Optimizing Remote Plasma Cleans Through Real-Time Process Monitoring

Nitrogen trifluoride (NF₃) is a stable and non-flammable gas that is commonly used in a chamber cleaning process in the manufacture of semiconductors. We conducted research at Infineon Technologies AG's 300 mm wafer manufacturing facility to optimize the NF₃-based process used to clean Novellus Altus Concept 3 CVD chambers following tungsten deposition. Our goal was to reduce gas costs and/or shorten the time required to clean the CVD chamber.

We monitored the chamber clean process downstream of the process pump using Fourier transform infrared (FTIR) spectroscopy and quadrupole mass spectrometry (QMS). These measurements determined process emissions and clean times for each chamber clean. Our strategy was to construct response surfaces for perfluorocompound (PFC) emissions and clean times as a function of NF₃ gas flow, Ar/NF₃ ratio, pressure and temperature. By analyzing these response surfaces, we identified processes that provided substantial reductions in NF₃ usage while maintaining the same clean time. Alternatively, the chamber was cleaned faster by using the same amount of NF₃.

By maintaining the same temperature as the baseline clean process, we demonstrated that NF₃ usage could be reduced by 22% while maintaining the same clean time. PFC (environmental) emissions were also reduced by 28-43%. By increasing the pedestal temperature to 460°C during the chamber clean, both faster cleans and lower NF₃ usage were possible. The etch rate increased by 38% in the higher-temperature clean compared with the baseline chamber clean, and PFC emissions were reduced by 13%.

The optimized chamber cleans removed the entire tungsten residue from the CVD chamber as indicated by measurements of the volumetric WF₆ emissions from the chamber.

The ISO tube trailer fill area at Air Products’ electronic specialty gas manufacturing facility in Hometown, Pa.

Experimental methodology

Tungsten films were deposited in a 300 mm Novellus Altus Concept 3 CVD chamber. To reduce the time needed for the design of experiments (DOE), the chamber was cleaned after a tungsten accumulation of 10 µm. A remote NF₃ plasma was then used to clean the chamber. The baseline process is:

NF₃ (2000 sccm), argon (3000 sccm), 3.0 Torr, 415°C, end point+10% optical end point (OE)

Optimized processes identified from the DOE were subsequently evaluated following a 50 µm tungsten accumulation.

Emissions measurements were made downstream of the process pump using extractive FTIR spectroscopy and QMS. The process effluents are diluted by the N₂ pump purge, since the process was sampled at the exhaust of the process pump. Sample gas was pumped through the FTIR cell and then across the QMS inlet before being returned to a toxic/corrosive exhaust. Sample pressure (~750 Torr) was controlled by a metering valve and measured with a capacitance manometer.

FTIR measurements of NF₃ and WF₆

NF₃ and WF₆ concentrations were determined using an online multigas FTIR spectrometer. This allows for real-time, quantitative emissions monitoring. The temperature and pressure of the gas cell was controlled at 150°C and 1.0 atmosphere. Reported concentrations are corrected for temperature and pressure during the measurement.

The absorbance region used to determine the NF₃ concentrations is 870-930/cm, and the reference spectrum used in the analytical method has a concentration of 100 ppm.m. The FTIR instrument was calibrated at Infineon for NF₃ using a 1% NF₃ gas standard. Calibration curves were measured using dynamic dilution methods to generate concentrations <10,000 ppm. The NF₃ calibration is linear over the range 0-10,000 ppm.

The absorbance region used for WF₆ measurements was 690-729/cm. The concentrations of the reference spectra used in the analytical method are 227, 465, 899, 1324, 1755, 2640 and 3786 ppm.m.

The chamber cleans were monitored with a differentially pumped UTI Qualitrace QMS having a 300 amu mass filter. The QMS was primarily used to monitor F₂ emissions and verify NF₃ emissions.

The QMS instrument was calibrated for NF₃ using a 1% gas standard. Calibration curves were measured using dynamic dilution methods to generate concentrations <10,000 ppm. The NF₃ calibration is linear over the range 0-10,000 ppm. The QMS instrument was calibrated daily to account for any changes in sensitivity.

Since the QMS was not calibrated for F₂, volumetric F₂ emissions (i.e., scc) cannot be determined. The F₂ profile is, however, a good end-point detector and can be used to measure clean times.

The purpose of the chamber clean is to volatilize the tungsten residue as WF₆ (Fig. 1 ). Cumulative WF₆ emissions are therefore a good end-point monitor. WF₆ concentrations were measured by FTIR. FTIR end point was defined as the time to remove 95% of the WF₆ emitted during the chamber clean (i.e., total WF₆ emissions were calculated, and the end point is when 95% of this value was achieved).

