such as Caffe and Tensorflow) have been made available to the public. Even though it has been demonstrated by a lot of artificial intelligence tasks that this framework is superior in optimizing networks propagations, this framework has not been utilized by the material science community in material structure reconstruction.
In this work, we present a computational graph approach that represents the two-point correlation function with stacked convolutional layers and nonlinear activation functions. The forward propagation of the computational graph leads to a fast characterization of microstructure, while the backward propagation of the graph efficiently provides gradient information that is needed in the reconstruction optimization. The gradient is then fed into a constrained optimization algorithm to obtain statistically equivalent microstructure realizations. We also demonstrate that, this approach has a faster convergence speed than the stochastic search methods and it is potentially capable of scalding up to larger 2D or 3D reconstructions.
11:00 AM - *CM7.5.06
Active Learning in High-Throughput Diffraction of Combinatorial Libraries
Ichiro Takeuchi 1
1 , University of Maryland, College Park, Maryland, United States
Show AbstractWe have demonstrated active learning and autonomous control of high-throughput diffraction experiments for rapid delineation of structural phase distribution across combinatorial libraries.
X-ray diffraction is carried out on composition spread wafers mapping ternary phase diagrams of metallic alloys. A clustering algorithm is used to delineate rough structural phase distribution based on similarity of diffraction data. Crystal structure data from experimental and DFT databases are used to guide the clustering algorithm toward optimal results. Graph-based endmember extraction and labeling (GRENDEL) is used as the unsupervised machine learning algorithm for clustering. After each subsequent diffraction measurement, the obtained structure information is used to update knowledge of structural phase distributions, and the instrument is automatically directed to measure the next successive spot which is predicted to best improve the accuracy of the learned structural phase distributions. Active learning has been demonstrated with in-house diffractometers as well as at a synchrotron beamline. This work shows that equipment as complex as synchrotron end-station instrumentation can be successfully folded into an active learning loop. The results so far indicate that the number of measurements can be reduced by a factor as large as 10 in order to obtain the same clustering information. The work is funded by ONR and NIST, and it was carried out in collaboration with A. Gilad Kusne, A. Mehta, F. Ren, S. Curtarolo, and T. Gao.
11:30 AM - CM7.5.07
Machine Learning Approaches for Experiment-Guided Atomistic Structure Determination
Spencer Hills 1 , Eric Schwenker 1 2 , Fatih Sen 1 , Alper Kinaci 2 1 , Kendra Letchworth-Weaver 1 , Maria Chan 1
1 , Argonne National Laboratory, Argonne, Illinois, United States, 2 , Northwestern University, Evanston, Illinois, United States
Show AbstractThe knowledge of atomistic structures is paramount for understanding a variety of chemical and physical processes. Systems with reduced symmetry, such as nanostructures and defects, are particularly challenging for atomistic structure prediction due to the large number of possible configurations. Atomistic modeling with first principles density functional theory (DFT) or empirical potentials has been increasingly used, in conjunction with machine learning techniques, to tackle this problem. However, such an approach can sometimes be plagued by the existence of multiple solutions and lack of experimental validaton. Therefore, the incorporation of experimental data containing atomic-scale information, such as x-ray pair distribution function (PDF) and transmission electron microscopy (TEM), may enable the determination of atomic structures more efficiently and accurately in these cases. In this talk, we will discuss the use of single and multi-objective evolutionary and basin-hopping approaches for experimentally-guided atomistic structure determination. Examples include noble metal nanoclusters for catalysis applications, as well as grain boundaries and other solid-solid interfaces in photovoltaic and electrocatalytic systems. We show that the combined use of energetic and experimental information is significantly more effective and efficient in arriving at the target solution.
11:45 AM - CM7.5.08
Machine Learning towards Better Growth of Materials—A Test Case with Epitaxial Oxide Thin Films
Rama Vasudevan 1 , Steven Young 1 , Sergei Kalinin 1 , Robert Patton 1
1 , Oak Ridge National Laboratory, Oak Ridge, Tennessee, United States
Show AbstractThe use of machine learning in the materials research community is currently experiencing a surge of interest especially in the realm of first principles theory; however, such techniques have not been extensively deployed towards understanding and optimizing growth of new materials. Furthermore, the lack of databases of experimental data very often hampers a thorough search through the existing knowledge base, remaining essentially inaccessible beyond simple searches. Here, we present an approach incorporating text mining, crowd-sourcing and machine learning to collate all the necessary growth/functional property information available in the published literature on a particular material subset, to determine optimum growth conditions. We use the test case of epitaxial thin films of commonly studied complex oxides, providing a large enough sample for useful statistics. We show that through modification of an open source annotation tool (BRAT) combined with regular expression queries, it is possible to specifically mark functional properties and growth conditions from papers in the extant literature. Through crowd-sourcing, the specific functional property is then matched to growth conditions in each paper, populating a database with thousands of entries specifying the unique characteristics of the grown films. Subsequent analysis of the database entries allows for exploration of the parameter space in which growth is feasible, highlights outliers, and could potentially be used for prediction of growth conditions for similar compounds. Research sponsored by the Laboratory Directed Research and Development Program of Oak Ridge National Laboratory, managed by UT-Battelle, LLC, for the U. S. Department of Energy (SEED #8203, RKV, SVK, SY, RP). A portion of this research was conducted at the Center for Nanophase Materials Sciences, which is a DOE office of Science User Facility.
