The process of chip manufacturing is being boosted by the very systems that those chips are designed to power. Artificial intelligence (AI) and digital technologies are helping semiconductor manufacturers cut back on waste, save on downtime and reduce their energy consumption.
These new technologies will enable the production of greener chips, contributing to the more sustainable growth of AI and alleviating some concerns about the climate impact of AI adoption.
Undoubtedly, such sustainability concerns have waned slightly in recent months as business leaders focus on economic and geopolitical concerns. We note, however, that these technologies also offer significant cost savings for companies.
As AI systems grow smarter and chips grow more efficient, we see AI significantly changing the way the semiconductor industry operates.
Making predictions, delivering insights
The use of AI starts long before the manufacturing process begins, with the designing and prototyping of chips. AI can optimise product designs and model energy consumption, which speeds up the design process and delivers the best possible chip.
On the manufacturing floor, we also see AI being used to monitor equipment and prompt predictive maintenance. This is significant because an hour of planned maintenance can result in a company saving on three to four hours of reactive maintenance later.
Manufacturing is an energy- and water-intensive process, and AI can help optimise the usage of both power and water. Our own research suggests the use of AI can help a fab reduce its carbon emissions by about 15 percent. Considering that a fab’s energy consumption can constitute up to a third of its operating expenditure, this can add up to significant savings.
AI can help manufacturers cut back on other resources too, including raw materials and tools used in the chip-making process. With the help of AI modelling, junior engineers can also gain more confidence and learn faster.
Real-world implications
Faced with such a long list of potentially useful applications, companies may struggle with selecting solutions to implement. If sustainability is not a priority, there may also be reluctance to invest in any implementation at all.
In our experience, however, AI adoption in the semiconductor manufacturing process can pay off substantially and quickly. We have seen customers achieve over US$1 million in annual savings from implementation in just one facility, or a return on investment within six months.
- Henri Berthe, Vice President, Semiconductors, Schneider Electric
Beyond the immediate cost barrier, we also see implementation challenges in the workforce. Employees may lack the skills to work with AI solutions, and some are also resistant to change. Companies that have achieved success will be the ones with an upskilling culture and strong change management programmes.
In the long run, companies that successfully implement AI solutions in their manufacturing processes are likely to be more attractive to employees, customers and suppliers. They will face fewer regulatory risks and are also more attractive to investors. These rewards are less tangible but no less beneficial.
The way forward towards sustainable AI
Of course, we would be remiss not to also address the sustainability challenges posed by the use of AI. The processes for AI can be energy intensive, and the industry has already drawn some criticism for its carbon footprint.
There is, however, a way forward in sustainable AI. This is a philosophy promoting energy-efficient algorithms, frugal hardware and low-impact system design to reduce resource consumption.
Sustainable AI creates a symbiotic relationship in which AI enhances renewable energy use and grid efficiency, enabling greener AI development. The use of AI in running data centres and making chips, for instance, will ultimately help reduce the carbon footprint associated with AI.
Over a trillion semiconductor chips are produced each year. They are found not just in AI data centres but also in smartphones, cars and even washing machines.
Our research estimates that semiconductor devices produced in 2021 will generate a lifetime carbon footprint of approximately 500 megatonnes (Mt) of CO₂ equivalent, with 63 percent of these emissions attributed to downstream Scope 3 sources.
Addressing emissions for this industry is critical not just for climate targets but for long-term viability. By 2030, the industry is expected to consume 237 TWh of electricity, equivalent to what Australia used in 2021.
As energy demand from the semiconductor industry rises, sustainable AI offers a pathway to manage this growth within planetary limits through renewables, efficiency gains and responsible design.
Henri Berthe is Vice President, Semiconductors, Schneider Electric