1. The concentration profile, measured by FTIR, during the chamber clean.

Clean times can also be determined using the QMS. Fluorine F₂ emissions are monitored during the chamber clean at 38 amu, and the clean time obtained from the concentration profile (Fig. 2 ). The start of the chamber clean is indicated by an NF₃ intrusion. While the chamber is being cleaned, little F₂ is emitted since fluorine is consumed by tungsten etching. As the CVD residue clears, however, the F₂ concentration increases. End point is defined as the time when the F₂ concentration reaches 95% of its overetch value.

2. The concentration profile, measured by QMS, during the chamber clean.

Two methods were used to measure the clean time: FTIR (WF₆) and QMS (F₂). Although these methods have different end-point criteria, and hence slightly different clean times, the consistency between these independent measurements is excellent. Clean time values could be in absolute agreement if small changes were made to each of the end-point criteria.

Because measurements are made downstream of the process pump, process byproducts are diluted by the N₂ pump purge. It is important to know the total gas flow to determine volumetric emissions (i.e., scc). The pump purge was determined by flowing NF₃ (2000, 1500, 1000 and 500 sccm) from the gas panel and measuring its concentration in the pump effluent. The dilution factor (50,340 sccm) is obtained from the slope of NF₃ concentration vs. NF₃ gas flow rate.

Byproducts of the NF₃ clean are non-utilized NF₃, WF₆, F₂ and HF. The FTIR (Fig. 1) and QMS (Fig. 2 ) profiles show the concentration of NF₃, WF₆, HF and F₂ during the baseline tungsten chamber clean. Both QMS and FTIR can be used for NF₃ measurements. There is excellent agreement between these independent methods.

Observing NF₃ in the effluent indicates the start of the chamber clean. Once the power is applied, there is a sharp increase in the WF₆ concentration as the chamber is cleaned. At the end point, the WF₆ concentration returns to baseline levels. F₂ is also generated as a byproduct of the NF₃ plasma etch. At end point, F₂ concentration achieves its overetch value.

Volumetric emissions are obtained by integrating under the concentration profiles (e.g., Figs. 1 and 2 ) and multiplying by the N₂ pump purge. Volumetric NF₃ and WF₆ emissions are calculated from the FTIR concentration measurements. Process emissions on a grams/wafer basis are obtained from the volumetric emissions using the molecular weight of NF₃ (71 amu). PFC emissions are reported as million metric tons carbon equivalent (MMTCE):

In this calculation, Q_NF3 (kg) is the amount of NF₃ emitted from the process and GWP₁₀₀ is the global warming potential (with a 100-year time horizon). There are no other PFC byproducts generated during the NF₃ chamber clean.

The baseline chamber clean is:

NF₃ (2000 sccm), argon (3000 sccm), 3.0 Torr, 415°C, end point+10% OE

3. Infrared absorbance spectrum measured during the chamber clean.

The FTIR absorbance spectrum and 100 amu mass spectrum collected during the baseline chamber clean is shown in Figures 3 and 4 , respectively. All FTIR spectral features are assigned to NF₃ and WF₆ (atmospheric CO₂ and H₂O absorbances are also observed). The byproducts NF₃ and F₂ are observed in the QMS spectrum (Fig. 4 ), as well as N₂, argon, H₂O and CO₂ from the N₂ pump purge. Concentration profiles (10 µm accumulation) are shown in Figures 1 (FTIR) and 2 (QMS). Process emissions and clean times for the baseline chamber clean following an accumulation of 10 µm (102,690 Å) and 53 µm (531,615 Å) are summarized in Table 2 . The amount of WF₆ emitted per unit accumulation (2959 ±43 scc/µm) and etch rate (3184 ±95 Å/min) are the same for these depositions.

4. Mass spectrum measured during the chamber clean.

Process optimization

The PECVD chamber was cleaned following a tungsten accumulation of 10 µm. Tungsten was accumulated while processing production wafers (400 nm W CVD process). The clean process was optimized using a central composite DOE methodology summarized in Table 3 . The NF₃ flow rate (1300-2300 sccm), pressure (2.0-4.0 Torr), temperature (415-475°C) and Ar/NF₃ ratio (0.5-1.5) were varied while measuring the process emissions and clean time. Plasma power is fixed. The NF₃ /Ar plasmas were stable under all of these conditions. Throughout the DOE, no problems were encountered with plasma stability.

