Tool and process improvements from MFC control system technology
04/01/2002
Overview
A new approach to MFC calibration links the physical parameters of nitrogen to the physical characteristics of various process gases. This precludes the conventional need for surrogate gases. What results is a physics-based tuning algorithm and enhanced digital control system that enables re-ranging and gas change of digital MFCs. The end result should be better process control through more accurate gas flow. The new method also decreases the number of MFC spare parts required to back up a fab.
 A Kinetics Fluid Systems fab worker. |
Mass flow controllers (MFCs) are highly complex, custom components used in front-end wafer-processing tools. Conventionally, they are manufactured to a specific gas and range, resulting in a high volume of backup inventory at a fab. For example, in a benchmarked fab complex there were 8000 individual MFCs with 2300 unique configurations. By optimizing the MFC control system, it is possible to reduce these unique configurations as much as 100 times (i.e., to 23 unique configurations in the benchmark example) with improved accuracy, response, and repeatability.
A conventional MFC consists of a thermal sensor, bypass, control valve, and electronics. The electronics drive the sensor, condition the sensor signal, and control the valve to maintain a desired mass flow rate. The thermal sensor can only measure a maximum flow of ~5-10sccm. Consequently, a bypass is used to divert the balance of the total flow around the sensor. The bypass design ensures that total flow is directly proportional to the sensor flow.
Today's MFC trends
Gas-specific variables, including viscosity, specific heat, density, thermal conductivity, and compressibility, affect MFC performance. These variables are pressure- and temperature-dependent, thus adding to the complexity of the control system. Users expect an MFC to provide precise gas delivery in a variety of process conditions where the delivery pressure can be a few torr to 50psig and the ambient temperature can range from 23-150°C. Due to these complexities, MFCs are custom-manufactured for a specific gas and range to maximize accuracy and performance specifications. Over the years, manufacturers of process equipment have accumulated a database of thousands of MFC part numbers depending on particular operating parameters such as gas and range. Fabs normally stock hundreds of different MFC part numbers to minimize machine downtime in case of failure.
With newer digital MFCs came the potential for inventory reduction. Manufacturers began offering multiple gas calibrations stored in memory. Improvements in control algorithms and accuracy allowed users to operate the MFC over a larger range. Simple computer programs helped users convert the digital MFC to a new gas and range.
To date, however, digital MFCs have not delivered on the promise of inventory reduction. Users must predict in advance the gas calibrations they may require. Although this may seem simple, the specific tool that goes into a new or expanding fab is often not known in advance and the process that will be run can change even after the tool is installed. Therefore, obtaining a precise list of gases and ranges for each MFC can be difficult and time-consuming. In addition, this approach severely limits flexibility because an MFC typically holds just 9-36 specific gas and range combinations. Moreover, the customer must pay for each of the additional calibrations. Therefore, despite inventory reduction, the total cost of using multigas MFCs is higher than using conventional MFCs. In general, the market has rejected the digital approach.
Why custom calibration?
A surrogate gas is an inert gas selected to match the properties of the target gas. Each MFC is calibrated on the surrogate gas, since it is neither practical nor cost-effective to use process gases during MFC manufacturing. A conversion factor defines the relationship between the process gas and the surrogate gas used for calibration.
Conversion factors are typically a constant number. This implies the MFC will maintain accuracy and response when flowing the process gas at different flow, temperature, and pressure conditions than those for which it was calibrated.
A constant conversion factor assumes a linear relationship (i.e., the bypass sensor relationship and response characteristics do not change) between the surrogate gas and the process gas. Any deviation from a linear relationship results in degradation in accuracy and response. Constant conversion factors published by all MFC manufacturers give users a false impression that properly applied conversion factors allow the use of any gas.
Realistically, gases of varying properties can rarely be substituted on a conventional MFC without adjusting tuning and calibration. Consider the effect of surrogate calibration on accuracy. Ideally, the bypass design maintains a constant flow ratio between the sensor and the bypass. Multiplying the sensor output by this ratio calculates the total flow. A poorly designed bypass creates a nonlinear relationship between sensor and bypass flow rates. In other words, the flow ratio between the sensor and bypass changes as the total flow through the MFC increases.
 Figure 1. Measurement accuracy and errors using N2 surrogate and conversion factors to calculate actual MFC flow for specific gases. |
Figure 1 shows the effect of a nonlinear bypass. The MFC was calibrated on N2 and then tested on several gases. A constant conversion factor was used to calculate the expected flow rate on several test gases. The actual flow rate was then measured while running test gases on the N2-calibrated unit. The setpoint error is the difference between actual flow rate and expected flow rate. The error lines would all collapse to zero if linear conversion factors accurately represented the system.
