The rush to boost production of memory chips to meet fast accelerating demand from artificial intelligence will add to the semiconductor sector’s climate footprint and risks lifting costs of managing emissions.
Higher manufacturing volumes, efforts to deliver more complex and resource-intensive products, and additional production in nations with fossil fuel-reliant power grids will combine to increase the industry’s pollution profile, according to Ottawa-based research firm TechInsights Inc.
Semiconductor manufacturing emissions will climb by about a third to 247 million metric tons of carbon dioxide equivalent by 2030, the researcher forecasts. That’s higher than annual emissions of scores of countries and roughly equivalent to the 2024 volume produced by Algeria.
Power consumption and the use of fluorinated gases to etch circuits on silicon wafers are among the main sources of the industry’s climate pollution. While foundry and logic chips, a category that includes Nvidia Corp’s AI accelerators, will continue to account for the largest overall share of emissions, the impact from manufacturing dynamic random-access memory, or DRAM – short-term memory used in data centres and computers – will increase at a faster rate, according to TechInsights.
Of specific concern are high-bandwidth memory chips, or HBM, a subset of DRAM that require more materials and which researcher Silicon Analysts estimates can consume as much as five times more energy per gigabyte than standard memory categories during production.
"The AI-driven surge in HBM and other advanced memory is likely to raise semiconductor manufacturing emissions in absolute terms,” said Stephen Russell, senior technical fellow at TechInsights. "It increases memory wafer starts and adds process complexity, even as leading manufacturers improve efficiency.”
A race for AI dominance is spurring the likes of Alphabet Inc, Amazon.com Inc, Meta Platforms Inc and Microsoft Corp to earmark hundreds of billions of dollars in spending this year. The pressure is now on manufacturers of memory, one of the biggest bottlenecks for training AI models, which demand ever-faster access to huge datasets.
AI is already facing scrutiny for driving up global electricity demand, intensifying water-stress in some regions, and contributing to higher US emissions. Data centre and semiconductor growth is also impacting green efforts by major firms, with Microsoft citing AI and cloud expansion after last year reporting total emissions had jumped almost a quarter since 2020.
Limiting emissions from chip production can be costly. Abatement systems to manage fluorinated gases typically need investments of hundreds of thousands to millions of dollars per fabrication line, said Pallav Purohit, a senior research scholar at the International Institute for Applied Systems Analysis in Austria.
Samsung Electronics Co, SK Hynix Inc and Micron Technology Inc, which collectively account for the vast majority of HBM production, are all taking steps to lower the impact of their manufacturing.
SK Hynix is "actively researching and implementing various measures” to reduce emissions, including installation of high-efficiency scrubbers, the company said in a statement. Scope 1 and 2 emissions intensity of the producer’s chips per gigabyte fell a third between 2021 and 2024, the latest year for which data is available.
Samsung "is continuously making efforts to reduce carbon emissions even as we expand our manufacturing facilities,” the company said in a statement. In the firm’s semiconductor segment that involves the installation of systems to treat process gases and development of alternative manufacturing inputs.
Micron reported lower Scope 1 and 2 emissions in fiscal 2024, and has a target to achieve a 42% reduction in Scope 1 emissions from a 2020 baseline by 2030. The producer is "confident in our trajectory toward meeting this target, even as we expand our global manufacturing footprint,” it said in a statement.
Moving manufacturing to cleaner grids could also help curb emissions, though in the short-term at least, the industry is likely to see more capacity growth in nations which remain dependent on fossil fuels for electricity generation, like China and South Korea.
Advanced chipmaking is a top priority for China as the government aims to lower reliance on the US and its allies. Beijing’s bid for technological supremacy is spurring local companies from Huawei Technologies Co to Semiconductor Manufacturing International Corp and Changxin Memory Technologies Inc to expand into advanced AI chips.
"Real-world mitigation is improving rapidly,” said TechInsights’ Russell. Still, "total emissions can still rise if volumes grow faster than reductions in emissions intensity.” – Bloomberg
