Managing Fluorinated Byproducts of CVD Chamber Cleans

Semiconductor fabs built prior to an introduction of NF₃-based CVD chamber cleans may be inadequately equipped to manage the increased load of fluorine, typically handled through a combination of exhaust scrubbing and wastewater treatment. This study examines the effects of F₂ emissions on various duct materials, wastewater treatment capacity and air emissions.

We assessed the effects on fab facility operations by quantifying and testing the F₂ and chemical byproducts emitted using mass spectrometry, Fourier transform infrared (FTIR) spectroscopy, and a fluorine-specific chemiluminescent sensor. We used a specialized duct composed of fiberglass-reinforced plastic (FRP), which was installed between the main header exhaust lateral and the output of two TEOS-based CVD chambers. The chambers were also equipped with remote plasma devices. In this way, the tool exhaust was segregated from other CVD tools or processes.

A damper allowed varied amounts of ambient air to be introduced to the duct to determine the effect on F₂ concentration. We studied reaction dynamics by installing numerous sampling ports along the exhaust stream duct at different distances from the injection point. A study of various polymeric duct materials determined the effects of long-term exposure to F₂ and immersion in HF.

The International Technology Roadmap for Semiconductors calls for proactive reduction of emissions that may cause global climate change. International SEMATECH participants have signed a voluntary agreement with the U.S. Environmental Protection Agency to work toward reducing PFC emissions. In April 1999, the World Semiconductor Council announced an international goal to reduce PFC emissions by 10% by 2010, relative to the 1995 baseline.

As a result, semiconductor manufacturers have been actively pursuing strategies to reduce PFC emissions from etch and CVD chamber cleaning processes, which account for up to 90% of emissions from a modern 200 mm fab. The industry developed remote plasma clean technology using NF₃ and argon, which has demonstrated >99% NF₃ destruction efficiency, virtually eliminating PFC emissions from chamber cleaning.¹

Another important development was that of in situ NF₃ -based clean processes.^2,3 However, one drawback is the increased generation of F₂. Current in situ C₂F₆ chamber cleans generate some F₂ byproduct, but NF₃-based cleans can emit up to 6× as much F₂ because of high NF₃ flows and greater destruction efficiencies. A major concern is the eventual fate of this large amount of F₂ in an exhaust system and its effect on the emission treatment infrastructure.

The remote clean device is a compact, lid-mounted, point-of-use (POU) fluorine generator that uses a mixture of NF₃ and argon (~1:2). The unit is mounted on the chamber lid of either a resistively heated or lamp-heated chamber. The device converts 95-99% of the NF₃ (0.1-4.0 std. L/min total gas flow, 1-8 Torr pressure) to molecular and atomic fluorine by coupling 400 kHz radio frequency to a low-field torroidal plasma.

Because the plasma is generated upstream of the chamber, the process kit is not bombarded with ions, increasing kit lifetime. In addition, atomic and molecular fluorine generated in the upstream plasma has the effect of cleaning remote areas of the chamber and foreline, resulting in slower accumulation of chamber deposits. Mean time between cleans of 27,000 wafers has been reported.⁴ NF₃'s higher ionization efficiency relative to C₂F₆ means clean times can also be reduced. Throughput improvements of 30% for a resistively heated chamber^5,6,7 and up to 65% for a lamp-heated chamber have been reported.⁸ To evaluate the fluorine emissions from NF₃ chamber clean processes, we installed a 4 in. (OD) FRP duct between the main header exhaust duct and the output of two TEOS-based oxide CVD chambers, each equipped with remote clean units. The oxide deposition chambers were each pumped by a 96 m³/hr mechanical pump. A mass flow controller controlled the nitrogen purge and ballast to 45 L/min to maintain dilution. A 5 cm (OD) stainless-steel (SS) duct conveyed the tool exhaust output to the FRP duct. Figure 1 shows the relationship between the source chambers, SS duct and FRP ductwork.

1. Fluorine samples are taken along the FRP duct between the main header exhaust duct and the output of two TEOS CVD chambers.

2. View of the 30 ft FRP test duct and sampling ports alongside the main exhaust header.

Airflow changes also affect the H₂O to F₂ ratio, thus changing the thermodynamic characteristics of the F₂ to HF reaction mechanism. Raising the airflow increases the availability of water vapor for reaction, but also decreases reaction time. Ambient air consisted of 45% RH, or ~6800 ppmv water vapor. Although it is estimated that the production of F₂ is limited stoichiometrically to 1.5× the NF₃ input, in practice, the fluorine emitted is considerably less — on the order of 1-1.1×.