12:00 PM - CM7.5.09
Accelerated Discovery of Solar Fuels Materials by Integrating High Throughput Characterization and Machine Learning Techniques
Santosh Suram 1 , Lan Zhou 1 , Yexiang Xue 2 , Junwen Bai 6 , Ronan Le Bras 2 , Sean Fackler 3 , Walter Drisdell 3 , Alpha N'Diaye 3 , Apurva Mehta 4 , Junko Yano 3 , Robert Van Dover 5 , Carla Gomes 2 , John Gregoire 1
1 Joint Center for Artificial Photosynthesis, California Inst of Technology, Pasadena, California, United States, 2 Department of Computer Science, Cornell University, Ithaca, New York, United States, 6 Zhiyuan College, Shanghai Jiao Tong University, Shanghai China, 3 , Lawrence Berkeley National Laboratory, Berkeley, California, United States, 4 , SLAC National Accelerator Laboratory, Menlo Park, California, United States, 5 Department of Materials Science and Engineering, Cornell University, Ithaca, New York, United States
Show AbstractHigh-throughput (HiTp) materials discovery approaches have traditionally focused on identifying optimal materials by enabling rapid composition-property mapping. However, to unravel the complex composition-structure-property relationships that provide the framework for directed accelerated discovery efforts it is essential to rapidly construct materials genomes that include relevant characterization data. Crystal and electronic structure characterization play a significant role in establishing composition-structure-property relationships, especially in the case of semiconductor materials necessary for solar fuels applications. In the case of structural characterization using X-Ray Diffraction (XRD), while instrumentation for HiTp XRD has been established, automated data analysis has remained a bottleneck for rapid phase analysis. In this context, we shall present discovery of structure-property relationships in a V-Mn-Nb oxide system by combining HiTp optical analysis with automated phase mapping from HiTp XRD data using a novel factor decomposition algorithm (AgileFD) that incorporates physical properties of a phase diagram.
In the case of mixed cation mixed anion systems that crystallize in the same phase for a large compositional range, crystal structure characterization is insufficient and complimentary information from electronic structure characterization is necessary to unravel structure-property relationships. In this context, we shall present discovery of a photoabsorber in the La-Ta-O-N system by combining combinatorial X-ray absorption near edge spectroscopy (XANES) with XRD and HiTp optical analysis.
12:15 PM - CM7.5.10
Statistical Learning of Kinetic Monte Carlo Models for Complex Chemistry from Molecular Dynamics
Qian Yang 1 , Carlos Sing-Long 2 , Enze Chen 1 , Evan Reed 1
1 , Stanford University, Stanford, California, United States, 2 , Pontificia Universidad Catolica de Chile, Santiago Chile
Show AbstractWe propose a statistical learning framework for building predictive and simplified kinetic Monte Carlo (KMC) models of complex chemistry from atomistic molecular dynamics (MD) data. We apply this scale-bridging technique to a high temperature, high pressure system of liquid methane under conditions similar to shock compression, and show that the KMC model is able to simulate the molecular concentration of methane observed in the MD to within 10% error, at a fraction of the computational time, cost, and system complexity.
Our method first uses bond length and duration criteria to coarse grain from atomistic molecular dynamics data in phase space to molecules and elementary reactions. We then use maximum likelihood estimation to fit this coarse grained data to a KMC model that solves a corresponding chemical master equation. We show that the bond duration should be chosen systematically to minimize the error between kinetic Monte Carlo simulations and the corresponding molecular dynamics.
We then apply three different model reduction methods to our KMC model: removing rare reactions, solving an integer program, and solving a convex relaxation of the integer program using L1 regularization. In particular, we build intuition for how the solution of the computationally efficient L1 regularization method compares to integer programming and the physically motivated solution of removing rare reactions. For our liquid methane system, less than 15% of the approximately 2000 reactions observed from molecular dynamics are necessary to reproduce the molecular concentrations over time of methane. This reduced set of reactions enables better analysis of the important chemistry in this complex system.
Finally, we show how this methodology may be used to build predictive chemical reaction networks. In particular, we demonstrate how a KMC model learned using this framework from a molecular dynamics simulation of isobutane at comparable temperature and pressure can predict the time evolution of our methane system to within reasonable error.
CM7.6: Simulation and Experiment Applied to Materials Discovery I
Session Chairs
Thursday PM, April 20, 2017
PCC North, 100 Level, Room 124 B
2:30 PM - *CM7.6.01
Materials Informatics—Computer-Aided Design of Novel Materials with the Desired Electronic and Physical Properties
Olexandr Isayev 1 , Corey Oses 2 , Taylor Moot 3 , James Cahoon 3 , Stefano Curtarolo 2 , Alexander Tropsha 1
1 UNC Eshelman School of Pharmacy, University of North Carolina, Chapel Hill, North Carolina, United States, 2 Center for Materials Genomics, Duke University, Durham, North Carolina, United States, 3 Department of Chemistry, University of North Carolina, Chapel Hill, North Carolina, United States
Show AbstractHistorically, materials discovery has been driven by a laborious trial-and-error process. The growth of materials databases and emerging informatics approaches offer the opportunity to transform this practice into data- and knowledge-driven rational design—accelerating the discovery of novel materials with the desired properties.
Using data from the AFLOW repository of high-throughput ab-initio calculations for inorganic molecules, we have generated Quantitative Materials Structure-Property Relationship (QMSPR) models to predict three critical material properties, namely metal/insulator classification, bulk modulus, Fermi energy, and band gap energy. To enable these calculations, we have developed novel materials descriptors such as universal property-labelled fragments (PLMF).[1] We have established that the accuracy of predictions obtained with machine learning models matches that of GGA DFT functionals yet model development requires a minute fraction of computational time as compared to ab initio calculations. Notably, due to the representation of materials with PLMF, the QMSPR models are broadly applicable to virtually any stoichiometric inorganic materials. This representation also affords straightforward model interpretation in terms of simple heuristic design rules that could guide rational design of novel materials.
As a proof-of-concept study we have employed a materials informatics approach to identify a novel photocathode material for dye-sensitized solar cells (DSSCs). By conducting a virtual screening of 50,000 known inorganic compounds, we have selected lead titanate (PbTiO3), a perovskite, as the most promising photocathode material, which differs substantially from the traditionally used base elements or crystal structures. Subsequent investigations have provided firm experimental support for this hypothesis.
