Officials see an opportunity to invest more in AI research and development to improve semiconductor manufacturing.
The White House sees promising potential in artificial intelligence to significantly impact semiconductor manufacturing technology, which is becoming increasingly complex and resource-intensive.
“We will see semiconductor materials that are co-optimized not only for performance, but that can now address important issues like supply chain resilience and critical sustainability, both for existing, known materials, but potentially for new materials that we haven’t even thought of before,” said Arati Prabhakar, director of the White House Office of Science and Technology Policy, at a June 13 White House event.
President Biden’s CHIPS and Science Act Prioritizes Semiconductor Technology Financingwhich makes up essential components of things like cars, household appliances and also defense weapons systems. The plan aims to increase U.S. leadership in Chip technology production and reduce our dependence on Asia for this.
Federal leaders gathered at the White House to discuss the potential of AI in semiconductor manufacturing, part of a broader plan called AI Aspirations to explore how AI can be used in research and development in other sectors including healthcare, education, weather and more.
In semiconductors, time is of the essence. It takes nearly a decade to design the new materials needed to make new chips. That’s where artificial intelligence could make a difference, executives say.
“We’re never going to get there the way we’re used to doing it, the way we’ve been taught. We need something like artificial intelligence. But even that’s not enough,” said Benji Maruyama, a senior materials research engineer in the Air Force Research Laboratory’s Materials and Manufacturing Directorate. “We need to invest more in experiments, but also in better understanding, more simulation, so we can have better insights.”
Integrating AI into the manufacturing process is quite a technical endeavor, and manufacturers would need huge amounts of data to feed and train AI models. Indeed, critical dimension scanning electron microscopes (CD-SEMs), a system dedicated to measuring dimensions on a semiconductor wafer, are becoming more complex and require larger amounts of data to map dimensions.
“The CD-SEMs need to be scanned and enough of the correct dimensions of the CD-SEM fed back into the AI model to achieve accuracy,” said Melissa Grupen-Shemansky, CTO and vice president of technology communities at industry association SEMI. “It takes an order of magnitude or two more than that for the actual data points needed.”
Francesca Tavazza, NIST’s group leader in data-driven materials science and AI, said materials and AI scientists need to think about the right features or ingredients when creating models. These models aren’t a black box where you add more data to try to create a better model, but rather follow scientific laws throughout the process, she said.
“We need to understand the source of the key ingredients that make physical phenomena happen, and that’s what we put as input into the model,” Tavazza said.
Dana Weinstein, OSTP’s Senior Deputy Director for Industrial Innovation and Special Advisor for Chip Research and Development set out a vision for a materials accelerator where AI could help reduce the time to deploy advanced materials from 20 years to 1-3 years. This will depend in part on strong partnerships.
Incentives for research labs, industry partners, and startups to join the conversation could help eliminate bias and other challenges from the start. Tom Kenny, a Stanford University engineering professor, noted that during his time at DARPA, he learned the importance of incentives to foster partnerships between government, industry, and academia.
“Often you put together a lot of government money and you do some matching, and you put that money in the middle between academia and industry,” Kenny said. “At DARPA, we’ve started programs primarily by identifying a gap between communities that we were eager to try to bring together. The CHIPS Act is a great opportunity to build teams out of a willingness to work together around substantial resources.”
