Virtual lab opens up machine learning possibilities for understanding a promising class of quantum materials

Thomas Edison famously tried hundreds of materials and failed thousands of times until he discovered that carbonized cotton threads burned long and brightly in an incandescent light bulb. Experiments are often time consuming (Edison’s team took his 14 months) and expensive (a successful combination cost about $850,000 in today’s dollars).

Developing quantum materials that revolutionize modern electronics and computing is costly and time consuming.

To enable the discovery of quantum matter, researchers look to detailed databases as virtual laboratories. A new database of understudied quantum materials created by researchers at the Pacific Northwest National Laboratory (PNNL) has discovered a new material that can power devices far more powerful than Edison’s light bulb. provide the means to

Beyond trial and error in the Edison era

“We wanted to understand a general class of materials that have the same crystal structure but have different properties depending on how they are combined and grown,” said materials scientist Tim Pope. said. This class of materials, known as transition metal dichalcogenides (TMDs), contains thousands of potential combinations, and it takes him a week to grow glitter-sized flakes of the material in each combination. I need a reaction.

Creating a material is just the first step in understanding what you can do with it. As PNNL computational scientist Mika Prange put it, each flake is “very small and very delicate” and its quantum features only appear when studied at very low temperatures. In essence, “the whole research program could go into each flake.”

Despite their difficulty in creating and measuring them, each combination has the potential to dramatically improve electronics, batteries, pollution remediation, and quantum computing devices.

]Prange said the flakes can be thought of as “fancy graphene with richer phenomenology and more practical potential.” Strong, light, and flexible, graphene is the material of the future, used in everything from aerospace to wearable electronics.

“The variety of properties across this class of materials means that as we get to know them better, we can pick one of the combinations for the desired properties and combine them precisely for the ideal application.” said Pope. “It can even be used for completely new applications.”

Future quantum material development

The construction of the database started with the PNNL Chemical Dynamics Initiative. This is an effort to use PNNL’s strengths in data science to fill knowledge gaps left by measurement challenges and experimental limitations.

These particular quantum materials are produced by combining 38 transition metals, such as tungsten and vanadium, with three elements from the sulfur family in varying proportions. It can also be grown in three different crystal structures. That means there can be thousands of combinations, all with different properties.

Using a form of modeling called density functional theory, the researchers calculated the properties of 672 unique structures with a total of 50,337 individual atomic arrangements. Prior to this study, fewer than 40 configurations had been studied and only a rudimentary understanding of their properties was available.

“The model can explain the quantum mechanics of how atoms are arranged,” Prange said. “From this, we can tell whether the material conducts electricity or is transparent, or how difficult it is to compress or bend.”

PNNL researchers used the database to reveal striking differences in electrical and magnetic behavior between different combinations. Importantly, as the researchers varied the transition metals, they also discovered other trends, including a new understanding of transition metal chemistry at the quantum level.

Quantum Combining for Machine Learning

“We overlaid the crystal structure with the database and it was a perfect match,” Pope said of the flakes grown at PNNL, where validation of modeling results has begun.

“The real idea was to develop a big data set of theoretical simulations so that we could understand these materials using data analytics,” Prange said. “The immediate value of this project is that we have done enough different cases to use machine learning efficiently.”

Open source datasets published at scientific dataprovides researchers with a powerful starting point to explore the relationships between initial structures and corresponding properties. You can use this information to select specific materials for study.

“This project is an example of how large computational datasets can be used to guide experimental research,” said Peter Suschko, CDI Principal Scientist. “Projects like this are important to the machine learning community. “This is exciting because it provides us with valuable data and has the potential to streamline materials development.” To be able to synthesize these materials with atomic precision, we need to consider what we need to understand next. ”