Evaluating the performance of digital mass-flow controllers

Mohamed Saleem and Sowmya Krishnan, Ultra Clean Technology

With the advent of 300-mm and single-wafer processing, the semiconductor industry has placed increasing demands on precise process control and repeatability. One area requiring stringent process control to meet the demands of decreasing device geometries is the delivery of gases to the process tools. Controlling gas delivery, in turn, requires accurate flow control devices such as mass-flow controllers (MFCs).

Typically, gas delivery is accomplished by means of an MFC located on a process tool among other gas-delivery components. Most MFCs operate on the principles of heat and mass transfer. Analog MFCs were previously the workhorse of the semiconductor industry. However, despite their widespread use, conventional analog MFCs have limitations in specific areas of gas-flow range. Most analog MFCs become less accurate at flows below 10% of the full-scale flow range. Their accuracy specification is normally a function of full-scale flow, and flow errors are magnified when the MFC operates at less than its specified full-scale flow. Most MFC manufacturers specify accuracy in terms of ±1% of full scale for 2–100% flow ranges. For example, an analog MFC rated to perform at 1000 std cm³/min has an error range of ±10 std cm³/min. If the unit operates at a setpoint of 10% (100 std cm³/min), the flow rate is 100 ±10 std cm³/min; in other words, the flow rate has a ±10% error range. The error is further magnified when the MFC operates below a setpoint of 10%.

Thus, while the flow accuracy of an analog MFC is about 1% near full scale, it drops several percentage points at low flow rates (<25% of full-scale flow). That loss of accuracy poses repeatability problems at low flow rates, limiting the overall functionality of analog MFCs. Additionally, analog MFCs cannot be reconfigured easily to accommodate changing flow or gas conditions, as required by process engineers in the fab.

To avoid the limitations of analog MFCs, digital MFCs are becoming increasingly visible in the semiconductor industry. They enjoy several advantages over analog MFCs, including improved accuracy over a wide range of setpoints (generally ±1% accuracy over a setpoint range of 25–100%) and ease of calibration and configuration in multigas/multiflow applications at end-user sites. By means of extensive modeling and gas characterization, MFC manufacturers have been able to develop user-configurable units that permit end-users to program them for virtually any gas or flow rate within a specified boundary.¹ That flexibility permits manufacturers to reduce their inventories of MFCs, eliminates recalibration costs arising from configuration changes, and minimizes or eliminates cancellation costs associated with configuration changes.

Digital MFCs have become increasingly accurate because they calibrate and store data at multiple points throughout their operating range. Their accuracy specification, based on percent of setpoint as opposed to percent of full scale employed by analog MFCs, leads to improved linearity and accuracy specifications.²

This article discusses the results of an evaluation of several digital MFCs from different manufacturers. Two digital MFCs (a 500–std cm³/min N₂ unit and a 10–std L/min N₂ unit) from each manufacturer were tested for particles, moisture drydown time, flow accuracy, and multiflow and multigas capability. While particle performance and moisture drydown characteristics are not directly related to the digital capabilities of MFCs, they are important for understanding how far MFCs have advanced technologically.

Test Protocol and Results

Particle Characteristics. Each MFC was tested for particles using a Micro-LPC-HS laser particle counter (LPC) from Particle Measuring Systems (Boulder, CO). The range of particle sizes tested was 0.05–1.0 µm. The LPC has a sampling rate of 2.8 std L/min and operates at atmospheric pressure. A pressure diffuser was used to reduce the upstream pressure and flow to the LPC. Particle tests were performed in both static and dynamic modes. Static tests were performed for 30 minutes, during which time the MFC remained fully open. Dynamic tests were performed for 10 minutes, during which time the MFC valve was cycled for 10 seconds every minute. In other words, with the MFC at 100% setpoint, the unit's power was turned off for 10 seconds and then turned back on again.

Particle test data show that nearly all of the MFCs under investigation appeared to be clean during 30 minutes of static testing in N₂. However, after undergoing 10 minutes of dynamic testing, a few MFCs exhibited moderate particle shedding. Particle shedding was attributed either to the internal design of the gas-wetted areas of the MFCs or to contamination arising during assembly or packaging. In addition, the MFCs used in this evaluation were test devices; they were not designed to be used in volume production or to be installed in gas boxes.

