Atomic-level defects and contamination: Impact on Power Consumption in Data Centers
Before rapid introduction of generative AI, power consumption in data centers was recognized to be a major issue, but nobody was capable to see how much energy would be needed to power AI systems. AI power demand is expected to surge 550% by 2026, from 8 TWh in 2024 to 52 TWh, before rising another 1150% to 652 TWh by 2030.
Semiconductor chip power consumption and cooling requirements are the main causes of data center energy consumption. This calls for considering energy efficiency at all stages of the chip development and manufacturing processes.
Atomic-level defects and contamination in semiconductor interfaces significantly impact the power consumption of data centers. These defects disrupt the normal flow of electrons, leading to increased leakage current and reduced energy efficiency.
Defects and contamination create additional energy states within the band gap of semiconductors, allowing electrons to flow more easily even when the device is supposed to be in an "off" state. This results in increased leakage current. In data centers, where thousands of servers are constantly running, increased leakage current can lead to significant energy losses.
Atomic-level impurities degrade the performance of transistors and other semiconductor devices, making them less efficient at switching and processing data. Lower energy efficiency means that more power is required to perform the same amount of work. This can lead to higher overall energy consumption in data centers. For instance, the average energy consumption was calculated to be 1.15 kWh/cm² based on a study on semiconductor manufacturing industry.
Consequences for data center finances
Increased power consumption directly translates to higher electricity costs. Data centers are already major consumers of electricity, and any inefficiency can significantly increase operating costs. Higher operating costs reduce profit margins and require additional budget allocations for energy expenses.
Higher power consumption generates more heat, which requires more cooling to maintain optimal operating temperatures. Increased cooling costs add to the overall financial burden, further straining the data center’s budget.
Consequences for sustainability
Data centers are significant contributors to carbon emissions due to their high energy consumption. Inefficiencies caused by defects and contamination exacerbate this issue. A larger carbon footprint can harm the data center’s sustainability goals and reputation, as environmental regulations and consumer expectations become more stringent.
Efforts to implement energy-saving measures, such as using renewable energy sources or optimizing power usage, may be undermined by inefficiencies at the semiconductor level. This can hinder progress towards sustainability targets and certifications, such as LEED or Energy Star ratings.
Consequences for operations
Defects and contamination can lead to unpredictable device failures and performance degradation over time. Lower reliability can result in more frequent downtime and maintenance requirements, affecting the data center’s ability to provide consistent and reliable services.
More frequent failures and performance issues require additional maintenance and repairs. Higher maintenance costs and increased workload for technical staff can strain operational resources and budgets.
Mitigation opportunities
To address these issues, data center operators and semiconductor manufacturers can employ several strategies.
High-Quality Semiconductor Materials: Using high-purity materials to minimize defects and contamination.
Regular Maintenance and Monitoring: Implementing rigorous maintenance schedules and real-time monitoring to detect and address issues early.
Energy-Efficient Designs: Incorporating energy-efficient hardware and software solutions to optimize power usage.
Renewable Energy Sources: Investing in renewable energy sources to reduce the carbon footprint and energy costs.
In addition, there is a major opportunity in the utilization of new innovative fabrication technologies, such as SisuSemi, that will clean semiconductor interfaces from atomic-level impurities and will make the surfaces ordered, significantly reducing leakage currents and energy consumption. Thus, data centres can mitigate the impacts of atomic-level defects and contamination, improving their financial, sustainability, and operational performance.