AI Used To Predict Synthesis of Complex Novel Materials – “Materials No Chemist Could Predict”

Date:

Machine learning makes it possible to discover materials.Credits: Northwestern University

AI machine learning provides a roadmap for defining new materials for all needs that impact green energy and waste reduction.

Scientists and institutions spend more resources each year discovering new materials to invigorate the world. As natural resources diminish and the demand for higher-value, higher-performance products increases, researchers are increasingly turning to nanomaterials.

Nanoparticles from energy storage and conversion Quantum computing And treatment. However, given the vast composition and structural tunability that nanochemistry allows, a set of experimental approaches to identify new materials imposes insurmountable limitations on discovery.

Now researchers Northwestern University Toyota Laboratories (TRI) has successfully applied machine learning to guide the synthesis of new nanomaterials and remove barriers associated with material discovery. Highly trained algorithms combine defined datasets to accurately predict new structures that may fuel processes in the clean energy, chemical, and automotive industries.

Chad Mirkin, a nanotechnology expert at Northwestern University and the corresponding author of the paper, said: “The machine predicted 19 possibilities, and after experimentally testing each one, we found that 18 predictions were correct.”

The study “Machine Learning-Accelerating the Design and Synthesis of Multi-Elemental Heterostructures” will be published in the journal on 22 December. Science Advances..

Markin is Professor George B. Rasman of Chemistry at Weinberg University of Arts and Sciences. Professor of Chemistry and Biotechnology, Biomedical Engineering, and Materials Science and Engineering at McCormick Institute of Technology. Professor of Medicine at Feinberg School of Medicine. He is also the founding director of the International Institute for Nanotechnology.

Material genome mapping

What makes this so important, according to Mirkin, is that machine learning models and AI algorithms are as good as the data used to train them, so they’re bigger than ever. Access to large, high quality datasets.

A data generation tool called the “Mega Library” was invented by Markin and dramatically expands the horizons of researchers. Each megalibrary contains millions or billions of nanostructures, each with slightly different shapes, structures, and compositions, all positioned and encoded on a 2 x 2 square centimeter chip. increase. To date, each chip contains more new inorganic materials than those previously collected and classified by scientists.

Mirkin’s team developed a megalibrary using a technique called polymer pen lithography (also invented by Mirkin). It is a massively parallel nanolithography tool that enables site-specific deposition of hundreds of thousands of features per second.

When mapping the human genome, scientists were tasked with identifying the combination of four bases. However, the broadly synonymous “material genome” includes the combination of nanoparticles of any of the 118 elements available in the periodic table and parameters such as shape, size, phase morphology, and crystal structure. By constructing smaller subsets of nanoparticles in the form of mega-libraries, researchers will approach completing a complete map of the material genome.

Markin said a variety of tools are needed to identify how to use and label materials, even if they are similar to the “genome” of the material.

“Even if we could make the material faster than anyone else on the planet, it’s a potential ocean drop,” Markin said. “We want to define and mine the material genome. The way we do that is through artificial intelligence.”

Machine learning applications are ideal for addressing the definition of material genomes and the complexity of mining, but are gated by the ability to create datasets for training algorithms in space. Markin said the combination of megalibraries and machine learning could ultimately eradicate the problem and lead to an understanding of which parameters drive specific material properties.

“Materials that chemists could not predict”

If Megalibraries provide maps, machine learning provides legends.

According to Mirkin, by using a megalibrary as a source of high-quality, large-scale material data for training artificial intelligence (AI) algorithms, researchers typically have a “sharp chemical intuition” that accompanies the material discovery process. And can be separated from continuous experiments.

“Northwestern had synthetic and state-of-the-art characterization capabilities to determine the structure of the materials we produce,” says Mirkin. “We worked with TRI’s AI team to create data inputs for AI algorithms that ultimately make these predictions for materials that chemists can’t predict.”

In this study, the team edited previously generated megalibrary structural data composed of nanoparticles with complex compositions, structures, sizes, and morphologies. They were asked to use this data to train the model and predict the composition of the four, five, and six elements that would result in specific structural features. With 19 predictions, the machine learning model correctly predicted the new material 18 times. This is about 95%. Accuracy ratio.

