Gas-flow instabilities and inaccurate delivery can decrease yields. To prevent such problems, mass-flow controllers (MFCs) are used for the accurate and repeatable delivery of process gases to the wafer. Incorporating a digital signal processor into an MFC further increases responsiveness, accuracy, and repeatability.^1–3 While previous generations of analog MFCs were typically used in the range of 25–100% of full-scale flow, digital MFCs with extended-range capability have been validated through extensive testing to perform at <10% of full scale while still providing accuracy of <0.275% of full scale. Moreover, while analog devices were limited to flow ratios of 10 to 1 and their accuracy was limited to l% of full scale, digital MFCs increased rangeability to 100 to 1.

Figure 1: Test results showing the impact of the calibration coefficient on the flow accuracy of an MFC with a flow range of 20 std cm3/min using C5F8.

Process recipes are evolving into multistep sequences to improve process control and performance. Increasing overall tool productivity through the implementation of in situ cleans also has enhanced the process capabilities of tool platforms. Process chamber cleans generally require high flow rates to reduce recipe-step times. These steps can require gas flows in the range of several hundred standard cubic centimeters per minute.

When these steps use the same gases as the process recipes, two gas sticks are generally installed in the gas box of a process tool. This arrangement often requires the installation of a second gas line leading to the gas box. Depending on its complexity, however, the installation of a second line can cost a facility thousands of dollars.

Figure 2: Comparison of the impact of the calibration coefficient on the flow accuracy of MFCs supplied by two different manufacturers.

As the complexity of process recipes increases, the need to minimize the number of redundant gas sticks is essential for optimal process tool performance. Digital, extended-range MFCs eliminate the need for redundant gas sticks and additional MFCs. In addition, they can provide accurate and repeatable flows at both high and low ranges, simplifying installations, allowing for more-flexible process tool platforms, and improving tool platforms' overall cost of ownership. Extended-range MFCs are calibrated over a wider range to improve accuracy at the low end.

Finally, providing MFCs with self-diagnostic capabilities helps engineers to monitor MFC performance and troubleshooting problems without having to remove the devices from the gas panel.

This article addresses MFC calibration issues, including the use of inert versus real gases to develop calibration coefficients. It continues by discussing the performance of an extended-range MFC on an oxide etch tool. In addition to addressing the impact of MFCs on process performance, the article highlights how the flexibility of an extended-range MFC helps process and equipment engineers to improve process control and development.

The use of on-tool MFC calibration is widespread. The practice is particularly important in processes where certain process gases such as hydrogen bromide (HBr) are used, since MFCs have been known to drift in such cases. The rate-of-rise technique is the primary means of performing calibration in process tool chambers because many platforms have built-in rate-of-rise capability. The article discusses several assumptions related to this technique. It also describes alternative calibration methods, including "golden MFCs," which are then compared with add-on rate-of-rise methods. The article concludes by addressing the yield impact associated with changing out a critical MFC.

The data presented in this article were gathered from a variety of tools provided by major semiconductor equipment suppliers. The processes under investigation included interlevel dielectric etch and various nitride etch steps, which utilize fluorinated hydrocarbons and corrosive gases. These applications are considered demanding from a process-requirement perspective. Thermal-based MFCs were used to conduct the study.¹

Traditionally, mass-flow controllers have been calibrated using inert surrogate gases. These inert gases are thermodynamically compared to reactive process gases and calibration coefficients are developed. Using process gases in the normal manufacturing process to calibrate MFCs would raise several concerns. First, if the reactive gases from the calibration process are not completely purged, they can contaminate the MFC and alter its performance characteristics when its wetted surfaces are exposed to air. Second, testing MFCs can pose safety and handling issues. Moreover, handling the wide variety of process gases can be expensive.

A formula relating these thermodynamic parameters to gas flow has been developed. The formula assumes that the gases will act in an "ideal" way, but that is not always a valid assumption. The calibration coefficents can be off by as much as 5% because of error stacking in the thermodynamic data. These errors can be especially problematic when the MFCs in question have low flow rates and use low-vapor-pressure gases such as C₅F₈.

