Computational Discovery and Design of Emerging Materials: Session II
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS: Computational Materials Science and Engineering Committee
Program Organizers: Arunima Singh, Arizona State University; Houlong Zhuang, Arizona State University; Sugata Chowdhury, National Institute of Standards and Technology; Arun Kumar Mannodi Kanakkithodi, Purdue University
Monday 2:30 PM
February 24, 2020
Room: 32B
Location: San Diego Convention Ctr
Session Chair: Sugata Chowdhury, National Institute of Standards and Technology
2:30 PM Invited
Identification of 11 New Solid Lithium-ion Conductors with Promise for Batteries Using Data Science Approaches: Austin Sendek1; Evan Reed1; 1Stanford University
We discover 11 new crystalline solid materials with fast single crystal Li ion conductivity at room temperature, discovered through density functional theory simulations guided by machine learning-based methods. The discovery of new solid Li superionic conductors is of critical importance to the development of safe all-solid-state Li-ion batteries. With a predictive universal structure−property relationship for fast ion conduction not well understood, the search for new solid Li ion conductors has relied largely on trial-and-error computational and experimental searches over the last several decades. In this work, we perform a guided search of materials space with a machine learning (ML)-based prediction model for material selection and density functional theory molecular dynamics (DFT-MD) simulations for calculating ionic conductivity. These materials are screened from over 12 000 experimentally synthesized and characterized candidates with very diverse structures and compositions.
3:00 PM
Predicting Organic Ligands Mechanical Behavior with Deep Neural Network and Understanding the Mechanism: Weiyi Zhang1; Chengxi Yang1; Alan Fern2; Matthew Campbell2; P. Alex Greaney1; 1University of California, Riverside; 2Oregon State University
We have developed an automated system for designing organic ligands computationally with a decision tree. This approach formalizes a design space of molecules that is astronomically large and beyond brute force exploration by computer. In order to accelerate the search of this space we have tested machine learning methods for forecasting several different forms of the lingands‘ mechanical and kinematic behavior from four different methods for representing the molecules’ structure. Success at prediction has provided an opportunity to learn the structure-property relationships that give rise to specific dynamic behaviors such as mechano-isomerization and high stiffness. This was achieved by examining the gradients in the response of trained neural networks in order to identify the elements of structure that have the strongest correlation with the property of interest. The results present a formal approach for learning structure-property mechanisms in molecular systems such as metal organic frameworks.
3:20 PM
Haber–Bosch Reaction Mechanism and Kinetics on Highly Reactive Iron Surface and Hierarchical High-throughput in Silico Screening Catalyst Design: Qi An1; Alessandro Fortunelli2; William Goddard3; 1University of Nevada, Reno; 2CNR-ICCOM,ThC2-Lab, Consiglio Nazionale delle Ricerche; 3Caltech
To discover new alloy catalysts that dramatically improve the efficiency of the Haber–Bosch (HB) process for ammonia synthesis, we employed quantum mechanics to determine the reaction mechanism and free energy reaction barriers for the most highly reactive Fe(111) surfaces under experimental single crystal reaction conditions. We determined the rates for all 10 important surface reactions on both surfaces. Then we used this information in a full kinetic Monte Carlo simulation to determine the kinetics. We predict that the Turn Over Frequency (TOF) for Fe(111) surface is TOF =18.7 s−1 per 2 × 2 surface site for 1.5 Torr NH3 pressure in excellent agreement with experiment. To dramatically improve the efficiency of HB reactions on these highly reactive surfaces, we applied hierarchical high throughput screening approach to sift the promising doping elements on these two highly reactive surfaces. We found several promising dopants that can significantly improve the TOF.
3:40 PM
Machine-learning based Discovery of Novel Scintillator Chemistries: Anjana Talapatra1; Blas Uberuaga1; Chris Stanek1; Ghanshyam Pilania1; 1Los Alamos National Laboratory
Perovskites are among the most common compounds and are employed in electronics, photonics, and energy detection technology. Their versatility stems from their propensity to accommodate a very large number of elemental combinations. This large compositional space naturally lends itself to modern learning paradigms to discover new and improved perovskites. Of special interest are double-halide-perovskites with mixed anion chemistries, which are known to be efficient scintillators. Scintillators have wide-ranging applications from medical imaging to global security. Despite a pressing need for improved scintillators for these diverse applications, the discovery and design of new scintillator materials has historically relied on laborious, time-intensive, trial-and-error approaches yielding little physical insight and leaving a vast space of potentially revolutionary materials unexplored. This talk presents an adaptive design framework coupling high-throughput experiments, first-principles computations and machine learning to efficiently screen a large chemical space of potential scintillator chemistries to identify materials with better scintillation properties.
