This article is part of our exclusive IEEE Journal Watch series in partnership with IEEE Xplore.
Many tech industry sectors are rapidly adopting AI and virtualization tools, even though there are some concerns about the amount of energy required to run those tools. But is it possible that in some situations, the environmental impacts of computer simulations are less than those from real-world experiments?
One of the largest semiconductor manufacturing companies in the world, Lam Research, did an internal deep dive analysis of their R&D operations to answer this question. The results show that despite the fact that AI modeling and virtualization requires significant energy, they can reduce carbon emissions by 20 to 80 percent compared to real-world experiments, depending on what aspect of semiconductor research is being done.
In 2021, Lam Research committed to goals of operating on 100 percent renewable electricity by 2030 and achieving net zero emissions by 2050.
David Fried, the corporate vice president of Semiverse Solutions at Lam Research, says for his team to meet this net-zero goal, “It was important that we tested a variety of our processes and measure the actual carbon footprint of our R&D activities to obtain the information needed to make the most meaningful changes in our business.”
In particular, the company examined the carbon emissions associated with research and development of the equipment and processes needed for nanoscale fabrication of semiconductors such as integrated circuits.
“Fabricating these minute devices is an extremely complex manufacturing process spanning hundreds of specialized steps that are extensively refined through the R&D process, nearly half of which involve intricate chemical plasma processes such as etching and deposition with nanoscale precision,” says Fried. “This requires significant resources in the form of materials, chemicals, gases, and high energy consumption.”
Replacing Lab Tests With Virtual Tests
Fried’s team sought to compare the carbon emissions associated with traditional research and development done in a physical lab to the same research done with AI and virtualization being used as much as possible in lieu of lab experiments. They assessed different areas of hardware prototyping, process optimization, and recipe development for wafers, the round thin discs used for fabricating semiconductors.
In each case, they compared the carbon footprint of computational modeling to the carbon footprint of physical experimentation. The results are described in a study published in the November issue of IEEE Transactions on Semiconductor Manufacturing.
The team reports that, while simulations yielded the same results as the lab experiments, simulations involved significant reductions in carbon emissions: Around 20 percent for many of the research scenarios, but for some scenarios such as plasma ion simulations, it was as high as 80 percent.
“Simulation appears to be almost universally less resource-intensive than physical experimentation, so we recommend that researchers look for opportunities to minimize experiments and attempt to solve their problems using computation methods,” Fried says.
Of note, the study revealed that production of just one full-loop wafer has a footprint of approximately 1,500 kilograms of carbon dioxide over its lifetime. By comparison, virtualization and AI tools running on a high end computer would need to run computations for 27,000 hours, or slightly more than 3 years, to achieve the same level of carbon emissions.
Fried points out that the environmental benefits of simulations extend beyond reduced carbon emissions, by conserving other vital resources, such as water and chemicals used extensively in semiconductor manufacturing, as well as helping to negate the release of air pollutants.
In terms of company operations, Fried says his team also found that simulations led to time savings, areduction in costs, and greater collaboration among teams.
“[Collectively] these promising results further point to virtual twins and simulation being a gamechanger for semiconductor manufacturing,” he says.
From Your Site Articles
Related Articles Around the Web
