Cuerpos Sin Edad, Mentes Sin Tiempo - Deepak Chopra
Cuerpos Sin Edad, Mentes Sin Tiempo - Deepak Chopra
Cuerpos Sin Edad, Mentes Sin Tiempo - Deepak Chopra
Co-chairs:
Executive Chair:
This workshop, held in Mexico City, was co-led by the Mexican Ministry of Energy (SENER)
and the U.S. Department of Energy (US DOE) in partnership with
the Canadian Institute for Advanced Research (CIFAR).
Mission Innovation
January 2018
Contents
Executive Summary....................................................................................i
Introduction..............................................................................................1
4.1 Opportunities..................................................................................... 24
4.2 Benefits and Impact........................................................................... 25
4.3 Relationship to the Other Goals of the Materials
Acceleration Platform........................................................................ 26
4.4 Materials-Specific Challenges and Approaches .................................. 27
5.1 Opportunities..................................................................................... 38
5.2 Benefits and Impact........................................................................... 39
5.3 Relationship to the Other Goals of the Materials
Acceleration Platform........................................................................ 39
5.4 Materials-Specific Challenges and Approaches................................... 40
7.1 Opportunities..................................................................................... 61
7.2 Benefits and Impact .......................................................................... 63
7.3 Relationship to the Other Goals of the Materials
Acceleration Platform........................................................................ 63
7.4 Materials-Specific Challenges and Approaches .................................. 64
8.1 Opportunities..................................................................................... 72
8.2 Benefits and Impact .......................................................................... 73
8.3 Relationship to the Other Goals of the Materials
Acceleration Platform ....................................................................... 74
8.4 Materials-Specific Challenges and Approaches................................... 74
Conclusion..............................................................................................87
Acknowledgments....................................................................................93
Workshop Participants..............................................................................95
Acronyms ...............................................................................................99
Executive Summary
M
aterials are an essential element of advanced energy
technologies. Accelerating the discovery of new materials,
and the associated research required for maturing these
technologies into deployment, will require a multidisciplinary and
international effort that brings together a wide variety of individuals
working effectively across their specialties, as well as across sector
and political boundaries. It will also require a radical departure from
traditional forms of discovery.
1
Mission Innovation is a global initiative comprising 22 countries and the European
Union that share the goal of accelerating clean energy innovation.
i
acceleration platforms will unleash a “Moore’s law for scientific discovery” that will
speed up the discovery of materials at least by a factor of ten—from 20 years to 1
to 2 years. This will catalyze a transition from an Edisonian approach to scientific
discovery to an era of inverse design, where the desired property drives the rapid
exploration, with the aid of advanced computing and AI, of materials space and
the synthesis of targeted materials. The inverse design of materials allows for their
accelerated scale-up into installed technologies, accelerating energy technology
innovation. This, in turn, will benefit all seven Innovation Challenges of Mission
Innovation.
Workshop participants identified six key priority research areas that comprise the
MAP elements. These scientific and engineering challenges require the expertise
of multidisciplinary, international teams for their successful implementation. These
research areas include the development of the following tools, research activities
and infrastructure:
5. New methodologies for bridging the length and timescales associated with
materials simulation; and
ii
1 Introduction
M
aterials discovery “There are many paths to cheap
and development zero CO2 free energy, what we need
crosscut the entire to do is fund the wild scientists
energy technology portfolio, who are looking at the early stages
from energy generation of these problems…I think we will
and storage to delivery and get increased investments from
end use. Materials are the government and private sector, I
foundation of every clean know that the scientific possibilities
energy innovation: advanced are pretty incredible.”
batteries, solar cells,
low-energy semiconductors, – Bill Gates [1]
thermal storage, coatings,
and catalysts for the conversion, capture, and use of CO2. In short, new
materials constitute one of the cornerstones for the global transition
to a low-carbon future.
Figure 1.1 Virtuous cycle: Laboratory automation and data collection enable machine
learning algorithms and computational methods to be more effective at planning
experiments.
1
molecule is run through simulation, synthesis, and characterization, with synthetic
procedures taking from 10 to 20 years at a very high cost. Materials discovery and
development, however, are at the cusp of a transformational change that could
reduce the time to design, optimize, and discover new materials by at least 10
times, cutting it down to one or two years. This enormous opportunity — and the
associated technical challenges of realizing it — inspired the members of the global
Mission Innovation (MI) initiative to launch the Clean Energy Materials Innovation
Challenge. As a first step, international experts attended a technical workshop in
Mexico City on September 11-14, 2017 to identify and explore the research and
development (R&D) challenges, and the most promising breakthrough opportunities
for accelerating the materials discovery process, with a long-term view towards
2030 and beyond.
Workshop Participants
Workshop participants shared a common interest in accelerated materials innovation.
They represented a broad mix of expertise spanning advanced theory, applied physical
chemistry and materials sciences, advanced computing, machine learning, and robotics.
The workshop drew 133 attendees:
2
Computation and design, however, are “Although each [Mission Innovation]
only the first step in bringing novel member will decide on priorities
materials to market. Materials synthesis for its own expenditure, each
and characterization have yet to benefit challenge will be promoted through
from automation and accelerated learning international conferences and
on a large scale. Integrating synthesis workshops, and it is anticipated that
and characterization with advanced many cross-country projects will be
computing, machine learning, and robotics funded. The creativity and ingenuity
would automate the entire materials that will flow from these projects
discovery process. Finally, closing the will produce market-facing solutions
loop to create a virtuous cycle by using AI through private finance and the
to direct experimentation and simulation engagement of private companies.
for optimal learning would result in an It is critically important that the
accelerated, comprehensive, end-to-end appropriate members of the research
materials innovation platform (Figure 1.1). community are fully involved in
these developments over the coming
1.1 ABOUT MISSION years.”
INNOVATION AND THE
INNOVATION CHALLENGES – Sir David King [9]
Transforming traditional materials discovery pipelines into an integrated platform
requires commitments from governments, academic research institutions, large
industries, and capital providers [5]. This ambitious effort fits well in the framework
of MI Innovation Challenges.
MI is a global initiative of 22 countries and the European Union with the goal of
accelerating clean energy innovation. Participating countries have committed to
seek to double their governments’ investments in clean energy R&D over five years.
They have also agreed to encourage greater levels of private-sector investment
in transformative clean energy technologies and participate in an information
sharing system to promote transparency, engage stakeholders, spur identification
of collaborative opportunities, and provide the private sector with actionable
information. These efforts will increase the availability of the advanced technologies
that will accelerate the transition to a global energy mix that is clean, affordable,
and reliable.
3
MI was announced on November 30, 2015,
as world leaders came together at the
21st Conference of the Parties (COP21)
in Paris to undertake ambitious efforts
to combat climate change. Clean energy
innovations and disruptive technological
breakthroughs are an essential element
of broader efforts to significantly reduce
greenhouse gas (GHG) emissions to
limit the increase of the global average
temperature to less than 2°C [6,7].
Figure 1.3 (Top) Workshop attendees at the
plenary session on the first day. (Bottom) (From
A year later, in November 2016 at COP22 left to right) Dr. Paul Durrant, Dr. Horst Simon, Dr.
Alan Bernstein, and Dr. Mario Molina participate
in Marrakech, MI members launched in a panel moderated by Dr. Hermann Tribukait.
seven Innovation Challenges, which are
global calls to action aimed at accelerating
research, development, and demonstration
in specific technology areas where increased
international attention would make a
significant contribution to meeting the
goals of the challenges. The following
Innovation Challenges were selected and
developed through a collaborative process
in which policy and technical experts from
all MI members presented proposals and Figure 1.4 Through interactive workshop sessions,
exchanged information about national participants identify and discuss critical R&D
priorities and gaps.
needs and priorities:
4
6. Clean Energy Materials: To accelerate “Perhaps of greatest interest to
the exploration, discovery, and use the theoretical physics, physical
of new high-performance, low-cost chemistry, and materials science
clean energy materials. communities that are working
alongside the machine learning,
7. Affordable Heating and Cooling of
robotics, and next-generation
Buildings: To make low-carbon heating
computing communities is the
and cooling affordable for everyone.
challenge of developing clean energy
A voluntary coalition of participating MI materials. The goal is to provide
members, under the co-leadership of an integrated end-to-end materials
two to four countries, advances each innovation approach, or platform, to
Innovation Challenge through detailed deliver the mix of solutions…Mission
work programs. Innovation is intended to spur the
interest of the creative community in
Mexico proposed and now leads the Clean
what is now the most urgent series
Energy Materials Innovation Challenge, with
of demands facing humanity. The
the United States as co-lead. Participating
opportunities are, simply, immense.”
MI members include Australia, Canada,
Denmark, European Commission, Finland,
– Sir David King [9]
France, Germany, India, Italy, Republic
of Korea, Netherlands, Norway, Saudi Arabia, Sweden, United Arab Emirates, and
United Kingdom [8].
5
investments. The workshops also identify
prospects for bilateral or multilateral R&D
collaborations that would benefit from
cooperation on high-impact scientific
research. The workshops are multi-day,
carefully planned, and structured working
meetings.
On the afternoon of the first day, attendees divided into three panels on:
The panels began by identifying the critical R&D priorities and gaps in each area
of the energy materials innovation chain. They then explored opportunities for
combining individual research pathways, high-throughput synthesis, high-throughput
calculation, and high-throughput characterization into an integrated materials
innovation approach or “platform”. Participants engaged in technical discussions on
combining advanced theoretical and applied physical chemistry with next-generation
computing, machine learning, and robotics. Discussions continued through the
second day of the workshop.
6
research areas — the Six Grand Goals for a materials innovation revolution — for
further discussion. Attendees reconvened in work groups according to their materials
focus areas to discuss how to leverage the cross-cutting themes of theory, synthesis,
and characterization.
The report is organized as follows. Chapter 2 covers the state of the art in materials
discovery. Chapter 3 presents the urgent need for a materials revolution and
introduces the Six Grand Goals of the Materials Acceleration Platform. Chapters 4
through 9 discuss each of the goals in turn, identifying promising opportunities,
the benefits and impact of achieving the goal, its relationship to the other five
goals, and materials-specific challenges and approaches for organic materials,
inorganic materials, and nanomaterials and composites. The conclusion and some
final reflections are provided in Chapter 10.
7
References for Chapter 1
1. Bill Gates, “Beating Nature at Its Own Game,” Gates Notes, March 14, 2017, www.gatesnotes.
com/Energy/Beating-Nature.
2. George I. Seffers, “Scientists Pick AI for Lab Partner: Robotic research systems may foster a
boom in scientific knowledge,” Signal, September 1, 2017, www.afcea.org/content/scientists-
pick-ai-lab-partner.
3. P. Raccuglia, K. C. Elbert, P. D. F. Adler, C. Falk, M. B. Wenny, A. Mollo, M. Zeller, S. A. Friedler, J.
Schrier and A. J. Norquist, “Machine-learning-assisted materials discovery using failed experiments,”
Nature 533 (2016): 73–77.
4. S. V. Kalinin, B. G. Sumpter and R. K. Archibald, “Big–deep–smart data in imaging for guiding
materials design,” Nature Materials 14 (2015), 973–980.
5. Jonathan D. Linton and Steven T. Walsh, “From bench to business,” Nature Materials 2 (2003):
287–289, DOI: 10.1038/nmat882.
6. United Nations Framework Convention on Climate Change (UNFCCC), Adoption of the Paris
Agreement, Report No. FCCC/CP/2015/L.9/Rev.1, 2015, unfccc.int/resource/docs/2015/cop21/
eng/l09r01.pdf.
7. Eliza Northrop, Hana Biru, Sylvia Lima, Mathilde Bouyé and Ranping Song, “Examining the Alignment
between the Intended Nationally Determined Contributions and Sustainable Development Goals,”
World Resources Institute, September 2016, www.wri.org/publication/examining-alignment-between-
intended-nationally-determined-contributions-and-sustainable.
8. Mission Innovation, “Innovation Challenges,” Mission Innovation (website), accessed 2017, mission-
innovation.net/our-work/innovation-challenges/.
9. David King, “Global clean energy in 2017,” Science 355 (6321), 111, DOI: 10.1126/science.aam7088.
8
2 State of the Art in Materials Discovery
T
his chapter briefly describes the successful implementation of
components of an autonomous platform in the biological and
medical sciences before identifying the key gaps preventing
the full-scale deployment of autonomous discovery to materials
science. It also summarizes the main materials-related approaches
and opportunities of a range of clean energy technologies, including
photovoltaics, batteries, and solar fuels.
9
Figure 2.1 Annual trend in the “Cost per genome” determined by NHGRI. Data for cost estimates are
presented alongside those predicted by Moore’s law, which benchmarks performance based on a
yearly doubling of computational power. [Source: U.S. National Human Genome Research Institute]
use automated synthesis to exhaustively search for compounds suitable for a given
medicinal application [1,2,3]. These methods have yielded many successes in both
medicine and materials science [4,5].
Despite these advances, the automation of synthesis and characterization still has
substantial room for improvement [11] and most of automated chemistry is based
on a “trial and error’’ approach. We are now entering an age where integrated
systems using increasing computer power, machine learning, and AI could not only
enable exhaustive searching for new materials, but could also provide intermediate
decision-making capabilities to accelerate the discovery and development of
materials with specific, targeted properties. In principle, such AI-based algorithms
can be adaptive, using feedback from experimental characterization procedures and
enabling a more rational route to research objectives—either optimizing properties
or even directly verifying hypotheses. If realized, such an approach would markedly
increase research productivity, reduce its cost, and enable entirely new experimental
paradigms that are critical for innovation [12].
10
In the broader context of pharmaceuticals and biotechnology, many start-up
business models that reflect the paradigm of integrating AI and robotics into the
discovery process have begun to appear [13, 14, 15]. For example, companies
such as Zymergen, Transcriptic, Gingko Bioworks, Emerald Cloud Lab, Citrine, and
BenevolentAI follow data-driven approaches and seek to maximally exploit existing
knowledge bases. These efforts often use sophisticated human language processing
methods to data mine existing literature. For example, ARES (Autonomous Research
System) was the first AI-driven experimentation system to demonstrate closed-loop,
iterative experimentation with in situ characterization for materials development
[16]. Despite this emerging paradigm, a generalized end-to-end platform has yet
to be developed. Moreover, decision science, the basic tool of discovery-enabling
AI, and data-driven approaches specifically for materials science have not been
sufficiently developed to support such an integrated platform.
Figure 2.2 Comparison of (left) technique-centred and (right) sample-centred approaches. The
technique-centred approach involves replicating growth procedures or shipping samples among
geographically separated institutes. The sample-centred approach brings the people and techniques
to a central growth/characterization facility, saving time and allowing techniques to be applied to
the identical sample arrays. [Source: Dr. Eli Rotenberg, Lawrence Berkeley National Laboratory]
11
illustrates a central facility that concentrates efforts on preparing the best samples
(or arrays of samples) to be shared among many co-located techniques, such as
synchrotron beamlines or high-resolution electron microscopy, that cannot be
replicated easily in small lab settings.
Databases underlie all machine learning and data-driven approaches. In the course
of the Materials Genome Initiative, databases such as NOMAD (Novel Materials
Discovery), OQMD (Open Quantum Materials Database), AiiDA (Automated
Interactive Infrastructure and Database for Computational Science), and AFLOW,
have been developed [17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27]. However, as with
characterization techniques, many of these databases are specialized to address
case-by-case issues; an international data ecosystem does not yet exist.
A final gap is the lack of understanding of how to most effectively apply machine
learning to materials discovery. Most of the machine learning methods applied to
date, both theoretical and experimental, are relatively straightforward adaptations
of methods originally developed for other problems, such as image recognition,
text generation, and translation [33, 34, 35, 36]. These methods include deep
neural networks and Bayesian optimization. However, the development of new AI
methods tailored for materials could provide a breakthrough that greatly increases
the effectiveness of these techniques, thus accelerating the discovery process.
Currently, materials discovery is still a “trial and error” process; however, developing
new machine learning algorithms that generate candidate materials, rather than
only predicting properties, could lead to an environment where the inverse design
of materials is possible. Therefore, investing in this area would help derive new
machine learning algorithms and better tune existing ones specifically for materials
discovery [37].
12
2.3 MATERIALS OPPORTUNITIES FOR CLEAN ENERGY
TECHNOLOGIES
Table 2.1 summarizes the current status of seven clean energy technologies and
the key opportunities associated with each from a materials perspective. Many
of the technologies need new materials to continue to grow because they are
fundamentally limited by their current materials. The table also lists important
targets published by governments and consortia.
Table 2.1: Current status and opportunities for major energy technologies
Energy
Current Status Materials Opportunities
Technology
Photovoltaics • PV module production is • Further reductions in cost of
dominated by a small number solar electricity require improved
of materials: polycrystalline and efficiency, longer module/
single crystalline Si and CdTe. electronics lifetimes, and reduced
efficiency degradation.
• Reducing capital and
manufacturing costs could • New materials and new
lower full levelized cost of solar mechanisms for solar energy
electricity to 0.03 US$/kWh (i.e., harvesting are required.
US DOE 2030 target).
• Hybrid perovskites are one of the
most promising classes of new
materials, with power conversion
efficiencies >22%, but suffer
from low long-term stability and
presence of toxic elements.
• New mechanisms for
boosting efficiency (e.g.,
photon up-conversion, carrier
multiplication, luminescence
concentrators) require new
materials, but performance of
current materials is lacking.
13
Batteries • Technology is dominated by • New materials architectures,
(for electric lithium-ion batteries. interfacial treatments, and cell
vehicles design are needed to improve
• While single electrode materials
and grid power density.
have demonstrated high-power
storage) capabilities, the current Li-ion cell • New materials with higher
suffers from power limitation due energy and power densities and
to interfacial reactions. improved chemical and structural
stability are required.
• Current materials have about half
the specific energy (200–250 Wh • Lower-cost, low supply-risk
kg–1) of the goal proposed by the element replacements for current
Battery500 Consortium: >10-year battery components are required
life and total mileage of ~150,000 to scale up battery storage.
miles.
• There have been rapid
• Current materials have problems developments in solid-state Li
of chemical and structural batteries, but interfacial reactions
stability. and cell design need further R&D.
• Grid storage applications require • Redox flow batteries show
performance stability, i.e., >6,000 promise for grid storage, but
cycles and >20-year lifetimes. solubility and stability of redox
molecules at any state of charge
must be improved.
• The cost of ion-exchange
membranes must decrease.
Solar fuels • Key reactions to produce fuels (by • More efficient catalysts based
light-induced or dark catalysis) on cheaper, earth-abundant
are limited in efficiency and elements are needed.
long-term cycling.
• Use of solar fuels (in fuel cells)
requires high temperatures, which
increase operating costs and
cause corrosion of cell materials.
Or, if run at low temperatures,
they rely on precious metal
catalysts, e.g., platinum.
Wind Power • Wind power is one of the fastest The following new materials are
growing renewable energy needed:
sources, currently providing
• erosion-resistant coatings and
2.8% of the world’s electricity
materials;
generation.
• hydrophobic, light-weight
• If growth continues, wind could
materials and coatings;
provide 20-30% of global power
generation, displacing 3 billion • impact-resistant materials for the
tons of CO2 per year. leading edge of wind blades; and
• To generate electricity, wind relies • high-performance permanent
on electric motors, which today magnet materials that do not
require permanent magnets. contain elements with supply risk
(critical materials).
• In cold and icy or sandy desert
climates, wind blades erode • Offshore installations (which
and foul, which can cause up to provide much more power
25-50% degradation in power than onshore) need improved
production. salt-water corrosion-resistant
materials.
14
Thermal • Harnessing waste heat would help • More efficient thermoelectric
energy mitigate global energy needs, materials are required to
conversion but most heat sources are not convert waste heat to electricity.
harnessed. Promising strategies include
nanostructures, heterostructures,
and alloying.