4:00 PM
A General Machine Learning Framework for Impurity Level Prediction in Semiconductors: Arun Kumar Mannodi Kanakkithodi1; Michael Toriyama1; Fatih Sen1; Michael Davis1; Robert Klie2; Maria Chan1; 1Argonne National Laboratory; 2University of Illinois Chicago
The quick and accurate prediction of the likelihood of formation of impurities and electronic levels created by them in semiconductors is critically important for optoelectronic and photovoltaic applications. In this work, we combine high-throughput density functional theory (DFT) and machine learning (ML) to develop general predictive models for the formation enthalpy and charge transition levels of impurities in two broad semiconductor classes: (a) ABX3 halide perovskites, and (b) group IV, III-V and II-VI semiconductors. Based on descriptors that represent any “semiconductor + impurity” combination in terms of elemental properties, coordination environment, and cheaper unit cell calculations, models are trained using methods like neural networks and random forest regression and are seen to be applicable to all possible compositions in the chemical space. The versatility of the machine learned-models provides an avenue to access the energetic and optoelectronic impact of any atomic impurity in any given semiconductor.
4:20 PM Break
4:40 PM Invited
High-Throughput Screening and Synthesis of Semiconductor Electrodes for Photocatalytic Water Splitting: Ismaila Dabo1; 1Pennsylvania State University
Solar energy is the most abundant energy source available to humankind, but this energy cannot be harnessed on demand due to the variability of sunlight. Artificial photosynthesis overcomes that variability through the photocatalytic storage of solar power into chemical fuels. Nevertheless, most of the stable photocatalysts rely on metal oxide semiconductors whose bandgap does not match the solar spectrum. This presentation will discuss the development of a computational-experimental protocol to understand, predict, and optimize visible-light-active materials that can split water into hydrogen and oxygen with a focus on solar compatibility using electronic-structure methods beyond density-functional theory [Timrov et al., Physical Review B 98, 085127 (2018)] and on electrochemical stability by exploiting quantum-continuum methods [Andreussi et al., Journal of Chemical Physics 136, 064102 (2012); Campbell et al., Physical Review B 95, 205308 (2017); Campbell et al., Physical Review Materials 3, 015404 (2019); Xiong et al., Physical Review Materials 3, 065801 (2019)].
5:10 PM
Machine Learning Guided Search for Single Phase High Entropy Oxides: Shruba Gangopadhyay1; Prasanna Balachandran1; 1University of Virginia
High entropy oxides (HEOs) are made up of (near) equi-molar solid solutions of four or more complex oxides (e.g., binary oxides) that forms a single phase, likely stabilized by configuration entropy. One of the intriguing characteristics of this materials class is that not all “parent” oxides have the same structure. Further, there are insights in the literature that indicate that the HEO is in a metastable state. This presents a formidable challenge for materials discovery because of the lack of a priori design rules for guiding experiments and the vast search space (~200,000) of possible oxide combinations. A data-driven machine learning (ML) approach will be discussed to accelerate the search for novel single phase HEOs. We constructed a dataset from surveying the literature and explored several ML strategies (unsupervised and supervised learning methods). New single phase HEOs are also predicted, which we recommend for experimental synthesis.
5:30 PM
Use of Atomistic-based Modeling and Materials Informatics to Design and Synthesize Ultra-thin Tunnel Junctions: Ridwan Sakidja1; Devon Romine1; Jagaran Acharya2; Ryan Goul2; Judy Wu2; 1Missouri State University; 2The University of Kansas
The combination of atomistic-based modeling and materials informatics can be used nowadays as powerful tools to design and synthesize a wide range of complex and novel structures. The modeling techniques allow us to provide a much deeper understanding on the synthesis mechanism and the dynamics within the structures. We applied these tools in conjunction with the experimental results to elucidate the detailed mechanisms of interfacial reactions and the formation of sub-nm amorphous alumina or crystalline MgO layer manufactured through Atomic Layer Deposition (ALD) process as a critical part of the ultra-thin tunnel barrier fabrication. The support from NSF-EPMD (Award No. 1809284) program is gratefully acknowledged.
5:50 PM
Designing High Glass Transition Temperature Polymers using Machine Learning: Chiho Kim1; Rohit Batra1; Lihua Chen1; Huan Tran1; Rampi Ramprasad1; 1Georgia Institute of Technology
Machine learning (ML) approaches have proven to be of great utility in modern materials innovation pipelines. ML models are trained on predetermined data and then used to make predictions for new cases. Active-learning is a paradigm in which ML models can direct the learning process itself through providing dynamic suggestions/queries for the next-best experiment. We demonstrate how an active-learning framework can aid in the discovery of polymers possessing high glass transition temperature (Tg). Starting from an initial small dataset, we use Gaussian process regression in conjunction with an active-learning framework to iteratively add candidate polymers to the training dataset. The active-learning workflow terminates once 10 polymers possessing Tg greater than a certain threshold temperature are selected. We statistically benchmark the performance of three strategies (exploitation, exploration, or balanced exploitation/exploration) for selection of the next-best experiment with respect to the discovery of high-Tg polymers for this particular demonstrative design challenge.