Such a design would require a nonlinear conversion factor to accurately calculate flow on a process gas. But nonlinear bypass ratios and conversion factors are complex functions of viscosity, specific heat, density, thermal conductivity, sensor drive electronics, sensor bypass geometry, and pressure drop across (or velocity through) the sensor and bypass. Gas properties are also temperature-dependent. Consequently, the degree of system nonlinearity changes with each gas.
Calibrating an MFC on a process gas or on a surrogate gas that closely matches a specific process gas overcomes system nonlinearity. It is clear, however, that such an MFC requires calibration for a specific gas and range.
 Figure 2. Response degradation when running gases of varying properties on an N2-tuned MFC. |
Just as gas properties affect MFC accuracy, they also affect response (Fig. 2). Thermal properties of a gas affect sensor response time, while density and viscosity affect valve-operating position. For example, an MFC tuned on N2 will oscillate on a lighter gas such as H2 while it appears over-damped on a heavier gas such as SF6. Additionally, response will further degrade at lower setpoints, thus limiting the operating range. For this reason, an MFC often requires retuning of the control system before it can run a different process gas.
It is clear users cannot run process gases through an MFC without compensating for calibration and tuning. Although users have, for the most part, rejected using multiple calibrations because it is expensive and too restrictive, they still demand a digital solution to reduce inventory costs and shorten tool repair time.
A new class of MFCs is emerging called "multigas, multirange." This class must cover 3sccm to 30slm with only nine mechanical configurations. To accomplish this, a multigas, multirange MFC must be re-rangeable by a factor of 3:1, and must provide accurate flow measurement across all process gases, as well as consistent response at any setpoint from 2-100% of the desired full-scale range.
For example, consider a multigas, multirange MFC mechanically configured to flow a maximum of 1000sccm of N2. The user would like to run Cl2. The N2 to Cl2 conversion factor is 0.851, meaning the maximum amount of Cl2 the MFC could flow is 851sccm. The user would then have the option to electronically re-range the MFC to a full-scale 284sccm Cl2 (851/3) and still perform at a setpoint of 2% of full scale. This means the 1000sccm N2 MFC must accurately control Cl2 flow at ~5.5sccm. That is a control range 15x greater than a conventional analog MFC.
Two alternatives
MFC manufacturers have pursued two alternative approaches to providing multigas, multirange MFCs. The first involves empirically testing each gas over a wide range of flows and mechanical configurations and performing characterizations over a wide range of flows, temperatures, pressures, etc. Empirically based corrections developed for each process gas can then be applied to the surrogate gas calibration.
This approach is impractical because every gas, bypass configuration, flow range, and operating pressure and temperature requires a nonlinear conversion factor and tuning coefficients. Mechanical variations across production lots will affect the accuracy of the characterization. There is no guarantee the empirical relationship will hold after changing the mechanical configuration. Any change in the mechanical system or gas property affects accuracy and response. Empirical solutions are not easily expandable. The empirical relationships are specific not only for each process gas but also for the valve, bypass, and sensor system; changing any one would require the determination of a completely new set of empirical relationships.
The second approach is to characterize an MFC system on a surrogate gas, develop a physical model of the control system to predict how gas properties affect the system, and intelligently adjust calibration and tuning parameters to compensate for changes in gas properties. Since the system is based on a physical model, it is valid for any gas and range over a variety of conditions. Unlike empirical solutions, adding new gases does not require revalidation. The technology is expandable or transferable to a different system (i.e., a modified MFC with a new valve or new sensor). Finally, when properly implemented, MFC-to-MFC process gas accuracy and response will be more consistent, that is, it can be more consistently manufactured.
The solution
We believe that the second approach is optimal. But there are three critical elements that must be included in the design of a multigas, multirange system: a linear sensor-bypass configuration, a wide-range valve, and a sophisticated digital control system. Without such enabling technology, MFC mechanics will limit rangeability and accuracy. If the MFC system is capable of wide-range performance, a physics-based model of the control system can then be developed.