• More thermally stable
thermoelectric materials that do
not experience degradation under
severe thermal cycling conditions
are needed.
• New materials for smart-windows
using electro-, photo-, and
thermochromic mechanisms
are needed to control infrared
radiation.
• Materials are needed to improve
building energy efficiency.
15
References for Chapter 2
1. Alexander G. Godfrey, Thierry Masquelin and Horst Hemmerle, “A remote-controlled adaptive
medchem lab: an innovative approach to enable drug discovery in the 21st Century,” Drug Discovery
Today 18, Issues 17–18 (2013): 795–802, DOI: 10.1016/j.drudis.2013.03.001.
2. Michael Shevlin, “Practical High-Throughput Experimentation for Chemists,” Merck & Co., Inc.,
Medicinal Chemistry Letters 8, Issue 6 (2017): 601–607, DOI: 10.1021/acsmedchemlett.7b00165.
3. Angelika Weber, Erich von Roedern and Hans Ulrich Stilz, “SynCar: An Approach to Automated
Synthesis,” Aventis Pharma Deutschland GmbH, Journal of Combinatorial Chemistry 7, Issue 2
(2005): 178–184, DOI: 10.1021/cc049838z.
4. Shevlin, op. cit.
5. Gurpur Rakesh D. Prabhu and Pawel L. Urban, “The Dawn of Unmanned Analytical Laboratories,”
TrAC Trends in Analytical Chemistry 88 (2017): 41–52, DOI: 10.1016/j.trac.2016.12.011.
6. Prof. Steven V. Ley, Daniel E. Fitzpatrick, Dr. Richard. J. Ingham and Dr. Rebecca M. Myers, “Organic
Synthesis: March of the Machines,” Angewandte Chemie International Edition 54, Issue 11 (2015):
3449–3464, DOI: 10.1002/anie.201410744.
7. Michael Maiwald, Holger H. Fischer, Young-Kyu Kim, Klaus Albert and Hans Hasse, “Quantitative
high-resolution on-line NMR spectroscopy in reaction and process monitoring,” Journal of Magnetic
Resonance 166, Issue 2 (2004): 135–146, DOI: 10.1016/j.jmr.2003.09.003.
8. Tzu-Chi Kuo, Niranjan A. Malvadkar, Ray Drumright, Richard Cesaretti and Matthew T. Bishop,
“High-Throughput Industrial Coatings Research at The Dow Chemical Company,” The Dow Chemical
Company, ACS Combinatorial Science 18, Issue 9 (2016): 507–526, DOI: 10.1021/acscombsci.6b00056.
9. Johannes Hepp, Florian Machui, Hans-J. Egelhaaf, Christoph J. Brabec and Andreas Vetter,
“Automatized analysis of IR-images of photovoltaic modules and its use for quality control of solar
cells,” Energy Science & Engineering 4, Issue 6 (2016) 363–371, DOI: 10.1002/ese3.140.
10. G. Zhou, A. Moment, S. Yaung, A. Cote and T.-E. Hu, “Evolution and Application of an Automated
Platform for the Development of Crystallization Processes,” Organic Process Research and Development
17, Issue 10 (2013): 1320–1329, DOI: 10.1021/op400187h.
11. S. Szymkuć, E. P. Gajewska andT. Klucznik. “Computer-Assisted Synthetic Planning: the End of
the Beginning,” Angewandte Chemie International Edition 55 (2016), DOI: 10.1002/anie.201506101.
12. Claudia Houben and Alexei A, Lapkin, “Automatic discovery and optimization of chemical processes,”
Current Opinion in Chemical Engineering 9 (2015) 1–7, DOI: 10.1016/j.coche.2015.07.001.
13. E. C. Hayden, “The Automated Lab,” Nature News 516 (2014): 131–132.
14. S. V. Ley, D. E. Fitzpatrick, R. J. Ingham and R. M. Myers, “Organic Synthesis: March of the
Machines,” Angewandte Chemie International Edition 54, Issue 11 (2015): 3449–3464.
15. J. Bohannon, “A New Breed of Scientist, with Brains of Silicon,” Science (2017), DOI: 10.1126/
science.aan7046.
16. Pavel Nikolaev, Daylond Hooper, Frederick Webber, Rahul Rao, Kevin Decker, Michael Krein, Jason
Poleski, Rick Barto and Benji Maruyama, “Autonomy in materials research: a case study in carbon
nanotube growth,” npj Computational Materials 2 (2016), DOI: 10.1038/npjcompumats.2016.31.
17. P. Raccuglia, K. C. Elbert, P D. F. Adler, C. Falk, M. B. Wenny, A. Mollo, M. Zeller, S. A. Friedler, J.
Schrier and A. J. Norquist, “Machine-learning-assisted materials discovery using failed experiments,”
Nature 533 (2016): 73–77, DOI: 10.1038/nature17439.
18. Scott Kirklin, James E Saal, Bryce Meredig, Alex Thompson, Jeff W Doak, Muratahan Aykol,
Stephan Rühl & Chris Wolverton, “The Open Quantum Materials Database (OQMD): assessing the
accuracy of DFT formation energies,” npj Computational Materials 1, Number 15010 (2015), DOI:
10.1038/npjcompumats.2015.10.
19. Pizzi et al, op. cit.
20. Dr. Jens S. Hummelshøj, Dr. Frank Abild-Pedersen, Dr. Felix Studt, Dr. Thomas Bligaard and Prof.
Jens K. Nørskov, “CatApp: A Web Application for Surface Chemistry and Heterogeneous Catalysis,”
Angewandte Chemie International Edition 51, Issue 1 (2012): 272–274, DOI: 10.1002/ange.201107947.
21. Ivano E. Castelli, Thomas Olsen, Soumendu Datta, David D. Landis, Søren Dahl, Kristian S.
Thygesena and Karsten W. Jacobsen, “Computational screening of perovskite metal oxides
for optimal solar light capture,” Energy & Environmental Science 5, Issue 2 (2012): 5814–5819,
DOI: 10.1039/c1ee02717d.
22. S. Curtarolo, W. Setyawan, S. Wang, J. Xue, K. Yang, R. H. Taylor, L. J. Nelson, G. L. W. Hart, S.
Sanvito, M. Buongiorno-Nardelli, N. Mingo and O. Levy, “AFLOWLIB.ORG: a distributed materials
properties repository from high-throughput ab initio calculations,” Computational Materials Science
58 (2012): 227–235, DOI: 10.1016/j.commatsci.2012.02.002.
23. A. Hjorth Larsen, J. Jørgen Mortensen, J. Blomqvist, I. E. Castelli, R. Christensen, M. Dułak,
J. Friis, M. N. Groves, B. Hammer, C. Hargus, E. D. Hermes, P. C. Jennings, P. Bjerre Jensen, J.
Kermode, J. R. Kitchin, E. Leonhard Kolsbjerg, J. Kubal, K. Kaasbjerg, S. Lysgaard, J. Bergmann
Maronsson, T. Maxson, T. Olsen, L. Pastewka, A. Peterson, C. Rostgaard, J. Schiøtz, O. Schütt,
M. Strange, K. S. Thygesen, T. Vegge, L. Vilhelmsen, M. Walter, Z. Zeng and K. W. Jacobsen,
“The atomic simulation environment–a Python library for working with atoms,” Journal of Physics:
Condensed Matter, 29, Number 27 (2017), DOI: 10.1088/1361-648X/aa680e.
16
24. J. Hachmann, R. Olivares-Amaya, S. Atahan-Evrenk, C. Amador-Bedolla, R. S. Sánchez-Carrera, A.
Gold-Parker, L. Vogt, A. M. Brockway and A. Aspuru-Guzik, “The Harvard Clean Energy Project:
Large-Scale Computational Screening and Design of Organic Photovoltaics on the World Community
Grid,” The Journal of Physical Chemistry Letters 2 (2011): 2241–2251, DOI: 10.1021/jz200866s.
25. G. Pizzi, A. Cepellotti, R. Sabatini, N. Marzari and B. Kozinsky, “AiiDA: Automated Interactive
Infrastructure and Database for Computational Science,” Computational Materials Science 111
(2016): 218–230, DOI: 10.1016/j.commatsci.2015.09.013.
26. “NOMAD Repository,” The NOMAD Laboratory (website), accessed 2017, nomad-repository.eu/.
27. A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek, S. Cholia, D. Gunter, D. Skinner,
G. Ceder and K. A. Persson, “Commentary: The Materials Project: A materials genome approach
to accelerating materials innovation,” APL Materials 1, 11002 (2013), DOI: 10.1063/1.4812323.
28. Johannes Hachmann, Roberto Olivares-Amaya, Adrian Jinich, Anthony L. Appleton, Martin A. Blood-
Forsythe, László R. Seress, Carolina Román-Salgado, Kai Trepte, Sule Atahan-Evrenk, Süleyman
Er, Supriya Shrestha, Rajib Mondal, Anatoliy Sokolov, Zhenan Baod and Alán Aspuru-Guzik, “Lead
candidates for high-performance organic photovoltaics from high-throughput quantum chemistry
– the Harvard Clean Energy Project,” Energy & Environmental Science 7, Issue 2 (2014): 698, DOI:
10.1039/C3EE42756K.
29. Steven A. Lopez, Edward O. Pyzer-Knapp, Gregor N. Simm, Trevor Lutzow, Kewei Li, Laszlo R.
Seress, Johannes Hachmann and Alán Aspuru-Guzik, “The Harvard organic photovoltaic dataset,”
Scientific Data 3 (2016), DOI: 10.1038/sdata.2016.86.
30. Hachmann et al. 2011, op. cit.
31. A. J. Lawson, J. Swienty-Busch, T. Géoui and D. Evans, “The Making of Reaxys—Towards Unobstructed
Access to Relevant Chemistry Information,” in ACS Symposium Series 1164 (2014): 127–148, Chapter
DOI: 10.1021/bk-2014-1164.ch008.
32. Reaxys (website), accessed 2017, www.reaxys.com/.
33. Rafael Gómez-Bombarelli, Jennifer N. Wei, David Duvenaud, José Miguel Hernández-Lobato,
Benjamín Sánchez-Lengeling, Dennis Sheberla, Jorge Aguilera-Iparraguirre, Timothy D. Hirzel,
Ryan P. Adams and Alán Aspuru-Guzik, “Automatic chemical design using a data-driven continuous
representation of molecules,” Cornell University Library, January 6, 2017, arXiv:1610.02415v2 [cs.LG].
34. David K. Duvenaud, Dougal Maclaurin, Jorge Iparraguirre, Rafael Bombarell, Timothy Hirzel, Alan
Aspuru-Guzik and Ryan P. Adams, “Convolutional Networks on Graphs for Learning Molecular
Fingerprints,” in the proceedings of Advances in Neural Information Processing Systems 28 (NIPS
2015), papers.nips.cc/paper/5954-convolutional-networks-on-graphs-for-learning-molecular-
fingerprints.
35. Benjamin Sanchez-Lengeling, Carlos Outeiral, Gabriel L. Guimaraes and Alan Aspuru-Guzik,
“Optimizing distributions over molecular space. An Objective-Reinforced Generative Adversarial
Network for Inverse-design Chemistry (ORGANIC),” Harvard University, Chem Rxiv, August 17,
2017, doi.org/10.26434/chemrxiv.5309668.v3.
36. Edward O. Pyzer-Knapp,a Gregor N. Simmb and Alán Aspuru Guzik, “A Bayesian approach to
calibrating high-throughput virtual screening results and application to organic photovoltaic
materials,” Materials Horizons 3, Issue 3 (2016) 226–233, DOI: 10.1039/C5MH00282F.
37. R. Ramakrishnan and O. A. von Lilienfeld, “Machine Learning, Quantum Chemistry, and Chemical
Space,” Chapter 5 in Reviews in Computational Chemistry 30, Abby L. Parrill and Kenny B. Lipkowitz,
Eds. (John Wiley & Sons, Inc., 2017), Chapter DOI: 10.1002/9781119356059.ch5.
38. Steven Chu, Yi Cui and Nian Liu, “The path towards sustainable energy,” Nature Materials 16 (2017):
16–22, DOI:10.1038/nmat4834.
39. Jarad A. Mason, Mike Veenstra and Jeffrey R. Long, “Evaluating metal–organic frameworks for
natural gas storage,” Chemical Science 5 (2014): 32–51, DOI: 10.1039/c3sc52633j.
17
18
The Urgent Need for a Materials
3 Acceleration Platform
T
hroughout history, humans have developed energy technologies
to maximize the energy available for human consumption and
increase productivity. This drive has led to transitions from wood
burning to the use of coal, oil, and natural gas, whose extraction by
machinery became far more efficient than harvesting trees through
human innovation. The development of nuclear power required
scientific understanding to harness the high-energy content of nuclear
fuel, while tackling persistent challenges in safety, cost, and waste
handling. With GHG emission reductions in mind, the current shift to
clean energy is the most challenging and potentially most rewarding
energy transition to date.
19
mind to assimilate the level of detail and magnitude of data that are now produced
by experiments and computational simulations. Computer simulations with current
computer power are approaching the limit to predict the properties of matter at the
relevant length and time scales. We have also reached the limit of any single research
group to own and operate the variety of synthetic and characterization equipment
needed to completely understand these materials. Moreover, we are approaching
the time limit for deploying worldwide ubiquitous clean energy technologies before
the environmental damage becomes irreversible.
The Materials Acceleration Platform, or MAP, aims to reduce the materials development
cycle from 10 to 20 years to 1 or 2 years. It builds on recent scientific breakthroughs
and the ability to program machines to assist the design of materials, moving away
from the drudgery of Edisonian discovery methods. Such breakthroughs allow
scientists to order whole families of new materials based on desired properties
20
without tedious hours of trial and experiments in the lab. The ability of AI to sift
through vast quantities of data can deliver new scientific insights to humans. The
development of new AI-assisted theoretical methods can speed up the simulation
and design of new materials. Achieving this, however, requires development of
machine intelligence and materials synthesis capabilities beyond the scale of any
human team.
MAP features six integrated priority research areas, the Six Grand Goals, which can
be approached and achieved more effectively and faster through strong international
cooperation (Figure 3.1):
21
for each of these length and time scales, systematic methods of connecting
results and ideas across these scales would enable transformative discoveries.
22
Goal #1: Closing the Loop in
4 Autonomous Discovery and Development
A
n autonomous platform for materials discovery would exploit
the “feedback” from measurements and guide new synthesis
and characterization experiments or simulations, i.e., enable
closed-loop research or a self-driven laboratory. It would ultimately
converge to an optimal materials design for target functionality, while
autonomously generating and validating hypotheses of the underlying
physical and chemical phenomena. The overarching challenge of
autonomous materials research combines (i) AI-based predictive theory
of new materials and their properties, (ii) autonomous robotic systems
for synthesis and experimental data collection, (iii) data analytics such
as feature extraction, (iv) machine learning-based classification and
regression of the results, and (v) decision modules to drive optimal
experimental design for subsequent experimental iterations. For
autonomous discovery and development, these elements need to be
integrated into a closed-loop platform for designing and performing
experiments, which can ultimately create new materials to meet society’s
needs for clean energy.
23
“successful” data and do not report data of “failed” experiments. The comprehensive
incorporation of all data, successful or failed, is one of the strengths of machine
learning approaches to AI for materials. With sufficient data, an AI-driven machine
can choose the next step in experiments or simulations more efficiently than
humans, speeding up the optimization of a given property.
4.1 OPPORTUNITIES
A key challenge in materials discovery, particularly in the organic materials domain, is
the limited availability of automated or closed-loop synthetic tools. Synthetic organic
molecules are an important form of matter in a huge range of human activities.
Organic molecules are used as medicines, functional coatings, energy storage, and
energy harvesting materials, to name a few applications. Despite the widespread
applications, synthesizing novel organic molecules is an activity reserved for highly
trained specialists. The limited pool of experts trained in synthetic chemistry
creates a significant bottleneck for researchers in other fields interested in using
synthetic molecules to achieve their R&D aims. To drive discovery and development
of new materials for clean energy, materials researchers need access to synthetic
molecules much faster and on a much larger scale than currently possible.
24
as those produced by automated synthesis techniques, could be more challenging.
As such, a new characterization paradigm is needed in which sample analysis is
performed in an automated fashion driven by machine learning and AI. There are
two types of automation loops, broadly described as “optimization” and “discovery”
(Section 4.4.4). The distinction between them is not sharp, and many investigations
would include aspects of both. What is clear, however, is that both approaches need
similar growth and characterization tools, but their workflows could differ owing
to different time scales of the characterization techniques.
On the individual lab scale, success in this goal would ultimately lead to a bench-top
device that is capable of autonomous synthetic chemistry. Its first iteration would
integrate the process of designing the synthetic route to the target molecule,
validating the routes and optimization of the resulting synthetic steps using integrated
analytical technologies. The device would also purify and characterize the target
molecule against the molecular properties desired. It would be able to learn from
the results of each experiment to further refine the AI that designs the synthetic
routes and the algorithms that are used to optimize the individual chemical steps.
Going beyond this, the long-range impact could be the development of an autonomous
tool that designs for function instead of target molecule. The current paradigm for
materials discovery is to design a molecule, then make the molecule, and then test
the molecule to see if it has the target properties. The ultimate goal, however, is not
the specific structure of the molecule, but obtaining of a material with the target
properties. The ideal device would begin with a description of the target properties
and then iterate on molecular architectures until it achieves these properties.
Closing the loop would also result in a paradigm where samples are no longer
replicated or transferred between separate technique-oriented facilities, but
become the centre; the techniques would be merged in national-scale facilities
dedicated to creating carefully curated sample arrays or libraries that are shared
25
among co-located techniques. To ensure that the characterizations have the most
information, the robotics could transfer the sample in tailored environments, such
as controlled atmosphere, ultra-high vacuum, temperature, etc. It is unlikely that
one facility could encompass the three classes of materials that are the focus of this
document (i.e., organic, inorganic, composite) along with others. However, several
facilities with similar organization, but with tools and environments optimized for
each class, could be developed. Some common questions for all materials classes
could lead to overlap in the characterization methods.
To integrate materials data in closed-loop discovery systems, the data stored must
go beyond the raw formats to include pipelines for feature extraction and automated
processing of characterization output. For example, extraction of domain sizes,
heterogeneity, microstructure, and other parameters from micrographs (e.g.,
scanning electron microscopy, SEM; transmission electron microscopy, TEM; atomic
force microscopy, AFM; scanning tunnelling, microscopy, STM) is often performed
on an ad hoc basis in a particular research group. However, tools are often not
shared or connected directly in machine-friendly data formats to calculated data
or experimental conditions. Therefore, along with the data, materials informatics
tools are expected to help accelerate the materials discovery process within an
integrated feedback loop.
26
other goals, the integration, management, and communication of the components
comprise a unique set of basic research challenges whose development would be
accelerated through shared resources and effective dissemination of know-how.
Decisions and research planning based on the data could initially be performed
with existing AI optimization algorithms. To further capitalize on the autonomous
infrastructure and enable autonomous laboratories to explore materials space
beyond the confines of previous research, the reasoning and intuitive decision
making of AI would be necessary.
27
Ridge National Laboratory (ORNL) group, Joint Center for Artificial Photosynthesis
(JCAP) [10,11]), even fewer are open access/open source. Ideally, data taken and
analyzed on the fly are referenced against open databases, both experimental
(e.g., ICSD) and theoretical (e.g., OQMD [12], NOMAD [13], AiiDA [14], Materials
Project [15], AFLOW [16]). Potential contradictions are also automatically flagged
for deeper experimental investigation and theoretical validation. Developing this
type of software requires databases to expose public application programming
interfaces (APIs) that are structured to facilitate use by software engineers from a
variety of backgrounds and development environments.
