MIT Researchers Unveil AI Model to Detect Atomic Defects in Materials
In the realm of biology, defects are usually seen as unfavorable. However, in the field of materials science, defects can sometimes be beneficial, serving as a way to enhance the properties of materials. This is a common practice in the production of steel, semiconductors, and solar cells, where atomic-scale defects are deliberately introduced to optimize their performance, improve strength, and manage electrical conductivity.
Nevertheless, accurately pinpointing different types of defects and their concentrations in finished products presents a significant challenge, particularly when trying to avoid damaging the material. Engineers run the risk of creating products with subpar performance or unexpected properties if they cannot correctly identify the defects in their materials.
Now, a team of researchers from the Massachusetts Institute of Technology (MIT) has developed an artificial intelligence (AI) model that can categorize and quantify specific defects using data from a nondestructive neutron scattering technique. The model, trained on data from 2,000 different types of semiconductors, can simultaneously identify up to six types of point defects in a material, something that would be impossible with traditional techniques alone. This breakthrough could lead to more precise applications of defects in products such as semiconductors, solar cells, microelectronics, and battery materials.
How the AI Model Works
The researchers built a computer database with 2,000 semiconductor materials for their experiment. They made pairs of samples of each material, doping one for defects and leaving the other without defects, and then used a neutron scattering technique that measures the different vibrational frequencies of atoms in solid materials. They trained a machine learning model based on the results.
“This created a basic model that covers 56 elements in the periodic table,” says lead author Mouyang Cheng, a doctoral student in the Department of Materials Science and Engineering. “The model uses the multihead attention mechanism, just like what ChatGPT uses. It similarly extracts the difference in data between materials with and without defects and gives a prediction about which dopants were used and at what concentrations.”
Advantages and Future Applications
The new AI model can be a game-changer for manufacturers. The current process involves invasive testing on a small percentage of products as they come off the assembly line, a slow procedure that limits their ability to detect any defect. With this AI model, manufacturers could potentially detect up to six defects simultaneously, with defect concentrations as low as 0.2 percent.
Despite this promising advancement, the researchers caution that it would not be easy for companies to quickly implement their technique for measuring vibration frequencies with neutrons in their quality control processes.
“This method is very powerful, but its availability is limited,” says Eunbi Rha, a master’s student. “Vibrational spectra is a simple idea, but in certain setups it is very complicated. There are some simpler experimental setups based on other approaches such as Raman spectroscopy that could be adopted more quickly.”
Looking forward, the researchers plan to train a similar model based on Raman spectroscopy data and expand their approach to detect features larger than point defects, such as grains and dislocations.
The Power of AI in Defect Detection
What makes this AI model particularly impressive is its ability to decipher mixed signals from different types of defects. To the human eye, these defect signals would look practically identical, but the AI’s pattern recognition capabilities can detect subtle differences and draw accurate conclusions. This opens up a new paradigm in defect science, marking a significant milestone in the field of materials science.
For more details on this study, you can find the article Here.
The research was supported in part by the Department of Energy and the National Science Foundation.