With little knowledge of chemistry and physics, and using only training data, the model was able to accurately predict complex structures that never existed on Earth.

“As these data suggest, the application of machine learning in combination with Megalibrary technology may ultimately be the way to define the material genome,” said Joseph Montoya, senior research scientist at TRI. increase.

Metal nanoparticles show the potential to catalyze industrially important reactions such as hydrogen generation and carbon dioxide (CO)2) Reduction and reduction and evolution of oxygen. The model is a large dataset built in the northwest to look for multimetal nanoparticles with parameters related to phases, sizes, dimensions, and other structural features that change the properties and functions of the nanoparticles. Was trained in.

Megalibrary technology also has the potential to facilitate discoveries in many areas of future importance, such as plastic upcycling, solar cells, superconductors, and qubits.

A tool that works over time

Prior to the advent of mega-libraries, machine learning tools were trained on incomplete datasets collected by different people at different times, limiting their predictability and generalizability. Mega libraries allow machine learning tools to do what they are best at. In other words, it takes time to learn and become smarter. Markin said the model is only good at predicting the correct material because it provides higher quality data collected under controlled conditions.

“Creating this AI feature is to be able to predict the materials needed for any application,” says Montoya. “The more data we have, the more predictive we are. When we start training AI, we start by localizing it to one dataset, and as AI learns, we add more and more data. It’s like getting a doctorate from kindergarten with your child. Combining experience and knowledge will ultimately determine how far you can go. “

The team is currently using this approach to find catalysts that are essential to the fuel supply process of the clean energy, automotive and chemical industries. By identifying new green catalysts, waste and abundant raw materials can be converted into useful substances, hydrogen is produced, carbon dioxide is used, and fuel cells can be developed. Production catalysts can also be used in place of expensive and rare materials such as iridium, the metal used to produce green hydrogen and CO.2 Reduction product.

Reference: “Machine Learning-Accelerating Design and Synthesis of Multi-Elemental Heterostructures” December 22, 2021 Science Advances..
DOI: 10.1126 / sciadv.abj5505

The study was supported by TRI. Additional support is from the Sherman Fairchild Foundation, Inc. , And from the Air Force Scientific Research Bureau (Prize Nos. FA9550-16-1-0150 and FA9550-18-1-0493). The northwestern co-authors are Carolina B. Wahl, a PhD student in materials science and engineering, and Jordan H, a PhD student in chemistry. Swisher, both members of the Mirkin Lab. Authors of TRI include Murata Hanei Col and Montoya.

This work used NU’s EPIC facility at Northwestern UniversityANCE Center supported by Soft and Hybrid Nanotechnology Experiment (SHyNE) Resources (NSF ECCS-1542205). MRSEC Program of National Institute for Materials Science (NSF DMR-1720139). International Institute for Nanotechnology (IIN); Keck Foundation; Illinois through IIN.

(function(d, s, id){
var js, fjs = d.getElementsByTagName(s)[0];
if (d.getElementById(id)) return;
js = d.createElement(s); js.id = id;
js.src = “//connect.facebook.net/en_US/sdk.js#xfbml=1&version=v2.6”;
fjs.parentNode.insertBefore(js, fjs);
}(document, ‘script’, ‘facebook-jssdk’));

AI Used To Predict Synthesis of Complex Novel Materials – “Materials No Chemist Could Predict” Source link AI Used To Predict Synthesis of Complex Novel Materials – “Materials No Chemist Could Predict”

The post AI Used To Predict Synthesis of Complex Novel Materials – “Materials No Chemist Could Predict” appeared first on California News Times.

Source link

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Share post:

Popular

More like this
Related

The real reasons behind Polkadot’s [DOT] 11% week are…

Polkadot has been on the up this week on...

Palestinian gunmen open fire on bus in Jerusalem, eight injured

Palestinian gunmen opened fire on a bus near Jerusalem’s...