HTE characterization techniques have been developed for a broad range of inorganic
materials functionalities including ferroelectric, magnetic, energy storage, catalysis,
and corrosion resistance. However, existing platforms tend to be application specific
and expensive. Making decisions about which techniques to include (e.g., Raman;
x-ray diffraction, XRD; x-ray absorption spectroscopy, XAS), depending on the
design metrics space, is a clear challenge of high-throughput characterization [18].
Another challenge for all inorganic materials is the automated investigation of a
given material’s microstructure, a key need in understanding performance. Nano
and atomic structure is confirmed only by electron microscopy, which is difficult
to automate. Advances in modular robotics would enable on-site, on-the-fly sample
transfer between synthesis and characterization, accelerating an already accelerated
process for materials discovery. Advances in on-the-fly machine learning during
synthesis and characterization are required to maximize the amount of knowledge
gained per experimental point and reduce instrument “dead time.”
28
Technologies enabling autonomous organic synthesis are rapidly advancing towards
the goal of a truly autonomous discovery machine. Although many of the pieces
required for such a machine exist today, they are not yet fully connected. Initiatives,
such as DARPA’s Make-It program [19] and European effort Dial-a-Molecule [20],
have made advances in this area.
29
Figure 4.2 Workflow for automated feedback optimization. [Source: ref 28]
An important component of self-optimizing systems is the online and inline analysis
tools available to characterize the results of the automated reactions. Spectroscopic
tools, such as NMR, infrared (IR), and Raman, along with high-performance liquid
chromatography (HPLC), have been successfully applied to enable automated
optimization, particularly in flow chemical systems [30,31]. Relatively simple and
static algorithms are used in all these automated optimization examples; enabling
learning algorithms would yield improvements.
Chan et al. have built WANDA, a robot for high-throughput nanocrystal synthesis,
which pioneered automated nanoparticle synthesis [32]. As mentioned in Chapter 2,
the state of the art in closed-loop autonomous synthesis is ARES, which uses AI and
closed-loop, iterative experimentation to learn to grow carbon nanotubes at targeted
rates [33]. It is the first autonomous research robot for materials development.
30
transition from carbon nanotubes to scaled production would directly affect the
clean energy materials goals because applying carbon nanotubes to lightweighting
transportation vehicles would improve power transmission efficiency, greatly
reducing energy consumption. Additionally, the transformation of hydrocarbons
(e.g., from conversion of captured CO2) to carbon nanomaterials would effectively
sequester the carbon indefinitely.
There are universal characterization techniques associated with all classes of materials:
scanning probe microscopies for atomic-scale information, x-ray nanoprobes for
spatially resolved chemical and electronic information, lasers and x-ray lasers for
time-resolved response, and tools integrated with the sample growth environments
to characterize growth dynamics.
31
value of this approach is diminished because it is by nature statistical. If the probes
lack sufficient spatial resolution, or are not applied to the same region of the same
samples, the information available from direct correlations is lost. Furthermore,
the samples are no longer in the same environment or in the same stage of their
lifetimes. Nominally identical, but distinct, samples may have been prepared under
similar, but not identical, conditions and therefore further weaken the statistical
strength of the enterprise. This greatly weakens the correlations between techniques,
and wastes a great deal of time (Figure 2.2).
Even more important is the fact that future discoveries are likely to be found in the
morphological and combinatorial modifications of materials. These would surely
require side-by-side comparison of tailored samples with identical and known
provenance. Creation and understanding of such sample arrays would create a
future bottleneck, and replicating these arrays in multiple laboratories would not
be generally possible or practical. This is especially true because many valuable
techniques are surface-sensitive, or the samples themselves are dominated by
surface effects.
Figure 4.3 shows such a workflow, where first, a large quantity of samples is analyzed
in an automatic manner for a specific functionality that is faster to measure, after
which a reduced number of highly contrasting samples is subjected to more detailed
analysis (middle section of the graph) to investigate the relationship between preparation
conditions and composition. This information is used to zoom in on the promising
samples that are subsequently subjected to in-depth and more time-consuming
analysis to understand the material with tools that yield the most important atomic,
energy, momentum, spin, and/or time resolution. Theoreticians can then use this
knowledge to refine predictive models. The left side of the workflow focuses on the
optimization process, which can be optimized for high-speed machine-controlled
operation. The right side focuses on the “why”, which, on its slower time scale, is
better suited to the human side of the process.
32
Figure 4.3 A typical progressive pathway to both optimization and understanding of material
functionality begins with left, a high-throughput screening of promising material families (theoretical
and/or experimental) to narrow down the phase space, followed by a more in-depth, narrower search
of phase space to optimize structural/functional characteristics. The final step (right) is focused
on gaining understanding through the slowest, most information-rich techniques with spatial,
energy, momentum, and spin resolution. [Source: ref 36, and based on original works presented
in refs 37,38,39,40]
33
References for Chapter 4
1. George I. Seffers, “Scientists Pick AI for Lab Partner: Robotic research systems may foster a
boom in scientific knowledge,” Signal, September 1, 2017, www.afcea.org/content/scientists-
pick-ai-lab-partner.
2. Martin L. Green, Ichiro Takeuchi and Jason R. Hattrick-Simpers, “Applications of high throughput
(combinatorial) methodologies to electronic, magnetic, optical, and energy-related materials,”
Journal of Applied Physics 113, Issue 23, Number 231101 (2013), DOI: 10.1063/1.4803530.
3. Emory M. Chan, Chenxu Xu, Alvin W. Mao, Gang Han, Jonathan S. Owen, Bruce E. Cohen and Delia
J. Milliron, “Reproducible, High-Throughput Synthesis of Colloidal Nanocrystals for Optimization in
Multidimensional Parameter Space,” Nano Letters 10, Issue 5 (2010): 1874–1885, DOI: 10.1021/
nl100669s.
4. Jonathan K. Bunn, Richard Z. Voepel, Zhiyong Wang, Edward P. Gatzke, Jochen A. Lauterbach and
Jason R. Hattrick-Simpers, “Development of an Optimization Procedure for Magnetron-Sputtered
Thin Films to Facilitate Combinatorial Materials Research,” Industrial & Engineering Chemistry
Research 55, Issue 5 (2016): 1236–1242, DOI: 10.1021/acs.iecr.5b04196.
5. Daniel J. Arriola, Edmund M. Carnahan, Phillip D. Hustad, Roger L. Kuhlman and Timothy T.
Wenzel, “Catalytic Production of Olefin Block Copolymers via Chain Shuttling Polymerization,”
Science 312, Issue 5774 (2006): 714-719, DOI: 10.1126/science.1125268.
6. Qimin Yan, Jie Yu, Santosh K. Suram, Lan Zhou, Aniketa Shinde, Paul F. Newhouse, Wei Chen, Guo
Li, Kristin A. Persson, John M. Gregoire and Jeffrey B. Neaton, “Solar fuels photoanode materials
discovery by integrating high-throughput theory and experiment,” Proceedings of the National
Academy of Sciences 114, Number 12 (2016): 3040–3043, DOI: 10.1073/pnas.1619940114.
7. Prasanna V. Balachandran, Dezhen Xue, James Theiler, John Hogden and Turab Lookman,
“Adaptive Strategies for Materials Design using Uncertainties” Scientific Reports 6 (2016), DOI:
10.1038/srep19660.
8. Jason R. Hattrick-Simpers, John M. Gregoire and A. Gilad Kusne, “Perspective: Composition–
structure–property mapping in high-throughput experiments: Turning data into knowledge,” APL
Materials 4, Issue 5 (2016), DOI: 10.1063/1.4950995.
9. Aaron Gilad Kusne, Tieren Gao, Apurva Mehta, Liqin Ke, Manh Cuong Nguyen, Kai-Ming Ho, Vladimir
Antropov, Cai-Zhuang Wang, Matthew J. Kramer, Christian Long and Ichiro Takeuchi, “On-the-fly
machine-learning for high-throughput experiments: search for rare-earth-free permanent magnets,”
Scientific Reports 4 (2014), DOI: 10.1038/srep06367.
10. Sergei V. Kalinin, Evgheni Strelcov, Alex Belianinov, Suhas Somnath, Rama K. Vasudevan, Eric J.
Lingerfelt, Richard K. Archibald, Chaomei Chen, Roger Proksch, Nouamane Laanait, and Stephen
Jesse, “Big, Deep, and Smart Data in Scanning Probe Microscopy,” ACS Nano 10, Issue 10 (2016):
9068–9086, DOI: 10.1021/acsnano.6b04212.
11. Santosh K. Suram, Yexiang Xue, Junwen Bai, Ronan Le Bras, Brendan Rappazzo, Richard Bernstein,
Johan Bjorck, Lan Zhou, R. Bruce van Dover, Carla P. Gomes, and John M. Gregoire, “Automated
Phase Mapping with AgileFD and its Application to Light Absorber Discovery in the V–Mn–Nb Oxide
System,” ACS Combinatorial Science 19, Issue 1 (2017): 37–46, DOI: 10.1021/acscombsci.6b00153.
12. Scott Kirklin, James E Saal, Bryce Meredig, Alex Thompson, Jeff W Doak, Muratahan Aykol,
Stephan Rühl & Chris Wolverton, “The Open Quantum Materials Database (OQMD): assessing the
accuracy of DFT formation energies,” npj Computational Materials 1, Number 15010 (2015), DOI:
10.1038/npjcompumats.2015.10.
13. “NOMAD Repository,” The NOMAD Laboratory (website), accessed 2017, nomad-repository.eu/.
14. G. Pizzi, A. Cepellotti, R. Sabatini, N. Marzari and B. Kozinsky, “AiiDA: Automated Interactive
Infrastructure and Database for Computational Science,” Computational Materials Science 111
(2016): 218–230, DOI: 10.1016/j.commatsci.2015.09.013.
15. A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek, S. Cholia, D. Gunter, D. Skinner,
G. Ceder and K. A. Persson, “Commentary: The Materials Project: A materials genome approach
to accelerating materials innovation,” APL Materials 1, 11002 (2013), DOI: 10.1063/1.4812323.
16. S. Curtarolo, W. Setyawan, S. Wang, J. Xue, K. Yang, R. H. Taylor, L. J. Nelson, G. L. W. Hart, S.
Sanvito, M. Buongiorno-Nardelli, N. Mingo and O. Levy, “AFLOWLIB.ORG: a distributed materials
properties repository from high-throughput ab initio calculations,” Computational Materials Science
58 (2012): 227–235, DOI: 10.1016/j.commatsci.2012.02.002.
17. A. Oliynyk, E. Antono, T. Sparks, L. Ghadbeigi, M. Gaultois, B. Meredig, A. Mar, “High-throughput
machine-learning-driven synthesis of full-heusler compounds,” Chemistry of Materials 28, Issue
20 (2016): 7324–7331, DOI: 10.1021/acs.chemmater.6b02724.
18. Fang Ren, Ronald Pandolfi, Douglas Van Campen, Alexander Hexemer, and Apurva Mehta, “On-the-Fly
Data Assessment for High-Throughput X-ray Diffraction Measurements,” ACS Combinatorial Science
19, Issue 6 (2017): 377–385, DOI: 10.1021/acscombsci.7b00015.
19. Anne Fischer, “Make-It,” Defense Advanced Research Projects Agency (DARPA), accessed 2017,
www.darpa.mil/program/make-it.
20. Victor Sans and Leroy Cronin, “Towards dial-a-molecule by integrating continuous flow, analytics
and self-optimisation,” Chemical Society Reviews 45, Issue 8 (2016): 2032–2043, DOI: 10.1039/
c5cs00793c.
21. “Wiley ChemPlanner,” Wiley, accessed 2017, www.chemplanner.com.
34
22. S. Szymkuć, E. P. Gajewska andT. Klucznik. “Computer-Assisted Synthetic Planning: the End of
the Beginning,” Angewandte Chemie International Edition 55 (2016), DOI: 10.1002/anie.201506101.
23. Nicholas J. Browning, Raghunathan Ramakrishnan, O. Anatole von Lilienfeld, and Ursula
Roethlisberger, “Genetic Optimization of Training Sets for Improved Machine Learning Models
of Molecular Properties,” The Journal of Physical Chemistry Letters 8, Issue 7 (2017): 1351–1359,
DOI: 10.1021/acs.jpclett.7b00038.
24. Alexander G. Godfrey, Thierry Masquelin and Horst Hemmerle, “A remote-controlled adaptive
medchem lab: an innovative approach to enable drug discovery in the 21st Century,” Drug Discovery
Today 18, Issues 17–18 (2013): 795–802, DOI: 10.1016/j.drudis.2013.03.001.
25. A. Adamo, R. L. Beingessner, M. Behnam, J. Chen, T. F. Jamison, K. F. Jensen, M. C. Monbaliu, A.
S. Myerson, E. M. Revalor, D. R. Snead, T. Stelzer, N. Weeranoppanant, S. Y. Wong and P. Zhang,
“On-demand continuous-flow production of pharmaceuticals in a compact, reconfigurable system,”
Science 352 (2016): 61–67, DOI: 10.1126/science.aaf1337.
26. Steven V. Ley, Daniel E. Fitzpatrick, Rebecca M. Myers, Claudio Battilocchio and Richard. J.
Ingham, “Machine-Assisted Organic Synthesis,” Angewandte Chemie International Edition 54
(2015):10122–10136, DOI: 10.1002/anie.201501618.
27. Richard J. Ingham, Claudio Battilocchio, Daniel E. Fitzpatrick, Eric Sliwinski, Joel M. Hawkins
and Steven V. Ley, “A Systems Approach towards an Intelligent and Self-Controlling Platform for
Integrated Continuous Reaction Sequences,” Angewandte Chemie International Edition 54 (2014):
144 –148, DOI: 10.1002/anie.201409356.
28. Brandon J. Reizman and Klavs F. Jensen, “Feedback in Flow for Accelerated Reaction Development,”
Accounts of Chemical Research 49, Issue 9 (2016): 1786–1796, DOI: 10.1021/acs.accounts.6b00261.
29. B. J. Reizman, Y-.M. Wang, S. L. Buchwald and K. F. Jensen, “Suzuki–Miyaura cross-coupling
optimization enabled by automated feedback,” Reaction Chemistry & Engineering 1, Issue 6 (2016):
658-666, DOI: 10.1039/c6re00153j.
30. David C. Fabry, Erli Sugiono and Magnus Rueping, “Self-Optimizing Reactor Systems: Algorithms,
On-line Analytics, Setups, and Strategies for Accelerating Continuous Flow Process Optimization,”
Israel Journal of Chemistry 54, Issue 5 (2014): 341–350, DOI: 10.1002/ijch.201300080.
31. D. C. Fabry, E. Sugionoa and M. Rueping, “Online monitoring and analysis for autonomous
continuous flow self-optimizing reactor systems,” Reaction Chemistry & Engineering 1, Issue 2
(2016): 129–133, DOI: 10.1039/c5re00038f.
32. Chan et al., op. cit.
33. Pavel Nikolaev, Daylond Hooper, Frederick Webber, Rahul Rao, Kevin Decker, Michael Krein, Jason
Poleski, Rick Barto and Benji Maruyama, “Autonomy in materials research: a case study in carbon
nanotube growth,” npj Computational Materials 2 (2016), DOI: 10.1038/npjcompumats.2016.31.
34. K. A. Stoerzinger, O. Diaz-Morales, M. Kolb, R. R. Rao, R. Frydendal, L. Qiao, X. R. Wang, N. B.
Halck, J. Rossmeisl, H. A. Hansen, T. Vegge, I. E. L. Stephens, M. T. M. Koper and Y. Shao-Horn,
“Orientation-Dependent Oxygen Evolution on RuO2 without Lattice Exchange,” ACS Energy Letters
2 (2017): 876–881, DOI: 10.1021/acsenergylett.7b00135.
35. G. Pariente, V. Gallet, A. Borot, O. Gobert and F. Quéré, “Space–time characterization of ultra-intense
femtosecond laser beams,” Nature Photonics 10 (2016): 547–553, DOI:10.1038/nphoton.2016.140.
36. Solving Scientific Challenges with Coherent Soft X-Rays: Workshop report on early science enabled by
the Advanced Light Source Upgrade, Advanced Light Source (ALS-U) Lawrence Berkeley National
Laboratory and U.S. Department of Energy, January 18–20, 2017, als.lbl.gov/wp-content/
uploads/2017/08/ALS-U-Early-Science-Workshop-Report-Full.pdf.
37. John M. Gregoire, Chengxiang Xiang, Xiaonao Liu, Martin Marcin and Jian Jin, “Scanning droplet
cell for high throughput electrochemical and photoelectrochemical measurements,” Review of
Scientific Instruments 84, 024102 (2013), DOI: 10.1063/1.4790419.
38. Dillon Wong, Jairo Velasco Jr, Long Ju, Juwon Lee, Salman Kahn, Hsin-Zon Tsai, Chad Germany,
Takashi Taniguchi, Kenji Watanabe, Alex Zettl, Feng Wang and Michael F. Crommie, “Characterization
and manipulation of individual defects in insulating hexagonal boron nitride using scanning
tunnelling microscopy,” Nature Nanotechnology 10 (2015): 949–953, DOI: 10.1038/nnano.2015.188.
39. Sara Barja, Sebastian Wickenburg, Zhen-Fei Liu, Yi Zhang, Hyejin Ryu, Miguel M. Ugeda, Zahid
Hussain, Zhi-Xun Shen, Sung-Kwan Mo, Ed Wong, Miquel B. Salmeron, Feng Wang, Michael F.
Crommie, D. Frank Ogletree, Jeffrey B. Neaton and Alexander Weber-Bargioni, “Charge density
wave order in 1D mirror twin boundaries of single-layer MoSe2,” Nature Physics 12 (2016): 751–756,
DOI: 10.1038/nphys3730.
40. Ruimin Qiao, Qinghao Li, Zengqing Zhuo, Shawn Sallis, Oliver Fuchs, Monika Blum, Lothar
Weinhardt, Clemens Heske, John Pepper, Michael Jones, Adam Brown, Adrian Spucces, Ken Chow,
Brian Smith, Per-Anders Glans, Yanxue Chen, Shishen Yan, Feng Pan, Louis F. J. Piper, Jonathan
Denlinger, Jinghua Guo, Zahid Hussain, Yi-De Chuang and Wanli Yang, “High-efficiency in situ
resonant inelastic x-ray scattering (iRIXS) endstation at the Advanced Light Source,” Review of
Scientific Instruments 88, 033106 (2017), DOI: 10.1063/1.4977592.
41. National Institute of Standards and Technology (NIST) Materials Genome Initiative (MGI),
“High-Throughput Experimental Materials Collaboratory,” NIST MGI: Gateway to Materials Genome
Information (website), last updated February 18, 2015, mgi.nist.gov/htemc.
35
36
Goal #2: Artificial Intelligence for
5 Materials
A
I is a broad field of computer science with the general goal of
making machines intelligent. Within the past decade, AI has
achieved breakthroughs in computer vision, speech, language
translation, and natural language understanding. AI-based computer
systems have surpassed humans for complex tasks such as image
recognition, cancer detection, and games (e.g., Chess, Go, Poker,
Jeopardy, video games). Active areas of AI research include self-driving
cars and the understanding of human speech. This chapter introduces
Artificial Intelligence for Materials (AI4M) as a Grand Goal at the
intersection of AI and materials research.
37
knowledge and understanding take precedence over quantitative results, rather
than an optimization or an image recognition task for which a data model can be
sufficiently scored by its goodness-of-fit or convergence rate.
5.1 OPPORTUNITIES
The current challenges in realizing AI4M range from technical to organizational.