[1] O. Isayev, C. Oses, S. Curtarolo, A. Tropsha. Universal Fragment Descriptors for Predicting Electronic Properties of Inorganic Crystals. arXiv:1608.04782
3:00 PM - *CM7.6.02
Functional 2D and 3D Borides
Paul Robinson 1 , Anastassia Alexandrova 1
1 , University of California, Los Angeles, Los Angeles, California, United States
Show AbstractWe will show how the theory of chemical bonding can be adopted, translated, and extended toward materials chemistry, and provide understanding and predictive power for their macroscopic properties. Our bonding models for materials are built up from small cluster fragments, which can be characterized spectroscopically, and treated at high level of ab initio theory (i.e. not DFT). The latter aspect is crucial to gain insight into strongly correlated solids. We will illustrate the approach on borides of transition metals: 3D super-hard alloys, topological Kondo insulators with mixed valency, surface boride catalysts, and 2D borides and supported boron materials. We will show how the promiscuous and fluxional nature of the metal-boron bond (directional or delocalized, covalent or ionic, and ionic with B being an anion or cation) governs all these interesting properties.
3:30 PM - CM7.6.03
The Thermodynamic Scale of Inorganic Crystalline Metastability
Wenhao Sun 1 2 , Stephen Dacek 1 , Shyue Ping Ong 3 , Geoffroy Hautier 4 , Anubhav Jain 2 , William Richards 1 , Anthony Gamst 3 , Kristin Persson 2 , Gerbrand Ceder 2 1 5
1 , Massachusetts Institute of Technology, Cambridge, Massachusetts, United States, 2 , Lawrence Berkeley National Laboratory, Berkeley, California, United States, 3 , University of California, San Diego, San Diego, California, United States, 4 , Université catholique de Louvain, Louvain-la-neuve Belgium, 5 , UC Berkeley, Berkeley, California, United States
Show AbstractThe space of metastable materials offers promising new design opportunities for next-generation technological materials such as complex oxides, semiconductors, pharmaceuticals, steels and beyond. Although metastable phases are ubiquitous in both nature and technology, only a heuristic understanding of their underlying thermodynamics exists. Here we report a large-scale data-mining study of the Materials Project, a high-throughput database of DFT-calculated energetics of ICSD structures, to explicitly quantify the thermodynamic scale of metastability for 29,902 observed inorganic crystalline phases. We reveal the influence of chemistry and composition on the accessible thermodynamic range of crystalline metastability for polymorphic and phase-separating compounds, yielding new physical insights that can guide the design of novel metastable materials. We further assert that not all low-energy metastable compounds can necessarily be synthesized, and propose a principle of “remnant metastability” – that observable metastable crystalline phases are generally remnants of thermodynamic conditions where they were once the lowest free-energy phase.
3:45 PM - CM7.6.04
Coarse Grained Microstructure Field Modeling From Fine Grained Features in Continuous Fiber Composite Structures
Jeff Simmons 1 , Samuel Sherman 2 1 , Stephen Bricker 1 , Przybyla Craig 1
1 Materials and Manufacturing Directorate, Air Force Research Laboratory, Dayton, Ohio, United States, 2 , Universal Technology Corporation, Dayton, Ohio, United States
Show AbstractDescriptions of microstructures in materials are highly influenced by the observations that can be made with existing technologies, which leads to a characterization bias. Microstructures are typically described in terms of features that can be observed within a micrograph, such as particles, interfaces, or morphologies, with the tendency to build bigger microscopes to observe smaller features. A less explored regime is the large scale structures formed by patterns of microscopically observable features. Typical approaches involve homogeneous averages to estimate particle size distributions, morphologies, or correlations. But, the process of forming homogeneous averages, by the Law of Large Numbers, eliminates rare or anomalous events from the description. The approach we use is to develop a model of a `mesoscale microstructure,' based on some sort of expected behavior. In our case, we model continuous fibers as independent streamers in non-turbulent fluid flows and extract larger scale features such as local chirality, shearing, expanding, or contracting groups of fibers, borrowing from fluid dynamics. We can estimate velocity fields that would be present if it were a fluid and, from that, estimate the velocity gradient tensor. From the characteristic features of the velocity gradient tensor, we can identify these features on a small enough scale to be allow for identification of local anomalies. Examples will be given from the SiC/SiC ceramic composite system.
4:30 PM - *CM7.6.05
Conjugated Polymers Unraveled—A Poly(3-Hexylthiophene) Case Study
Nils Persson 1 , Ping Hsun Chu 1 , Michael Mcbride 1 , Elsa Reichmanis 1 , Martha Grover 1
1 , Georgia Tech, Atlanta, Georgia, United States
Show AbstractWhile semiconducting performance of a conjugated polymer system is highly dependent upon polymer molecular structure, molecular structure is not the only determinant of macroscale charge carrier mobility. The molecules must reliably organize into suitable interconnected assemblies that facilitate charge transport. From a morphological perspective, perfectly crystalline solids are expected to afford the most favorable charge transport performance; however, macroscale crystalline solids are rarely achieved in polymeric semiconductors. The materials are generally semicrystalline, where grain boundaries act as charge traps that limit mobility over a long range.
P3HT is perhaps the canonical semicrystalline conjugated polymer to investigate the mechanisms by which semiconducting polymers self-assemble into tightly packed, ordered structures that promote charge transport. Through the development of solution-based processes, we have deposited P3HT thin films from a base mobility of ~10-3 cm2 V-1 s-1 to 10-1 and beyond, without changing material chemistry, device architecture, or deposition method. Since the solution phase of P3HT is notoriously difficult to characterize, we find clues to the processes at play in solution through analysis of changes in the thin film microstructure. A large database containing uniformly specified information was assembled from research in the group, and included process parameters, quantitated AFM morphology, and carrier mobility. Mining of this data helped to uncover underlying relationships.
