National laboratory researchers accelerate chip design processes with artificial intelligence

Researchers at Argonne National Laboratory have uncovered new avenues and continue to explore new ways to advance a semiconductor chip design technique using artificial intelligence.

In a recently published study, they present several AI-based approaches to optimize atomic layer deposition or ALD processes. The process produces superfine films of material that are around an atom thick.

It also partially supports the manufacture of computer chips, which are now at the center of a global supply chain shortage that is driving up prices for all types of electronics.

“The effort precedes the current issues of chip scarcity, but we have long been concerned with semiconductor processing and its manufacturing challenges,” said Angel Yanguas-Gil, principal materials scientist at ARNL, to Nextgov on Thursday.

Yanguas-Gil has been pursuing AI-oriented semiconductor innovations for a long time and is helping, among other things, with the development of a state-of-the-art neuromorphic computer chip that is modeled on insect brains. He stated that the hunt for fundamental breakthroughs that could profoundly impact advanced manufacturing is a priority for Argonne. Yanguas-Gil’s group received funds from the Department of Energy’s Technologist in Residence program, among other funds, which he said “encourages.” [them] the industry as a whole. ”The lab has a strong ALD program, he noted, and its established connections with the private sector helped insiders fully understand the key challenges that existed.

“This resulted in an internally funded research program specifically focused on the application of AI in manufacturing and supporting current research,” said Yanguas-Gil.

A press release recently published by the laboratory describes the complexity of the new experiments.

ALD can be used to grow thin films for various applications. It occurs in a chemical reactor where “precursors” or two different chemical vapors adhere to a surface and form a fine film over time. According to the laboratory’s press release, “the technology is superior to the growth of precise nanoscale films on complex 3D surfaces” and motivates scientists to research the manufacture of new ALD materials for the next generation of devices.

Developing and optimizing emerging ALD processes is incredibly tedious. So Yanguas-Gil and the lab team examined three optimization strategies: random, expert system and Bayesian optimization.

“We can use a cooking metaphor to explain the three models,” said Yanguas-Gil.

Bayesian optimization is an algorithm that – while trying different conditions and receiving feedback – learns an internal model that helps it understand exactly which conditions are most promising to try next, he noted. When it comes to cooking, this algorithm has absolutely no prior information about what “cooking” means – it just starts with a collection of ingredients and an oven setting.

“He’s really good at quickly finding out the proportions of ingredients that make up a good recipe based on the feedback he receives,” explains the scientist.

In contrast, the expert system builds on previous types of AI that attempt to codify some expert information about trends that it should expect during the optimization process. To find the optimal condition, the system uses these expectations or rules along with the feedback it receives after trying a new condition. Given the cooking analogy, “It’s like we’re telling the algorithm to focus on the main ingredient proportions first because you can always correct the seasonings once the dish is cooked,” said Yanguas-Gil.

And the random system is the basic case, he added. By doing so, they essentially “just randomly pick up conditions in the hope that one of them will be a good one”. At some point the system will end up at a point that may be close enough to the target – “but you have to be willing to try really gross dishes while doing it,” says Yanguas-Gil.

“The random system helps us get a feel for how hard – and how gross – it is to come up with a really good recipe,” he noted, “but it’s not a method that is really intended for the user.”

Laboratory researchers evaluated the three optimization strategies in order to find, among other things, conditions that lead to “high and stable film growth in the shortest possible time”. The work included optimization algorithms, a simulated system and more. The researchers have also established a closed-loop system through which a simulation concludes an experiment and feeds the results into an AI tool. The AI ​​tool then interprets them and recommends the next experiment – all without human input.

The AI-based approaches determined optimal timing elements for various simulated ALD processes. The study is “one of the first to show that thin-film optimization is possible in real time with AI,” confirmed the laboratory’s press release.

For Yanguas-Gil, coupling in-situ techniques – or investigating events where they take place – with machine learning algorithms to drive optimization seems “like child’s play”. But he and his team were unable to identify meaningful examples of ALD in the scientific literature prior to this work.

“I think that’s partly because you need a team with a wide range of specialist knowledge that works together: experts in technology, both in terms of fundamentals, in-situ characterization and manufacturing, experts in machine learning, modeling and simulation “Instrumentation experts,” said Yanguas-Gil. “We happened to have that in Argonne.”

Research provides a means to accelerate the integration of new processes into manufacturing. It also opens up the chance that such approaches could help American manufacturers save time and money in chip development.

“The actual impact will depend on whether tool makers or specialty factories take these ideas and apply them to their own problem,” noted the researcher. “Our mission is to bring the technology to market and to help the industry in every possible way to bring new capacity online as quickly as possible.”

For the future, he and his team have several concepts for how they can “advance this research”. And they are also thinking about technology transfer. With this in mind, Yanguas-Gil added that one advantage of the new approach is that it works with standard characterization tools.

“We’re talking to industry – companies interested in understanding how this research can help them optimize their processes,” he said.