Test results from an MFC supplied by manufacturer A with a flow range of 20 std cm³/min using C₅F₈ are shown in Figures 1 and 2. The data in Figure 1, showing the impact of the calibration coefficient on flow accuracy, were collected from a process tool. These data indicate that an MFC-related discrepancy was observed when the process was first qualified. The equipment engineer was concerned that the original data were at the 5% limit, a potentially unacceptable risk level indicating that intervention from the equipment engineering group was necessary. Manufacturer A reacted by adjusting the calibration coefficient. However, while that adjustment resulted in a slightly improved error rate, it was not significant enough to generate acceptable flow results to improve the overall accuracy of the MFC.

The equipment engineering group also tested MFCs from another supplier, who uses process-gas testing to validate gas correction factors.³ Figure 2, which compares the calibration error rate of MFCs from manufacturers A and B, demonstrates the impact of accurate calibration coefficients on controller effectiveness. This figure illustrates that the use of a process gas, rather than an inert surrogate, to calibrate MFCs improves the devices' accuracy over a range of flow rates.

Process recipes include complex combinations of gas flows. One method of addressing different flow-range requirements is to use multiple gas sticks for any given gas in the gas box. This approach can be expensive and also may be limited because of the number of available locations within the gas box. It also can require that a second gas line be installed outside the tool—an expensive proposition.

An alternative approach is to use an extended-range MFC, which makes it possible to use one controller for multiple steps within a process recipe. Figure 3, showing data from a test using O₂ at flow rates of 10–500 std cm³/min, illustrates the capability of an extended-range MFC over the range of 2.5–100% of full scale. (Traditional mass-flow controllers function with limited accuracy below 10%.) This test demonstrates that the improved accuracy provided by a digital device can be used to reduce hardware requirements.

The chart in Figure 3 was compiled from process tool and laboratory data and illustrates the relative reading error for each set point measured. While on-tool calibration generates significant errors, probably amounting to an accuracy of no better than 5% of reading, laboratory calibration has an accuracy rate of 1–2% of reading. The lower flow settings, because of their improved calibration accuracy, can be used in critical process steps, while the higher flow settings, despite their lesser accuracy, can be used in other process steps. The ability of an extended-range digital MFC to calibrate set points at higher flow settings makes it superior to analog or conventional digital devices.

Validation checks are commonly performed on the MFCs associated with plasma process tools. Typically part of monthly preventive maintenance in production fabs, such checks frequently uncover problems with MFCs. In particular, shifts in calibration that can affect process recipes may be detected.

Figure 3: Comparison of data from process tool and laboratory tests illustrating the capability of a mass-flow controller over the range of 2.5–100% of full scale.

Figure 4: Test results showing the calibration error of MFCs using HBr over a period of more than a year.

As part of this study, data from an oxide etch system were collected on a monthly basis for more than 1 year to test the calibration errors of MFCs using HBr, CF₄, and CHF₃. Based on measurements obtained from a rate-of-rise tool, Figure 4 presents calibration errors for HBr, while Figure 5 presents calibration errors for CF₄ and CHF₃. The trend change around November 15 seen in Figure 5 occurred when the MFCs were adjusted to compensate for calibration shifts.

The functional tests that can be performed on typical process tools include rate-of-rise calibration capability. The rate-of-rise test uses chamber volume, pressure transducers (usually capacitance diaphragm gauges), and a timer to measure the pressure rise that occurs as gas flows into the chamber. The rate-of-rise methodology, although simple, has inherent limitations. First, chamber volume is a variable that must be measured very accurately to develop a precise model for the rate-of-rise measurement. However, this volume can be affected by differences between replaceable parts in process chambers. For example, ceramic shields are variable components that are typically replaced during preventive maintenance. Second, chamber volume can be affected by changes to the foreline components or gas distribution systems. Third, adsorption of gases on chamber wall surfaces can decrease the accuracy of the rate-of-rise technique.