The story that emerges is that of quantifiable improvements in thin film microstructure through the rational design of solution-based aggregation processes. Polymer chains can be manipulated by solute-solvent interactions, by the application of ultrasonic irradiation, by exposure to low dose UV, or simply allowing them to age with time. These processes lead to increased conjugation lengths, greater orientational alignment and inter-grain connectivity, resulting in correlated increases in OFET device performance on a standardized platform.
5:00 PM - *CM7.6.06
Construction of Standardized Neural Network-Based Interatomic Models for Structure Prediction Acceleration
Alexey Kolmogorov 1
1 , Binghamton University, Binghamton, New York, United States
Show AbstractNeural networks (NNs) are proving to be attractive alternatives to traditional interatomic potentials. Being general and flexible learning machines, they show encouragingly accurate description of interatomic interactions in different elemental and multicomponent systems. In an effort to build a standardized library of NN models, we have developed a hierarchical training in which NNs for multicomponent systems are obtained by sequential training from the bottom up: first unaries, then binaries, and so on. The stratified procedure and a new automated data generation protocol implemented in MAISE have been used to produce sets of accurate NNs models consistent across a large block of chemical elements. Our tests indicate that the NN interaction models are efficient enough to accelerate ab initio prediction by orders of magnitude and reliable enough to identify overlooked stable materials.
5:30 PM - CM7.6.07
Structure Prediction for Sn2N2, a New Metastable Binary Nitride
Stephan Lany 1 , Elisabetta Arca 1 , Aaron Holder 1 , Chris Caskey 1 , Andriy Zakutayev 1
1 , NREL, Golden, Colorado, United States
Show AbstractRecent advances in theoretical structure prediction methods and high-throughput computational techniques are revolutionizing experimental discovery of the thermodynamically stable inorganic materials. Metastable materials represent a new frontier for these studies, since even simple binary non-ground state compounds of common elements may be awaiting discovery. However, there are significant research challenges related to non-equilibrium thin film synthesis and crystal structure predictions. An interesting example of a metastable material is Sn2N2, a mixed valence Sn(II)/Sn(IV) tin nitride, which was discovered only recently [1]. Despite being a simple binary nitride, Sn2N2 remained elusive due to its metastability relative to metallic Sn, N2, and Sn3N4. This metastability presents a challenge for computational structure prediction, as common ground state search strategies are guided by energy minimization, which will eventually lead to a phase-separated configuration (Sn+N2) instead of the desired Sn2N2 compound. Initial structure sampling has identified a number of candidate structures [1], but did not lead to an unequivocal assignment. We present the results of a new hybrid structure sampling approach predicting a bilayer structure with the possibility of polytypism.
[1] C.M. Caskey et al., J. Chem. Phys. 144, 144201 (2016).
5:45 PM - CM7.6.08
Automated Tools to Enable High-Throughput Calculations of Intrinsic Point-Defect Properties—Applications to Halide Perovskite Compounds
Danny Broberg 1 , Bharat Medasani 2 3 , Nils Zimmermann 2 , Andrew Canning 2 , Maciej Haranczyk 2 , Mark Asta 1 , Geoffroy Hautier 4
1 Department of Materials Science and Engineering, University of California, Berkeley, Berkeley, California, United States, 2 Computational Research Division, Lawrence Berkeley National Laboratory, Berkeley, California, United States, 3 Physical and Computational Sciences Directorate, Pacific Northwest National Laboratory, Richland, Washington, United States, 4 Institute of Condensed Matter and Nanosciences (IMCN), Universite Catholique de Louvain, Louvain-la-Neuve Belgium
Show AbstractDensity-functional-theory (DFT) based methods for the calculation of charged point-defect properties in semiconductors and insulators have been extensively developed over the past three decades. These efforts have led to advances in computational approaches for increasing the accuracy of such calculations, to correct for errors inherent in the predicted band gap, as well as the spurious defect interactions introduced by the use of periodic boundary conditions. In this talk we summarize the recent development of a computational tool, PyCDT, which automates the application of these methods, to enable high-throughput calculations of charged point-defect properties [1]. We describe the computational framework implemented in this tool, and demonstrate its application in the calculation of point-defect properties in inorganic halide-perovskite materials.
Halide perovskite materials have drawn significant recent attention as a result of the demonstration of their application as photoabsorbers in high efficiency solar cells [2]. Many outstanding challenges remain for widespread application of these materials, such as thermodynamic stability and the presence of Pb in the highest performing materials. Theoretical methods have been employed extensively to assess improved stability as well as search for favorable characteristics like absorption coefficient and effective mass [3]. In addition to these important bulk properties, intrinsic point defects are known to have a large influence on the efficiency of a solar cell device, as they lead to changes in the carrier concentrations and introduce electron/hole traps within the fundamental gap. While calculations of intrinsic defects have been carried out for several halide perovskite materials, the consideration of the trends in these properties over a broad class of the halide chemistries has not yet been undertaken. In this talk we focus specifically on results for charged defect calculations for the inorganic perovskite-based CsSnI3 and CsPbI3 compounds. Emphasis will be placed on assessing the reliable quantities and trends that can be derived from charged point-defect properties within the PyCDT framework. The lessons learned from this study provide a basis for scaling defect analysis to the screening of defect properties in halide perovskites more generally. This work was intellectually led by the Materials Project Center, supported by the Office of Basic Energy Sciences (BES) of the U.S. Department of Energy (DOE) under Grant No. EDCBEE.
[1] D. Broberg, B. Medasani, N. Zimmermann, A. Canning, M. Haranczyk, M. Asta, and G. Hautier. “PyCDT: A Python toolkit for modeling point defects in semiconductors and insulators” in preparation.
[2] A. Kojima, K. Teshima, Y. Shirai and T. Miyasaka, J. Am. Chem. Soc. (2009).
[3] W. J. Yin, T. Shi, and Y. Yan. "Unique properties of halide perovskites as possible origins of the superior solar cell performance." Advanced Materials (2014).