The necessary AI algorithms cannot be derived or adapted from the present AI
field, largely because development of these algorithms did not consider materials
science. Algorithm inputs for materials science still look less like big data and more
like sparse, heterogeneous data that can only be understood in the context of the
chemical and physical laws that constrain the accessible portions of the vast, multi-
dimensional materials parameter space. AI for materials must be enlightened by
these chemical and physical laws. This constitutes an entirely new field of research
that can only be pioneered at the intersection of the AI and materials communities,
with top researchers from both working side by side. The need for collaboration
requires an evolution of the organizational and funding mechanisms of stakeholders
and research institutions. New mechanisms for cross-discipline education are also
needed to enable an informed research community to develop multidisciplinary
research programs that can simultaneously advance the materials and AI fields.
Second, generative models allow for the generation of samples from probability
distributions. Traditional methods, such as the Monte Carlo method, have been
used for some time. Other methods, specifically autoencoders and generative
adversarial networks, have rapidly advanced other areas of machine learning for the
generation of artificial data. Examples include computer-generated art and video.
These methods are relevant to AI4M as they employ a latent-space representation,
i.e., a relatively low dimensional vector space to represent the problem of interest.
The points in this space can be clustered according to properties of interest, and can
lead to the understanding and generalization of concepts from the materials data.
Generating models that not only give the correct predictions, but also lead to new
understandings of materials chemistry and physics is an important challenge in
machine learning. The overall problem of generating human-interpretable machine
learning results is an active area in computer science research (e.g., Fairness,
Accountability, and Transparency in Machine Learning (FAT/ML) initiative [1], DARPA’s
Explainable AI program [2]). Attaining “understanding” of materials phenomena
requires translation of knowledge across different physical models and types of
computational and experimental data; therein lie not only the initial barriers to
38
development of materials-specific AI algorithms, but also the opportunities for
these algorithms to supersede human capabilities of deriving understanding from
data. Moreover, the many external constraints, such as experimental cost, hazards,
desirability, and regulatory concerns, must be appropriately incorporated into
material-specific AI. In the short term, human-machine cooperation may be the
most effective way to capture the tacit knowledge, as demonstrated in a recent
application to the optimization of nuclear fusion experiments [3]. Materials-specific
software platforms that facilitate rapid and broad collection of human expertise,
e.g., via crowdsourcing, are required for building the datasets needed to train AI4M
to reproduce scientific expertise [4].
39
Reasoning and AI model interpretation offer a direct insight into and facilitate
inverse design (Goal 4). AI also provides a necessary technology to facilitate a
dramatic breakthrough in modular materials robotics (Goal 3) and bridging length
and time scales (Goal 5).
One example of the state of the art in this area is the use of machine learning to
predict previously unknown similarities between inorganic materials (Figure 5.1),
40
then to use these similarities to rapidly predict the properties of materials. For
example, AFLOW consortium data make it possible to find similarities that can
predict important properties relating to light absorption and mechanical strength
(e.g., metal/insulator classification, band gap energy, bulk/shear moduli, Debye
temperature, heat expansion coefficient, heat capacities), which can be used to
predict these properties for virtually any new inorganic crystalline solid [12]. In
turn, this prediction can be used to extract simple heuristic rules that explain “why”
a particular material has these properties, in terms of thermodynamic properties
of the atoms and chemical bonds comprising the material. These simple rules
not only reduce the time for estimating properties from >1000 hours/material to
<0.1 second/material, but also provide human-interpretable insight and facilitate
inverse design (Goal 4, Chapter 7).
A second example is the discovery of new materials for catalysis. Starting from a
set of detailed density functional theory simulations of catalyst surfaces and their
interaction with the substrates, it was possible to use machine learning methods
to extract the key quantities that predict the catalytic activity. To date, these key
quantities are generalizations of existing theoretical models of catalysis, such
as finding a relationship to the d-band centre [13] and d-band width [14] of the
material, coordination number of the surface catalytic atoms [15], and atomic-orbital
specific coordination numbers of those surface atoms [16]. These relationships
“explain” the detailed simulation results in terms of existing theories, and also
serve as simple rules-of-thumb for the design of new materials. Moving forward,
the challenges in this field include (i) using developing AI methods that can use this
type of information to plan the most promising “next move” in unexplored chemical
space that will yield enhanced performance, and (ii) developing machine learning
methods that can go beyond existing catalysis concepts.
41
5.4.2 Organic materials
Quantum mechanics computation, combined with the advent of supercomputing, has
drastically increased our ability to understand and identify materials computationally.
AI4M offers a new revolution in computational efficiency and efficacy with two main
high-impact avenues: (i) enhancing the speed and accuracy of calculations for a
family of materials, and (ii) using data and understanding of different methods to
design and automatically execute computation campaigns.
In the first area, phenomenal success has been recently demonstrated in the
computational modelling of organic molecules through adoption of machine learning
methods. Instead of performing expensive quantum mechanics calculations on large
molecular screening libraries, calculations for a subset of the library were used to
train a more efficient machine learning model for discovering organic light-emitting
diodes (LEDs) [17]. Neural networks have also been used to learn an appropriate
quantum mechanical model of organic molecules that, when incorporated into mass
computation, offered a million-fold increase in computational efficiency (Figure
5.2) [18]. For fast calculations to generate scientific knowledge, the calculation
accuracy must be optimized and quantified. Machine learning methods such as
Bayesian statistics have also enabled rapid progress in this area, by identifying
when fast calculations can retain the accuracy of more time-consuming methods
and by learning how to identify and correct systematic errors in the calculations
[19, 20, 21, 22, 23, 24].
Figure 5.2 Comparison between speed and accuracy of neural network versus DFT in energy
prediction for organic molecules.
42
[25]. ORGAN can be trained to produce more diverse molecules through a reward
mechanism that highly resembles classical conditioning in psychology, which can
be good at producing outliers. Variational autoencoder representations are another
recent development widely applied to sentence and image generation [26]. These
methods attempt to map discrete molecules to a continuous space. Once in this
continuous space, it is possible to move around in the chemical space and find
new molecules outside of the training set. In the example above, the authors found
two molecules that scored better than any of the molecules in their training set.
Figure 5.3 (a) Finding the shortest path between cities connected by a set of roads is a challenging
computer science problem, although efficient algorithms exist for finding nearly optimal solutions.
(b) Optimizing the series of chemical reactions needed to synthesize a new organic molecule is
an analogous problem. Individual types of molecules correspond to “cities” and the reactions that
interconvert them correspond to “roads”; often these are “one-way streets” because the reactions
are irreversible. [Source: adapted from ref 31]
43
above, machines can suggest completely new reactions, opening new pathways in
the synthesis network or, in the language of Google Maps, create new roads and
highways to reach new destinations, i.e., novel materials.
This process revealed three previously unknown hypotheses, which have subsequently
been verified in the laboratory, about the experimental conditions and properties
needed to produce these materials. Future work should focus on (i) extending these
types of explanations to include more general types of experimental descriptions and
use various forms of characterization and computational results as justifications,
and (ii) closing the loop by allowing AIs to plan future experiments that optimally test
these hypotheses under the direction of human experts. Both of these require the
development of materials-specific AI methods that can present evidence convincingly
to materials scientists and chemists.
Machine learning has already been used for microstructure [33] and thermal
properties optimization of stanene nanostructures [34]. While exciting developments
are taking place in this field, the key challenges remain unresolved. Algorithms must
be developed to make autonomous decisions to discover new materials rather than
optimize properties out of known ones. Most of the studies so far have shown the
importance of machine learning for specific properties and materials systems.
Standardized methods and best practices for using AI tools for nanocomposite
design and optimization are still lacking. Furthermore, the reliability of developed
methods for new applications, properties, and materials systems has not been
assessed properly. A limited set of machine learning tools cater to the prediction
of morphologies and kinetic behaviour such as phase transformation. Addressing
this challenge requires better theoretical methods to resolve time-scale issues
in atomistic modelling (Goal 5, Chapter 8) of polymers and composite materials.
Machine learning tools do not yet utilize sophisticated multi-scale methods that
could resolve the kinetics challenge. Finally, since current tools are not open source
or integrated into open source packages, they are not easily accessible. As Goal
6 (Chapter 9) envisions open data access, open access to AI4M algorithms and
concepts is critical for meeting MAP’s ambitious vision.
44
Figure 5.4 Teaching humans what AIs know about the conditions needed to synthesize new materials.
Although the AI uses a model that is too complicated to be interpreted by humans, it can be used to
create a “model of a model” that represents its knowledge as a tree of “if-then” decisions in terms
of chemical properties and reaction conditions. (Left Inset) Human experts can read the colour-
coded path of decisions that lead to successfully making a new material, and from this obtain new,
previously unknown hypotheses about the reaction formation conditions (shown schematically in
the right inset). [Source: ref 32]
45
References for Chapter 5
1. Fairness, Accountability, and Transparency in Machine Learning (FAT/ML) (website), accessed 2017,
www.fatml.org/.
2. David Gunning, “Explainable Artificial Intelligence (XAI),” Defense Advanced Research Projects
Agency (DARPA), accessed 2017, www.darpa.mil/program/explainable-artificial-intelligence.
3. E. A. Baltz, E. Trask, M. Binderbauer, M. Dikovsky, H. Gota, R. Mendoza, J. C. Platt and P. F. Riley,
“Achievement of Sustained Net Plasma Heating in a Fusion Experiment with the Optometrist
Algorithm,” Scientific Reports 7, Number 6425 (2017), DOI:10.1038/s41598-017-06645-7.
4. Michael Nielson, Reinventing Discovery: The New Era of Networked Science (Princeton University
Press, 2011).
5. Demis Hassabis, Dharshan Kumaran, Christopher Summerfield and Matthew Botvinick,
“Neuroscience-Inspired Artificial Intelligence,” Neuron 95, Issue 2 (2017): 245-258, DOI: 10.1016/j.
neuron.2017.06.011.
6. Jason R. Hattrick-Simpers, John M. Gregoire and A. Gilad Kusne, “Perspective: Composition–
structure–property mapping in high-throughput experiments: Turning data into knowledge,” APL
Materials 4, Issue 5 (2016), DOI: 10.1063/1.4950995.
7. L. Ward, R. Liu, A. Krishna, V. I. Hegde, A. Agrawal, A. Choudhary and C. Wolverton, “Including crystal
structure attributes in machine learning models of formation energies via Voronoi tessellations,”
Physical Review B 96, Issue 2, Number 024104 (2017), DOI: 10.1103/PhysRevB.96.024104.
8. Yuma Iwasaki, A. Gilad Kusne and Ichiro Takeuchi, “Comparison of dissimilarity measures for
cluster analysis of X-ray diffraction data from combinatorial libraries,” npj Computational Materials
3, Number 4 (2017), DOI: 10.1038/s41524-017-0006-2.
9. C. J. Long, D. Bunker, X. Li, V. L. Karen, and I. Takeuchi, “Rapid identification of structural phases
in combinatorial thin-film libraries using x-ray diffraction and non-negative matrix factorization,”
Review of Scientific Instruments 80, 103902 (2009), DOI: 10.1063/1.3216809.
10. A. G. Kusne, D. Keller, A. Anderson, A. Zaban and I. Takeuchi, “High-throughput determination of
structural phase diagram and constituent phases using GRENDEL,” Nanotechnology 26 (2015),
DOI: 10.1088/0957-4484/26/44/444002.
11. Santosh K. Suram, Yexiang Xue, Junwen Bai, Ronan Le Bras, Brendan Rappazzo, Richard Bernstein,
Johan Bjorck, Lan Zhou, R. Bruce van Dover, Carla P. Gomes, and John M. Gregoire, “Automated
Phase Mapping with AgileFD and its Application to Light Absorber Discovery in the V–Mn–Nb Oxide
System,” ACS Combinatorial Science 19, Issue 1 (2017): 37–46, DOI: 10.1021/acscombsci.6b00153.
12. Olexandr Isayev, Corey Oses, Cormac Toher, Eric Gossett, Stefano Curtarolo and Alexander
Tropsha, “Universal fragment descriptors for predicting properties of inorganic crystals,” Nature
Communications 8 (2017), DOI: 10.1038/ncomms15679.
13. Xianfeng Ma, Zheng Li, Luke E. K. Achenie and Hongliang Xin, “Machine-Learning-Augmented
Chemisorption Model for CO2 Electroreduction Catalyst Screening,” The Journal of Physical Chemistry
Letters 6, Issue 18 (2015): 3528–3533, DOI: 10.1021/acs.jpclett.5b01660.
14. Juhwan Noh, Jaehoon Kim, Seoin Back and Yousung Jung, “Catalyst design using actively learned
machine with non-ab initio input features towards CO2 reduction reactions,” Cornell University
Library, September 14, 2017, arXiv:1709.04576 [cond-mat.mtrl-sci].
15. Federico Calle-Vallejo, David Loffreda, Marc T. M. Koper and Philippe Sautet, “Introducing structural
sensitivity into adsorption–energy scaling relations by means of coordination numbers,” Nature
Chemistry 7 (2015): 403–410, DOI:10.1038/nchem.2226.
16. Xianfeng Ma and Hongliang Xin, “Orbitalwise Coordination Number for Predicting Adsorption
Properties of Metal Nanocatalysts,” Physical Review Letters 118, 036101 (2017), DOI: 10.1103/
PhysRevLett.118.036101.
17. Rafael Gómez-Bombarelli, Jorge Aguilera-Iparraguirre, Timothy D. Hirzel, David Duvenaud, Dougal
Maclaurin, Martin A. Blood-Forsythe, Hyun Sik Chae, Markus Einzinger, Dong-Gwang Ha, Tony Wu,
Georgios Markopoulos, Soonok Jeon, Hosuk Kang, Hiroshi Miyazaki, Masaki Numata, Sunghan
Kim, Wenliang Huang, Seong Ik Hong, Marc Baldo, Ryan P. Adams and Alán Aspuru-Guzik, “Design
of efficient molecular organic light-emitting diodes by a high-throughput virtual screening and
experimental approach,” Nature Materials 15 (2016): 1120–1127, DOI:10.1038/nmat4717.
18. J. S. Smith, O. Isayev and A. E. Roitberg, “ANI-1: an extensible neural network potential with DFT
accuracy at force field computational cost,” Chemical Science 8 (2017): 3192–3203, DOI: 10.1039/
C6SC05720A.
19. Felix Brockherde, Leslie Vogt, Li Li, Mark E. Tuckerman, Kieron Burke and Klaus-Robert Müller,
“By-passing the Kohn-Sham equations with machine learning,” Cornell University Library, June 15,
2017, arXiv:1609.02815 [physics.comp-ph].
20. Rune Christensen, Heine A. Hansen and Tejs Vegge, “Identifying systematic DFT errors in catalytic
reactions,” Catalysis Science & Technology 5 (2015): 4946-4949, DOI: 10.1039/C5CY01332A.
21. J. Wellendorff, K. T. Lundgaard, A. Mogelhoj, V. Petzold, D. D. Landis, J. K. Norskov, T. Bligard
and K. W. Jacobsen, “Density functionals for surface science: Exchange-correlation model
development with Bayesian error estimation,” Physical Review B 85, Issue 23, 235149 (2012),
DOI: 10.1103/PhysRevB.85.235149.
46
22. Christensen et al., op. cit.
23. Rune Christensen, Heine A. Hansen, Colin F. Dickens, Jens K. Nørskov and Tejs Vegge, “Functional
Independent Scaling Relation for ORR/OER Catalysts,” The Journal of Physical Chemistry C 120,
Issue 43 (2016): 24910–24916, DOI: 10.1021/acs.jpcc.6b09141.
24. Andrew A. Peterson, Rune Christensen and Alireza Khorshidi, “Addressing uncertainty in atomistic
machine learning,” Physical Chemistry Chemical Physics 19, Issue 18 (2017): 10978–10985, DOI:
10.1039/c7cp00375g.
25. Gabriel Lima Guimaraes, Benjamin Sanchez-Lengeling, Carlos Outeiral, Pedro Luis Cunha Farias,
Alán Aspuru-Guzik, “Objective-Reinforced Generative Adversarial Networks (ORGAN) for Sequence
Generation Models,” Cornell University Library, November 29, 2017, arXiv:1705.10843 [stat.ML].
26. Rafael Gómez-Bombarelli, Jennifer N. Wei, David Duvenaud, José Miguel Hernández-Lobato,
Benjamín Sánchez-Lengeling, Dennis Sheberla, Jorge Aguilera-Iparraguirre, Timothy D. Hirzel,
Ryan P. Adams and Alán Aspuru-Guzik, “Automatic chemical design using a data-driven continuous
representation of molecules,” Cornell University Library, January 6, 2017, arXiv:1610.02415v2 [cs.LG].
27. Jennifer N. Wei, David Duvenaud and Alán Aspuru-Guzik, “Neural Networks for the Prediction of
Organic Chemistry Reactions,” ACS Central Science 2, Issue 10 (2016): 725–732, DOI: 10.1021/
acscentsci.6b00219.
28. Bowen Liu, Bharath Ramsundar, Prasad Kawthekar, Jade Shi, Joseph Gomes, Quang Luu Nguyen,
Stephen Ho, Jack Sloane, Paul Wender and Vijay Pande, “Retrosynthetic Reaction Prediction Using
Neural Sequence-to-Sequence Models,” ACS Central Science 3, Issue 10 (2017): 1103–1113, DOI:
10.1021/acscentsci.7b00303.
29. Marwin Segler, Mike Preuss and Mark P. Waller, “Towards ‘AlphaChem’: Chemical Synthesis
Planning with Tree Search and Deep Neural Network Policies,” Cornell University Library, January
31, 2017, arXiv:1702.00020 [cs.AI].
30. Marwin H. S. Segler and Prof. Mark P. Waller, “Neural-Symbolic Machine Learning for Retrosynthesis
and Reaction Prediction,” Chemistry: A European Journal 23, Issue 25 (2017): 5966–5971, DOI:
10.1002/chem.201605499.
31. S. Szymkuć, E. P. Gajewska andT. Klucznik. “Computer‐Assisted Synthetic Planning: the End of
the Beginning,” Angewandte Chemie International Edition 55 (2016), DOI: 10.1002/anie.201506101.
32. P. Raccuglia, K. C. Elbert, P. D. F. Adler, C. Falk, M. B. Wenny, A. Mollo, M. Zeller, S. A. Friedler, J.
Schrier and A. J. Norquist, “Machine-learning-assisted materials discovery using failed experiments,”
Nature 533 (2016): 73–77.
33. Ruoqian Liu, Abhishek Kumar, Zhengzhang Chen, Ankit Agrawal, Veera Sundararaghavan and Alok
Choudhary, “A predictive machine learning approach for microstructure optimization and materials
design,” Scientific Reports 5, Number 11551 (2015), DOI: 10.1038/srep11551.
34. Mathew J. Cherukara, Badri Narayanan, Alper Kinaci, Kiran Sasikumar, Stephen K. Gray, Maria K.
Y. Chan and Subramanian K. R. S. Sankaranarayanan, “Ab Initio-Based Bond Order Potential to
Investigate Low Thermal Conductivity of Stanene Nanostructures,” The Journal of Physical Chemistry
Letters 7, Issue 19 (2016): 3752–3759, DOI: 10.1021/acs.jpclett.6b01562.
47
48
6 Goal #3: Modular Materials Robotics
T
oday, synthesizing materials is an activity reserved for highly
trained specialists. Identifying synthetic pathways is intellectually
different from producing material variations in an effort to develop
critical structure-property relationships and create new materials.
Even among experts, the level of methodological specialization for
synthesis and characterization limits a researcher’s ability to explore
all possible materials for satisfying an application-specific functionality.
