JOM, Vol. 70, No.

9, 2018
[Link]
Ó 2018 This is a U.S. government work and its text is not subject to copyright protection in the United States; however, its text may be subject to foreign copyright protection

ICME - 10 YEARS LATER: SUCCESS AND CHALLENGES

Evolution of a Materials Data Infrastructure

1,3 2
JAMES A. WARREN and CHARLES H. WARD

1.—Material Measurement Laboratory, National Institute of Standards and Technology,

Gaithersburg, MD 20899, USA. 2.—Materials and Manufacturing Directorate, Air Force Research
Laboratory, Wright-Patterson AFB, OH 45433, USA. 3.—e-mail: [Link]@[Link]

The field of materials science and engineering is writing a new chapter in its
evolution, one of digitally empowered materials discovery, development, and
deployment. The 2008 Integrated Computational Materials Engineering
(ICME) study report helped usher in this paradigm shift, making a compelling
case and strong recommendations for an infrastructure supporting ICME that
would enable access to precompetitive materials data for both scientific and
engineering applications. With the launch of the Materials Genome Initiative
in 2011, which drew substantial inspiration from the ICME study, digital data
was highlighted as a core component of a Materials Innovation Infrastructure,
along with experimental and computational tools. Over the past 10 years, our
understanding of what it takes to provide accessible materials data has ma-
tured and rapid progress has been made in establishing a Materials Data
Infrastructure (MDI). We are learning that the MDI is essential to eliminating
the seams between experiment and computation by providing a means for
them to connect effortlessly. Additionally, the MDI is becoming an enabler,
allowing materials engineering to tie into a much broader model-based engi-
neering enterprise for product design.

done, through the tighter integration of computa-

INTRODUCTION
tion, experiment, and digital data (Fig. 1). The
The benefit of providing wide and easy access to subsequent MGI strategic plan explicitly called for a
materials data in digital form is not a new concept. Materials Data Infrastructure (MDI), and many of
The need was identified, and significant activity was the developments described here rose from this
undertaken as early as the 1980s.1,2 However momentum.7
visionary the early efforts were, they were ahead Fundamentally, an MDI is needed to manage
of their time as the technology available curtailed data and the associated metadata across the mate-
significant progress. Since then, several reports rials data lifecycle, even if it’s just a local instanti-
have continued to highlight the need for community ation, or node, for a research group or individual
access to materials data, culminating in the company. The local MDI can be networked, as
National Research Council report on Integrated needed, to be widely accessible via the Internet.
Computational Materials Engineering (ICME).3–6 Data are just the output of a scientific undertaking,
So what has changed over the past 10 years that and could include almost any type of digital infor-
has finally enabled significant progress in providing mation, whether it be from experiment or computa-
access to materials data? In short, a combination of tion. Metadata are nothing more, essentially, than
advancements in materials, computer, and data context about the data, and can include such things
technologies as well as new government policy and as the data’s provenance (the data’s creator, meth-
funding have spurred progress. A primary acceler- ods, location, time, associated models, software
ant for substantive action came in 2011 with the used, etc.) as well as quantitative characterizations
rollout of the US Materials Genome Initiative of the data, such as uncertainty, sensitivity, and
(MGI), with a goal of accelerated discovery, design, other hard numbers related to the quality of the
and deployment of new materials. MGI sought to information. The line between data and metadata
work across government to activate a paradigm depends on many factors, and the researchers who
shift in how materials discovery and design could be generate the data may not draw the line in the same

1652 (Published online June 11, 2018)

