Screenshot of one of the Materials Project’s new tools based on a growing and dynamic database, the Lithium Battery Explorer. Credit: Materials Project.

Robert F. Service, one of Science magazine’s best writers, has finally written a lengthy piece (subscription req’d) on the Materials Genome Initiative, which contains a nice interview with MIT’s Gerbrand Ceder — who rightfully deserves a lot of credit for pushing this initiative and demonstrating how it could pay off in the form of rapid identification of new and advanced materials — plus interviews with other research groups about how they are using data and software techniques to narrow in on promising materials.

Everyone seems to agree that the tipping point came when computing power caught up with existing software and algorithms. Faster computing, in turn, opened up even more software and analytical approaches.

Service captures much of the excitement that Ceder has for MGI-type efforts. For example, Service writes this about Ceder’s own ideas and captures how advanced computing techniques right now can provide extreme acceleration to the research process;

[Faster computing speeds] opens the door to computing the properties of a wide range of materials that once seemed unapproachably complex, Ceder says. Among the more tractable problems should be advances in catalysts, battery materials, and thermoelectrics, which convert heat to electricity. And it should be relatively straightforward to make a big impact on materials research quickly. There are between 50,000 and 100,000 known inorganic compounds, depending on whose figures you believe, Ceder notes. Crunching the numbers for all the computable properties of a single known material—including crystal structure, stability, and ionic mobility—takes the equivalent of 1 day for a standard computer chip, known as a CPU. To make that calculation for all known inorganic compounds would take between 2 million and 3 million CPU hours, a job one of the most advanced supercomputers could carry out in just a day and a half. Examining a good swath of the possible unknown materials out there would still take only half a billion CPU hours, Ceder predicts. “That’s just a drop in the bucket” of the computing power available, Ceder says. “We don’t know most things about most materials,” he says. “Materials scientists are hungry for this data.” [emphasis added]

Ceder and the rest of the advocates for MGI-type work quickly acknowledge that hands-on lab work and actual materials preparation and testing will still be needed, but the whole idea is to whittle down the work to focus on what the calculations suggest will be the most promising materials.

But, Ceder warns that computational approaches work best when first-principles knowledge already exists. Using the example of the challenge of developing construction materials made of new lightweight, high-strength metal alloys, Service quotes Ceder as saying, “We don’t even know the underlying science of how these materials work. …So you’ve got to pick the right problems,”

Interestingly, Service reports that some researchers take a different slant and are pursuing computational methods that don’t depend on first principles knowledge. I don’t know if I can adequately provide a brief description here (Service writes about it at length), but, as I understand it, a researcher team led by Krishna Rajan, a computational materials scientist at Iowa State University, Ames, use a “machine-learning” system to narrow in on a promising set of materials. In particular, while working on developing improved piezoeletric materials, they built and leveraged a database of, according to Service “30 observations on different materials variables, such as bond distances between pairs of atoms in a would-be crystal, and the affinity of different elements for electrons.”

Rajan’s group simultaneously worked to identify and quantify how these different variables related to each other. They then use a special algorithm to match the relationships with the variables in the observations database. Eventually, they refined the set of observations and relationships and ended up with a set of design rules to find optimal piezo materials.

So, it’s nice to see some examples emerging of how advanced computational methods can accelerate materials research.

But, I think there is also a growing feeling (somewhat echoed in Ceder’s comments, above) that there may be an advantage to focusing MGI work, and that is where, in my opinion, one cannot overemphasize the importance of the work, begun this week, to identify the “Grand Challenges for Ceramics Materials Research.”

The MGI and the Grand Challenges initiatives start from different places. The former from the micro world of properties and data and the later from the macro view of societal and industrial priorities. However, both represent efforts to strategically prioritize and narrow in on the most promising work. The Grand Challenges aspect is necessary, ultimately, to get buy-in from funding agencies and industry. Industry — both manufacturers and users — needs to know what some of the deliverables are going to be, how much of the MGI work is going to involve pre-competitive research, where IP lines are going to be drawn, etc. Funding agencies need to know what makes the science new, uncharted territory, challenging.

I suspect, however, that both the MGI work and the Grand Challenges will go a bit slower than advocates hope. Based on my studies of similar efforts in the glass science and industry community aimed at developing a fundamental and universally available understanding of intrinsic glass strength and flaw development properties, joint focused work gets bogged down in frustrating, but important, details, such as membership roles, IP policies, funding mechanisms, roadmaps, etc. But, even with this group, progress is being made, and it is hoped, will be a template on how to focus on other breakthroughs. It also may mean that coalition-building skills may be as important (and, initially, maybe more important) as computational skills.

In the meantime, everyone needs to be aware of another potential roadblock to MGI-type initiatives. Who is capable of doing the work? For example, are science and engineering students being adequately trained to work in situations where advanced computing techniques are going to be the norm? Along these lines, several professional societies, including The Minerals, Metals & Materials Society, The American Ceramic Society and ASM International, along with the University Materials Council, co-sponsored a meeting a few weeks ago on the topic, “Equipping the Next Generation Workforce for Materials Innovation.”

As part of that meeting, a panel of industry representatives said changes needed to be made. According to a report from the meeting, “The industrial panelists thought that workforce development was critical to the success of MGI or ICME in the U.S. Their experiences suggested that it was hard to find U.S. graduates and even harder to find U.S. graduates in computational materials science. The education of such students requires governmental support of academia and partnering of academia with industry through internships, design projects etc. The problem is compounded by the fact that foreign nationals that come to the U.S. are often pushed out by immigration policies.”

So, the reality is that the MGI isn’t just one thing, but an effort with a lot of moving parts and gears that must mesh together. Good structural ideas have been sketched out, but now we have to figure out how to put it all together.

CTT Categories

  • Modeling & Simulation