Symposium Organizers
Carson Meredith, Georgia Inst of Technology
Sergei Kalinin, Oak Ridge National Laboratory
Momoji Kubo, Tohoku University
Artem Oganov, Skolkovo Institute of Science and Technology
Symposium Support
Applied Materials, Inc.
The Dow Chemical Company
Georgia Institute of Technology
CM7.7: Simulation and Experiment Applied to Materials Discovery II
Session Chairs
Toyohiro Chikyow
Jason Hattrick-Simpers
Friday AM, April 21, 2017
PCC North, 100 Level, Room 124 B
9:00 AM - *CM7.7.01
Entropy Descriptors as the Key for Synthesizability
Stefano Curtarolo 1 , Cormac Toher 1 , Jose Plata 1 , Demet Usanmaz 1 , Pranab Sarker 1 , Eric Perim 1
1 , Duke University, Durham, North Carolina, United States
Show AbstractIn this presentation we give some examples on the success of entropy descriptors for rational prediction of new materials systems.
9:30 AM - CM7.7.02
Improve the Simulation Accuracy with Abandoned Information
Ying Zhang 1 , Chen Ling 1
1 , Toyota Research Institute of North America, Ann Arbor, Michigan, United States
Show AbstractSimulation is becoming more and more powerful in materials science, thanks to the rapidly improved computing capability and continuously developed methodology. However, for the foreseeable future, the simulations cannot completely replace the experiments which yield the definitive measurements of materials properties. A feasible target for simulation is to predict accurately enough to guide the experimental design. Unfortunately, in most cases the improvement of accuracy also implies higher computational cost. A crucial challenge that lies at the center of computational materials science is to develop methods for accurate prediction without significantly increasing computing cost.
In this talk we propose a new scheme that utilizes abandoned information, which is the information not being utilized in simulation, to improve the accuracy of the prediction. Our target is to reconcile the well-known “band gap discrepancy” in density functional theory, in which the computationally friendly LDA/GGA fails to correctly predict the gaps between valance and conduction band. By introducing supervised machine learning techniques including linear regression, LASSO and SVR, our scheme yields greatly improved prediction of the band gap values for binary semiconductors. Key steps in our scheme such as identifying features of abandoned information and avoiding the likelihood of overfitting will be discussed. Our results provide a general method that can potentially decrease the discrepancy between experiments and simulations for all kinds of materials properties.
9:45 AM - CM7.7.03
Holistic Computational Structure Screening of more than 12,000 Candidates for Solid Lithium-Ion Conductor Materials
Austin Sendek 1 , Qian Yang 1 , Karel-Alexander Duerloo 1 , Yi Cui 1 , Evan Reed 1
1 , Stanford University, Stanford, California, United States
Show AbstractWe present a new type of large-scale computational structure screening approach for identifying promising candidate materials for solid state electrolytes for lithium ion batteries that is capable of screening all known lithium containing solids. To be useful for batteries, high performance solid state electrolyte materials must satisfy many requirements at once, an optimization that is difficult to perform experimentally or with computationally expensive ab initio techniques. We first screen 12,831 lithium containing crystalline solids for those with higih structural and chemical stability, low electronic conductivity, and low cost. We then train a data-driven ionic conductivity classification model using logistic regression for identifying which candidate structures are likely to exhibit fast lithium conduction based on experimental measurements reported in the literature. This screening reduces the list of candidate materials from 12,831 down to 21 promising structures, few of which have been examined experimentally. We discover that none of our simple atomistic descriptor functions alone provide predictive power for ionic conductivity, but when we mine for the most descriptive combination of features we develop a multi-descriptor model with a useful degree of predictive power. We also find that screening for structural stability, electrochemical stability, and low electronic conductivity eliminates 92% of all candidate materials, and then screening for high ionic conductivity eliminates a further 93% of the remainder. Meanwhile, screening for ionic conductivity alone only eliminates 89% of all materials, suggesting that ionic conductivity is not the most restrictive contraint across the full materials space. Our screening utilizes structure and electronic information contained in the Materials Project database.
10:00 AM - CM7.7.04
High-Throughput Design of Two-Dimensional Electron Gas Systems Based on Perovskite Oxide Heterostructures
Kesong Yang 1
1 , University of California, San Diego, La Jolla, California, United States
Show Abstract
The perovskite-based oxide heterointerfaces between two wide-band-gap insulators such as LaAlO3 and SrTiO3 are attracting increasing interests because of their novel electronic properties such as the two-dimensional electron gas (2DEG) at the interface that have potential applications in the next-generation nanoelectronic devices. In this talk, we show that a group of combinatorial descriptors such as the polar character, lattice mismatch, band gap, and the band alignment between the perovskite-oxide-based band insulators and the SrTiO3 substrate, can be introduced to realize a high-throughput (HT) design of SrTiO3-based 2DEG systems using perovskite-oxide-oriented quantum materials database. By using these combinatorial descriptors, we have carried out a HT screening of all the polar perovskite compounds, uncovering 42 compounds of potential interests. Our approach, by defining materials descriptors solely based on the bulk materials properties, and by relying on the perovskite-oriented quantum materials repository, opens new avenues for the discovery of perovskite-oxide-based functional interface materials in a HT fashion.
10:15 AM - CM7.7.05
Supercomputer Post-K Project “Challenge of Basic Science” in Japan
Momoji Kubo 1
1 , Tohoku University, Sendai Japan
Show AbstractSupercomputer K was constructed on 2011 in Kobe, Japan. Supercomputer K contains over 80,000 nodes and 700,000 cores and its calculation speed is 1016 flops. Then, it won a championship in the supercomputer world ranking on June, 2011. However, in the recent supercomputer world ranking on June, 2016, supercomputer K ranks fifth. Then the ministry of education, culture, sports, science, and technology in Japan plans to construct new supercomputer post-K on 2020. The above ministry advertises “Post-K” projects on the development of Japanese-made simulation software for supercomputer Post-K. Then, our proposed project “Challenge of Basic Science - Exploring Extremes through Multi-Physics and Multi-Scale Simulations” was adopted on June, 2016 and started on August, 2016. In this symposium, we introduce our supercomputer Post-K project and discussed its impact on the material genome approach. In addition to the super-large scale and highly-accurate calculations, the high-throughput screening and statistical analysis are also important subjects in the supercomputer “Post-K” project. The high-throughput screening and statistical analysis required much amount of computational facilities and a large number of calculations. We also discuss how to use supercomputer “Post-K” for the material genome approach.