Evolution of a Materials Data Infrastructure 1653

worldwide to discover, access, and combine data

from ground- and space-based telescopes, maximiz-
ing re-use of high-value data and opening up new
opportunities for discovery. The VO approach points
the way forward for the materials science and
engineering community.
For the same reasons that an MDI is desirable,
infrastructures and supporting policies for research
data publication have risen in prominence. The
drivers for these changes came from several quar-
ters.11 Of course, one was the availability of the
Internet, which made the notion of sharing seem
trivial, even when it wasn’t. There were several
other reasons driven by policy, including a desire for
transparency in how government funds are spent,
as well as an acknowledgment that the products of
taxpayer-funded research should be made as
broadly available as practical.12 These pressures
led to cross-government mandates for research data
management plans for all funded work, as well as
infrastructures such as [Link], a data repository
with web access holding a wide variety of govern-
Fig. 1. Materials innovation infrastructure of the materials genome ment-owned data, including research data. These
initiative.
drivers were amplified by a so-called ‘‘reproducibil-
ity crisis’’ in science, particularly in psychology and
biology, where small sample sizes and other prob-
lems with statistical controls have made many
place as a subsequent consumer of the data. Data at results difficult to interpret.13–15
an early stage of the research process will likely The world-wide research community has
become metadata further along the materials life- responded in a number of ways to these global
cycle (e.g., a material’s composition). Regardless of drivers. Existing entities such as the Committee on
how the data and metadata are classified, the MDI Data for Science and Technology of the Interna-
is essential to manage this information. tional Council for Science (CODATA) joined with
Better management of data has many obvious new entities, such as the Research Data Alliance
benefits. Some of these have been around for as long (RDA), to address the broad challenges concerning
as researchers have been collecting data, while the sharing of research data at both a policy and an
others are relatively new. Advantages include bet- infrastructural level. The open science movement
ter reproducibility of the work, better continuity also gained significant prominence, and the Open
within labs as graduate students move on or Science Framework, developed by the Center for
professors retire, easier discovery and retrieval of Open Science, provides a platform that supports
data sets relevant to a line of research, and fusion of data-rich collaboration and publication. Scientific
disparate data sets to improve statistics and pre- communities convened to develop the so-called
dictive power for a given model. Perhaps the most ‘‘FAIR Principles’’ to guide data management: Find-
exciting opportunity is enabling new material dis- able, Accessible, Interoperable, and Reusable.16
coveries and insights through the application of These principles were meant to provide a basic set
data science techniques by researchers who may not of guidelines in the stewardship of research data.
even have been involved in the original data
acquisition.8–10 With all these benefits, the argu- THE MATERIALS DATA INFRASTRUCTURE
ment for an MDI becomes compelling.
Of course, none of these ideas are unique to Although the MGI strategic plan called for cre-
materials science and engineering.11 There are ation of an MDI, its definition was only recently
some extremely successful preceding examples more fully articulated through a National Science
including astronomy and genomics. The astronomy Foundation (NSF)-sponsored study conducted by
community has demonstrated the benefits of The Minerals, Metals and Materials Society (TMS),
research data management and sharing, and, Building a Materials Data Infrastructure.17 At its
importantly, has implemented the Virtual Observa- heart, the MDI is enabled by the existence of
tory (VO) infrastructure to facilitate this sharing. connected networks and supporting tools, which
By federating distributed data resources through an make its ultimate utility orders of magnitude more
internationally agreed set of data descriptors and compelling than the call for mere databases in the
data access protocols, the VO enables astronomers past. The TMS report defines the MDI as (Fig. 2):
1654 Warren and Ward