Thus, large portions of materials systems remain unexplored, not
due to an inability to imagine islands of promising functionality, but
rather to an inability to apply the appropriate tools to scale these
lofty barriers. Decoupling methodological expertise from materials
discovery would enable researchers to fully explore materials space
to target functionality. Even if this were achieved, the lack of scale in
synthesis and characterization would prevent rapid discovery in novel
materials spaces. Therefore, a flexible and scalable infrastructure is
necessary to translate the understandings developed in the other five
goals into systems that design, synthesize, characterize, and iterate
until discovery.
Many of the general challenges surrounding synthesis revolve around the development
of new technologies that can provide the necessary flexibility and modularity for an
adaptive system. This requires an earnest effort to develop generic interfaces and
protocols that many groups can use to develop tools and techniques. Data formats
used for characterization are often proprietary or cryptic, making integration a
near impossibility. While standards are not required, incentives are needed to
motivate industry to work together on the components of a modular robotic system.
Although significant resources are poured into robotics and automation in other
fields, their development focuses on orthogonal necessities such as bipedal motion
or large-scale stabilization. These, however, are not the most pressing needs for
well-structured research settings that are at the limits of technologies, such as
manipulators or precision and reproducibility in position.
50
the discovery process. In many ways, it would mirror the rapid explosion of interest
and activity in additive manufacturing with the proliferation of three-dimensional
(3D) printing via the maker movement. This type of organic growth naturally lends
itself to self-directed ecosystems that enable all users to leverage not only the
infrastructure, but also the expertise of participants. It could potentially lead to an
exponential increase in materials discovery and deployment for clean energy goals.
51
Figure 6.1 Cluster tools for high-throughput synthesis of inorganic thin films. [Source: refs 2,3]
52
automating synthesis. The wafer geometry dictates much of the design criteria
from manipulators to synthesis chambers to characterization methodologies.
Moving away from this one overarching constraint is difficult because much of
the available sample manipulation technology in research also depends on wafer
geometry. General manipulator systems that provide the precision necessary for
extremely precise and reproducible characterization would enable other technologies
to rapidly automate and bring down the cost of auxiliary technologies for further
development in synthesis automation.
Just as the development of new materials can improve the performance of existing
technologies, a range of transformational new technologies is possible if materials
can be developed with properties that make them economically viable. Thermoelectric
materials could make it possible to harvest electricity from heat-flow or move
heat from the flow of electricity. Magnetocaloric materials could enable super-
efficient heating and cooling of buildings, if materials with the right combination of
properties were identified [4]. Both these classes of materials lack effective scale-up
mechanisms for synthesis and characterization. Scaling is a key challenge to the
employment of any new material, but particularly so if industry lacks experience
with that specific class, chemistry, geometry, etc. Automated facilities could be a
key de-risking technology to industry adoption.
Many opportunities exist for new materials to contribute to energy efficiency and,
correspondingly, many approaches to developing the necessary materials. However,
the approaches and labs that pursue them are largely unconnected today. An
enormous range of synthesis methods is applied to inorganic materials, including
techniques that take place in the solid, liquid, vapour, and plasma phases. In some
cases, the methods are necessarily batch processes and not easily automated. While
it is possible, for example, to contemplate the creation of robotics to undertake
mass-customized tasks such as glass-blowing for the encapsulation of materials to
make single crystal specimens for detailed characterization, it is more profitable to
address processes that are immediately amenable to automation. New paradigms
for solving the same synthesis problems may be the only way to automate a task
such as glass-blowing.
53
6.4.2 Organic materials
Organic molecules are used as medicines, functional coatings, energy storage, and
energy harvesting, to name but a few applications. To design a universal organic
materials synthesis machine is a massive undertaking due to the huge number of
possible molecular structures. An important research question is to determine the
minimal space of molecular architecture that can address the broadest range of
properties, i.e., the functional degeneracy of organic molecules. If a high degree
of functional diversity can be achieved with a small set of structural diversity or
from a set of molecules that can all be obtained using a small set of reactions, the
complexity of the design of an automated device is greatly reduced. Research is
needed to determine the structural space that would enable the broadest functional
space.
The automation of chemical synthesis has been accomplished for a few structural
classes. Peptide synthesis was first automated in the 1960s through the development
of solid-phase peptide synthesis technology. More recently, the synthesis of
polynucleotides and oligosaccharides has been automated. New methods are
being developed to automate the synthesis of polyketide-type products. For all
these methods, the products produced are built up from the iterative use of a
single reaction with structurally related starting materials. There have been efforts
to develop an automated synthesis machine that can perform a broad range of
reactions to assemble pharmaceutical drugs through a multi-step synthetic process
[5]. The ability to automate the mechanics of synthesis has been even extended to
the synthesis of biological molecules such as DNA and proteins [6].
Any high-throughput research pipeline that takes new functional organic compounds
to a fully optimized and tested device integration must incorporate automated
materials characterization tools to achieve a time-efficient feedback loop.
Advanced characterization tools integrated with modular materials robotics, such
as spectroscopies (e.g., NMR, infrared, ultra-fast spectroscopy), are essential to
close the loop of materials design and automated synthesis.
54
materials for hydrogen storage [8], and catalysts for generating fuel from sunlight
[9]. Nanostructured materials of this sophistication must be programmed by multi-
step chemical synthesis, where heterogeneous components are assembled in a
particular order, each under specific conditions. Much like the modular synthesis
of peptides and organic molecules, the atomic structure of inorganic nanomaterials
can be controlled by the regulation of a few key synthesis variables in sequence.
Modular robotics has the potential to enable high-throughput synthesis and evaluation
of these materials to gather and trend data. These automated workflows can potentially
find the limits of a given synthetic method in producing a nanomaterial with a given
chemical or physical property. The ability to “fail fast” using high-throughput,
automated synthesis would enable rapid evaluation of different synthesis recipes
for achieving a target material. Convergence on a single and robust protocol
would discourage proprietary recipes and allow faster entry of newly discovered
nanomaterials into technology space.
55
The batch synthesis of nanoparticles in parallel arrays offers a more flexible platform
for automated nanocrystal synthesis. However, due to the more extreme reaction
conditions, only a few custom-developed robots have been used to automate
the synthesis, characterization, and high-throughput screening of nanoparticles.
The Molecular Foundry, a U.S. DOE nanoscience user facility, pioneered the
automated, high-throughput synthesis of inorganic nanocrystals, introducing its
first-generation robot, WANDA, in 2008 [13]. (See Figure 6.2 for a photograph
of HERMAN, WANDA’s successor.) WANDA and other synthesis robots perform
typical chemistry operations such as dispensing reagents, heating and stirring
reactions, and sampling aliquots into arrays of 8 to 96 reactor wells. Modular
characterization tools such as high-throughput microplate readers measure the
absorption, luminescence, x-ray diffraction, Raman scattering, and dynamic light
scattering of products synthesized by upstream robots.
Figure 6.2 HERMAN. Robotic deck of HERMAN, the High-through Experimentation Robot for
Multiplexed Automation of Nanochemistry. HERMAN, the successor of WANDA, is used to synthesize
colloidal inorganic nanomaterials and nanocomposites at the Molecular Foundry (Lawrence Berkeley
National Laboratory, USA). Each module of HERMAN (horizontally distributed deck elements)
performs fundamental operations such as weighing solid reagents, dispensing liquid solutions,
heating and stirring chemical reactions, sampling products over time, and purifying products, all
inside an inert glovebox atmosphere.
56
Figure 6.3 Microplate. Photoluminescence micrograph of robotically synthesized cadmium selenide
and cadmium telluride nanoparticle solutions in a 96-well microplate under ultraviolet illumination.
The color of each nanoparticle reaction (each well) varies with their size and growth conditions.
[Credit: Emory Chan, Molecular Foundry]
57
industry and a handful of automated transmission electron microscopes have been
custom built for biomedical research, these rare and expensive tools have not been
integrated modularly into a robotic synthesis and screening workflow. Ultimately, the
automation of 2D and 3D TEM characterization modules to match the production
rate of automated synthesis workflows would dramatically accelerate the discovery
of atomically engineered nanomaterials for clean energy and catalysis.
58
References for Chapter 6
1. M. L. Green, C. L. Choi, J. R. Hattrick-Simpers, A. M. Joshi, I. Takeuchi, S. C. Barron, E. Campo, T.
Chiang, S. Empedocles, J. M. Gregoire, A. G. Kusne, J. Martin, A. Mehta, K. Persson, Z. Trautt1,
J. Van Duren, and A. Zakutayev, “Fulfilling the promise of the materials genome initiative with
high-throughput experimental methodologies,” Applied Physics Reviews 4, Issue 1, 011105 (2017),
DOI: 10.1063/1.4977487.
2. Cluster tool diagram (website), accessed 2017, https://www.nrel.gov/pv/assets/images/ap-diagram.jpg.
3. Cluster tool image (website), accessed 2017, http://www.sentech.com/files/1/0/0/10708.jpg.
4. K. A. Gschneidner and V. K. Pecharsky, “Thirty years of near room temperature magnetic cooling:
Where we are today and future prospects,” International Journal of Refrigeration 31 (2008): 945–961,
DOI: 10.1016/j.ijrefrig.2008.01.004.
5. Junqi Li, Steven G. Ballmer, Eric P. Gillis, Seiko Fujii, Michael J. Schmidt, Andrea M. E. Palazzolo,
Jonathan W. Lehmann, Greg F. Morehouse and Martin D. Burke, “Synthesis of many different
types of organic small molecules using one automated process,” Science 347, Issue 6227 (2015)
1221–1226, DOI: 10.1126/science.aaa5414.
6. Craig Venter in Nature Biotechnology 35, 672–675 (2017) doi:10.1038/nbt.3859
7. Anna Llordés, Guillermo Garcia, Jaume Gazquez and Delia J. Milliron, “Tunable near-infrared and
visible-light transmittance in nanocrystal-in-glass composites,” Nature 500 (2013): 323–326, DOI:
10.1038/nature12398.
8. Ki-Joon Jeon, Hoi Ri Moon, Anne M. Ruminski, Bin Jiang, Christian Kisielowski, Rizia Bardhan
and Jeffrey J. Urban, “Air-stable magnesium nanocomposites provide rapid and high-capacity
hydrogen storage without using heavy-metal catalysts,” Nature Materials 10 (2011): 286–290,
DOI: 10.1038/nmat2978.
9. Vismadeb Mazumder, Youngmin Lee and Shouheng Sun, “Recent Development of Active Nanoparticle
Catalysts for Fuel Cell Reactions,” Advanced Functional Materials 20 (2010): 1224–1231, DOI:
10.1002/adfm.200902293.
10. Maria Ibáñez, Zhishan Luo, Aziz Genç, Laura Piveteau, Silvia Ortega, Doris Cadavid, Oleksandr
Dobrozhan, Yu Liu, Maarten Nachtegaal, Mona Zebarjadi, Jordi Arbiol, Maksym V. Kovalenko and
Andreu Cabot, “High-performance thermoelectric nanocomposites from nanocrystal building
blocks,” Nature Communications 7, Number 10766 (2016), DOI: 10.1038/ncomms10766.
11. Emory M. Chan, Richard A. Mathies, and A. Paul Alivisatos, “Size-Controlled Growth of CdSe
Nanocrystals in Microfluidic Reactors,” Nano Letters 3, Issue 2 (2003): 199–201, DOI: 10.1021/
nl0259481.
12. Samuel Marre and Klavs F. Jensen, “Synthesis of micro and nanostructures in microfluidic systems,”
Chemical Society Reviews 39 (2010): 1183–1202, DOI: 10.1039/B821324K.
13. Emory M. Chan, Chenxu Xu, Alvin W. Mao, Gang Han, Jonathan S. Owen, Bruce E. Cohen and Delia
J. Milliron, “Reproducible, High-Throughput Synthesis of Colloidal Nanocrystals for Optimization in
Multidimensional Parameter Space,” Nano Letters 10, Issue 5 (2010): 1874–1885, DOI: 10.1021/
nl100669s.
14. Jonathan S. Owen, Emory M. Chan, Haitao Liu and A. Paul Alivisatos, “Precursor Conversion
Kinetics and the Nucleation of Cadmium Selenide Nanocrystals,” Journal of the American Chemical
Society 132, Issue 51 (2010): 18206–18213, DOI: 10.1021/ja106777j.
15. Michael D. Clark, Sanat K. Kumar, Jonathan S. Owen and Emory M. Chan, “Focusing Nanocrystal
Size Distributions via Production Control,” Nano Letters 11, Issue 5 (2011): 1976–1980, DOI:
10.1021/nl200286j.
16. Joseph P. Feser, Emory M. Chan, Arun Majumdar, Rachel A. Segalman and Jeffrey J. Urban,
“Ultralow Thermal Conductivity in Polycrystalline CdSe Thin Films with Controlled Grain Size,”
Nano Letters 13, Issue 5 (2013): 2122–2127, DOI: 10.1021/nl400531f.
17. Jacob D. Kilbane, Emory M. Chan, Christian Monachon, Nicholas J. Borys, Elizabeth S. Levy, Andrea
D. Pickel, Jeffrey J. Urban, P. James Schuck and Chris Dames, “Far-field optical nanothermometry
using individual sub-50 nm upconverting nanoparticles,” Nanoscale 8, Issue 22 (2016): 11611–11616,
DOI: 10.1039/c6nr01479h.
18. Saahil Mehra, Amy Bergerud, Delia J. Milliron, Emory M. Chan, and Alberto Salleo, “Core/Shell
Approach to Dopant Incorporation and Shape Control in Colloidal Zinc Oxide Nanorods,” Chemistry
of Materials 28, Issue 10 (2016): 3454–3461, DOI: 10.1021/acs.chemmater.6b00981.
19. Chan et al. 2010, op. cit.
20. Chan et al. 2010, op. cit.
21. Mehra et al., op. cit.
22. Chan et al. 2010, op. cit.
23. Emory M. Chan, Gang Han, Joshua D. Goldberg, Daniel J. Gargas, Alexis D. Ostrowski, P. James
Schuck, Bruce E. Cohen, and Delia J. Milliron, “Combinatorial Discovery of Lanthanide-Doped
Nanocrystals with Spectrally Pure Upconverted Emission,” Nano Letters 12, Issue 7 (2012):
3839–3845, DOI: 10.1021/nl3017994.
24. Daniel J. Gargas, Emory M. Chan, Alexis D. Ostrowski, Shaul Aloni, M. Virginia P. Altoe, Edward
S. Barnard, Babak Sanii, Jeffrey J. Urban, Delia J. Milliron, Bruce E. Cohen and P. James Schuck,
“Engineering bright sub-10-nm upconverting nanocrystals for single-molecule imaging,” Nature
Nanotechnology 9 (2014): 300–305, DOI: 10.1038/NNANO.2014.29.
25. Feser et al., op. cit.
59
60
7 Goal #4: Inverse Design
T
he current approach to materials discovery centres on the use
of human intuition as the starting point and driver for desired
functionality. That intuition is the basis for a search on structure,
composition, and properties, which, in a linear fashion, reduce into
hypotheses, testing methodologies, and iterative processes that update
the initial intuition until the desired target functionality is achieved.
Computational screening has accelerated the “guess and check” process,
but it will always be hampered by the limits of this Edisonian paradigm.
7.1 OPPORTUNITIES
The fundamental challenges in realizing generic inverse design strategies
emanate from the strong correlation of steps involved in conventional
design processes. Several paradigm changes are required to ensure
that this ideology does not become a wrapper around a still inefficient
Edisonian process, but rather disrupts the perceptions of materials
discovery.
61
Figure 7.1 Schematic representation of a computational inverse materials design process. After
devising design principles and generating a set of candidates that satisfy them, theory and experiment
are employed in a high-throughput manner to rapidly screen the candidates down to a short list.
Subsequent advanced characterization is conducted and all aspects of the approach are integrated
to aid in further rational inverse design. [Source: Based on ref 1]
Even if all the data and models actually existed, there is very little understanding
on how researchers should mix structured and unstructured data. Methods of
incorporating the unknown are necessary to have a “built-in” ability to discover
transformational materials that do not exist in the current models of understanding,
e.g., the discovery of quasi-crystals. There is an opportunity to expand the inverse
design process to take materials discovery into new regions of understanding,
opening up all possible functionality, rather than only what we currently grasp.
62
breakthrough materials are non-equilibrium in a global sense. Key to surmounting
this challenge is incorporating fingerprints that represent synthesis to serve as
proxies for complex dynamical considerations that could alleviate complexity and
make the inverse design process tractable. These fingerprints would be synthesis,
chemistry, structural, and domain specific to ensure that a variety of dynamical
considerations, from kinetics of atomic motion to macroscopic degradation such
as corrosion and fatigue, could be adequately incorporated into the search.
More generally, inverse design processes can improve the efficiency of existing
infrastructure for synthesis and characterization. Fingerprints are not limited to
fundamental materials building blocks, but rather any combination of variables that
define the search space for an outcome. Thus, an inverse design process can be
used to direct the most appropriate characterization conditions instead of searching
for optimal parameters via brute force. Similarly, in synthesis, the best parameters
would be directed rather than randomly searching for them, reducing the waste
and ensuring that the global optimum is achieved rather than local optima, as is
often the case when a human is directly optimizing a process. This would enable
crucial scaling necessary for a materials breakthrough to translate from a lab-scale
curiosity to an industrially made and applied breakthrough.
63
building blocks to the appropriate scope and ensuring consideration of all aspects
of materials parameters, from atomic to engineering scale.
Basic inverse design methodology has already been applied to several challenges
in inorganic semiconductor materials [5]. For example, silicon and germanium are
the most widely used materials for producing solar cells, but they are inherently
inefficient at absorbing light due to their indirect band gaps. In principle, specially
designed alternating layers of silicon and germanium (Si/Ge heterostructures) can
have a direct band gap that more efficiently absorbs light, thus reducing the amount
of material — and hence the cost — of solar cells. Here, the modular building blocks
are Si and Ge atoms, which can be arranged in an astronomical number of possible
stoichiometries and configurations, e.g., solid solutions, mixtures, layers, varying
layer thicknesses. Zunger and co-workers demonstrated that low-computational-
cost proxy models could be used to assess the stability of the arrangements and
light absorption properties, and genetic algorithms used to direct the search for
spatial arrangements with this property [6]. A similar approach can be used to
design layers in Si/Ge nanowires with optimal solar light-absorbing properties [7],
and even Si/Ge materials with optimal performance for quantum computers [8].
64
Figure 7.2 A schematic illustration of an informatics-driven loop enabling design of novel DSSC
material. [Source: ref 9]
In both these examples, human intuition still played a key role in inverse design.
Fully automated inverse design would require minimal definition from humans on
structure-property relationships and find the most appropriate connections based
on the available data. Nevertheless, these examples showcase paradigms that can
be templated with simple substitutions, reducing the human involvement in decision
making, and expanding scope from simple structural and property models to fully
integrated atomistic- to engineering-scale descriptions.
A key challenge is the ability to handle more than the structure of single molecules
and encompass the full spatial-scale, non-equilibrium arrangements of the molecules
65
Figure 7.3 Example of inverse design for improved power output in organic photovoltaic materials
by increased interfacial roughness. [Source: Dr. Geoffrey Hutchison, University of Pittsburgh]
66
Figure 7.4 Generic algorithms for prediction of the global optimum icosahedral Ni@Cu core-shell
configuration for 2 nm CuNi nanoalloys. [Source: ref 17]
The next step towards inverse nanocatalyst design is identifying the specific particle
size, composition, and structure that yield the highest total particle activity. This is
already within reach through use of materials descriptors and scaling relations as a
fingerprint of the catalytic activity. Using a Gaussian process surrogate model trained
on adsorption energies and group additivity fingerprints, Nørskov et al. recently
demonstrated that it is possible to predict the most important reaction step on a
simple metal surface to be calculated explicitly without computationally demanding
electronic structure theory [20]. An inverse design scenario for nanoparticle catalysts
would also need to include the ability to assess, differentiate, and generate multiple
possible reaction pathways on a wide distribution of ill-defined, under-coordinated,
and strained sites, which remains beyond existing techniques.
