National laboratory researchers improve chip design processes with artificial intelligence

0


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

They present several AI-based approaches to optimize atomic layer deposition, or ALD, processes in a recently published paper. study. The process produces films of ultra-thin material, one atom thick.

It also partly underpins the manufacturing of computer chips, which are now at the center of a global supply chain shortage that has driven up the prices of all kinds of electronics.

“The effort predates current chip shortage issues, but we have long looked at semiconductor processing and its manufacturing challenges,” said Angel Yanguas-Gil, senior materials scientist at ARNL. Nextgov Thursday.

Yanguas-Gil has a long history of pursuing AI-aligned semiconductor innovation, including helping to develop an advanced neuromorphic computer chip modeled from insect brains. He explained that researching fundamental breakthroughs that could have a profound impact on advanced manufacturing is a priority for Argonne. Yanguas-Gil’s group received funding from the Department of Energy’s residency technologist program, among others, which, he said, “encouraged [them] to look at the industry as a whole. The lab has a strong ALD program, he noted, and their established links with the private sector have helped initiates fully grasp the key challenges that exist.

“This led to an internally funded research program specifically focused on the application of AI to manufacturing that supported current research,” Yanguas-Gil said.

A Press release recently published by the lab details the complexity of the new experiments.

ALD can be used to grow thin films for various applications. It occurs inside a chemical reactor, where “precursors”, or two different chemical vapors, cling to a surface and, over time, form a thin film. The technique “excels in growing precise nanoscale films on complex 3D surfaces,” according to the lab’s statement, motivating scientists to explore making new ALD materials for next generations of devices.

Training and optimizing nascent ALD processes is incredibly time consuming. Yanguas-Gil and the laboratory team therefore traveled to explore three optimization strategies: random optimization, the expert system and Bayesian optimization.

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

Bayesian optimization is an algorithm that, as it tries various conditions and receives feedback, learns an internal model that helps it understand precisely which conditions are most promising to try next, has t he noted. In terms of cooking, this algorithm has absolutely no prior information on what it means to “cook” – it just starts with a collection of ingredients and the oven setting.

“It’s really good to quickly determine the proportions of ingredients that lead to a good recipe based on the feedback it receives,” the scientist explained.

In contrast, the expert system relies on earlier types of AI that attempt to codify certain expert information about what trends to 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. Considering the cooking analogy, “it’s like telling the algorithm that it should focus on the proportion of the main ingredients first, because you can always rectify the spices after the dish is cooked”, Yanguas-Gil said.

And the random system is the benchmark case, he added. Because of this, they “just pick up conditions at random, hoping that one of them is the right one.” Eventually, the system will land on a point that may be quite close to the target – “but you have to be prepared to taste some really disgusting food in the process,” according to Yanguas-Gil.

“The random system helps us understand how difficult – and crude – it is to come up with a really good recipe,” he noted, “but it’s not a method that is really meant for the job. ‘user.

Researchers at the lab evaluated the three optimization strategies, aiming to find the conditions that led to “high and stable film growth in the shortest possible time”, among other goals. The work involved optimization algorithms, a simulated system and more. The researchers also set up a closed-loop system, through which a simulation completes an experiment, feeds the results to an AI tool. Then the AI ​​tool interprets them and recommends the next experiment, all without any human input.

AI-based approaches determined optimal timing elements for different simulated ALD processes. The study is “one of the first to show that real-time thin-film optimization is possible using AI,” the lab’s statement confirmed.

For Yanguas-Gil, coupling in situ techniques – or examining events where they occur – with machine learning algorithms to stimulate optimization “seems obvious”. But he and his team could not identify any significant examples applied to ALD in the scientific literature prior to this work.

“I think part of the reason is that you need a team with a wide range of expertise working together: experts in the technique, both in terms of fundamentals, in-situ characterization and manufacturing, machine learning, modeling and simulation experts, instrumentation experts, ”Yanguas-Gil said. “We had this at Argonne.

Research provides a way to accelerate the integration of new processes into manufacturing. It also introduces the possibility that such approaches could help US manufacturers save time and money in chip development.

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

Going forward, he and his team have several concepts on how they can “take this research further”. And they also think of technology transfer. With this in mind, Yanguas-Gil added that an advantage of the new approach is that it works with standard characterization tools.

“We are having conversations with industry – companies interested in understanding how this research can help them optimize their processes,” he said.


Share.

About Author

Leave A Reply