REPOSITORIES
Repositories provide the place where people can
house their data. Usually they are more than a
simple database as they often provide a host of
capabilities and an application programming inter-
face (API) that improve the utility of the data. In
fact, a best practice is to provide a modern API, e.g.,
Representational State Transfer API (RESTful),18
with a repository to facilitate query and data
extraction. There have been materials databases
available for decades, and they’ve been widely
accessible on the Internet since the late 1990s.
Institutions such as the National Institute of Stan-
dards and Technology (NIST) in the US and the
National Institute of Materials Science (NIMS) in
Japan are two government entities that have strong
mandates in this area. Resources such as the
Thermodynamics Research Center ([Link]
gov/) and MatNAVI ([Link]
[Link]) are exemplar efforts in this vein. Comple-
mentary to government efforts, professional soci-
eties like ASM International, nonprofit
Fig. 2. Pictorial representation of the materials data infrastructure. organizations such as the Cambridge Crystallo-
Reprinted from Ref. 17. graphic Data Centre, as well as companies such as
Granta Design have developed successful business
models around selling access to materials data.
The Materials Data Infrastructure (MDI) con- Most of the legacy efforts from the 1990s are
sists of three core digital components—repos- largely characterized by the delivery of very high-
itories, tools, and e-collaboration platforms—as quality, fundamental reference materials data that
well as the technology, policies, incentives, have been curated and have some guarantee of
standards, people, and activities necessary to quality. While this makes this information extremely
plan, acquire, process, analyze, store, share, valuable, the amount of available data is necessarily
reuse, and dispose of materials data. small as the data is usually hand-curated from the
literature. In contrast, the new repository infras-
Repositories allow for the storage, either locally tructure supporting research needs to meet a number
where the data was generated, or in the ‘‘cloud.’’ of additional demands including ease of use for
Tools include a host of capabilities to enable the nonexperts, assignment of a persistent identifier
capture, processing, curation, dissemination, (like a DOI, or digital object identifier, which enables
search, discovery, and analysis of data, while e- citation and better tracking of data provenance),
collaboration platforms allow these tools and data to broad accessibility, sophisticated searchability, APIs
be joined into a cohesive workflow, allowing for full that allow machine-to-machine interactions with
tracking of the data lifecycle, and enabling minimal human involvement, and the capacity to be
increased reuse of data by both the researcher that federated across multiple instances in geographically
generated the work as well as the wider research disparate locations. This host of requirements has
community. been implemented in a number of institutions,
The MDI is nascent and will continue to evolve for including more general purpose repositories such as
the foreseeable future. The snapshot provided the NIST Materials Data Repository ([Link]
herein will be likely out-of-date in a relatively short [Link]), as well as the Materials Data
time given the rapid pace of technology advance- Facility ([Link] sponsored
ment and adoption by the community, but the basic, by the Center for Hierarchical Materials Design (a
underlying, paradigm of repositories, tools, and e- NIST center of excellence), the Department of
collaboration platforms will persist. These ideas Energy (DOE)-funded Predictive Integrated Struc-
need not be distinct, but they should serve as a tural Materials Science (PRISMS) Materials Com-
useful way to frame the existing state of the mons ([Link] and several
infrastructure. Regardless, the ideal data infras- resources associated with the DOE Energy Materials
tructure is one that, like all good infrastructure, is Network ([Link]
invisible to the user unless it fails. A seamless MDI network/energy-materials-network).19–21 There are
would allow for the capture, storing, sharing, numerous other governmental, nonprofit, and pub-
searching, analysis, and learning from data spread lishing-based repositories such as Zenodo (https://
over multiple sites. [Link]/), Figshare ([Link] and
Evolution of a Materials Data Infrastructure 1655