67
References for Chapter 7
1. Center for Inverse Design (website), accessed 2017, http://www.centerforinversedesign.org/
approach.html
2. Sang Moon Kim, Chi-Yeong Ahn, Yong-Hun Cho, Sungjun Kim, Wonchan Hwang, Segeun Jang,
Sungsoo Shin, Gunhee Lee, Yung-Eun Sung and Mansoo Choi, “High-performance Fuel Cell with
Stretched Catalyst-Coated Membrane: One-step Formation of Cracked Electrode,” Scientific Reports
6, Number 26503 (2016), DOI: 10.1038/srep26503.
3. Millicent B. Smith and Josef Michl, “Singlet Fission,” Chemical Reviews 110, Issue 11 (2010):
6891–6936, DOI: 10.1021/cr1002613.
4. Chun-Te Wu, Wen-Pin Liao and Jih-Jen Wu, “Three-dimensional ZnO nanodendrite/nanoparticle
composite solar cells,” Journal of Materials Chemistry 21, Issue 9 (2011): 2871–2876, DOI: 10.1039/
c0jm03481a.
5. I. E. Castelli and K. W. Jacobsen, “Designing rules and probabilistic weighting for fast materials
discovery in the Perovskite structure,” Modelling and Simulation in Materials Science and Engineering
22, Number 055007 (2014), DOI: 10.1088/0965-0393/22/5/055007.
6. Mayeul d’Avezac, Jun-Wei Luo, Thomas Chanier and Alex Zunger, “Genetic-Algorithm Discovery of
a Direct-Gap and Optically Allowed Superstructure from Indirect-Gap Si and Ge Semiconductors,”
Physical Review Letters 108, Number 027401 (2012), DOI: 10.1103/PhysRevLett.108.027401.
7. Lijun Zhang, Mayeul d’Avezac, Jun-Wei Luo and Alex Zunger, “Genomic Design of Strong Direct-Gap
Optical Transition in Si/Ge Core/Multishell Nanowires,” Nano Letters 12, Issue 2 (2012): 984–991,
DOI: 10.1021/nl2040892.
8. Lijun Zhang, Jun-Wei Luo, Andre Saraiva, Belita Koiller and Alex Zunger, “Genetic design of enhanced
valley splitting towards a spin qubit in silicon,” Nature Communications 4, Number 2396 (2013),
DOI: 10.1038/ncomms3396.
9. Taylor Moot, Olexandr Isayev, Robert W. Call, Shannon M. McCullough, Morgan Zemaitis, Rene Lopez,
James F. Cahoon and Alexander Tropsha, “Material informatics driven design and experimental
validation of lead titanate as an aqueous solar photocathode,” Materials Discovery 6, Issue C
(2016), DOI: 10.1016/j.md.2017.04.001.
10. Rafael Gómez-Bombarelli, Jorge Aguilera-Iparraguirre, Timothy D. Hirzel, David Duvenaud, Dougal
Maclaurin, Martin A. Blood-Forsythe, Hyun Sik Chae, Markus Einzinger, Dong-Gwang Ha, Tony Wu,
Georgios Markopoulos, Soonok Jeon, Hosuk Kang, Hiroshi Miyazaki, Masaki Numata, Sunghan
Kim, Wenliang Huang, Seong Ik Hong, Marc Baldo, Ryan P. Adams and Alán Aspuru-Guzik, “Design
of efficient molecular organic light-emitting diodes by a high-throughput virtual screening and
experimental approach,” Nature Materials 15 (2016): 1120–1127, DOI:10.1038/nmat4717.
11. Johannes Hachmann, Roberto Olivares-Amaya, Adrian Jinich, Anthony L. Appleton, Martin A. Blood-
Forsythe, László R. Seress, Carolina Román-Salgado, Kai Trepte, Sule Atahan-Evrenk, Süleyman
Er, Supriya Shrestha, Rajib Mondal, Anatoliy Sokolov, Zhenan Baod and Alán Aspuru-Guzik, “Lead
candidates for high-performance organic photovoltaics from high-throughput quantum chemistry
– the Harvard Clean Energy Project,” Energy & Environmental Science 7, Issue 2 (2014): 698, DOI:
10.1039/C3EE42756K.
12. Rafael Gómez-Bombarelli, Jennifer N. Wei, David Duvenaud, José Miguel Hernández-Lobato,
Benjamín Sánchez-Lengeling, Dennis Sheberla, Jorge Aguilera-Iparraguirre, Timothy D. Hirzel,
Ryan P. Adams and Alán Aspuru-Guzik, “Automatic chemical design using a data-driven continuous
representation of molecules,” Cornell University Library, January 6, 2017, arXiv:1610.02415v2 [cs.LG].
13. Benjamin Sanchez-Lengeling, Carlos Outeiral, Gabriel L. Guimaraes and Alan Aspuru-Guzik, “Optimizing
distributions over molecular space. An Objective-Reinforced Generative Adversarial Network for
Inverse-design Chemistry (ORGANIC),” ChemRxiv, chemrxiv.org/articles/ORGANIC_1_pdf/5309668.
14. Marcus Müller and De-Wen Sun, “Process-directed self-assembly of block copolymers: a computer
simulation study,” Journal of Physics: Condensed Matter 27, Number 19 (2015), DOI: 10.1088/0953-
8984/27/19/194101.
15. Ryan S. Gebhardt, Pengfei Du, Olga Wodo and Baskar Ganapathysubramanian, “A data-driven
identification of morphological features influencing the fill factor and efficiency of organic
photovoltaic devices,” Computational Materials Science 129 (2017): 220–225, DOI: 10.1016/j.
commatsci.2016.12.020.
16. Alexander Y. Piggott, Jan Petykiewicz, Logan Su and Jelena Vučković, “Fabrication-constrained
nanophotonic inverse design,” Scientific Reports 7, Number 1786 (2017), DOI: 10.1038/s41598-
017-01939-2.
17. Steen Lysgaard, David D. Landis, Thomas Bligaard and Tejs Vegge, “Genetic Algorithm Procreation
Operators for Alloy Nanoparticle Catalysts,” Topics in Catalysis 57 (2014): 33–39, DOI: 10.1007/
s11244-013-0160-9.
18. Steen Lysgaard, Jón S. G. Mýrdal, Heine A. Hansen and Tejs Vegge, “A DFT-based genetic algorithm
search for AuCu nanoalloy electrocatalysts for CO2 reduction,” Physical Chemistry Chemical Physics
17 (2015): 28270–28276, DOI: 10.1039/c5cp00298b.
19. Tao Cheng, Yufeng Huang, Hai Xiao and William A. Goddard III, “Predicted Structures of the Active
Sites Responsible for the Improved Reduction of Carbon Dioxide by Gold Nanoparticles,” The Journal
of Physical Chemistry Letters 8, Issue 14 (2017): 3317–3320, DOI: 10.1021/acs.jpclett.7b01335.
20. Zachary W. Ulissi, Andrew J. Medford, Thomas Bligaard and Jens K. Nørskov, “To address surface
reaction network complexity using scaling relations machine learning and DFT calculations,” Nature
Communications 8, Number 14621 (2017), DOI: 10.1038/ncomms14621.
68
Goal #5: Bridging Length and Time
8 Scales
M
aterials systems frequently demand understanding and
control of properties that span wide ranges of length and
time scales. For example, mechanical properties of materials
depend on atom-atom adhesion at the nanoscale, crack formation
and grain-boundary interactions at the mesoscale, and stress/strain
relations at the macroscopic scale. In addition to determining bulk
material properties, a range of different length scales dominate the
function of complete devices (Figures 8.1-8.3), and complex interfacial
interactions between combinations of materials are crucial to the
design and function of most devices, such as solar cells, batteries,
organic light-emitting devices, and others. Such length scales can
range over 10 orders of magnitude, from sub-nanometre to multiple
metres or larger. Similarly, event time scales for materials can range
from femtoseconds, e.g., in the case of photoexcitation in solar cells,
to years and decades in the case of long-term photodegradation and
fatigue. Since important phenomena at vastly different scales are
frequently interrelated, methods that can bridge all relevant scales
consistently are needed to achieve a comprehensive and predictive
understanding of material properties.
Figure 8.1 Range of length scales that must be considered to understand the
performance characteristics of a device (left), down to its material components
(right). [Source: ref 1]
69
Figure 8.2 The importance of interfacial interactions between materials for a device. [Source: ref 2]
Computationally, techniques for specific length and time scales include atomic-
level quantum mechanical calculations, nanoscale classical molecular dynamics
simulations, and mesoscale and macroscopic continuum field methods (Figure
8.4). Such techniques are usually implemented in separate software programs
and written in different programming languages, which are not easily integrated.
Parameterization and validation of theoretical models and methods are often
narrowly scoped and unautomated. Model development to bridge scales is usually
Figure 8.3 The performance of a flexible organic solar cell depends on electronic processes that
occur on a wide range of length scales, down to the atomic level.
70
undertaken on a case-by-case basis, such as in the development of multi-scale
coarse-grained (MS-CG) molecular dynamics models from atomistic models [3].
Hybrid-resolution methods, such as quantum mechanics/molecular mechanics (QM/
MM) [4], and particle/continuum [5] approaches that can bridge some scales have
been developed, but current implementations do not allow for general application
or automation. Different computational methods are adapted for different scales,
but no general method exists for partitioning space or time between different
scales to be treated by different techniques. Some progress, however, has been
made, such as in the mapping of atomistic to coarse-grained molecular models
[6]. General-purpose software for some methods/scales does not exist or is still
in the early stages of development, as in the case of field-theoretic or dynamical
(classical) density functional theory simulations of mesoscale to macroscopic
properties [7]. A general, integrated theoretical framework and computational
platform for treating all relevant length and time scales to simulate and predict
material properties does not exist, nor do data or software standards for interfacing
software and experiments designed to treat different scales.
To date, most theory and simulation have been devoted to calculating equilibrium
properties, whereas materials and devices in energy applications often operate
out-of-equilibrium. Basic theories of non-equilibrium processes are much less
Figure 8.4 Breadth of different time and length scales encompassing various levels of theory and
interactions. Multi-scale theoretical and computational methods used for materials model development
and computer simulations. [Source: ref 14]
71
developed, and fewer computational algorithms for addressing non-equilibrium,
time-dependent, or excited-state behaviour are available compared with equilibrium
properties, particularly for phenomena that span many scales [8].
8.1 OPPORTUNITIES
The development of new, automated experimental and computational bridging
tools that give a holistic view of a material is critical to accelerating the design of
better functional materials and devices. Improved data standards and software
interfaces between simulation codes and between experimental instruments
would enable the integration of tools adapted to different scales. A combination
of fundamental theoretical basic research, software development, and advances
in machine learning could allow time and length scales to be systematically and
reliably partitioned between models of different resolution, making possible more
efficient, accurate, and automated multi-scale or hybrid-resolution simulations of
materials. AI tools could also help to bridge both spatial and time scales, e.g., in
the sampling of rare events [15].
72
the wavelengths can be quite different since they are typically long for acoustic
phonons and short for electrons in metals. Thus, the different excited states interact
differently with their surroundings, suggesting new schemes to tailor the electron-
phonon coupling by hierarchically structuring the materials on the nanoscale.
In the next 10 years, synchrotrons are expected to evolve towards fully coherent x-ray
sources, opening up new approaches. When coherent x-rays scatter from samples,
the inhomogeneity is encoded in the scattered wavefronts, and this information
can be analyzed with spatial, chemical, magnetic, and temporal resolution.
From the temporal point of view, chemical processes bridge spatial scales from
the atomic/ultra-fast (bond breaking and formation) to the mesoscale (diffusion)
to the higher-level structures on the microscale. Coherent soft x-ray wavelengths
cover the entire range of length scales, provide chemical sensitivity, and scale from
seconds to nanoseconds. These tools will revolutionize the information that can be
obtained from heterogeneous, multi-step chemical processes.
It may be possible to develop a system that uses synchrotron x-rays as the critical
characterization tool in an optimization loop that explores the morphology of materials.
Such a system could be used to optimize new materials for various applications.
However, to maximally leverage available knowledge of chemical characterization,
advanced characterization techniques should be more closely integrated. Schemes
for sharing, storing, and relating data collecting for characterization at the many
length and time scales are critical for the success of such a program. In addition,
surface and interfacially sensitive characterization techniques are not well developed
enough to integrate cleanly in such a system and fundamental research into their
development is needed. Finally, the framework for integrated characterization
must be adaptable to understanding the impact of defects and impurities, since
these are critical to the performance of many energy materials, e.g., catalysts and
photo-absorbers.
73
fatigue and decay) and solar energy conversion (from individual photo-excitations to
photodegradation). Such tools would accelerate discovery and improve predictions
for inverse design. Rather than waiting for years to determine if a new material will
be stable, predicting this long-term behaviour during short experiments would be
possible. The deeper and broader fundamental understanding of materials (including
heterogeneous, disordered, and non-equilibrium phases, interfaces, and defects),
and of how properties on different length and time scales relate would enable new,
emergent structure-processing-property relationships to be predicted, would lead
to new physics and chemistry (e.g., multi-functional or active materials) that could
be exploited for energy applications. New in situ approaches for characterization of
materials and devices across scales would enable more appropriate characterization
of critical properties relevant to performance under operation conditions. The
development of an integrated multi-scale platform for modelling and characterization
would also create a more multidisciplinary community and foster relationships and
collaboration schemes, enabling and accelerating complementary work.
Achieving this goal would provide an essential building block for many of the other
goals. More efficient experimental and computational tools for bridging scales
would accelerate materials characterization and prediction, thereby speeding up
the design cycles needed for autonomous discovery and development (Goal 1). AI
for materials (Goal 2) needs to make inferences and provide explanations that span
length and time scales. A better fundamental understanding of how interactions
on small length and time scales influence large-scale behaviour would enable more
accurate predictions of functional material properties, such as device performance,
stability, and durability needed for inverse design (Goal 4). Finally, advances in
knowledge of relationships across scales could lead to improvements in how data
infrastructure (Goal 6) is organized.
74
phenomena, with applications ranging from the design of more durable aircraft
alloys to semiconductor chips in modern computers. For example, the length of
solid-electrolyte interface is critically related to both the lifetime and performance
of lithium-ion batteries. Understanding the formation of this layer at multiple
length scales and understanding the time scales of its nucleation and longer-term
consequences is critical to designing more efficient energy storage materials
[19,20]. In another example, perovskites Ba1-xNdxCuO2 and CaCuO2 are both
insulating, but they can only be combined to form a high temperature (high Tc)
superconductor in super lattice form [21].
75
8.4.3 Nanomaterials and composites
Nanomaterials and composites are inherently heterogeneous both in composition
and organization. Functional properties are typically sensitive to interactions between
molecular components, as well as to material interfaces, defects, and disorder on
a range of length scales. For example, organic molecules affixed to quantum dots
suspended in a liquid crystal environment modify the interfacial properties between
the solid and the liquid medium, modulating dispersion and driving self-assembly
(Figure 8.5) [27]. Ultimately, this governs how individual components of the liquid
crystal organize themselves and, therefore, controls function. The heterogeneous
nature of nanomaterials (ligand/particle) or composites requires theoretical
understanding and characterization data covering both molecular components and
nanoscale domains and interfaces. Time-dependent phenomena (e.g., nucleation and
aging) that span a huge range of scales are crucial to understanding and controlling
the synthesis of nanomaterials and composites, as well as to ensuring long-term
stability in applications. Understanding time-dependent behaviour is also important
for designing novel active materials. Understanding of heterogeneity, disorder,
and interfaces inherent to nanomaterials and composites would be enhanced by
the development of new, automated tools to study buried, multi-phase interfaces
(e.g., organic-inorganic, solid-liquid, solid-gas, triple-phase), i.e., translating 2D
to 3D imaging and translating local to ensemble properties. Currently, no tools
exist for chemically mapping surfaces. Heterogeneity and disorder also present a
challenge for the reliable and reproducible synthesis of nanomaterials. Methods
for characterizing and controlling heterogeneity and disorder on a range of scales
are important for designing specific materials for specific applications.
Figure 8.5 Nanoparticle quantum dot (QD) “shells” arranged around a liquid crystal (LC) core. The
arrangement and functionality of the QDs in the shell are dictated by interactions between nanoparticle
surfaces, which may be functionalized by ligand adsorption, and the LC phase. [Source: ref 28]
76
References for Chapter 8
1. Pavel Škarvada, Lubomír Škvarenina, Pavel Tománek, Dinara Sobola, Robert Macků, Jitka
Brüstlová, Lubomír Grmela, Steve Smith, “Multiscale experimental characterization of solar cell
defects,” Proceedings: 20th Slovak-Czech-Polish Optical Conference on Wave and Quantum Aspects
of Contemporary Optics Volume 10142 (2016), DOI: 10.1117/12.2259884.
2. American Chemical Society (ACS), “How a Solar Cell Works,” ChemMatters Online, accessed
2017, www.acs.org/content/acs/en/education/resources/highschool/chemmatters/past-issues/
archive-2013-2014/how-a-solar-cell-works.html.
3. M. G. Saunders and G. A. Voth, “Coarse-graining methods for computational biology,” Annual
Review of Biophysics 42 (2013): 73–93, DOI: 10.1146/annurev-biophys-083012-130348.
4. A. Warshel and M. Levitt, “Theoretical Studies of Enzymic Reactions: Dielectric, Electrostatic and
Steric Stabilization of the Carbonium Ion in the Reaction of Lysozyme,” Journal of Molecular Biology
103 (1976): 227–249, DOI: 10.1016/0022-2836(76)90311-9.
5. Ronald E. Miller and E. B. Tadmor, “A unified framework and performance benchmark of fourteen
multiscale atomistic/continuum coupling methods,” Modelling and Simulation in Materials Science
and Engineering 17, Number 053001 (2009), DOI: 10.1088/0965-0393/17/5/053001.
6. Anton V. Sinitskiy, Marissa G. Saunders and Gregory A. Voth, “Optimal Number of Coarse-Grained
Sites in Different Components of Large Biomolecular Complexes,” The Journal of Physical Chemistry
B 116, Issue 29 (2012): 8363–8374, DOI: 10.1021/jp2108895.
7. Akash Arora, Jian Qin, David C. Morse, Kris T. Delaney, Glenn H. Fredrickson, Frank S. Bates and
Kevin D. Dorfman, “Broadly Accessible Self-Consistent Field Theory for Block Polymer Materials
Discovery,” Macromolecules 49, Issue 13 (2016): 4675–4690, DOI: 10.1021/acs.macromol.6b00107.
8. Marcus Müller and Juan J. de Pablo, “Computational Approaches for the Dynamics of Structure
Formation in Self-Assembling Polymeric Materials,” Annual Review of Materials Research 43 (2013):
1-34, DOI: 10.1146/annurev-matsci-071312-121618.
9. H. Simons, A. King, W. Ludwig, C. Detlefs, W. Pantleon, S. Schmidt, F. Stöhr, I. Snigireva, A. Snigirev
and H. F. Poulsen, “Dark-field X-ray microscopy for multiscale structural characterization,” Nature
Communications 6 (2015), DOI: 10.1038/ncomms7098.
10. A. Junay, S. Guézo, P. Turban, S. Tricot, A. Le Pottier, J. Avila, S. Ababou-Girard, P. Schieffer and
F. Solal, “Effective Metal Top Contact on the Organic Layer via Buffer-Layer-Assisted Growth: A
Multiscale Characterization of Au/Hexadecanethiol/n-GaAs(100) Junctions,” The Journal of Physical
Chemistry C 120, Issue 42 (2016): 24056–24062, DOI: 10.1021/acs.jpcc.6b05729.
