Abstract
In semiconductor manufacturing, the precise calibration of mass flow controllers (MFCs) is critical for process stability. The "Rate of Rise" (RoR) method is commonly used for in-situ verification, but it suffers from measurement inaccuracies due to non-uniform gas temperatures and sensor hysteresis during high-speed charging. This paper presents a 3D Computational Fluid Dynamics (CFD) approach to predict the average gas temperature inside a fixed chamber. By modeling the dynamic temperature changes for various flow rates and gases, the researchers achieved a "virtual measurement" that compensates for physical sensor limitations. Experimental results demonstrate that this method reduces flow rate measurement errors from 2.8% to within ±0.5% across a range of 5 to 2850 sccm.
Introduction: The Challenge of In-Situ Calibration
Mass Flow Controllers (MFCs) are the gatekeepers of chemical delivery in chip fabrication. To maintain accuracy, they require periodic in-situ calibration using the pVTt (pressure, Volume, Temperature, time) method. By charging a known volume and measuring the rate of pressure and temperature rise, the actual flow rate can be calculated.
However, the "T" component—the average gas temperature—is notoriously difficult to capture. During rapid charging, "flow work" generates heat, creating a highly non-uniform temperature field. Standard sensors, often wall-mounted or shielded, exhibit significant hysteresis and fail to reflect the instantaneous average temperature of the gas bulk. Traditionally, this forced a trade-off: either wait for thermal stabilization (reducing throughput) or accept significant measurement errors.
Methodology: Virtual Temperature Measurement
To overcome these physical sensing barriers, a "virtual measurement" approach was developed using 3D CFD simulations.
CFD Modeling and Simulation
The researchers utilized ANSYS FLUENT to perform unsteady simulations of a 2.83L standard container. A high-density mesh was employed, with specific refinement at the sonic nozzle to accurately capture the complex flow physics and heat transfer during the charging process. Simulations were conducted for Nitrogen ($N_2$) at flow rates ranging from 5 to 2850 sccm.
Experimental Setup
A physical test rig was constructed to validate the simulation results. The system included a pressure regulating valve, a high-precision flow standard, and the standard container equipped with multiple temperature and pressure sensors.
To make the CFD insights applicable in real-time without the need for live, heavy computation, the researchers derived mathematical relationships from the simulation data. They found that the temperature rise ($\Delta T$) follows a linear relationship with flow rate at low flows (5–275 sccm) and a cubic polynomial relationship at higher flows (325–2850 sccm).
Key Insights: The Sensor Blind Spot
The experimental data confirmed a significant discrepancy between the actual gas state and sensor readings. While the pressure inside the chamber rose linearly during charging, the physical temperature sensors remained nearly constant or showed delayed, noisy responses.
This "sensor blind spot" is the primary source of error in traditional RoR calibration. The CFD model revealed that while the temperature field is non-uniform, the average temperature stabilizes within approximately 0.6 seconds of the start of the flow. By using the predicted average temperature instead of the lagged sensor data, the calibration can proceed dynamically without waiting for the system to reach thermal equilibrium.
Results and Discussion
The implementation of CFD-based temperature prediction yielded a dramatic improvement in calibration accuracy:
- Error Reduction: Without correction, the relative error in flow rate measurement was as high as 2.8%. With the proposed dynamic temperature compensation, the error was reduced to within ±0.5%.
- Operational Range: The method proved robust across a wide range of flow rates, from 5 sccm to 2850 sccm.
- Efficiency Gains: By eliminating the stabilization wait time, the throughput of in-situ MFC verification is significantly increased, a vital factor for high-volume semiconductor manufacturing.
Conclusion
The integration of 3D CFD simulations into the RoR calibration process provides a powerful solution to the problem of dynamic temperature measurement. By treating temperature as a predictable variable rather than a poorly sensed one, the researchers have enabled faster and more accurate in-situ verification of MFCs. Future work will focus on extending this model to other process gases and further refining the mesh to minimize residual discrepancies between simulation and experiment.
Reference: Zhang, Y., et al. (2025). Dynamic Temperature Prediction for the Charging Process of the Fixed Chamber. Proceedings of the 12th International Conference on Fluid Flow, Heat and Mass Transfer (FFHMT 2025). Paper No. 160.
Summary updated