Furthermore, process gases that may be considered reactive behave differently under different conditions. For example, HBr and CF₄ are affected by the condition of the chamber. To compensate, the on-tool control limit for testing these gases is ±5%. But testing of each MFC over the range of flows is time-consuming and labor-intensive. Since such testing for a typical chamber requires several hours of downtime, it results in decreased throughput.

In addition to the chamber, the other major component used to measure rate-of-rise calibrations is a capacitance diaphragm gauge. However, such gauges have had zero-drifting problems associated with the deposition of etch by-products on the sensing diaphragm, which affects gauge accuracy.

Figure 5: Test results showing the calibration error of MFCs using CF4 and CHF3 over a period of more than a year.

Figure 5: Test results showing the calibration error of MFCs using CF4 and CHF3 over a period of more than a year.

Two other approaches are commonly used to perform on-tool MFC calibrations: the golden MFC method and the fixed-volume method. Golden MFCs offer higher calibration accuracy than standard ones. Used to deliver inert gases such as N₂, golden MFCs serve as standards to calibrate other, less-accurate MFCs. While the use of a golden MFC allows engineers to calibrate all other MFCs without encountering discrepancies in chamber volume and capacitance diaphragm gauges, the reliance on one MFC is risky. If a golden device is inaccurate, its errors will be transferred to all other MFCs in the system. Moreover, if the golden MFC occupies a position in the gas box, the number of positions available for the additional process gases required for new process recipes is reduced. Generally, one of the inert gas sticks needed for the functioning of the chamber is used to house a golden MFC. Gas-dependent variations also have been observed with golden MFCs, with corrosive gases exhibiting wider fluctuations than inert gases.

Fixed-volume rate-of-rise systems, which have been added to some process tools, employ a fixed-volume cylinder of known size for deriving rate-of-rise measurements. The major advantage of this method is that it requires a lower volume of gas to perform calibrations than does the built-in rate-of-rise method, expediting the calibration procedure. However, the former method does not necessarily lead to more-accurate results than the latter. Furthermore, if process gases used with the MFC are incompatible with the components in the fixed-volume system, gas adsorption can result in gas-dependent calibration errors. Finally, the installation of a fixed-volume system on a process chamber requires a significant initial capital investment.

Improving equipment performance by replacing a critical mass-flow controller can have an impact on multiprobe yield. Figure 6 illustrates a normalized product yield for the etch process step that was studied.

Four wafer lots each were run before and after a critical MFC was replaced. Since the multiprobe yield process was under development while this investigation was taking place and there were large variations between lots, the average of each set of four lots was calculated. More important than the absolute numbers is the fact that a positive yield shift was observed. Although not all of the yield improvement was attributable to the installation of a new MFC, improving the accuracy of a critical MFC in a process tool clearly does have an impact on process yield.

Thermal-based mass-flow controllers are widely used in the semiconductor industry. Nevertheless, the proliferation of new gases and complex recipes for oxide etch processes is requiring greater innovation with MFC technology. One example of such innovation is the use of digital technology to control and monitor devices. The widespread implementation of digital MFCs is one step in the process of enhancing MFC technology to meet increasing demands.

Digital technology can enable the implementation of improved calibration coefficients and allows for the implementation of nonlinear calibration coefficients over the flow range of a device. The implementation of on-tool calibrations also has led to improvements in MFC accuracy.

Despite their advantages, modifications to on-tool calibration coefficients may not always be enough to correct for calibration coefficient errors. Actual data from tests with process gases is essential to completely develop flow models for MFCs over their operational ranges.

Rate-of-rise measurements from process chambers contribute to the effectiveness of mass-flow controllers. Being able to quickly check MFC calibration helps engineers to identify dramatic calibration shifts. Although rate-of-rise systems decrease calibration time and can have a positive impact on uptime, they do not necessarily improve calibration accuracy.

Process yields are affected by the accurate control of gas flow to the process chamber. Attention to MFC performance is required to achieve the ever-tightening demands on process windows. It has been shown that the multiprobe yield of the etch process discussed in this article improved when a critical MFC was changed out. The improved accuracy of the device had a measurable impact on process performance.