Accelerate ALD with AI
The US Department of Energy’s Argonne National Laboratory (DOE) has developed various ways of making atomic layer deposition (ALD) more efficient through the use of artificial intelligence (AI).
ALD is a deposition technique in which materials are deposited in layers on chips. ALD has been used in the manufacture of DRAMs, logic components and other products for years.
In operation, wafers are inserted into a chamber within an ALD system. A chemical or a precursor is pumped into the chamber. The wafers are processed. Then the chemicals are flushed out of the system. The process is repeated, sometimes with a different chemistry.
However, ALD is a slow process. In addition, each process requires different materials and foils. And adapting the process for each new material can take some time, according to researchers at Argonne National Laboratory.
Other factors also play a role. Every chemistry is complex with different variables. There are several providers of ALD systems on the market. Each can have a different reactor design with different settings and conditions. In some cases, vendors need to use a time-consuming trial and error process to determine the optimal conditions.
In response, Argonne hopes to make ALD more efficient. The researchers evaluated three “optimization strategies” for ALD random, expert system and Bayesian optimization. The latter two use different machine learning approaches. Machine learning, a form of AI, is a neural network that processes data and identifies patterns. It then matches certain patterns and learns which of these attributes are important.
The researchers evaluated the three strategies by comparing the dosing and flushing times of the two precursors used in ALD. “Dosing time refers to the amount of time a precursor is added to the reactor, while purge time refers to the time it takes to remove excess precursor and gaseous chemical products,” said Argonne researchers.
All three “optimization approaches” accelerated the ALD process. But the two AI approaches effectively determined the optimal dose and flush times for various simulated ALD processes using a closed-loop control system.
“All of these algorithms allow you to converge to optimal combinations much faster because you don’t spend time putting a sample in the reactor, taking it out, taking measurements, and so on, as you normally would. Instead, you have a real-time loop connected to the reactor, ”said Angel Yanguas-Gil, a senior materials scientist at Argonne.
“In a closed-loop system, the simulation carries out an experiment, receives the results and feeds them into the AI tool. The AI tool then learns from it or interprets it somehow and then suggests the next experiment. And all of this happens without human intervention, ”said Noah Paulson, a computer scientist at the Argonne.
“This is exciting because it opens up the possibility of using such approaches to quickly optimize real ALD processes, a step that could potentially save manufacturers valuable time and money in the development of new applications in the future,” said Jeff Elam, chief chemist at Argonne.
Researchers at Carnegie Mellon University and St. Petersburg State University developed an algorithm called MolDiscovery that would help scientists categorize unknown molecules using mass spectrometry data.
MolDiscovery would help save time and money as scientists would not spend resources studying molecules that have already been identified.
Mass spectrometry is a technique that measures the mass-to-charge ratio of a molecule and provides information about the atomic mass of the molecule. The same element can have different atomic masses due to different numbers of neutrons in the nucleus, so the atomic mass of the molecule is shown in a spectrum.
However, there can be hundreds of thousands of molecules in an environment, and identifying individual, unknown molecules is a challenge.
MolDiscovery is an algorithm that searches millions of molecular data using a mass spectral database search method. Using the mass spectrum data of a molecule, MolDiscovery uses a probabilistic model to predict what the molecule is up to. By looking at the molecular structure, MolDiscovery creates metabolite diagrams and breaks down a molecule into fragmentation diagrams. Because the algorithm knows the mass spectra and graphs of the molecule, it can predict the molecule to see if it was previously discovered.
MolDiscovery is more efficient and accurate compared to previous methods of molecule identification. “The existing approaches are based on knowledge in the field of chemistry and cannot explain many of the peaks in mass spectra of small molecules. A search of over 8 million spectra of Global Natural Product Social’s molecular network infrastructure shows that MolDiscovery correctly identifies six times more unique small molecules than previous methods, ”writes Hosein Mohimani, assistant professor and researcher at Carnegie Mellon University, in an article published in Nature Communications . Others contributed to the paper.
MolDiscovery can help scientists and researchers in the medical and pharmaceutical industries, as well as in the discovery of other novel natural substances. By identifying new molecules early on, scientists can save money on drug development, microbial characterization, and disease diagnosis.
Using machine learning technology, Carnegie Mellon found a way to accelerate the development of natural drugs to treat cancer, viral infections, and other diseases.
Others contributed to the work, including the University of California at San Diego, the University of Saint Petersburg, the Max Planck Institute, Goethe University, the University of Wisconsin at Madison, and the Jackson Laboratory.
Many antibiotics, antifungal, and anti-tumor drugs come from natural products. They are considered safe, so it is imperative to expedite the development of these natural drugs.
In response, researchers developed a machine learning algorithm called NRPminer, a platform that helps scientists isolate natural substances. More specifically, NRPminer accelerates the discovery of non-ribosomal peptides (NRPs), an important type of natural product used in the manufacture of many antibiotics, cancer drugs, and other drugs.
NRPs are difficult to spot and identify. “Natural products are still one of the most successful avenues for drug discovery,” says Bahar Behsaz, project scientist at Carnegie Mellon. “And we believe we can go further with an algorithm like ours. Our calculation model is orders of magnitude faster and more sensitive. “
“The unique thing about our approach is that our technology is very sensitive. It can recognize molecules with an abundance of nanograms, ”said Mohimani of Carnegie Mellon. “Our hope is that we can drive this forward and discover other natural drug candidates and then develop them into a phase that would be attractive to pharmaceutical companies.”