Moisture Drydown Characteristics. Moisture drydown tests were performed using a UG-240A atmospheric pressure ionization mass spectrometer (APIMS) from Hitachi Instruments (Tokyo). The tests were performed on 10-std L/min MFCs only, since the APIMS requires 1.2 std L/min of flow for testing. The MFCs were purged with nitrogen for 12–48 hours until their initial moisture level dropped below 1 ppb. To prevent compromising their sensitive electronic circuitry, the MFCs were not baked to remove hydrocarbon contaminants. The units were challenged with 200 ppb of moisture and equilibrated for 20 minutes. Drydown was monitored by switching to zero gas (200-ppt-purity nitrogen), which generated data indicating moisture drydown performance after the device had been subjected to a controlled moisture spike. The tests were performed according to SEMI Standard F58-1000.

Figure 1: Moisture drydown characteristics for different MFCs.

The MFCs' moisture drydown performance was evaluated using APIMS after a 200-ppb moisture challenge lasting 20 minutes. The results of that test, shown in Figure 1, indicate that the different MFCs exhibited markedly different moisture drydown capabilities. The time required for the devices to reach a moisture level of 10 ppb ranged from 2 minutes in the best case to 8 minutes in the worst. The time required for the MFCs to reach the 1-ppb moisture level ranged from 62 minutes in the best case to 124 minutes in the worst. A few MFCs did not reach the 1-ppb level even after 5 hours of testing. However, the SEMI drydown specification for most gas panels is 20 ppb of moisture after 6 hours in dry nitrogen at at flow rate of 1 std L/min. Therefore, it was expected that most of the MFCs tested would meet current moisture drydown specifications.

Flow Verification. All flow-verification tests were performed in nitrogen using a Cal-Bench automated primary gas-flow calibration system from Sierra Instruments (Monterey, CA). The system's flow range is 1 std cm³/min to 50 std L/min and is based on primary measurements of length and time as stipulated by the National Institute of Standards and Technology. Tests for digital accuracy were conducted at 25, 50, 75, and 100% setpoints.

Figure 2: Flow-accuracy plots for 10-std L/min N2 MFCs tested in as-received digital mode.

Of the low-flow (500–std cm³/min N₂) MFCs that were tested as received, nearly all met the accuracy specification in the range of 25–100%, although some of the MFCs were marginally outside specification at the 25% setpoint.

Figure 2 shows the percentage error per setpoint for different high-flow (10-std L/min N₂) MFCs tested at 25, 50, 75, and 100% setpoints in as-received condition in digital mode. The results for those MFCs were much more scattered than the results for the low-flow devices. Less than half of the 10-std L/min devices met accuracy specifications at all setpoints. The remaining devices did not meet accuracy specifications even at 50% setpoint. Those results clearly indicate a need for improvement, especially at low setpoints.

MFC Multiflow and Multigas Features

Unlike analog devices, digital MFCs can be configured to accommodate multiple flow rates (generally down to 30–60% of the original full-scale flow rate, depending on the type of MFC). The multigas and multiflow capabilities of MFCs enable IC manufacturers to significantly reduce inventories and to increase MFC mean time to repair or replace rates.

When the gas-flow range of an analog MFC is changed, the device must be recalibrated. Recalibration involves sending the device to the manufacturer for resizing and testing. The resizing process results in increased turnaround times and added production costs. In contrast, because digital MFCs have multiflow capability, their flow range can be changed simply by changing their flow limits, an operation that can be accomplished during system testing or in the fab. In addition, fab personnel can use one MFC with a variety of gases simply by reconfiguring the device. Most digital MFCs can be programmed to accommodate a range of gases.