Dryad ([Link] A most telling The registry is designed so that entries can be
example of the rising value of materials data is the added in minutes by the resource host (e.g., owner
emergence of for-profit institutions that provide data or curator), and, if an institution so desires, a local
services. For example, Citrine Informatics provides version of the Materials Resource Registry can be
free data repository services for those researchers instantiated onsite (the source of the software can
willing to share their data with others.22 Citrine’s be found on GitHub). The multiple instances of the
business model is to provide data analysis tools and Materials Resource Registry, coupled with a proto-
services to commercial customers using the vast col, such as the OpenArchive Initiative Protocol for
wealth of materials data they have accumulated. Metadata Harvesting, enables metadata from each
For more specific types of data, highly valuable instance of the Materials Resource Registry to be
and user-friendly resources have been constructed. harvested by the other registry instances, yielding
For example, a huge success for the materials data global awareness of available resources.
repository effort has been in the creation of Addressing the other end of the FAIR spectrum
databases with massive numbers of density func- (accessibility, interoperability, reusability) is a
tional theory (DFT) calculations, which can then be much more challenging proposition. Success will
mined to facilitate materials discovery. Some of the not be achieved for all materials data at once;
DFT repositories include the Materials Project, rather, each subdiscipline will have its own needs
AFLOWLIB, Open Quantum Molecular Database and challenges. Solutions will vary by technical
(OQMD), and the European efforts MAX (http:// area and will evolve toward relatively stable meta-
[Link]/) and NOMAD ([Link] data schemas only after a considerable shake-out
[Link]/).23–26 Similarly, NIST provides a period. For relatively simple systems, schemas
repository of interatomic potentials ([Link] already exist (e.g., the Crystallographic Information
[Link]/potentials/), and the NSF has sup- File (CIF) format used by crystallography28).
ported the Open Knowledgebase of Interatomic Others, such as a Physical Information File, have
Potentials (OpenKIM) ([Link] There been proposed for more universal instances.29 For a
are also databases for catalysts, polymer properties, general approach, NIST has developed the Materi-
phase data, electronic properties, with more on the als Data Curation System (MDCS), which attempts
way. to address the problem of how to better curate
materials data quite generally.30 To do this, the
curation system requires the subdiscipline to define
TOOLS
a schema that describes the type of data to be
The data found in these repositories represent an curated. NIST also maintains a schema library, so a
enormous improvement over the information that researcher interested in a specific area (e.g., x-ray
must be extracted by hand from a traditional scattering), can use an existing schema from the
archival publication, where the data is often library or fine-tune it to meet their own workflow.
encoded in a graph and not easily machine readable. There are a large number of early adopters of the
Sometimes, these data are provided as supporting MDCS, including the NextFlex manufacturing
material for an archival publication, in which case institute, the NanoMine project, the LightMat
the manuscript provides significant metadata about energy materials network, and the Shanghai Mate-
the data, but not in the most easily used form. rials Genome Institute.
To address issues around findability and accessi- Beyond efforts mentioned above, however, lie
bility, NIST has worked with the international larger questions about the role standards might
community to create the Materials Resource Reg- play in reducing the manifest complexities sur-
istry, an open source software system that allows for rounding data sharing. Indeed, in 1985, ASTM
the discovery of useful resources for the materials International established the E49 committee on
researcher.27 The registry allows organizations to ‘‘Computerization of Materials Property Data’’ that
make their resources findable by using a simple, focused on such issues, releasing proposed stan-
internationally agreed upon (vetted through the dards throughout the 1990s, but these standards
Research Data Alliance) schema to describe their have since been withdrawn as they were premature
resource. Based on a similar system used in the compared to the technological ability to practice
astronomy community to enable the Virtual Obser- FAIR principles.31 Similarly, the European Union
vatory, organizations can register metadata about has proposed a number of infrastructural efforts,
organizations, data collections, application pro- including approaches to materials data standards,
gramming interfaces, informational sites, and mate- but with the exception of a few subdisciplines there
rials software. Once the resource is registered, is still only weak adoption.32 A related and impor-
searches will discover the resource with a closer fit tant consideration is having a means of defining
to purpose than the typical internet search result. data quality, an essential element in the context of
The Registry also allows for federation by enabling reusability. At its core, data quality must be
multiple institutions to install an instance of the described through the pedigree and provenance
registry across multiple sites and allowing them to metadata associated with the data. Quantitative
communicate with one another. measures of quality include data validation (for
1656 Warren and Ward

