Embedded Files
Accelerated discovery of materials is essential to develop efficient materials necessary for rapid realization of several emerging technologies. While computational materials theory approaches have made tremendous progress in rapid prediction of materials, accelerated experimental approaches have typically lagged. This is primarily due to challenges in development of strategies for a) throughput-linked characterization of materials, b) efficient theory-experiment integration, c) accelerated screening/characterization, and d) application of machine learning algorithms to convert large datasets into actionable knowledge.
My contributions in the above aspects of accelerated materials discovery resulted in publications on various topics such as:
- Combinatorial materials synthesis of metals, metal oxides, and mixed anion materials.
- High-throughput optical, (photo)-electrochemical screening.
- Application of machine learning for rapid phase mapping, throughput-linked characterization of materials, quantification of nanoscale 3D crystal structure information from atom probe tomography (APT) data.
- Discovery of 12 new photoanodes for solar fuels devices, doubling the number of known photoanodes.
- Automated discovery of band-gap engineering.
In the following paragraphs, I provide a brief description of a selection of these projects that are likely to have a broad impact on accelerated materials discovery research.
Throughput-linked characterization of materials:
High-throughput screening techniques have enabled rapid measurement of application specific engineering property of materials. However, investigation of the fundamental mechanism responsible for materials performance typically requires time consuming, expensive materials characterization that can be performed only on select samples, creating a throughput-gap for materials discovery. To address this issue, we developed a machine learning approach based on information theory and multi-tree genetic programming to identify distinct composition-property relationships present in a combinatorial library that guide selection of samples for detailed characterization and enable rapid discovery of property-mechanism relationships. This methodology aided in discovery of the role of multiphase nanostructure for high oxygen evolution catalytic activity in a (Ni−Fe−Co−Ce)Ox library.
ACS combinatorial science 2015, 17, (4), 224-33.
Rapid discovery of solar fuels photoanodes via integrating theory and experiment:
Although several high-throughput computational studies have suggested candidate photoanode materials for solar fuels devices, experimental realization of these suggestions has not been very successful primarily due to limitations in the rapid prediction of electronic structure for vast material spaces, and typically off-stoichiometric composition of the optimal performance material. Using a VO4 motif based hypothesis, our theory collaborators calculated valence, conduction band positions with high accuracy for all thermodynamically stable compounds that possessed this motif. Compounds with relevant band positions were synthesized using combinatorial reactive sputtering, screened for band gap and photo-electrochemical activity using custom high-throughput setups. Using this integrated pipeline, we identified 15 solar fuels photoanode phases, of which only three were previously known. These 12 phases doubled the number of known photoanode phases. Combinatorial synthesis and screening approaches revealed the off-stoichiometric compositions of optimal materials, and also identified phases that are photo-electrochemically active but were not considered in the pipeline, providing a feedback mechanism for hypothesis driven theory-experiment integration based discovery of materials.
PNAS 2017 114: 3040-3043.
Rapid phase mapping:
Characterization of crystal structure is necessary to establish structure-property relationships, and also to serve as a critical link between theory and experiment. Further, accelerated characterization of phase diagrams in higher-order composition spaces is necessary given that many binary phase diagrams are known but few ternary or higher-order phase diagrams have been explored. While 103-104 XRD patterns can now be routinely captured per day at synchrotron facilities, automated phase analysis of large scale datasets has remained a challenge for over a decade due to challenges in encoding phase-diagram properties such as peak shifting, Gibbs’ phase rule etc. We demonstrated the first application of a scalable algorithm that models properties of a phase diagram to generate physically relevant basis patterns and phase concentration maps. Development of such automated materials characterization will significantly accelerate discovery of materials.
ACS Comb. Sci., 2017, 19 (1), pp 37–46.
Automated discovery of band gap engineering:
Band gap energy is the primary figure-of-merit for several light based applications. However, methodologies for rapid, automated estimation of band gap energies do not exist. To address these challenges, we use a high-throughput on-the-fly UV-Vis spectroscopy instrument to measure transmission, reflection spectra at a rate better than 1s per sample. Using constraints that mimic the judgment of an expert scientist, we developed a scalable and automated algorithm that reliably estimates band gap from optical spectra. An example of estimation of band gap for thousands of compositions in the Bi-V-Fe oxide system is shown in the adjacent figure. By integrating automated phase mapping and automated band gap estimation, we enabled automated discovery of band gap engineering.
ACS Comb. Sci., 2016, 18 (11), pp 682–688.
Google Sites
Report abuse