MFC Multiflow Capabilities. To evaluate the MFCs' multiflow capabilities, the investigators reprogrammed both low-flow (500–std cm³/min N₂) and high-flow (10–std L/min N₂) devices using the software provided by the MFC vendors to accommodate the lowest flow rates. The reprogrammed flow rates of the low-flow MFCs ranged from 280 to 325 std cm³/min. For example, one 500–std cm³/min MFC was reprogrammed to a new full-scale flow rate of 280 std cm³/min N₂.

Figure 3: Flow-accuracy plots from multiflow tests of low-flow MFCs in digital mode.

Figure 3 plots the percentage error versus setpoint for the low-flow MFCs. The figure indicates that the percentage error was largest at the lower setpoints for most MFCs and decreased to within acceptable limits at the higher ones. Less than half of the MFCs tested met accuracy specifications at all setpoints. A few MFCs were out of specification at 25 and 50% setpoints. Most high-flow MFCs, on the other hand, did not meet specifications at any setpoints, although some of them were only marginally off.

Figure 4: Flow-accuracy plots from multigas tests of low-flow MFCs in digital mode.

Multigas Capabilities. MFCs with multigas capability can be configured for any gas listed in SEMI's gas tables (SEMI Standard E52-0302). To test that capability, the investigators reprogrammed both low- and high-flow devices for argon gas. Figure 4 presents an error plot indicating that all of the low-flow MFCs except one were within accuracy specifications at all setpoints.

Figure 5 shows an error plot for high-flow MFCs that were tested after being reprogrammed for argon gas. The results were consistent with the results from the as-received and multiflow tests. Only one MFC met accuracy specifications at all setpoints, while the others were outside specifications at nearly all setpoints. That test demonstrated the performance limitations of the MFCs under investigation.

Figure 5: Flow-accuracy plots from multigas tests of high-flow MFCs in digital mode.

Conclusion

Digital MFCs have significant advantages over analog MFCs. They are more accurate and easier to calibrate and reconfigure in multiflow/multigas applications at end-user sites. They can also reduce fab inventories. These features were demonstrably evident in the studies presented here. All of the digital devices tested were more accurate than their analog counterparts. However, many of the devices did not meet accuracy specifications at setpoints ranging from 25 to 100% and in multiflow/multigas conditions. Those limitations may be overcome by improving the modeling of gas characteristics and control algorithms. While digital MFCs should be improved in some areas, they are superior to analog MFCs. They can potentially reduce costs and offer flexibility to meet the configuration requirements of delivering gas to the process tool.

Acknowledgments

The authors would like to thank Dave Celli and Chris Burkhart from Novellus Systems for supporting the study described in this article. They would also like to acknowledge the MFC suppliers Advanced Energy Industries, Brooks Instrument, Mykrolis, MKS Instruments, Unit (now Celerity), and Horiba STEC for their participation in this program.

References

1. CF Drexel, "Digital Mass Flow Controllers Come of Age," Solid State Technology 39, no. 11 (1996): 99–106.

2. P Rudent, "Calibration Method and New Developments for High-Performance Mass Flow Controllers," Semiconductor Fabtech (8th ed.): 167–170.

---

Mohamed Saleem, PhD, is a senior technology development engineer at Ultra Clean Technology (Menlo Park, CA). An active participant in the standards activities of gas-panel, corrosion, surface-analysis, and stainless-steel task forces of SEMI, Saleem has published several papers in the field of materials science and has presented many papers at major conferences. He received an MS in chemical engineering from Tufts University in Medford, MA, and a PhD in materials science and engineering from the University of Florida in Gainesville. (Saleem can be reached at 650/323-4100 or msaleem@uct.com.)

Sowmya Krishnan, PhD, is CTO of Ultra Clean Technology, where she oversees the technology and product development areas for semiconductor gas- and liquid-delivery systems. She has chaired sessions on contamination-free manufacturing and gas delivery at Semicon West, Semicon Southwest, and CleanRooms West. She has also chaired the standards task force at SEMI. She has authored several technical papers and editorials in the fields of microcontamination and semiconductor processing. She received MS and PhD degrees in chemical engineering from Clarkson University in Potsdam, NY. (Krishnan can be reached at 650/323-4100 or skrishnan@uct.com.)