11:00 AM - *CM7.7.06
Exploration of Photo-Functional Materials Using Ab Initio Evolutionary Searching
Junjie Wang 1 2 , Naoto Umezawa 2 3 , Tomofumi Tada 1 , Hideo Hosono 1
1 , Materials Research Center for Element Strategy, Tokyo Institute of Technology, Yokohama Japan, 2 International Center for Materials Nanoarchitectonics (MANA), National Institute for Materials Sciences, Tsukuba, Ibaraki, Japan, 3 Center for Materials Research by Information Integration (CMI2), National Institute for Materials Science, Tsukuba, Ibaraki, Japan
Show AbstractUsing ab initio evolutionary searching, we demonstrated that mixed valence tin oxides (SnxOy) can be used as efficient solar light absorbers. Our computational study has identified new crystal structures for unknown phases as well as for previously reported phases, all of which possess layered structures bound by dispersion forces, and thus, they are classified as van der Waals (vdW) materials. Based on our comprehensive theoretical studies on the electronic structures and photo-absorption coefficients, we determined that these newly discovered SnxOy oxides show promise in photocatalytic H2 evolution and photovoltaic solar energy conversion. Interestingly, the band gaps of SnxOy materials exhibit a linear dependence on the interlayer distance, and their photo-absorption peaks cover a broad range of the solar spectrum. These results indicate the possibility of creating multi-junction solar cells simply by stacking layers of various SnxOy compositions upon one another.
Next, we present our attempts to determine the unresolved crystal structure of graphitic carbon nitride (g-C3N4) by evolutionary searching. We have theoretically revealed the crystal structures of g-C3N4 corresponding to different synthesis conditions: thermal condensation and salt melt synthesis (SMS). It was found that the most stable structure is a form that is distorted from the commonly accepted planar structure, and its band gap depends strongly on the distortion of the heptazine units. This finding leads to a mechanism for the previously observed temperature dependence of the band gap of g-C3N4. We also explored a series of highly complicated structures corresponding to the SMS condition. Notably, we determined that the previously reported sample is a compound consisting of C, N, H, Li, and Cl. From our careful analysis of the crystal structures of each composition, we identified C6N9H3-LiCl (Cmcm) as the most plausible crystal phase and got validated by X-ray diffraction pattern. A clear roadmap for synthesizing highly crystalline g-C3N4 was proposed based on a systematic investigation.
Finally, we present our recent work on the prediction of the structures of 2D MBx. Using the ab initio evolutionary search, we predicted a series of MBx structures, which consist of earth-abundant transition metals (M) and boron atoms in 2D. These structures are dramatically stabilized by the electron transfer from the metal atom; therefore, they show a much better stability than reported 2D boron sheets. In particular, the most stable structure presents stability with respect to the α boron sheet and well-known borides, and still exhibits a Dirac cone with massless Dirac Fermions. This work is the first report on the stability and electronic structure of 2D MBx materials. The high feasibility of experimental synthesis and novel electronic properties makes 2D MBx a promising new member for photoelectronic applications.
11:30 AM - CM7.7.07
A High-Throughput Experimental Approach to Study Transition Metal Ternary Chalcogenides
Ankita Bhutani 1 , Awadhesh Narayan 1 , James Eckstein 1 , Lucas Wagner 1 , Daniel Shoemaker 1
1 , University of Illinois at Urbana-Champaign, Urbana, Illinois, United States
Show AbstractIn order to capitalize computational modeling and predictions, fast and accurate experimental and characterization techniques are needed for quick validation. In this study, we investigate transition metal chalcogenides using microscopy and high-throughput experimental techniques, such as temperature and time-resolved in-situ x-ray diffraction, powered by computational predictions [1]. High-temperature in-situ x-ray diffraction accelerates materials discovery by allowing us to watch a chemical reaction in real time and identify new stable/metastable phases. It provides useful insights into the thermodynamics and kinetics of reactions. Transition metal chalcogenides are particularly interesting because of their understudied nature compared to oxides and d electron correlations which leads to various interesting properties like superconductivity, meta-magnetic metallic behavior, and quantum phase transitions. We screened 27 new empty ternary chalcogenide systems for novel stable phases with interesting electronic properties as suggested by DFT calculations in 6 months using both conventional and high-throughput synthesis. We studied ternary systems of the form XYZ, where the cation X = Ba, Ca, Sr, La, K, Bi, Pb; Y is a 3d transition metal; and Z = S or Se. We were able to integrate and synchronize the experimental and computational efforts together in a combinatorial approach to bridge the gap between computation and experiment.
1. Narayan, Awadhesh, Ankita Bhutani, Samantha Rubeck, James N. Eckstein, Daniel P. Shoemaker, and Lucas K. Wagner. “Computational and Experimental Investigation for New Transition Metal Selenides and Sulfides: The Importance of Experimental Verification for Stability.” Physical Review B 94, no. 4 (July 5, 2016): 45105. doi:10.1103/PhysRevB.94.045105.