We placed gas sampling ports along the central axis of the FRP duct and inserted a Teflon extraction line probe to sample the center of the duct. We used extractive FTIR for real-time detection of chamber clean byproducts such as HF, CF₄, COF₂, SiF₄ and NF₃. An extractive fluorine chemical sensor (FCS), developed by URS Corp., provided fast response and sensitive real-time detection of F₂. An extractive residual gas analyzer (RGA) provided a fluorine concentration check to FCS measurements. We mapped the fluorine reaction during an extended chamber clean with constant gas flow recipes (700 sccm NF₃, 1400 sccm argon) over a 900 sec period, and sampling at various locations.

The FCS was designed to monitor fluorine for the semiconductor industry. It allows for continuous, real-time detection of F₂ in gaseous streams,⁹ exhibits high sensitivity, wide dynamic range and rapid response, yet avoids cross contamination of other reactive species. The fluorine detection range is 2% to ~20 ppb.

The FCS extractive configuration allows a slipstream to be drawn through a small sample cell at moderate flows (~5 L/min). The cell body is made of nickel-plated aluminum to minimize interaction of fluorine (or HF) with cell walls. It is also designed such that the gas can uniformly interact with a proprietary organic substrate deposited on a window surface. The window's back end is near a head-on photomultiplier tube (PMT), a cascading electronic device that provides a sensitive measurement of visible light emitted from the chemiluminescent interaction of F₂ with the substrate.

A picoammeter detects the current output from the PMT, which is interfaced to a laptop computer with fast data acquisition software. The chemical interaction mechanism has been proven to be reproducible and easily characterized so that a mathematical relationship between detector output and F₂ concentration is established.

The FCS was calibrated at constant flow across a concentration range of 1.7 ppm to 1.9% using certified F₂ gas standards. We plotted photomultiplier response vs. time for a given concentration. The third-order polynomial expressed in Figure 3 indicates the PMT current levels plotted against the known F₂ concentration produce a calibration function that mathematically defines the detector's response. A fourth-order polynomial function was used below 50 ppm to effect a better fit.

3. Current FCS levels, taken over the concentration range of 1.7 ppm to 1.9%, demonstrate a third-order polynomial relationship between sensor response and F2 concentration.

4. Photomultiplier sensor results indicate F2 concentration at four duct positions (0.91-8.63 m from the F2 injection point) and four damper settings (12.5-75% open). The steep rise and fall are indicative of rapid sensor response.

Fluorine concentration as a function of damper position and sampling location (Fig. 5a and b) shows that, as the damper is opened, the reaction time at a sampling location is reduced due to the increased duct velocity, but the increased airflow reduces the fluorine concentration. However, the increased water composition within the duct enhances the reaction to form HF. For the same damper setting, fluorine concentration is also reduced as the sampling location is moved downstream because of increased reaction time and contact time with duct surfaces. This time- and position-dependent reaction mechanism accounts for the high fluorine DRE.

5. Increased airflow reduces the F2 concentration, as does the greater measurement distance downstream of the exhaust inlet. Excellent conversion is achieved with >25% damper open (a). DRE is 98-99% with the exception of the 12.5% open damper position, where DRE is 91% (b).

We conducted a material compatibility test to measure how well various exhaust duct materials withstand prolonged exposure to F₂ and HF. Test coupons (2-3 in.²) were fashioned out of 10 materials that were threaded, weighed and positioned on a section of fiberglass all-thread rod perpendicular to the exhaust flow. Because of the frequent use of the remote clean systems associated with the test duct, the coupons experienced regular exposure to F₂ for six months.

We evaluated the coupons based on weight, appearance (pitting, deformation) and integrity. There was no evidence of degradation as a result of exposure to F₂, most likely because of low F₂ levels. All coupons were subsequently immersed in 10% HF for 24 hr and re-weighed. Only the FRP and vinyl ester resin coupons underwent mass change from HF exposure. SEM analysis indicated that significant decomposition of the phenolic resin in the FRP occurred, exposing the underlying fiberglass cloth. The site of HF erosion in the vinyl ester resin was the fiberglass interface that was exposed when the coupon was drilled for installation on the threaded rod, and not the resin itself.

