To ensure process transparency when replacing mass flow controllers (MFC) or when employing new types of MFCs in a particular process, it is often desired to have the flow rate of the replacement unit match the flow rate of the previously installed MFC. Similar in nature is chamber-to-chamber matching to achieve the same process capability and results. Traditional approaches to matching flow rates rely on the inherent accuracy of MFCs, calibrated traceable to national standards. This article focuses on alternative methods, often using tool-flow verification methodology, to recalibrate the replacement MFC “on-tool” to match the output of the previous MFC.
It is common practice in the semiconductor industry to utilize the process tool to measure the flow from MFCs installed on the gas panel. The process tool can be used as the flow measurement standard and obtain flow data on MFCs prior to their removal. When replacements are installed, the same data can be taken from the tool and used to adjust the new MFCs to the same output as the originals.
In addition to providing process transparency, the replacement MFC may be adjusted on-tool to ensure that its flow measurements agree with the process-tool flow measurements to within a specified tolerance. Inherent to the technique of adjusting the MFC on the tool is the belief that flow measurements established by the tool are the best metric for improved process repeatability and reproducibility. This article will not attempt to validate this claim but will focus on the best methodology to achieve the most accurate delivery of gas precursors.
Figure 1. Schematic of a rate-of-rise (ROR) system.
In measuring the flow from MFCs, process tools utilize a constant volume chamber and change-in-pressure technique, sometimes referred to as the rate-of-rise (ROR) method for flow determination (Fig. 1). This technique is widely used by both flow standards laboratories and industry for flow measurements in vacuum-based semiconductor processes. However, wafer fab tools are not optimized to make flow measurements. Systematic and random errors are introduced by tools that diminish the accuracy of flow measurements. Flow accuracy as measured with semiconductor tools depends on the tool design, flow rate, and usage, but the accuracy is generally in the range of 3-10% of setpoint. The equation used to measure the flow is
and is derived from the ideal equation of state for the gas, where ΔP is the pressure change over time interval Δt. The chamber volume is given by V, the absolute gas temperature by T, and R is the universal gas constant. Equation 1, for determining the flow, is in molecules/unit time and can be converted to conventional units, such as sccm . Errors in system volume have an uncertainty sensitivity factor of 1 (1% error in volume results in a 1% error in flow measurement). Errors in temperature have a sensitivity of 0.3, and errors in pressure change have a sensitivity factor of 1. Equation 1 assumes all gas input into the system remains in the gas phase, which is not always true for reactive gases, gases with low vapor pressures, or systems that may “pump” certain gases via chemical reactions with chamber surfaces or via physisorption. Given this, it is easy to see how cumulative errors can approach 10% for systems not optimized for flow measurements.
In spite of these limitations, it is often desired to have an agreement - within certain limits - between the flow as indicated by the MFC and the measurements by the tool flow verification method. The process engineer often uses this agreement to qualify process tool/gas delivery system performance. Depending upon the criteria (acceptance limits), disagreements between the MFC flow and tool flow verification method may occur. If the disagreement is within a certain range (typically ±10% for noncritical processes and ±5% for critical processes), the MFC would not be replaced, but may be adjusted to match the flow measured by the tool.
Tool-based flow adjustment for matching
Most process tools have the capability to compensate for the difference between the MFC flow and that measured by the tool. This is performed by using a single fixed-gain type correction factor, which effectively changes the setpoint to the MFC during processing. The technique can adequately compensate for differences that have constant errors as a percentage of setpoint, but is poor for situations where nonlinearities exist. Typically, tools are not equipped to make nonlinear adjustments and the MFC-tool agreement may not meet the expected level of matching with the tool-based adjustment. In this case, other approaches must be utilized within the MFC itself to improve flow matching.
MFC-based flow adjustment for matching
On-tool linearity adjustments of MFCs were not practical until the advent of digital MFCs. To make adjustments with the previous generation of analog MFCs, gas panels had to be opened and the MFC case removed. In some installations, this was nearly an impossible task. Once accessible, the changes would then require an iterative process and hours of tool time for implementation.
Digital MFCs allow the possibility of remote calibration through command transmission, which effectively changes the calibration constants. In some cases, these digital commands can be transmitted through the tool controller and, in others, may be accomplished through diagnostic ports on the MFCs via other computers.
Once the MFC’s settings are adjusted to the flow measured from the tool, the accuracy of the device can no longer be certified by the MFC manufacturer. With these adjustments, the accuracy of the MFC instrument will only be as accurate as the tool-flow measurement capability (3-10% of setpoint). Process engineers could find themselves in a dilemma, having to choose between good flow matching or verifiable accuracy, when the desired goal is to achieve both. The preferred path would be to understand and eliminate the discrepancies, which requires a thorough understanding of the MFCs and the chamber flow-measurement capability.