11:45 AM - CM7.7.08
Fabrication and Characterization of High-Coercivity L10-FeNi Films Using a Combinatorial Sputtering Approach
Andreas Kaidatzis 1 , Giorgos Giannopoulos 1 , Vassilios Psycharis 1 , Dimitrios Niarchos 1
1 , NCSR Demokritos, Aghia Paraskevi Greece
Show AbstractL10-type magnetic compounds, including FeNi, possess promising technical magnetic properties of both high magnetization and large magnetocrystalline anisotropy energy and thus offer potential in replacing rare earth permanent magnets in some applications. L10-FeNi was first discovered in an iron meteorite, formed by the long-range thermal diffusion of Fe and Ni in an asteroid’s core over a period of 4.6 billion years [1]. It was first artificially made in the L10-type structure with a stoichiometry of Fe50Ni50 by neutron bombardment [2] and estimated the order-disorder transformation temperature to be around 3200C, which is very low compared to other L10-alloys. This results in very low diffusion of Fe and Ni atoms and makes the transformation extremely sluggish. This transformation can be enhanced either by the creation of vacancies, core-shell FeNi/L10-AuCu nanoparticles [3], or in the case of thin films by a strain mediated process [4].
Here we have employed a combinatorial sputtering process in order to study the conditions for fabricating the L10-FeNi phase and measure its magnetic properties. We have used Si(100) 100 mm wafers as substrates and deposited multilayers of the following type: wafer/Cr(10 nm)/Cu3Au(70 nm)/combi-CuAuNi/NiFe(40 nm), where combi-CuAuNi is a compositional spread layer of varying stoichiometry and thickness, deposited using combinatorial sputtering and serving as a seed layer for L10 NiFe growth. The rest of the layers have homogeneous composition and thickness. The final deposition of FeNi was done at a temperature between 200 and 2800C, by co-sputtering Fe and Ni to a stoichiometry of 50/50 at%. We perform magnetic properties mapping of the multilayer by means of high-throughput polar Kerr effect magnetometry and we find that the coercivity increases from approximately 0.3 kOe to 1 kOe as the Au content of the combinatorial seed layer decreases. High-throughput x-ray diffraction measurements map the structural properties of the stacks, indicating the presence of L10 NiFe. A thorough study of the structural and magnetic properties of the materials libraries will be presented.
ACKNOWLEDGEMENTS
Funding from the European Commission is acknowledged (Grant No. 686056-NOVAMAG and 691235-INAPEM).
REFERENCES
[1] J. Albertsen et al. Nature 273 (1978) 453
[2] L. Neel et al., J. Appl. Phys. 35 (1964) 873
[3] M. Gong et al., Chem. Mater., 2015 (22) 7795
[4] T. Kojima et al., J. Phys.: Cond. Mat. 26 (2014) 064207
12:00 PM - CM7.7.09
Assessment of Trends in the Hydrogen Evolution Reaction and CO2 Reduction on Metal Nanoparticles
Dominic Alfonso 1 , Douglas Kauffman 1
1 , National Energy Technology Laboratory, Pittsburgh, Pennsylvania, United States
Show Abstract
We have undertaken a large-scale screening based on density functional theory (DFT) calculations to analyze trends in the activity of Ag, Au, Cu, Ir, Ni, Pd, Pt and Rh nanoparticles to assess the competition between electrochemical CO2 reduction and hydrogen evolution reaction on these materials. We have looked at different particle size (n=13, 55, 147 and 309) to investigate its influence on the activity. Our preliminary results indicate that many of these materials are potentially active for CO2 reduction as the relevant COOH and CHO intermediates exhibit an abnormal adsorption behavior since their adsorption strengths do not show linear correlation with that of CO. In general, the adsorption of COOH and CHO is enhanced with respect to that of CO, compared to that on the packed (111) metal counterparts with the scaling relations predicted to vary with system size. However, the calculations also predict that these nanoparticles prefer to bind hydrogen. That is, these materials appear to be relatively more active for hydrogen evolution reaction since the H intermediate competes better than COOH and CHO for active sites. Complementary experimental efforts are underway to validate key predictions obtained from the DFT calculations.
12:15 PM - CM7.7.10
Cheminformatics-Inspired Materials Discovery Platform
Olexandr Isayev 1 , Alexander Tropsha 1
1 , University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, United States
Show AbstractThe Materials Genome Initiative is transforming Materials Science into a data-rich discipline. These developments open exciting opportunities for knowledge discovery in materials databases using informatics approaches to inform the rational design of novel materials with the desired physical and chemical properties. Statistical and data mining approaches have been successfully employed in both chemistry and biology leading to the development of cheminformatics and bioinformatics, respectively. However, until recently their application in materials science has been limited due to the lack of sufficient body of data.
In this work we showcase a pilot materials informatics web platform capable of (i) instantaneously query and retrieve the necessary material information in the desired form, (ii) identify, visualize and study important data patterns, and (iii) generate experimentally-testable hypotheses by building predictive machine learning (ML) models based on materials’ characteristics. Our computational approach relies on cheminformatics methodologies that one of our groups has developed and employed successfully to enable rational design of organic compounds with desired properties (e.g., drug candidates).
CM7.8: Simulation and Experiment Applied to Materials Discovery III
Session Chairs
Momoji Kubo
Carson Meredith
Friday PM, April 21, 2017
PCC North, 100 Level, Room 124 B
2:30 PM - *CM7.8.01
High Throughput Experimentation and Materials Informatics
Toyohiro Chikyow 1 2 , Takahiro Nagata 1 , Satoshi Itoh 2 , Kiyoyuki Terakura 2 , Keiji Ishibashi 3 , Setsu Suzuki 3
1 , MANA National Institute for Materials Science, Tsukuba Japan, 2 , MI2I National Institute for Materials Science, Tsukuba Japan, 3 , COMETInc, Tsukuba Japan
Show AbstractIt has passed more than 15 years since the modern combinatorial materials science appeared as a new tool for high throughput materials screening. In this decade, the combinatorial synthesis and related systems are improved and it becomes sophisticated ones. Especially full automated combinatorial sputtering system enhanced the materials screening speed and well-designed control system enables to expand the combinatorial sputtering users. We have been demonstrating new materials development and discovery in electronics by these tools.. However the most time consuming process has been the combinatorial measurements. All point measurements was the basic way and we have been focusing on how fast we could measure the points. At present, we found that materials informatics is going to change the basic methodology. From the beginning the materials informatics has also been expected to be an innovative tool to encourage to combine combinatorial screening, computation, simulation and data mining. In this case, the materials informatics was not used effectively for new materials discovery, but rather for data analysis. Recently due to the rapid progress in machine learning, the essential innovation in materials science is going to happen. The Materials Genome Initiative started in 2011 and a lot of new challenge were reported. In this challenge, the machine learning was applied to a combinatorial X-ray measurement and succeeded in reducing the measurement points without accuracy. In this talk, modern high throughput materials synthesis and characterization with combinatorial methodology are discussed in relation with materials informatics or MGI aspect.