The Novellus software logs the tungsten accumulation between each chamber clean. A cumulative etch rate can therefore be calculated from the measured clean times:

From these experimental results, response surfaces were constructed using a neural network analysis. There was excellent agreement between the experimental measurements and the calculated response surfaces (Tables 4 and 5 ). All but one (94%) of the measured etch rates are within 5% of those predicted by the response surface (a single observation represents 6% of the values).

Response surfaces for the tungsten etch rate are shown in Figures 5 and 6 . These response surfaces show how the etch rate varies as a function of NF₃ flow rate, pressure, temperature and Ar/NF₃ ratio. Each response surface shows the sensitivity to two factors (e.g., NF₃ flow rate and pressure); the other factors (e.g., temperature and Ar/NF₃ ratio) have a fixed value. Higher etch rates, and hence faster cleans, are favored by higher NF₃ flow rates, lower Ar/NF₃ ratios, lower pressures and higher temperatures.

5. Response surface showing how the etch rate varies as a function of pressure and Ar/NF3 ratio (the temperature and NF3 flow rate are fixed).

Maintaining the same temperature as the baseline clean (415°C), the clean time and/or NF₃ usage can be reduced by lowering the Ar/NF₃ ratio and pressure. If higher temperatures are possible, considerable reductions in both the clean time and NF₃ usage are possible. However, although faster cleans are favored by higher temperatures, 90% of the benefit is achieved by 460°C.

6. Response surface showing how the etch rate varies as a function of temperature and NF3 flow rate (the pressure and Ar/NF3 ratio are fixed).

There was concern that some of the DOE processes do not remove the entire CVD residue (i.e., incomplete clean). The WF₆ emissions were used as a probe of clean effectiveness. WF₆ emissions would be lower for incomplete chamber cleans. WF₆ emissions for each of the DOE and standard chamber cleans were compared. Total WF₆ emissions are normalized to the tungsten accumulation (i.e., scc/µm). The amount of WF₆ during the DOE cleans (3073 scc/µm) is the same as the baseline process (3012 scc/µm), indicating that these conditions effectively remove all of the residue from the CVD chamber.

Response surfaces for etch rate (Figs. 5 and 6) allow a number of optimized processes to be recommended (Table 1 ). The processes LOW1 and LOW2 maintain a temperature of 415°C (same as baseline clean). NF₃ usage for LOW1 and LOW2 is predicted to be 22% lower while maintaining the same clean time (etch rate 3370 Å/min). LOW2 has a less aggressive reduction in the Ar/NF₃ ratio (1.0) compared with LOW1 (0.5). Although there were no problems with plasma stability throughout the DOE, the higher Ar/NF₃ ratio should be more robust. The chamber clean FAST allows a higher temperature (460°C) during the clean. By increasing the temperature to 460°C, substantial reductions in both NF₃ usage (-20%) and clean times (-27%) are predicted.

Since the response surface uses an empirical model to predict etch rates and clean times, it is necessary to verify the reductions for these optimized processes. Actual measurements were made for the recommended chamber clean processes following a 50 µm accumulation. When used in production, the Novellus Altus chambers are typically cleaned following an accumulation of 50 µm. Volumetric emissions and etch rates for the optimized chamber clean process are summarized in Table 6 . All emissions include a 10% overetch that is added to the clean time.

There is excellent agreement between the predictions of the response surfaces and the measured etch rates in Table 6 . The etch rate of the baseline clean was predicted to be 3200 Å/min. The etch rate measured for the baseline clean is 3279 Å/min. Similarly, the predicted and measured etch rates for the recommended clean FAST are 4500 and 4512 Å/min, respectively. Response surfaces cannot predict etch rates for the processes LOW1 and LOW2 because they are outside the range of the DOE parameters: pressure (1 Torr) and Ar/NF₃ ratio (0.4).

7. Volumetric WF6 emissions for the optimized chamber cleans (FAST, LOW1 and LOW2) are the same as the baseline clean (STD), indicating that they remove an equivalent amount of residue and effectively clean the chamber.

While clean times are the same (LOW1 and LOW2) or faster (FAST) as the baseline process (STD), the optimized processes must also effectively clean the CVD chamber (i.e., complete clean). Volumetric WF₆ emissions were used as a monitor of completeness. If the entire tungsten residue is not removed during the clean process, then volumetric WF₆ emissions will be lower than for the baseline chamber clean process. Volumetric WF₆ emissions for the Infineon and optimized processes are shown in Figure 7 following a 50 µm accumulation. The WF₆ emissions are normalized to the tungsten accumulation (i.e., scc/µm). The optimized WF₆ recipes produce the same amount of WF₆, and hence remove the same amount of tungsten residues. This indicates that the optimized processes effectively remove the entire tungsten residue from the CVD chamber.