Artificial Intelligence based control of Czochralski process
The Czochralski (CZ) method up to day remains as one of the most established techniques for producing high-quality crystals for strategic sectors in Europe, such as defense or semiconductor manufacturing. The traditional rule-based and PID control strategies to guarantee the crystal quality in this cost and time intensive process increasingly struggle to maintain process conditions at optimum in long production runs. Hence, Artificial Intelligence (AI)-enabled process control is emerging as an alternative innovative solution to replace conventional approaches.
In the CZ process, crystal quality is governed by complex interactions between thermal gradients, pulling rate, rotational speed, and ambient conditions. Small disturbances can lead to severe defects such as internal dislocations due to solid/liquid interfacing, oxygen concentration variations (i.e. porosity), or non-uniform diameter distribution. To avoid generating such defects, conventional control systems rely on physics- or mathematics based models and fixed setpoints, which are often insufficient to capture nonlinear, time varying dynamics. Instead, AI offers the ability to learn these relationships directly from process data, or synthetic data generated by digital models, replicating the same process conditions. AI-based controllers leverage historical and real-time sensor data to predict future process states. Artificial Neural Networks (ANNs) can estimate key quality characteristics (KQCs) that are difficult or impossible to measure directly, such as interface shape or internal stress, then adjust corresponding process parameters: pulling speed, heater power, or rotation rates to maintain optimal growth conditions.
One of the most significant advantages of AI-enabled control is its capacity for continuous improvement. As more growth runs are completed, models can be retrained to reflect equipment aging, raw material variation, and new operating regimes; or, to find synergic solutions in other industries facing similar challenges.
While the aforementioned challenges are not fully resolved yet, the adoption of AI in Czochralski crystal growing is accelerating. By complementing physical understanding with data-driven intelligence, AI-enabled process control represents a key step toward smarter, more autonomous crystal growth. Smart-Growth is one of the pioneering projects in this f ield, supported by European Commission’s I3 funding, connecting the leading institutes from Germany, Italy and Romania.
Figure 1. Cyber-Physical System (CPS) Architecture for AI-based CZ process control
Written by: Dr. Ozan Emre Demir
Assistant Professor in Department of Mechanical Engineering
Politecnico di Milano