computationally derived data), verification, uncer- from scientific instrumentation or simulation, track-
tainty, and sensitivity. These are currently loosely ing of research provenance, manual or automated
defined within materials science and engineering, data analysis, data search, and data publication
and best practices have not generally been devel- (either internally or externally). They provide an
oped for each element or, where developed, are not integration platform through which both reposito-
in widespread use.11 ries and tools can be accessed. These types of
The challenges associated with standards adop- platforms are an advanced evolution of laboratory
tions are, of course, deeply linked with the type of management information systems (LIMS).
standard being considered, and it is likely that this Examples of these platforms include the Air Force
issue is at the heart of the widespread absence of Research Laboratory’s (AFRL) HyperThoughtTM, the
materials data standards. NIST is widely known for University of Michigan’s Materials Commons, the
the standards it helps to develop, and NIST has University of Illinois at Urbana-Champaign’s 4CeeD
taken an active role in both support of efforts like ([Link] Georgia Institute of
the MGI, and other efforts that encourage the Technology’s MATIN ([Link] and
sharing for information where it might benefit US Johns Hopkins University’s SciServer.20,33,34 Given
industry. Yet, NIST also recognizes that standards the similar objectives between these platforms, AFRL
are best developed using a voluntary, consensus- and NIST have sponsored a series of ‘‘hackathons’’
based approach. In this vein, NIST has vigorously meant to help developers reach consensus on common
pursued standards for materials data sharing, but approaches for communication protocols, tool inte-
in a bipartite manner: a lightweight standard for gration, and data exchange. As noted, NIST’s lead in
sharing high-level resources that are already developing schemas will be extremely helpful in pro-
approaching a consensus-based solution, and a viding common touchstones for describing and
low-level, research domain-driven sets of standards exchanging data. However, the prevalence of propri-
that will take years to evolve into a stable form. etary data formats used in scientific instrumentation
Recognizing the need for improvements in data remains a critical impediment to the initial capture of
sharing (as well as the rise in the use of data data. The difficulty in extracting experimental data,
science), in 2014, NIST created the Office for Data and even more often the supporting metadata, from
and Informatics, and brought in leadership from the instrumentation makes data capture a monumental
astronomy community to help apply ideas and best challenge for each investigator and a barrier for
practices to the problems in material, chemical, and widespread implementation of an MDI.
biological data.
Of course, NIST is only one player in the devel- CONCLUSION
opment of the MDI, and standards are far from the
The Internet, advanced computational power, and
only role that NIST plays. Many other parties have
modern data science techniques, including machine
complementary roles to play in driving the develop-
learning and artificial intelligence, are providing an
ment of the MDI, both through mandates for better
exciting new avenue for the materials researcher to
data management plans as a condition for funding,
complement the use of physics-based understanding
as well as developing specific programs to develop
of material behavior. Data-driven techniques have
the MDI. For example, the NSF has targeted efforts
already demonstrated the ability for a rapid and
to accelerate materials discovery by more tightly
efficient means to augment materials discovery,
integrating materials data with computation and
design, and deployment. Ample examples now per-
experiment through its Designing Materials to
meate the literature across all stages of the materials
Revolutionize and Engineer our Future (DMREF)
lifecycle, whether it be discovering new material
program, as well as the Data Infrastructure Build-
compositions by mining DFT databases, relating
ing Blocks program. The Department of Defense
processing conditions to microstructure evolution,
(DoD) and DOE each have numerous programs with
linking microstructure to material properties through
mission-specific performance requirements for
deep learning, or linking processing-structure-prop-
materials, which in turn foster the same types of
erty-performance relationships in component design,
data management challenges and include incentives
materials scientists and engineers now have a new
for creative solutions.
tool at their disposal. Specifically, application of
machine learning and artificial intelligence tools have
E-COLLABORATION PLATFORMS
been applied to complex materials data to predict new
Management of data starts at the source of data thermoelectric compounds, new aluminum alloy com-
generation and initial use. In order to facilitate this positions for additive manufacturing, and microstruc-
activity, several organizations have been developing ture development from additive manufacturing
e-collaboration platforms that use web-based tech- processing conditions.35–37 More recently, research-
nologies to manage the materials data lifecycle and ers have been combining these tools with robotics and
implement the FAIR principles. The goal of these in situ process monitoring and characterization to
platforms is to enable data workflow management build autonomous research apparatuses. Such an
through the facile ingestion of data and metadata example is the Autonomous Research System (ARES)
Evolution of a Materials Data Infrastructure 1657

