SK Specialty, the world's largest producer of nitrogen trifluoride (NF3), has published in an international journal a technology it developed in-house to reduce the environmental burden of its core product. With greenhouse gas reductions emerging as a task across industries amid climate change, attention in the industry is focusing on the fact that corporations have taken the lead in establishing technology to search for alternative materials.
On the 19th, SK Specialty researchers unveiled an artificial intelligence (AI)-based global warming potential (GWP) prediction framework in the Journal of Vacuum Science & Technology B, published by the American Institute of Physics (AIP). The technology can pre-assess the warming impact of alternative substances to nitrogen trifluoride based on molecular structure information. GWP is a measure that compares a particular gas's contribution to climate change with that of carbon dioxide.
Nitrogen trifluoride is an ultra-potent greenhouse gas with a GWP about 17,000 times that of carbon dioxide. Although it is an essential gas for etching semiconductors or keeping equipment interiors clean, even small leaks into the atmosphere have a significant impact on climate change. Nevertheless, the use of nitrogen trifluoride has expanded more than 20-fold over the past 30 years because there has been no clear alternative in terms of process performance and stability.
SK Specialty is in a dual position of being the corporation that produces the most nitrogen trifluoride while also having to directly develop technology to reduce dependence on nitrogen trifluoride. In the semiconductor supply chain, where ESG (environmental, social and governance) management pressure is growing, securing process gases that reduce environmental burdens has become a strategic task not only for customers but also for suppliers.
To find potential alternative materials, the researchers developed a Machine Learning framework that calculates GWP based on molecular structure. The model learns data on about 200 molecules released by the Intergovernmental Panel on Climate Change (IPCC) and calculates the GWP of new molecules.
GWP is calculated by considering how effectively the gas traps heat in the atmosphere (radiative efficiency) and how long it remains in the atmosphere (atmospheric lifetime). Previously, complex experiments and lengthy computations were required, but by applying Machine Learning, the GWP of candidate substitute substances for nitrogen trifluoride can be calculated quickly.
SK Specialty researchers added that they expect this prediction technology to become an important tool for reducing dependence on nitrogen trifluoride and exploring new eco-friendly process gases.
Park Sung-woo, a researcher at SK Specialty, said, "The AI we developed does not stop at simply predicting environmental indicators. Ideally, it should evaluate eco-friendliness and process performance at the same time to find the optimal alternative substance," and added, "Even if a material is eco-friendly, semiconductor manufacturers, who are the customers, will not accept it if performance is poor."
Park continued, "The key question is how accurately we can predict GWP when data are very limited," and said, "This study is the first step to answering that question."
Recently, semiconductor corporations such as SK hynix and Samsung Electronics have also been working to reduce nitrogen trifluoride. SK hynix is developing alternative gases in collaboration with materials and equipment companies and is reducing nitrogen trifluoride use by optimizing more than 100 processes. Samsung Electronics has introduced nitrogen trifluoride reduction facilities into semiconductor production lines since 2018 and has improved decomposition efficiency to more than 95%.
References
Journal of Vacuum Science & Technology B (2025), DOI: https://doi.org/10.1116/6.0004715