Origin and nature of flow matching errors
Resolution of the differences between the tool measured flow and that of the MFC requires insight into the probable causes and the nature of the disagreement. If readings from one MFC disagree with those coming from all other MFCs on the same process chamber, it may be safe to assume that the performance of the single MFC is poor, given the caveats of the previous section (i.e., differential absorption of low-vapor pressure gases and possible chemical reactions).
A failure can be caused by many contributing factors, including instrument/component failure, poor factory-calibration accuracy, or poor accuracy in the model used to relate calibration gas to process gas. All but the latter typically can be determined through subsequent failure analysis. If the failure analysis reveals the MFC is performing within specifications, then the chamber flow measurement should be further investigated. Alternately, if all of the flow measurements of the MFCs made by the tool are skewed in a particular direction, it may indicate a systematic error in the tool. More often than not, the tool chamber volumes are not certified and the temperature used for the calculations may not be representative of the gas temperature. In addition, the pressure measurement devices are not typically recertified at regular intervals and may also be outside their respective tolerances.
Figure 2. Cross-section of an IntelliFlow digital MFC.
To mitigate these factors, many users have adopted the practice of using a single MFC as a benchmark standard on the tool. The selected MFC is typically configured to run an inert gas, such as argon or nitrogen, and is close to the mean flow used for processing. The flow of this MFC is measured by the tool, and the ratio of the tool-measured flow to the MFC-indicated flow is used as a scaling factor to adjust subsequent flows measured by the tool with other gas/range combinations. This approach, used by many OEMs and device manufacturers, has been given a variety of names and is often known as the “golden MFC” technique. The advantage is that the golden MFC effectively calibrates or compensates for fixed errors due to incorrect chamber volumes and temperature. Intuitively, one could surmise that this technique combines the best of two worlds: the accuracy of the MFC with its calibration gas, and the tool’s ability to measure the flow of other gases, based on a calibrated correction from the golden MFCs.
Digital MFC calibration
Digital MFCs, such as the IntelliFlow (Fig. 2), are calibrated by direct comparison to traceable flow standards set by the US National Institute of Standards and Technology (NIST). Typically the flow standards are laminar flow meters, which have been directly calibrated with NIST-traceable standards for the particular gas. The calibration is performed using the methodology of Semi E56-1296. The flow standards are certified to have an accuracy of ±0.25% of reading or better. The flow measurements from the standard are recorded at predetermined flows (typically 10%, 25%, 50%, 75%, and 100% of the MFC full scale). The digitized signal output of the MFC sensor is also recorded at these set points. Figure 3 shows a simple schematic of the signal acquisition path. A relationship can then be established between the MFC signal (sensor output) and the measured flow.
Figure 3. MFC control schematic.
In general, this can be determined with a functional fit as given in the equation
In practice, the relationship between the actual flow and the thermal-flow sensor output is relatively linear and seldom requires high-order fits. For example, IntelliFlow digital MFCs utilize a third-order fit to accommodate the nonlinearities particular to some process gases used in semiconductor manufacturing. A number of methods are available for determination of best-fit coefficients. Among these, the method of least squares  is the most used and accepted in instrumentation. It should be observed that Eqn. 2 does not contain a constant or zero-order coefficient because, by definition, the flow should be zero when the MFC sensor output is zero. For a first-order fit, the least squares coefficient is given by the equation
The generalized form of the least squares expression is given in matrix notation of this equation
After the coefficients are calculated, they are downloaded into the MFC’s electrically erasable/programmable EEPROM device via an RS-485 communications port. The control loop then uses these coefficients to convert the sensor signal into a calibrated flow output that can be communicated digitally through a number of industry standard protocols. The calibrated flow output also can be converted back to an analog signal through digital-to-analog converters resident in the digital signal processor of the control loop.
In lieu of using a continuous calibration function, many MFCs often employ piecewise linear calibration, which effectively creates a lookup table that identifies a flow output for a given sensor signal. To obtain flows at sensor signals intermediate to sensor values in the table, linear interpolation is utilized.
Process tool-MFC agreement
Establishing an agreement between the MFC and the tool is typically done at tool startup. To do this, measurements are performed over the intended range or at a prescribed interval, such as 10% of full scale to 100% of full scale. For best results, the MFC typically is zeroed and the chamber evacuated. The MFC is isolated from the vacuum pump, and the “leak-up” rate is measured. If it is acceptable, the MFC is valved into the chamber and given a setpoint. The pressure rate of change is then measured and the flow is calculated based on the equation cited at the start of this article. The time required to perform this measurement can vary from tens of seconds to 5+ min, depending on the flow rate and protocol used. This process is repeated for each setpoint of the MFC and each MFC in the gas panel. Depending on the number of MFCs and test points, the process can take from a few minutes to a few hours.