developed by AFRL researchers to determine opti- 11. C.H. Ward, J.A. Warren, and R.A. Hanisch, Integr. Mater.
mum growth conditions for carbon nanotubes.38 Manuf. Innov. 3, 22 (2014). [Link]
014-0022-8.
These systems stand to dramatically increase the 12. J.P. Holdren, Increasing Access to the Results of Federally
productivity of materials research. It is clear we are Funded Scientific Research (Washington, DC: The White
evolving to a state where the neatly defined historical House, 2013). [Link]
boundaries between individual domains of the Mate- default/files/microsites/ostp/ostp_public_access_memo_2013.
pdf.
rials Innovation Infrastructure depicted in Fig. 1 are 13. Open Science Collaboration, Science 349, 6251 (2015).
being dissolved. Thus, it’s becoming increasingly [Link]
difficult to separate individual elements of an MDI; 14. C.G. Begley and L.M. Ellis, Nature 483, 531 (2012). https://
all components must function seamlessly as a tool for [Link]/10.1038/483531a.
the materials scientist and engineer. 15. M. Baker, Nature 533, 452 (2016). [Link]
533452a.
An important goal of the MDI is to create a 16. M.D. Wilkinson, et al., Sci. Data 3, 160018 (2016). https://
‘‘digital thread’’ through the material lifecycle, from [Link]/10.1038/sdata.2016.18.
discovery through recycling, where critical informa- 17. Building a Materials Data Infrastructure (Pittsburgh, PA:
tion is seamlessly passed digitally along the lifecycle The Minerals, Metals & Materials Society, 2017). https://
[Link]/10.7449/mdistudy_1.
in order to reduce barriers of information transfer, 18. R.T. Fielding (Ph.D. dissertation, University of California,
thus reducing the re-learning currently needed at Irvine, 2000).
each stage. In fact, a description of materials and 19. A. Dima, S. Bhaskarla, and C. Becker, et al., JOM 68, 2053
processes by digital means has at least helped lead (2016). [Link]
to a vision of replacing the traditional means of 20. B. Blaiszik, K. Chard, J. Pruyne, R. Ananthakrishnan, S.
Tuecke, and I. Foster, JOM 68, 2045 (2016). [Link]
describing materials in terms of document-based 10.1007/s11837-016-2001-3.
specifications and design curves by model-based 21. B. Puchala, G. Tarcea, E.A. Marquis, M. Hedstrom, H.V.
definitions of materials and processes.39 Achieving Jagadish, and J.E. Allison, JOM 68, 2035 (2016). https://
this vision would bring the field of materials science [Link]/10.1007/s11837-016-1998-7.
22. J. O’Mara, B. Meredig, and K. Michel, JOM 68, 2031
and engineering in line with other disciplines that (2016). [Link]
have adopted a paradigm of model-based engineer- 23. A. Jain, S.P. Ong, G. Hautier, W. Chen, W.D. Richards, S.
ing. However, a fully functional MDI is essential in Dacek, S. Cholia, D. Gunter, D. Skinner, G. Ceder, and
order to realize this vision. K.A. Persson, APL Mater. 1, 011002 (2013). [Link]
10.1063/1.4812323.
24. S. Curtarolo, W. Setyawan, S. Wang, J. Xue, K. Yang, R.H.
REFERENCES Taylor, L.J. Nelson, G.L.W. Hart, S. Sanvito, M. Buon-
giorno-Nardelli, N. Mingo, and O. Levy, Comput. Mater.
1. J.H. Westbrook and J.R. Rumble Jr, Computerized Mate- Sci. 58, 227 (2012). [Link]
rials Data Systems (Fairfield Glades: National Bureau of 2012.02.002.
Standards, 1983). 25. S. Curtarolo, W. Setyawan, G.L.W. Hart, M. Jahnatek,