11. Nicolas Besnard, Aurélien Etiemble, Thierry Douillard, Olivier Dubrunfaut, Pierre Tran-Van,
Laurent Gautier, Sylvain Franger, Jean-Claude Badot, Eric Maire and Bernard Lestriez, “Multiscale
Morphological and Electrical Characterization of Charge Transport Limitations to the Power
Performance of Positive Electrode Blends for Lithium-Ion Batteries,” Advanced Energy Materials 7
(2017), DOI: 10.1002/aenm.201602239.
12. D. R. Baer, D. J. Gaspar, P. Nachimuthu, S. D. Techane and D. G. Castner, “Application of surface
chemical analysis tools for characterization of nanoparticles,” Analytical and Bioanalytical Chemistry
396 (2010): 983–1002, DOI: 10.1007/s00216-009-3360-1.
13. Gorgi Kostovski, Udayakumar Chinnasamy, Sasani Jayawardhana, Paul R. Stoddart and Arnan
Mitchell, “Nanoimprinting: Sub-15nm Optical Fiber Nanoimprint Lithography: A Parallel, Self-aligned
and Portable Approach,” Advanced Materials 4 (2011), DOI: 10.1002/adma.201002270.
14. Marius Stan, “Discovery and design of nuclear fuels,” Materials Today 12, Issue 11 (2009): 20–28,
DOI: 10.1016/S1369-7021(09)70295-0.
77
15. Tim Mueller, Aaron Gilad Kusne and Rampi Ramprasad, “Machine Learning in Materials Science:
Recent Progress and Emerging Applications,” Chapter 4 in Reviews in Computational Chemistry
29 (2016), Abby L. Parrill and Kenny B. Lipkowitz Eds. (John Wiley & Sons, Inc., 2017), Chapter
DOI: 10.1002/9781119148739.ch4.
16. J. S. Smith, O. Isayev and A. E. Roitberg, “ANI-1: an extensible neural network potential with DFT
accuracy at force field computational cost,” Chemical Science 8 (2017): 3192–3203, DOI: 10.1039/
C6SC05720A.
17. X. Zianni, “Effect of electron-phonon coupling on the thermoelectric efficiency of single-quantum-dot
devices,” Physical Review B 82, 165302 (2010), DOI: 10.1103/PhysRevB.82.165302.
18. Heng Wang, Yanzhong Pei, Aaron D. LaLonde and G. Jeffrey Snyder, “Weak electron–phonon
coupling contributing to high thermoelectric performance in n-type PbSe,” Proceedings of the
National Academy of Sciences 109, Issue 25 (2012): 9705–9709, DOI: 10.1073/pnas/1111419109.
19. A. A. Tahmasbia, T. Kadyka and M. H. Eikerling, “Statistical Physics-Based Model of Solid Electrolyte
Interphase Growth in Lithium Ion Batteries,” Journal of the Electrochemical Society 164, A1307
(2017), DOI: 10.1149/2.1581706jes.
20. J. Vetter, P. Novák, M. R. Wagner, C. Veit, K.-C. Möller, J. O. Besenhard, M. Winter, M. Wohlfahrt-
Mehrens, C. Vogler and A. Hammouche, “Ageing mechanisms in lithium-ion batteries,” Journal of
Power Sources 147, Issue 1 (2005): 269–281, DOI: 10.1016/j.jpowsour.2005.01.006.
21. B. Freelon, A. Augustsson, J.-H. Guo, P. G. Medaglia, A. Tebano and G. Balestrino, “Electron
Correlation and Charge Transfer in [(Ba0.9Nd0.1)CuO2+δ]2/[CaCuO2]2 Superconducting Superlattices,”
Physical Review Letters 96, Number 017003 (2006), DOI: 10.1103/PhysRevLett.96.017003.
22. Marcel Schreier, Laura Curvat, Fabrizio Giordano, Ludmilla Steier, Antonio Abate, Shaik M.
Zakeeruddin, Jingshan Luo, Matthew T. Mayer and Michael Grätzel, “Efficient photosynthesis of
carbon monoxide from CO2 using perovskite photovoltaics,” Nature Communications 6, Number
7326 (2015), DOI: 10.1038/ncomms8326.
23. Joseph H. Montoya, Linsey C. Seitz, Pongkarn Chakthranont, Aleksandra Vojvodic, Thomas F.
Jaramillo and Jens K. Nørskov, “Materials for solar fuels and chemicals,” Nature Materials 16
(2017): 70–81, DOI: 10.1038/nmat4778.
24. M. Cerny, M. Sob, J. Pokluda and P. Sandera, “Ab initio calculations of ideal tensile strength and
mechanical stability in copper,” Journal of Physics: Condensed Matter 16, Issue 7 (2004): 1045–1052,
DOI: 10.1088/0953-8984/16/7/004.
25. Geoffrey R. Hutchison, Mark A. Ratner and Tobin J. Marks, “Electronic Structure and Band Gaps
in Cationic Heterocyclic Oligomers. Multidimensional Analysis of the Interplay of Heteroatoms,
Substituents, Molecular Length, and Charge on Redox and Transparency Characteristics,” The
Journal of Physical Chemistry B 109, Issue 8 (2005): 3126–3138, DOI: 10.1021/jp046579v.
26. S. B. Ghugare, S. Tiwary, V. Elangovan and S. S. Tambe, “Prediction of Higher Heating Value of
Solid Biomass Fuels Using Artificial Intelligence Formalisms,” BioEnergy Research 7, Issue 2 (2014):
681–692, DOI: 10.1007/s12155-013-9393-5.
27. Andrea L. Rodarte, Blessing H. Cao, Harmanpreet Panesar, Ronald J. Pandolfi, Makiko Quint,
Lauren Edwards, Sayantani Ghosh, Jason E. Hein and Linda S. Hirst, “Self-assembled nanoparticle
micro-shells templated by liquid crystal sorting,” Soft Matter 11, Issue 9 (2015): 1701–1707, DOI:
10.1039/C4SM02326A.
28. Ibid.
78
Goal #6: Data Infrastructure and
9 Interchange
A
utonomous materials discovery requires the involvement of
many disparate communities ranging from materials science,
chemistry, physics, and engineering to AI and robotics. Each
community has a unique terminology and set of standards. Further, the
rich and rapidly growing data on materials and their property, synthesis,
and characterization remain largely uncurated, unintegrated, and
unexplored due to the compartmentalization of this information, e.g.,
such data remain in the primary research lab. The full potential of the
effective exploitation of both textual data and published chemical data
often is untapped, mostly due to the lack of tools and focused methods
to curate, integrate, and transform the data into new, experimentally
testable hypotheses.
79
on poor or failed experiments or samples would be deposited along with data on
successful experiments.
9.1 OPPORTUNITIES
As the essential input to AI and machine learning, data directly influence the predictive
capability envisioned for MAP. Integrating existing and future materials data,
originating from theory, synthesis, and characterization, could therefore accelerate
the materials discovery process. Given the depth, breadth, and heterogeneity of
materials property and process data, standardized data formats, data exchange
protocols, and workflows are needed. Building schema conversion and migration
tools could also leverage these protocols to facilitate an open data infrastructure
accessible both to users and machine learning algorithms. This section identifies
some of the opportunities that could help close the gap in data integration and
knowledge sharing.
Incentivizing data contribution and participation could help facilitate the data
infrastructure described above. In particular, contributions from individual groups,
analogous to crowdsourcing, would drive expansion and large-scale use of integrated
materials datasets. Participation could be achieved by requiring compliance with
the mandates of funding agencies or journals or by giving researchers easy-to-use
data tools to help them store, analyze, visualize, and share data while, at the same
time, providing credit attribution. A number of efforts are attempting to do this, e.g.,
MPContribs [3], Citrination [4,5], Materials Data Curation System (MDCS) [6,7,8,9],
Materials Data Facility (MDF) [10,11], Materials Commons [12,13]. Some of these
platforms currently cater more to using machine learning and AI (Citrination), while
others focus on data management (MPContribus, MCDS, MDF, Materials Commons),
particularly large data (MDF).
80
Searchability tools that provide end-to-end tracking of individual contributions
to synthesis, characterization, and workflow development could also encourage
contributions. Industrial researchers should be encouraged to form pre-competitive
data sharing collaborations, similar to existing initiatives for pharmaceutical
development. In short, to grow data infrastructure, incentives to participate should be
offered for all types of research. Beyond capturing future contributions of research
groups, a large corpus of experimental and characterization data, particularly in
areas such as synthesis, is embedded in publications and awaits conversion to
machine-readable, standardized knowledge, for example, via language processing
as demonstrated recently [14].
81
material structure, the first step is access to materials’ characteristics, or property
sets, with interchange of data formats across fields. A major component of bridging
length and time scales (Goal 5) is uniting data across different computational tools,
which would benefit from standardized data storage, availability, and exchange
protocols. Similarly, autonomous discovery and development (Goal 1) require retaining
and interconverting data and metadata at all stages of the research progress to
match characterization and simulation, enabling continuous optimization. For
synthetic production from modular materials robotics (Goal 3), retaining data and
metadata related to particular samples, including starting materials, processing
conditions, purification, quality, and time of production, is needed for integrated
end-to-end autonomous research. Similarly, the development of AI for materials
(Goal 2) requires reliable data, including both positive and negative results.
Figure 9.1 Example of Materials Project repository for inorganic materials, including visualization
of structure and properties. [Source: https://materialsproject.org/]
82
9.4.2 Organic materials
Currently, many open repositories (see Figure 9.2 for two examples), often in the
organic computational quantum chemical space, have little connection to synthesis
or characterization data, or to pharmaceutical compounds. Some proprietary and
commercial repositories (e.g., internal industrial data) do not allow interconversion
or automated queries. Data types include reactions (Reaxys, SciFinder); molecular
geometries (e.g., crystallography through CCSD, Crystallographic Open Database);
spectroscopy (IR, NMR, Mass Spec); and density functional calculations. While
standard molecular identifiers, such as the IUPAC International Chemical Identifier
(InChI) strings, can connect molecular data, more connections are needed, even
between open databases. A large quantity of proprietary data in organic molecules
and polymers (e.g., solubilities, melting points, spectra) could be contributed in a
pre-competitive manner if proper incentives were designed.
Figure 9.2 Examples of molecular repositories including (top) Pitt Quantum Repository [Source:
https://pqr.pitt.edu/] and (bottom) Harvard Clean Energy Project (Molecular Space) [Source: http://
www.molecularspace.org].
83
In contrast to other areas of materials science, for organic materials some protocols
and methods have been established in the cheminformatics community for data
quality, consistency, and interoperability [16,17,18]. Although there are some barriers
to polymer applications, standardization tools could be adapted from other organic
material applications [19].
The incompatibility of data types with current repositories has led to limited
integration of the results of research studies. The creation of a set of standards
would allow the seamless drawing of insights across components and result in a
massive acceleration in discovery and understanding of energy materials. Further, the
integration of data would permit autonomous systems and AI programs to identify
trends, understand behaviours, and predict materials targets, thus enabling complex
multi-dimensional research decisions for human-robot research teams (Figure 9.3).
Figure 9.3 Example of integrated autonomous research, enabled by data interchange across
synthesis/processing, characterization, and analysis.
84
References for Chapter 9
1. The Minerals, Metals & Materials Society (TMS), “Building a Materials Data Infrastructure: Opening
New Pathways to Discovery and Innovation in Science and Engineering,” www.tms.org/mdistudy.
2. Allotrope Foundation, “Solution,” Allotrope Foundation (website), accessed 2017, www.allotrope.
org/solution.
3. Patrick Huck, Dan Gunter, Shreyas Cholia, Donald Winston, Alpha N’Diaye and Kristin Persson,
“User Applications Driven by the Community Contribution Framework MPContribs in the Materials
Project,” Concurrency and Computation: Practice and Experience 28, Issue 7 (2015): 1982–1993,
DOI: 10.1002/cpe.3698.
4. Citrine, “Citrine’s public platform for data-driven materials and chemistry research,” accessed
2017, citrination.com/.
5. Jordan O’Mara, Bryce Meredig and Kyle Michel, “Materials Data Infrastructure: A Case Study of
the Citrination Platform to Examine Data Import, Storage, and Access,” JOM 68, Issue 8 (2016):
2031–2034, DOI: 10.1007/s11837-016-1984-0.
6. Alden Dima, Sunil Bhaskarla, Chandler Becker, Mary Brady, Carelyn Campbell, Philippe Dessauw,
Robert Hanisch, Ursula Kattner, Kenneth Kroenlein, Marcus Newrock, Adele Peskin, Raymond
Plante, Sheng-Yen Li, Pierre-François Rigodiat, Guillaume Sousa Amaral, Zachary Trautt, Xavier
Schmitt, James Warren and Sharief Youssef, “Informatics Infrastructure for the Materials Genome
Initiative,” JOM 68, Issue 8 (2016): 2053–2064, DOI: 10.1007/s11837-016-2000-4.
7. NIST GMI, “Materials Data Curation System,” NIST MGI: Gateway to Materials Genome Information
(website), last updated February 18, 2015, mgi.nist.gov/materials-data-curation-system.
8. NIST GMI, “Facilitating the Development of Modular Data Models in Materials Science,” NIST MGI:
Gateway to Materials Genome Information (website), last updated February 18, 2015, mgi.nist.gov/
facilitating-development-modular-data-models-materials-science.
9. Mark D. Wilkinson et al., “The FAIR Guiding Principles for scientific data management and
stewardship,” Scientific Data 3, Number 160018 (2016), DOI: 10.1038/sdata.2016.18.
10. Materials Data Facility (MDF) (website), accessed 2017, materialsdatafacility.org/.
11. B. Blaiszik, K. Chard, J. Pruyne, R. Ananthakrishnan, S. Tuecke and I. Foster, “The Materials
Data Facility: Data Services to Advance Materials Science Research,” JOM 68, Issue 8 (2016):
2045–2052, DOI: 10.1007/s11837-016-2001-3.
12. Brian Puchala, Glenn Tarcea, Emmanuelle. A. Marquis, Margaret Hedstrom, H. V. Jagadish and
John E. Allison, “The Materials Commons: A Collaboration Platform and Information Repository
for the Global Materials Community,” JOM 68, Issue 8 (2016): 2035–2044, DOI: 10.1007/s11837-
016-1998-7.
13. Materials Commons: Projects, Published Data (website), accessed 2017, materialscommons.org.
14. Edward Kim, Kevin Huang, Alex Tomala, Sara Matthews, Emma Strubell, Adam Saunders, Andrew
McCallum and Elsa Olivetti, “Machine-learned and codified synthesis parameters of oxide materials,”
Scientific Data 4 (2017), DOI: 10.1038/sdata.2017.127.
15. Paul Raccuglia, Katherine C. Elbert, Philip D. F. Adler, Casey Falk, Malia B. Wenny, Aurelio Mollo,
Matthias Zeller, Sorelle A. Friedler, Joshua Schrier and Alexander J. Norquist, “Machine-learning-
assisted materials discovery using failed experiments,” Nature 533 (2016): 73–76, DOI: 10.1038/
nature17439.
16. Noel M O’Boyle, Michael Banck, Craig A. James, Chris Morley, Tim Vandermeersch and Geoffrey
R. Hutchison, “Open Babel: An open chemical toolbox,” Journal of Cheminformatics 3: 33 (2011),
DOI: 10.1186/1758-2946-3-33.
17. Denis Fourches, Eugene Muratov and Alexander Tropsha, “Curation of chemogenomics data,”
Nature Chemical Biology 11, Issue 8 (2015): 535, DOI: 10.1038/nchembio.1881.
18. Denis Fourches, Eugene Muratov and Alexander Tropsha, “Trust, But Verify: On the Importance of
Chemical Structure Curation in Cheminformatics and QSAR Modeling Research,” Journal of Chemical
Information and Modeling 50, Issue 7 (2010): 1189–1204, DOI: 10.1021/ci100176x.
19. Ibid.
20. Diana de la Iglesia, Stacey Harper, Mark D. Hoover, Fred Klaessig, Phil Lippell, Bettye Maddux, Jeffrey
Morse, Andre Nel, Krishna Rajan, Rebecca Reznik-Zellen and Mark T. Tuominen, “Nanoinformatics
2020 Roadmap,” InterNano (website) (2011), eprints.internano.org/607/.
85
86
10 Conclusion
A
chieving the urgent “As the discovery processes for
transition to a advanced materials accelerate,
low-carbon economy the potential economic benefits
requires the development will grow. Thus, private-sector
of new, high-performance, stakeholders that enter early will
low-cost materials that are presumably have a first-mover
safe for humans and the advantage: the know-how to adjust
environment, recyclable, and and gain a larger share of these
use abundant elements so growing benefits.”
that they can be deployed
globally. Today, advanced – Dr. Hermann Tribukait
materials represent about
50% of the manufacturing cost of clean energy, and this is expected
to increase to 80% in the near future. This challenge is not merely an
engineering problem. New fundamental scientific advances are required
to design and organize matter from the atomic scale to the systems
scale. Therefore, the cross-cutting goal of the Clean Energy Materials
Innovation Challenge is crucial: to accelerate the pace of materials
discovery and develop new high-performance, low-cost clean energy
technology solutions.
87
The report may also inform the development of a multidisciplinary workforce for the
discovery, production, and integration of advanced materials. The ultimate success
of these efforts would depend on the support and leadership of all stakeholders,
including academia, governments, industry, non-governmental organizations
(NGOs), foundations, and multilateral organizations. The continuous support of
R&D initiatives, such as the Materials Genome Initiative, can aid the development
and deployment of a discovery workforce that is ready for the challenges ahead
[1,2]. Undoubtedly, international coalitions built around particular topics would
advance the proposed agenda and produce results more effectively.
This final chapter briefly presents the report’s conclusions, beginning with the
opportunities and near-term advances in the six priority areas of the Materials
Acceleration Platform. It then emphasizes the importance of communicating the
benefits of investing in materials science and of educating tomorrow’s students in
multidisciplinary materials design. The chapter closes with some final reflections.
• Artificial Intelligence for Materials: AI models for materials that include constraints,
including constraints dictated by physics for theoretical predictions of properties
as well as experimental constraints, should continue to be developed. Development
and increased usage of novel machine learning algorithms and architectures,
specifically built to accelerate materials discovery, can result in much better
performing models for materials, especially in “low-data” scenarios. Purpose-built
AI tailored for materials discovery is needed to provide breakthroughs.
•
Engaging Students
• Embedding multidisciplinary
Modular Materials experiences
Robotics: Work at should
in this area all levels at universities
focus would
on the development of
educate students and
more generalized promote
interfaces andprograms
protocolsthat supportsystems
for robotic interdisciplinary
in both synthesis
research, spanning climate
and characterization. On theand energy
robotics policy,
front, science (e.g.
the processes thatmaterials
are most science,
amenable
computer science),
to automation needtechnology, engineering,
to be identified. andefforts
Additional mathematics (STEM).
dedicated to identifying
the best ways to characterize the “building blocks” of materials will accelerate
One mechanism
development in to boost
this theThe
area. engagement of both
development of high school and undergraduate
robotic/automation hardware
students to the field of materials discovery would be to implement a transformative
capable of executing research workflows for the synthesis and testing of materials
program similar
is crucial for to the
broader International Genetic Engineered Machines (IGEM)
adoption.
competition in synthetic biology. This initiative engages students from a wide
• range of scientific disciplines and backgrounds. Students could be further
• engaged throughFurther
Inverse Design: studentdevelopment
exchange programs and international
and refinement competitions
of generative machine
with recognized prizes.
learning models will continue to help advance the theoretical side of inverse
88
design. Continued development of discriminative models will also be key due to
the accelerated rate for which properties can be predicted.