 Figure 3. Multigas, multirange accuracy from a 750sccm N2 MFC at setpoints from 10% to 110% full scale. |
We have found that a linear sensor-bypass design dramatically simplifies gas-to-gas accuracy across a wide range. Constant conversion factors can accurately predict flow on any process gas. Figure 3 shows the impact of a linear sensor-bypass on accuracy of an N2-calibrated MFC running various gases. This MFC was calibrated on N2 only. As with the data shown in Fig. 1, a constant conversion factor was used to calculate the expected flow rate. Note the dramatic reduction in flow error when compared to the MFC represented in Fig. 1. The multigas, multirange sensor design provides minimum drift and maximum signal-to-noise ratio, enabling precise and repeatable flow measurement over a range of 150:1.
A wide-range valve is a key component. Since the mechanical configuration cannot be changed in the field, the valve must be able to handle a wide variety of gases and volumetric flow rates. The rangeability of the valve enables the MFC to handle heavy and light gases across a wide range of operating pressures. The valve significantly reduces the number of mechanical configurations required to cover user applications.
The speed of response on a conventional MFC typically degrades at lower setpoints (see Fig 2). An MFC with a response of <1 sec at 100% will slow down to 2-3 sec at a 5% setpoint. A sophisticated digital control algorithm ensures consistent response across all setpoints from 2-100%. This provides a response of <1 sec across the entire operating range from 2-100%, even when re-ranged by a factor of three. This allows re-ranging without degrading response performance.
We have found that a multigas, multirange MFC also needs a physical model of the control system. A physics-based model must predict the effect of viscosity, density, conductivity, and specific heat on the response of the system. The model can be used to predict performance on the desired gas and adjust MFC tuning "filters" accordingly. The ideal filter coefficients can be calculated from gas properties. With the model validated, a user can confidently predict and run any future process gas if gas properties are known.
In our work, we have found that a multigas, multirange MFC can easily be reconfigured through software control: The user selects the desired gas and range. The software then indicates the required "standard configuration" and corrects the calibration and adjusts the tuning coefficients for the desired process gas. Several checks are automatically performed to ensure proper MFC configuration. Flowing gas is not required, since there is no recalibration after the MFC leaves the factory. The entire reconfiguration process takes <2 min. Once complete, the MFC is ready for installation on a tool.
Proving the technology
To prove the validity of our multigas, multirange MFC approach, we have worked with several OEMs, production fabs, and an independent third party to qualify the methodology. At each, the technology was tested for accuracy and focused on the rangeability of the digital product. Several test gases were selected to test a wide range of gas properties. The multigas, multirange MFCs were configured for the test gas and re-ranged using the configurator software. The average calibration uncertainty (3σ) was <1% of reading from below 10% to 100% of full scale. Response was consistently <1 sec at all setpoints.
In tests that digitally changed the tested MFC from 750sccm N2 to 825sccm N2 to 250sccm N2 to 825sccm H2, accuracy was well within the 1% setpoint error band (the dotted blue line in Fig. 3). The precise flow accuracy achieved validates the linearity of the sensor-bypass design and conversion factors. Therefore, N2 calibration accurately predicts performance on process gas. This shows that the MFC operates well within the accuracy specifications for very light, average, and very heavy gases.
 Figure 4. Typical response of a multigas, multirange MFC when reconfigured via software to run H2, N2, or SF6. |
We also found that multigas, multirange MFCs provide <1-sec response and finely honed flow stability across the entire operating range from 2-100% (Fig. 4). This allows re-ranging these MFCs without degrading response performance.
Conclusion
By optimizing an MFC control system, it is possible to use a very small number of standard-configuration multigas, multirange MFCs that concurrently have improved accuracy, response, and repeatability performance. OEMs, production fabs, and an independent third party have validated this control system for repeatability, accuracy, and response. The system relies on a stable, wide-range bypass and sensor to adjust for a large number of variables that affect MFC performance. This control system requires a consistent sensor-bypass relationship and provides a 1:1 backup with just 2% more installed MFCs as backup.
Acknowledgments
MultiFlo II (a multigas, multirange MFC approach) is a registered trademark of Kinetics Fluid Systems. Kinetics has five patents (granted or pending) associated with MultiFlo II technology covering a wide array of specific technical advances.
Peter Friedli is a graduate of the University of Texas, Austin, and received his masters in business administration from Case Western Reserve University. He is product marketing manager at Kinetics Fluid Systems, 22600 Savi Ranch Pkwy., Yorba Linda, CA 92887; ph 714/921-2640, fax 714/921-0804, e-mail [email protected].
Bill Valentine received his BS in mechanical engineering from Cal Poly, San Luis Obispo, and his masters in mechanical engineering from Purdue University. He is director of engineering at Kinetics Fluid Systems.
Peter Friedli, Bill Valentine, Kinetics Fluid Systems, Yorba Linda, California