2. J.S. Glassman and J.R. Rumble, eds., Computerization and R.V. Chepulskii, R.H. Taylor, S. Wang, J. Xue, K. Yang, O.
Networking of Materials Data Bases, ASTM STP 1017 Levy, M. Mehl, H.T. Stokes, D.O. Demchenko, and D.
(Philadelphia: American Society for Testing and Materials, Morgan, Comput. Mater. Sci. 58, 218 (2012). [Link]
1989). 10.1016/[Link].2012.02.005.
3. National Research Council, Materials Research to Meet 26. J.E. Saal, S. Kirklin, M. Aykol, B. Meredig, and C.
21st-Century Defense Needs (Washington: The National Wolverton, JOM 65, 1501 (2013). [Link]
Academies Press, 2003). [Link] s11837-013-0755-4.
4. National Research Council, Accelerating Technology 27. The NIST Materials Resource Registry. [Link]
Transition: Bridging the Valley of Death for Materials and gov/materials-resource-registry/.
Processes in Defense Systems (Washington: The National 28. S.R. Hall, F.H. Allen, and I.D. Brown, Acta Cryst. A47, 655 (1991).
Academies Press, 2003). [Link] 29. K.J. Michel and B. Meredig, MRS Bull. 41, 617 (2016).
5. International Assessment of Research and Development in [Link]
Simulation-Based Engineering and Science (Baltimore, MD: 30. The Materials Data Curation System. [Link]
World Technology Evaluation Center, Inc., 2009. http:// materials-data-curation-system/.
[Link]/sbes/SBES-GlobalFinalReport_BW.pdf. 31. J. Rumble, E-Materials Data. ASTM International. Stan-
6. National Research Council, Integrated Computational dardization News (2014). [Link]
Materials Engineering: A Transformational Discipline for zation-news/perspective/[Link].
Improved Competitiveness and National Security (Wash- 32. T. Austin, C. Bullough, D. Gagliardi, D. Leal, and M.
ington, DC: The National Academies Press, 2008). https:// Loveday, Int. J. Digit Curation 8, 5 (2013). [Link]
[Link]/10.17226/12199. 10.2218/ijdc.v8i1.245.
7. Materials Genome Initiative Strategic Plan, The White 33. M.D. Jacobsen, J.R. Fourman, K.M. Porter, E.A. Wirrig,
House, Washington, DC (2014) [Link] M.D. Benedict, B.J. Foster, and C.H. Ward, Integr. Mater
[Link]/sites/default/files/microsites/ostp/NSTC/mgi_ Manuf. Innov. 5, 12 (2016). [Link]
strategic_plan_-_dec_2014.pdf. 016-0055-2.
8. S.R. Kalidindi, Int. Mater. Rev. 60, 150 (2015). [Link] 34. N.S. Carey, T. Budavri, N. Daphalapurkar, and K.T. Ra-
org/10.1179/1743280414Y.0000000043. mesh, Integr. Mater. Manuf. Innov. 5, 7 (2016). [Link]
9. S.R. Kalidindi and M. de Graef, Annu. Rev. Mater. Res. 45, org/10.1186/s40192-016-0049-0.
171 (2015). [Link] 35. M.W. Gaultois, T.D. Sparks, C.K.H. Borg, R. Seshadri,
020844. W.D. Bonificio, and D.R. Clarke, Chem. Mater. 25, 2911
10. J. Hill, A. Mannodi-Kanakkithodi, R. Ramprasad, B. Mer- (2013). [Link]
edig, Computational Materials System Design, eds. D. 36. J.H. Martin, B.D. Yahata, J.M. Hundley, J.A. Mayer, T.A.
Shin, J. Saal (New York, NY: Springer, 2018). [Link] Schaedler, and T.M. Pollock, Nature 549, 365 (2017).
org/10.1007/978-3-319-68280-8_9. [Link]
1658 Warren and Ward

37. E. Popova, T.M. Rodgers, and X. Gong, et al., Integr. Mater. Comput, Mater. 2, 16031 (2016). [Link]
Manuf. Innov. 6, 54 (2017). [Link] npjcompumats.2016.31.
017-0088-1. 39. D.U. Furrer, D.M. Dimiduk, J.D. Cotton, and C.H. Ward,
38. P. Nikolaev, D. Hooper, F. Webber, R. Rao, K. Decker, M. Integr. Mater. Manuf. Innov. 6, 249 (2017). [Link]
Krein, J. Poleski, R. Barto, B. Maruyama, and N.P.J. 10.1007/s40192-017-0102-7.