We estimated the annual generation of F₂ that requires air/water treatment based on a consumption level of 25,000 lb annually. Fabs that use more or less NF₃ can scale the results.

It is important to determine the worst-case impact to air and wastewater. Reactive fluoride byproducts include HF, F₂, silicon tetrafluoride (SiF₄) and carbonyl fluoride (COF₂). The dominant species by far is F₂ (85-95%), followed by SiF₄ (5-10%) and HF (1-2%). Given that the fab can effectively transfer the increased fluoride load from the gas phase to the liquid phase (HF), the impact to wastewater treatment becomes a concern.

The impact of the additional load from NF₃ -based chamber cleans depends heavily on the site's discharge limit. Typical municipal fluoride discharge limits range from 2 to 65 mg/L, but are generally low. Because fluorides are ubiquitous in silicon processing, most sites have some fluoride level even without the added load from NF₃ chamber cleans.

Figure 6 describes the relationship between air and wastewater impacts at various control efficiencies. As efficiency approaches 100%, the fluoride is almost completely transferred to wastewater. Although typical fabs have some form of end-of-pipe scrubbers for acid gas emissions control, performance varies widely. Scrubber efficiency is 40-99% depending on scrubber design, make-up water quality and operations and maintenance practices. Scrubber performance and likely fluorine emissions levels will define the site's treatment strategy.

6. Fluorinated waste that is not managed by the scrubbers results in higher concentration of HF in the wastewater. Based on 25,000 lb/yr NF3 usage, 100% scrubber efficiency and 1000 gal/min wastewater flow, the time-averaged increase in fluoride load from NF3 cleans is 4.6 mg/L. The worst case is 46 mg/L. Practically, load fluctuates around ~2× the time-averaged level.

Another consideration is the relationship between air pollution control and wastewater treatment. The design and operation of the air pollution control system can have a pronounced effect on the concentration and volume of fluoride-containing wastewater that requires subsequent treatment. Specifically, a scrubber designed and operated with continuous blowdown will generally generate a lower-concentration, higher-volume fluoride-bearing waste stream than scrubbers designed and operated with intermittent blowdown. Another factor that controls the scrubbers' operational efficiency is pH. Generally, higher-pH water is more effective at removing airborne acids.

A POU or localized air pollution control strategy typically has the effect of averaging the fluoride impact over a larger volume of wastewater. An end-of-pipe approach can maximize the fluoride concentration in the wastewater and minimize the volume. Higher concentrations and lower volumes will usually make fluoride-bearing wastewater handling and treatment easier and more cost-effective. The site's air pollution control strategy will impact its fluoride-in-wastewater handling and treatment approach. This relationship is important if unexpected capital expenditures are to be avoided. Commonly employed handling/treatment approaches include:

• Fluoride concentrates collection and metering to the wastewater stream (to avoid spike loading).
• Concentrates collection and on-site treatment via precipitation.
• Concentrates collection and off-site disposal, and in some cases industrial wastewater treatment.

The existing wastewater and/or air pollution control approach and infrastructure may not be well suited to the increased F₂ load from NF₃ -based chamber cleans. Thorough evaluation of the site's air pollution control strategy, existing wastewater handling/treatment infrastructure and capacity, and F₂ discharge limits should be completed prior to proliferation of NF₃ -based chamber cleans.

Remote plasma clean technology represents a new approach to PFC emissions reduction. As NF₃ is substituted for C₂ F₆, which produces considerable amounts of the greenhouse gas CF₄, fabs must be prepared for the associated increase in F₂ emissions using NF₃. Special consideration should be given to selection of exhaust duct materials and design and operation of wastewater treatment facilities.

A new solid-state fluorine chemical sensor was used to conduct duct measurements of F₂ concentration from remote NF₃ chamber cleans. Results indicate 98-99% conversion of F₂ to HF in the FRP duct. Primary factors affecting F₂ to HF conversion are the water-to-fluorine ratio and downstream distance. A material compatibility study indicated that the phenolic resin in FRP substantially degraded upon exposure to HF. Before proliferating NF₃ -based chamber cleans, users should fully evaluate expected changes in fluoride ion load in the wastewater treatment facility and air pollution control strategy.