Figure 4. a) MFC-tool verification plot for 300mm dielectric chamber, and b) tool agreement using the golden MFC method.
A sample MFC-tool verification plot is given in Fig. 4a for a 300mm dielectric etch chamber. The range of full-scale flows for MFCs varies from a few liters/min for argon to tens of sccm for oxygen. The data show agreement to within ±3%. While the data is within the acceptance criteria there are distinct trends. The fluorocarbon gases CHF3 and C4F6 are lower than other higher vapor-pressure gases. In this particular instance, the gas line for C4F6 was heated to 40°C while the MFC calibration was optimized for 20°C. After this adjustment was made along with appropriate zeroing of the remaining MFCs, the agreement between the tool and MFC was within a ±1.5% tolerance band.
A different outcome would result if the golden MFC approach was utilized with the effective chamber volume being adjusted upward by 0.4% to align with the nitrogen MFC at 50% of full scale. In this instance, the use of the golden MFC approach actually induces a failure of the C4F6 MFC by biasing the data in the positive direction, as shown in Fig. 4b. The golden MFC approach only reduces failure rates in instances where there is a systematic error in the chamber volume or the chamber temperature is not known to an acceptable level. In all other instances, the golden MFC can add the variation of the MFC itself to the measurement system.
Figure 5. Distribution of a) an MFC-tool flow verification agreement for all gases and b) for nitrogen gas only.
The previous example examined the MFC-tool agreement at a chamber-specific level, where the engineer wants to maintain agreement for a particular chamber. When expanded to a section of a fab, such as etch, the amount of data and interpretation of the data become more difficult. Figure 5 shows the distribution of MFC-tool flow errors for dielectric etch (~10 tools). The distribution of errors for nitrogen MFCs is similar in variation to that of the gases as a whole. The data collection was predominantly made at the extremes of the process range (lowest flow rates) to isolate MFC-tool disagreements.
Due to the disagreements in measurements cited previously, a select number of MFCs were returned to Mykrolis and tested with the process gas with NIST-traceable flow standards. It was not possible to verify the tool-measured errors (in the range of 3-4%). Subsequent work determined that standardized protocols for tool-MFC data collection were not always followed, and this added unnecessary variation. The following protocols should be implemented to ensure the best MFC performance and tool-MFC agreement.
- MFCs should be zeroed on tool before process qualification using standard protocols.
- If heated lines are employed, the tool owner should verify the calibration accuracy with the MFC vendor.
- If not employing a golden MFC approach, the chamber volumes should be verified.
- The best known methods for establishing chamber temperature should be followed. (Typically, this requires a certain amount of chamber cool-down time for plasma process tools before flow verification commences.)
- Performance of the capacitance manometers should be verified using cross-comparison on the same chamber or other methods.
- Ensure use of best known methods for accumulation time. (ROR test of pressures below 300mtorr is generally discouraged.)
Adjusting MFCs to achieve MFC-tool agreement
In an alternative approach, process engineers may align the MFCs to agree with the tool flow measurement when the tool is commissioned and subsequently utilize the MFC-tool disagreement as a measure of process drift. This MFC adjustment is often termed “relinearization.” While relinearization may reduce the accuracy of the MFC, it may be desirable for other reasons such as chamber-to-chamber matching.
In other instances, process engineers utilize the tool alone as the measurement standard and adjust the MFCs whenever the disagreement exceeds an established threshold. This practice is the worst implementation of MFC-tool agreement, because it negates the usage of the data to diagnose problems and incorporates flow-verification process errors or systematic flow-measurement errors into the MFC. Many processes are much less sensitive to variation in total pressure than they are to chemical balance changes, which are induced through improper adjustment of the MFCs.
Summary and future directions
The best technique for ensuring chamber-to-chamber or tool-to-tool reproducibility is not positively known and may vary by application. As in many processes, proper execution of the chosen strategy is extremely important. The best method to ensure flow accuracy would be embedding a flow standard on the process tool, which could verify the MFC performance and subsequently make adjustments if required. These types of systems are available but require tight integration with the tool process-control software for full utilization. Adoption of this technology will not only help eliminate wasted efforts in utilizing process tools as flow standards, but also will increase tool availability due to reduced discrepancies in MFC-tool flow measurements. While promising, the adoption of this technology will be incremental because its integration into existing tools is not a trivial task.
The author would like to acknowledge the data collection efforts of the Mykrolis applications development group and the field applications group. IntelliFlow is a registered trademark of Mykrolis.
- Semi standard E12-96.
- Neter et al., 4th Edition of Applied Linear Statistical Models, McGraw-Hill Companies, New York, 1996.
Stuart Tison is the business unit manager for the Gas Delivery Business at Mykrolis Corp., 915 Enterprise Blvd., Allen, TX 75013; ph 972/359-4401, fax 972/359-4100, e-mail Stuart_Tison@mykrolis.com.