3:00 PM - *CM7.8.02
A Materials Genome Approach to the Discovery of Novel Multi-Principal Component Alloys
Jonathan Bunn 1 2 , Benjamin Ruiz-Yi 1 2 , Travis Williams 1 2 , Logan Ward 3 , Fang Ren 4 , Apurva Mehta 4 , Christopher Wolverton 3 , Jason Hattrick-Simpers 1 2 5
1 Department of Chemical Engineering, University of South Carolina, Columbia, South Carolina, United States, 2 , South Carolina SmartStateTM Center for the Strategic Approaches to the Generation of Electricity Columbia, Columbia, South Carolina, United States, 3 Department of Materials Science and Engineering, Northwestern University, Evanston, Illinois, United States, 4 Stanford Synchrotron Radiation Lightsource, SLAC National Accelerator Laboratory, Menlo Park, California, United States, 5 National Institute of Standards and Technology Gaithersburg, Materials Measurement and Science Division, Gaithersburg, Maryland, United States
Show AbstractOver the past 10 years there has been a resurgent interest in the development of novel metallic alloys, largely spurred by the discovery of a new solid solution alloy class and the creation of data-centric rules for the discovery of metallic glasses. Since both types of alloy systems reside in large composition-processing-property parameter spaces their exploration is a combinatorial problem. However, successfully mapping alloy systems that can include more than 6 elements and all their combinations is not possible with traditional high-throughput experimental approaches. Here, I will discuss an approach that seeks to address the rational exploration of multi-principal component space by combining theory, experiment and data science. Our approach is to use insights from theory or scientifically guided data mining to identify the regions of parameter space most likely to yield technologically interesting materials. We then employ computationally guided high-throughput synthesis techniques to obtain composition spread samples that systematically probe composition and processing space; investigating not just the material put forward to be interesting but also those that are compositionally, and hence structurally, similar. Rapid structural and corrosion resistance studies are then performed using in situ synchrotron diffraction studies, yielding thousands of data sets containing information regarding the evolution of the alloy phase and corrosion products. Supervised and semi-supervised machine learning techniques are then used to automate the extraction of phase dynamics during annealing/oxidation studies. With this information, we are able to assess and update the models used to generate the initial lead materials and plan the next material system to study. In this talk, I will emphasize our recent work using these techniques to investigate phase stability in high entropy alloys and metallic glasses.
3:30 PM - CM7.8.03
Controlled Growth of 2D Materials via Simulation Guided Experiments
Kasra Momeni 1 , R A Vila 2 , Joshua Robinson 2 , Long-Qing Chen 2
1 Mechanical Engineering, Louisiana Tech University, Ruston, Louisiana, United States, 2 Materials Science and Engineering, The Pennsylvania State University, State College, Pennsylvania, United States
Show AbstractChemical Vapor Deposition (CVD) technique is one of the main techniques for the synthesis of 2D materials due to its versatility and relatively low cost. However, despite the success in the synthesis of various 2D materials using this technique, a reproducible and controlled growth is still challenging. Furthermore, any attempt to grow new materials involves many try-and-error experiments. This is due to the complex nature of the multi-physics involved and the coupling among them, e.g. heat transfer, fluid flow, and chemical reactions. To overcome this challenge, we developed a coupled multiphysics model of the CVD process for the growth of 2D materials, such as MoS2 and GaSe, which is utilized to provide guidance for experiments. The numerical framework for solution of this model is developed and tuned using a small set of test experiments. The optimal experimental configuration and conditions for growth of the 2D MoS2 nanofins are determined, which utilized as the guideline for the new set of experiments. The developed numerical model gives a better understanding of the fundamental physics governing the growth by enabling us to perform thought experiments, which might be hard or oven technically impossible and to measure quantities that are experimentally inaccessible. Furthermore, it circumvents many trial-and-error experimentations and thus significantly reduces the time and costs for growing high-quality 2D materials.
3:45 PM - CM7.8.04
Materials Informatics for Magnetic Properties
Hitoshi Fujii 1 , Tetsuya Fukushima 2 , Tamio Oguchi 1 3
1 Materials Research by Information Integration Initiative, National Institute for Materials Science, Tsukuba-shi, Ibaraki, Japan, 2 , Institute for NanoScience Design, Toyonaka, Osaka, Japan, 3 Department of Theoretical Nanotechnology, Institute of Scientific and Industrial Research, Ibaraki-shi, Osaka, Japan
Show AbstractIn the field of material science, “informatics” is a new and powerful tool not only to accelerate finding new materials with target properties but also to understand the origin of the properties. We have applied this informatics technique to magnetic materials, such as 3d transition metal alloys, Heusler alloys, and transition metal oxides to extract a set of physically meaningful parameters, i.e., descriptors, which determine their magnetic properties. In this study, we calculated the saturation magnetization of ferromagnetic 3d transition metal alloys, i.e., the Slater-Pauling (SP) curve, by using the AkaiKKR code which is based on the Korringa-Kohn-Rostoker coherent potential approximation (KKR-CPA) method. Our calculations successfully reproduced the experimental SP curve. By using LASSO (least absolute shrinkage and selection operator) method contained in a library of R package, we also performed regression analysis in order to find a set of descriptors.