This use case develops AI-supported scheduling methods for batch-processing machines in semiconductor wafer fabs, especially energy-intensive tools such as furnace equipment. The goal is to reduce environmental impact and energy cost by shifting operations to periods with cheaper or greener energy while still maintaining production performance. Genetic algorithms, genetic programming, dispatching rules, Pareto-optimal schedules, and discrete-event simulation are used to balance energy consumption, total weighted tardiness, cycle time, and fab-wide KPIs. The expected result is an energy-aware decision-support tool that improves on-time delivery and reduces energy consumption and operating cost.
| # |
Technology Name |
Description |
| 1 |
Energy-aware scheduling |
Scheduling approaches are used to save energy and improve the on-time delivery performance of lots on batch processing machines. |
| 2 |
Intelligent energy management |
Machine learning methods are used to better understand and forecast energy consumption across factory buildings, cleanrooms, and equipment. The approach addresses the lack of equipment- or product-level energy breakdown in current semiconductor infrastructure by training ML models on power, production, and weather data. |
| KPI # |
Related Technology |
KPI Target |
| 1 |
Energy-aware scheduling |
Improve on-time delivery cost and total electricity cost by up to 20% using genetic programming. |
| 3 |
Energy savings |
Demonstrate that more than 750 kWh could be saved per day in case of high energy peaks. |
| 4 |
Cost savings for hardware |
Identify hardware cost savings for the golden tool strategy, with 2.8 million EUR savings identified for smart meters at Infineon Dresden. |
| KPI # |
Baseline |
| 1 |
No energy-aware dispatching rules are available; only manual stops of high-energy equipment are used. |
| 2 |
No hardware, such as energy meters for semiconductor equipment, is available at IFD and IFAT. |
| 3 |
No ML methods for energy management are available. |
| KPI # |
Validation Method |
| 1 |
Design, implement, and assess the performance of energy-aware scheduling algorithms. Evaluation is based on randomly generated problem instances for parallel machines and two-stage flexible flow shops, simulation experiments for complex job shops, and a limited field test using real-world parallel-machine problem instances from IFD. |
| 2 |
Collect and analyze energy-related data using smart power meters installed at individual tools and busbars, flowmeters for gas and other media at tools, and temperature sensors in chambers at individual tools. |
| 3 |
Develop and train ML models on historical power, production, and weather data using data science tools. |
| 4 |
Extend the approach to other media usage by applying various ML models. |
- Industrial process phases made more effective and precise through integrated and tailored AI algorithms, methods, and tools.
- 20% improvement in prediction precision.
- Reduced manufacturing costs in the long term.
- Improved overall productivity of the manufacturing factory by approximately 20%.
- Providing documented methods for dynamic application of tools to improve sustainability.
- Decreased energy consumption for PoC above 10%.
The use case contributes to more effective and precise industrial process phases by introducing AI-supported and energy-aware scheduling methods for batch-processing machines in wafer fabs. Genetic programming-based scheduling aims to improve both on-time delivery cost and total electricity cost by up to 20%, while ML-based energy management improves the understanding and forecasting of energy consumption across factory buildings, cleanrooms, and equipment. The use case also supports sustainability objectives by demonstrating potential daily energy savings of more than 750 kWh during high energy peaks and by identifying substantial hardware cost savings for the golden tool strategy. The combination of scheduling optimization, energy forecasting, anomaly detection, and validation frameworks contributes to lower energy consumption, improved operational efficiency, and more sustainable semiconductor manufacturing.
| AI Toolbox Tool |
Use within the Use Case |
| Lag-Llama |
Used to forecast energy consumption patterns from historical power, production, and weather data. It supported the intelligent energy management component by helping predict future energy demand across factory buildings, cleanrooms, and equipment. |
| Time-Series Anomaly Detection Tool |
Used to detect unusual energy-consumption patterns, high-energy peaks, and abnormal behavior in equipment or facility-level energy data. It supported the identification of situations where energy-aware scheduling or corrective action could reduce consumption. |
| Tabular Prediction Tool |
Used to train prediction models on structured power, production, equipment, and weather datasets. It supported energy forecasting, energy-cost estimation, and the analysis of relationships between production schedules and electricity consumption. |
| Explainability Tool |
Used to interpret the energy forecasting and scheduling-support models, helping users understand which production, equipment, or weather-related factors contributed most strongly to predicted energy demand or cost peaks. |