• Bridging Length and Time Scales: Automating the current state of the art in multi-
scale simulation will make simulations more accessible to general users. Machine
learning methods must be integrated into all aspects of multi-scale simulation;
integrating characterization tools that bridge length and time scales must also
be developed.
• Closing the Loop: Designing and developing automated platforms that integrate in
situ analysis of a material, either during synthesis and/or after fabrication, into
a single unit will solve some of the current analytical bottlenecks.
Ultimately, all six areas of MAP can be contained in a single facility, housing
“well-to-wheel” materials discovery operations similar to the operational structure
of Tesla’s manufacturing facilities, including the Gigafactory. Autonomous facilities
(or “materials factories”) will accelerate the discovery process by increasing the
pace of experiments performed, harnessing and storing controlled data sets, and
seamlessly integrating these with computational predictions and machine learning
algorithms with robotic and automation tools. Thus, traditionally siloed competence
areas, such as simulation, characterization, and synthesis, can be integrated by
emerging computer science, machine learning, data infrastructure, and robotic
tools. Operations can be tightly tied together, enabling researchers to quickly
receive and incorporate feedback from adjacent processes to make both high-level
decisions and necessary adjustments to optimize design and development. The
proposed platform could also quantify uncertainty in materials discovery to help
decide whether a particular set of experiments is likely to yield promising results.
It could also elevate human creativity away from the details of materials synthesis
and calculation, towards a higher-level control of goals-driven design of theoretical
and experimental studies, where it is most needed.
89
Public engagement efforts could be enhanced by creating strong awareness of
past successes and exciting breakthroughs enabled by public funding of materials
research and development. Telling relatable “stories” about how new materials
discoveries have improved living standards and produced substantial societal
benefits would increase awareness and support, including for efforts dedicated to
training future generations in the quest for better technologies.
One example is the benefits health research to which the general public and
stakeholders can personally relate. In this context, the priority would be to highlight the
advances in areas that connect directly to societal improvements, such as medicine,
information technologies, and transportation—advances that would not have been
possible without crucial contributions from the fundamental sciences: physics,
chemistry, and materials. For example, MRI (materials resonance imaging) relies
on the inventions of NMR and superconductivity, and their practical implementation
on highly functional materials such as magnets and superconductors. Although
these materials are probably best known for their benefits to medicine, they could
also enable a breakthrough in the energy sector. Today, superconductors only
operate at a very low temperature, which incurs tremendous cost. If materials with
high-temperature superconductivity were designed and discovered, the resulting
cables would radically improve efficiency in power transmission and distribution.
Developing new materials with these properties would revolutionize the electric
grid, facilitate wider use of renewable energy resources and result in significant
cost-savings from lower transmission losses.
From solar cells, to lithium-ion batteries and electric vehicles to microchips and the
shale gas revolution, there are many more examples of scientific breakthroughs and
new materials that can help communicate the relevance and impact of materials
discovery. The benefits of developing an integrated, end-to-end materials platform,
such as MAP, can be the subject of many more success stories.
91
92
Acknowledgments
We would like to acknowledge the support and leadership of Mexico’s Energy Secretariat
(Secretaría de Energía, SENER) and the United States Department of Energy (U.S.
DOE), as the lead and co-lead of the Clean Energy Materials Innovation Challenge,
as well as the 16 additional Mission Innovation members that are participating in
this initiative. Sincere appreciation goes to Deputy Secretary Leonardo Beltrán and
Carlos Ortiz of SENER, as well as to Dr. Robert Marlay and Maureen Clapper of the
U.S. DOE, for their valuable leadership and support. As the lead for the Clean Energy
Materials Challenge of Mission Innovation, Dr. Hermann Tribukait of Mexico’s Energy
R&D Innovation Funds served as Executive Chair of the workshop. The Canadian
Institute for Advanced Research (CIFAR) supported numerous fellows and other
experts who contributed to the workshop and this report, and we are thankful for
their committed partnership and the leadership of CIFAR’s CEO and President,
Dr. Alan Bernstein. Dr. Cherry Murray, formerly of the U.S. DOE, presented the
Basic Research Needs (BRN) model at a Mission Innovation meeting in London
and offered her support to this Innovation Challenge, and Dr. Linda Horton of the
U.S. DOE provided valuable guidance, advice and support that were instrumental
to adapt the BRN workshop model of the U.S. DOE to this international effort.
93
production, and support throughout the process provided by Matt Antes, Anna
Mosby, and Julie Chappell of Energetics and Charlotte Franchuk of Oak Ridge National
Laboratory. The team received outstanding support from Elli Saldaña leading the
workshop logistics efforts together with Sostener’s team (Oscar González, Cristian
Dávila, Jorge Hinojosa, Alejandra Villegas, Maritza Soto, and Alonso Ortega). Clare
Walker and Kurt Kleiner of CIFAR edited the report and greatly improved its flow
and readability.
The workshop and report would not have been possible without the expertise
and tireless efforts of the rapporteurs and panel co-chairs, who are also report
co-authors (named below). The report was led by Professors Aspuru-Guzik and
Persson and co-authored by Dr. Alfredo Alexander-Katz of Massachusetts Institute
of Technology; Dr. Carlos Amador and Dr. Diego Solis-Ibarra of UNAM; Matt Antes
and Anna Mosby of Energetics; Dr. Murat Aykol of Toyota Research Institute; Dr.
Emory Chan, Dr. Shyam Dwaraknath, Dr. Joseph Montoya, and Dr. Eli Rotenberg
of LBNL; Dr. John Gregoire of California Institute of Technology; Dr. Jason Hattrick-
Simpers of National Institute of Standards and Technology; Dr. David M. Huang of
University of Adelaide; Dr. Jason Hein of University of British Columbia; Dr. Geoffrey
Hutchison of University of Pittsburgh; Dr. Olexandr Isayev of University of North
Carolina; Dr. Yousung Jung of Korea Advanced Institute of Science and Technology;
Dr. Jari Kiviaho of VTT Technical Research Center of Finland; Dr. Christoph Kreisbeck,
Dr. Loïc Roch, Dr. Semion Saikin, and Dr. Daniel Tabor of Harvard University; Dr.
Juliette Lambert of Centro de Innovación Aplicada en Tecnologías Competitivas;
Dr. Susan Odom of University of Kentucky; Dr. Joep Pijpers of Instituto Nacional
de Electricidad y Energías Limpias; Dr. Michael Ross of University of California-
Berkeley; Dr. Joshua Schrier of Haverford College; Dr. Rachel Segalman of University
of California-Santa Barbara; Dr. Matthew Sfeir of Brookhaven National Laboratory;
Dr. Hermann Tribukait; and Dr. Tejs Vegge of Technical University of Denmark. The
authors are deeply appreciative of the time commitment and insights provided by
the expert workshop participants from around the world.
DISCLAIMER
The content of this report reflects the views and opinions of the participating
workshop experts and report authors and do not necessarily state or reflect those
of their respective government, institution, or any agency thereof or of Mission
Innovation members.
94
Workshop Participants
Name Affiliation Role
Aburto Anell, Jorge Arturo Instituto Mexicano del Petróleo Observer
Alexander-Katz, Alfredo Massachusetts Institute of Technology Panel Chair
Alvarez, Gerardo Observer
Alvarez-Franyutti, Gerardo Mexichem Observer
Alyamani, Ahmed King Abdulaziz City for Science and Technology MI Representative
Amador-Bedolla, Carlos Universidad Nacional Autónoma de México Local Chair
Andrade, Arturo Universidad Nacional Autónoma de México Observer
Antes, Matt Energetics Staff
Arjona, Diego Observer
Aspuru-Guzik, Alán Harvard University and CIFAR Workshop Co-Chair
Aykol, Muratahan Toyota Research Institute Panel Chair
Barrueta, Nancy Universidad Nacional Autónoma de México Observer
Bárzana, Eduardo Universidad Nacional Autónoma de México Observer
Becker, Jill S. 02139 Inc Participant
Beltrán, Leonardo Secretaría de Energía Invited Speaker
Instituto Nacional de Electricidad y Energías
Beltrán Zarza, Norma Angélica Observer
Limpias
Bentley, Stephen Cardiff University Observer
Berlinguette, Curtis University of British Columbia and CIFAR Participant
Bernstein, Alan CIFAR Invited Speaker
Bio, Matthew Snapdragon Chemistry Participant
Burke, Marty University of Illinois at Urbana Champaign Invited Speaker
Calva, Nacx Energía Mx Observer
Camacho, Emilio California Energy Commission Observer
Carrillo, Arturo Sostener Observer
Instituto Nacional de Electricidad y Energías
Castañón, Tatiana Romero Observer
Limpias
Ceder, Gerbrand University of California, Berkeley Invited Speaker
Cernuschi, Federico Ricerca sul Sistema Energetico MI Representative
Chan, Emory Lawrence Berkeley National Laboratory Panel Chair
Clapper, Maureen U.S. Department of Energy MI Representative
Cogdell, Richard Glasgow University Participant
Contreras, César Observer
Contreras, Rafael Observer
Cruz, Rosalba Observer
de Pablo, Juan University of Chicago Invited Speaker
Des Rosiers, Frank Natural Resources Canada Invited Speaker
U.K. Department of Business, Energy and
Dodson, Jennie MI Representative
Industrial Strategy
Domenella, Sergio Universidad Nacional Autónoma de México Observer
U.K. Department of Business, Energy and
Durrant, Paul MI Representative
Industrial Strategy
Dwaraknath, Shyam Lawrence Berkeley National Laboratory Participant
95
E. Heinonen, Jarmo Tekes Observer
Early, Bryan California Energy Commission Observer
Fabrizio, Monica CNR ICMATE Participant
Instituto Nacional de Electricidad y Energías
Fernández Rodríguez, Daniel Observer
Limpias
Fischer, Peter Lawrence Berkeley National Laboratory Participant
Instituto Nacional de Electricidad y Energías
Flores, Jose Roberto Observer
Limpias
Flores-Leonar, Martha Universidad Nacional Autónoma de México Participant
Flores-Moreno, Roberto Universidad de Guadalajara Observer
Garcia, Karl Universidad Nacional Autónoma de México Observer
Gieb, Martin European Commission MI Representative
Golden, Ariel Student Energy Summit Observer
Gómez, Fernando World Economic Forum Participant
Instituto Nacional de Electricidad y Energías
González, Francisco Arturo López Observer
Limpias
González, Abelardo ITAER Observer
Grageda, Federico A. Centro de Desarrollo y Estrategia Empresarial S.C. Observer
Gregoire, John California Institute of Technology Panel Chair
Instituto Nacional de Electricidad y Energías
Guillén, Mario Román Díaz Observer
Limpias
Hasmath, Reza CIFAR Observer
Hattrick-Simpers, Jason National Institute of Standards and Technology Panel Chair
Hein, Jason University of British Columbia Panel Chair
Holst Jørgensen, Birte Technical University of Denmark MI Representative
Huang, David M. University of Adelaide Participant
Hutchison, Geoffrey University of Pittsburgh Panel Chair
Isayev, Olexandr University of North Carolina, Chapel Hill Participant
Jaffer, Shaffiq TOTAL SA Participant
Juaristi, Eusebio CINVESTAV Participant
Korea Advanced Institute of Science and
Jung, Yousung Panel Chair
Technology
Kanellis, Pamela CIFAR Staff
Kendall, Michaela Adelan Observer
Korea Institute of Energy Technology Evaluation
KIM, Bong Gyun Observer
and Planning
King, Alex Critical Materials Institute Participant
Kiviaho, Jari VTT Technical Research Centre of Finland Participant
Kozdras, Mark Natural Resources Canada MI Representative
Kreisbeck, Christoph Harvard University Participant
Laguna, Humberto Universidad Nacional Autónoma de México Participant
Centro de Innovación Aplicada en Tecnologías
Lambert, Juliette Participant
Competitivas
Lindh, Roland Uppsala University Participant
Løvvik, Ole Martin SINTEF Materials and Chemistry Participant
Magadán, Jose Manuel Martínez Instituto Mexicano del Petróleo Observer
96
Maldonado, José-Luis Centro de Investigaciones en Óptica Observer
Instituto Nacional de Electricidad y Energías
Malo Tamayo, José María Observer
Limpias
Maruyama, Benji U.S. Air Force Research Laboratory Participant
Molina, Mario Universidad Nacional Autónoma de México Invited Speaker
Montone, Amelia ENEA MI Representative
Montoya, Joseph Lawrence Berkeley National Laboratory Participant
Morales, José Luis Universidad Nacional Autónoma de México Observer
Mosby, Anna Energetics Staff
Noguez, Cecilia Universidad Nacional Autónoma de México Participant
Odom, Susan University of Kentucky Participant
Olmedo, Jose Observer
Ortiz, Carlos Secretaría de Energía Invited Speaker
Pai, Ranjith Krishna Indian Ministry of Science and Technology MI Representative
Palma, Julio Pennsylvania State University Participant
Peón Peralta, Jorge Universidad Nacional Autónoma de México Observer
Persson, Kristin University of California, Berkeley, LBNL Workshop Co-Chair
Instituto Nacional de Electricidad y Energías
Pijpers, Joep Participant
Limpias
Pulido, Natalia Tecnológico de Monterrey, Campus Monterrey Observer
Instituto Nacional de Electricidad y Energías
Rejón, Leonardo Observer
Limpias
Roberts, Jeff Lawrence Livermore National Laboratory Participant
Roch, Loïc Harvard University Participant
Rocha Rinza, Tomás Universidad Nacional Autónoma de México Observer
Instituto Nacional de Electricidad y Energías
Rodas Grapain, Arturo Observer
Limpias
Ross, Michael University of California, Berkeley Participant
Rotenberg, Eli Lawrence Berkeley National Laboratory Participant
Ruiz-Morales, Yosadara Instituto Mexicano del Petroleo Observer
Saikin, Semion Harvard University Participant
97
Saldaña Talvitie, Elli Universidad Nacional Autónoma de México Staff
Instituto Nacional de Electricidad y Energías
Salinas Bravo, Víctor Manuel Observer
Limpias
Sanchez, Rafael Observer
Sánchez, Mariano Universidad Nacional Autónoma de México Participant
Sargent, Edward University of Toronto and CIFAR Participant
Schrier, Joshua Haverford College Panel Chair
Schuster, Frédéric Commissariat à l’Énergie Atomique Observer
Segalman, Rachel University of California, Santa Barbara Panel Chair
Serrano, Siria Harvard University Staff
Sfeir, Matthew Brookhaven National Laboratory Participant
Simon, Horst Lawrence Berkeley National Laboratory Invited Speaker
Singh, Chandra Veer University of Toronto Observer
Solis-Ibarra, Diego Universidad Nacional Autónoma de México Panel Chair
Soto, Maritza Sostener Staff
Spadaccini, Christopher Lawrence Livermore National Laboratory Participant
Energy Materials Industrial Research Initiative/
Stassin, Fabrice Observer
Umicore
Subramaniam, Velumani CINVESTAV Participant
Tabor, Daniel Harvard University Participant
Tao, Andrea University of California, San Diego Participant
Instituto Nacional de Electricidad y Energías
Torres, Ramon Carlos Observer
Limpias
Tribukait-Vasconcelos, Hermann Mexico Energy R&D Innovation Funds Executive Chair
van Grondelle, Rienk Vrije Universiteit Amsterdam and CIFAR Participant
Vegge, Tejs Technical University of Denmark Participant
Villegas, Alejandro Sostener Staff
Warren, Jeff Simon Fraser University Participant
Wilhelm, Martin Geopolis Energy Partners Observer
Williams, Daryl Imperial College Participant
Zeron, Paulino Observer
98
Acronyms
2D Two-Dimensional
3D Three-Dimensional
AC Alternating Current
AFLOW Automatic Flow (database)
AFM Atomic Force Microscopy
Ag Silver
AI Artificial Intelligence
AI4M Artificial Intelligence for Materials
AiiDA Automated Interactive Infrastructure and Database for Computational Science
ANI Accurate NeurAl networK engINe for Molecular Energies (ANAKIN-ME or ANI)
API Application Programming Interface
ARES Autonomous Research System
ARPES Angle-Resolved Photoemission Spectroscopy
Ba Barium
BEIS Department for Business, Energy & Industrial Strategy (United Kingdom)
C Carbon
Ca Calcium
CCSD Cambridge Crystallographic Structural Database
Cd Cadmium
CEO Chief Executive Officer
CH4 Methane
CIFAR Canadian Institute for Advanced Research
CO2 Carbon Dioxide
COP21 21st Conference of the Parties
COP22 22nd Conference of the Parties
Cu Copper
DARPA U.S. Defense Advanced Research Projects Agency
DC Direct Current
DFT Density Functional Theory
DNA Deoxyribonucleic Acid
DOE U.S. Department of Energy
DSSC Dye-Sensitized Solar Cell
EMIRI Energy Materials Industrial Research Initiative
eV Electron-Volt(s)
FAT/ML Fairness, Accountability, and Transparency in Machine Learning
g Gram(s)
Ge Germanium
GHG Greenhouse Gas
GSP Genome Sequencing Program (NHGRI)
H Hydrogen
h Hour(s)
HERMAN High-throughput Experimentation Robot for the Multiplexed Automation of
Nanochemistry
high Tc High-Temperature (superconductivity)
HPLC High-Performance Liquid Chromatography
HTE High-Throughput Experimental
ICSD Inorganic Crystal Structure Database
iGEM The International Genetically Engineered Machine
InChI International Chemical Identifier
IP Intellectual Property
IR Infrared
IUPAC International Union of Pure and Applied Chemistry
JCAP Joint Center for Artificial Photosynthesis
99
kg Kilogram(s)
kJ Kilojoule(s)
kWh Kilowatt-Hour(s)
LBNL Lawrence Berkeley National Laboratory
LC Liquid Crystal
LED Light-Emitting Diode
LHASA Logic and Heuristics Applied to Synthetic Analysis (computer program)
Li Lithium
MAP Materials Acceleration Platform
MDCS Materials Data Curation System
MDF Materials Data Facility
MI Mission Innovation
mol Mole(s)
MP Materials Project
MRI Materials Resonance Imaging
MS-CG Multi-Scale Coarse-Grained
N Nitrogen
Natl. Lab National Laboratory
Nd Neodymium
NGO Non-Governmental Organization
NHGRI U.S. National Human Genome Research Institute
Ni Nickel
NIMS National Institute for Materials Science
NIST National Institute of Standards and Technology
nm Nanometer(s)
NMR Nuclear Magnetic Resonance
NOMAD Novel Materials Discovery
NSF National Science Foundation (United States)
O Oxygen
OQMD Open Quantum Materials Database
ORGAN Objective-Reinforced Generative Adversarial Networks
ORNL Oak Ridge National Laboratory
Pb Lead
PEEM Photoemission Electron Microscopy
PL Photoluminescence
PV Photovoltaic(s)
QD Quantum Dot
QM/MM Quantum Mechanics/Molecular Mechanics
R&D Research and Development
RIXS Resonant Inelastic X-ray Scattering
Se Selenium
SEM Scanning Electron Microscopy
SENER Mexican Ministry of Energy (Secretaría de Energía)
Si Silicon
SMILES Simplified Molecular-Input Line-Entry System
STM Scanning Tunnelling Microscopy
STXM Scanning Transmission X-ray Microscopy
Te Tellurium
TEM Transmission Electron Microscopy
Ti Titanium
UNAM Universidad Nacional Autónoma de México
W3C World Wide Web Consortium
WANDA Workstation for Automated Nanomaterials Discovery and Analysis
Wh Watt-Hour(s)
XAS X-ray Absorption Spectroscopy
XRD X-ray Diffraction
Zn Zinc
100