Acid Gas, Acid Aerosol and Chlorine Emissions from Trichlorosilane Burning Processes

This study was set out to investigate the emission characteristics of HCl (in both particle (HClp) and gaseous (HClg) forms), and Cl2 during the trichlorosilane (TCS) burning process under various relative humidity conditions (RH; range = 55%–90%) which might exist at its storage area. All experiments were conducted in a test chamber. We found that HClp was consistently as the most dominant contaminant (= 1.30 × 10–1.46 × 10 mg/m), followed by the HClg (= 9.03 × 10– 11.4 × 10 mg/m) and Cl2 (= 1.91 × 10–2.18 × 10), emitted from the TCS burning process for the all selected RH conditions. The particle sizes of HClp fell to the range of the accumulation mode (MMADs = 0.808–1.04 m; GSDs = 2.13–3.50). Fractions of emitted HClp reaching to the alveolar region (= 85.8–88.8%) were much higher than that of the tracheobronchial region (= 6.53–8.80%) and head region (= 4.67–5.40%). It is concluded that more ill-health effects on the deep lung region can be expected than other regions as workers exposed to the contaminants emitted from TCS burning processes.


INTRODUCTION
According to statistical data, the annual world production of trichlorosilane (TCS) is ~350,000-380,000 tons and ~60-65% of TCS is consumed by the semiconductor industry as an alternative silicon source gas (Howe-Grant, 1997;Williams, 2000). TCS is also used as the basic ingredient in the production of solar cells, optical fibers, and the manufacturing of PV grade polysilicon and silane gas. For the above industries, many study have been conducted to address their fugitive emissions at process areas Shih et al., 2010) or pollutant removal efficiencies of various air pollution control devices Shiue et al., 2011). Though many studies have been conducted to address the emissions arising from combustion related activities (Choosong et al., 2010;Li et al., 2010;Ning et al., 2010), very limited have been conducted to characterize the emissions of raw materials during the accidental fire or explosion cases.
TCS is known with a boiling point of 31.85°C, flash point of -28°C, and flammability range of 7%-82% (Laurence, 1990). Apparently, its major hazard might simply lie on its flammability. Therefore, TCS emergency response guidelines proposed by different organizations, such as the Silicones Environmental, Health and Safety Council of North America (SEHSC), Centre Européen des Silicones (CES), Silicone Industry Association of Japan (SIAJ), concern still mostly on its fire and explosion effects, rather on the impact caused by its resultant combustion by-products (e.g., hydrogen chloride (HCl) and chlorine (Cl 2 )) (Higgins et al., 1999;CES, 2003). However, the impact of its combustion by-products has never been investigated.
In principle, the generation of HCl from the TCS burning process can be described as (Mores, 1984): 5SiHCl 3 + 6O 2 5SiO 2 + HCl + 7Cl 2 + 2H 2 O (1) In addition, part of emitted Cl 2 could further react with H 2 O to form HCl: From the Eq. (1), the generated HCl could be presented simultaneously in both gaseous (i.e., HCl g ) and aerosol (i.e., HCl p ) forms. Here, HCl p might be formed directly via the condensation of HCl g on particles (such as SiO 2 ) via the heterogeneous condensation process, or by directly dissolving HCl g into H 2 O coated on the surface of SiO 2 particle or water droplets. From Eq. (2), the generated HCl might directly dissolve water droplets to form HCl p .
Among these emitted contaminants, Cl 2 is known for its acute health effect on the respiratory system (Agabiti et al., 2001;Rabinowitz and Siegel, 2002;Uyan et al., 2009) and acute ischemic stroke (Kose et al., 2009). HCl g is an upper respiratory tract irritant because of its high water-solubility (Salem, 2005). HCl p , on the other hand, might penetrate into the deep lung of the repiratory tract and cause more serious ill-health effects depending on its particle size (ICRP, 1994). Therefore, simultaneously measuring Cl 2 , HCl g and HCl p are considered a better approach to characterize the release of toxic chemicals from the TCS burning process.
In this study, a chamber study was conducted to investigate the emission characteristics of Cl 2 , HCl g and HCl p during the TCS combustion process. In order to simulate the real situation for TCS storage in the indoor gas yard, four relative humidity (RH) conditions were used based on RH records provided by semiconductor industries in Taiwan. The results obtained from this research will provide information for the government, semiconductor and optoelectronic industries to generate appropriate strategies for TCS emergency response process. Fig. 1 shows the stainless steel 316 test chamber used in the present study. The chamber comprised three parts, Fig. 1. Schematic of the test chamber used in this study.

Sampling train
From air cleaning and conditioning system Honeycomb including an air inlet section (i.e., the lower pyramid part), a TCS burning section, and an air outlet section (i.e., the upper pyramid part). Before the supply air entering into the lower pyramid, it was treated by an air cleaning and conditioning system (Fig. 2). The system consisted of a compressor, a dryer, a surge tank, an activated charcoal, a high-efficiency particulate air (HEPA) filter, a home-made heating tank, and a humidifier (model FC-125, Perma Pure Inc., NJ, USA). The mass flow controller (MFC, Side-Trak® Model 840, Sierra Instruments Inc., CA, USA) was also installed in the system to ensure to provide a designated air flow with a preset humidity condition. The flat top of the air inlet section was installed with a honeycomb to provide a uniformly distributed laminar flow for the TCS burning section. This burning section was a 0.42 m 3 cube installed with one glass door, and three stainless steel side-walls. On the opposite wall of the glass door was installed with a TCS feeding tube (diameter = 0.64 cm) to transport a fixed amount of TCS (ACS grade, ~99.9% purity) from the TCS storage cylinder to the test pan (diameter =5.0 cm, height = 3.0 cm). In order to know the consumption rate of TCS, the test pan was placed on a digital balance (accuracy = ± 0.02 g, model SKY-600, Javeder Scale Co., Taiwan) to continuously send weight signals to the laptop per 1/30,000 second during each combustion test run. All air samples were collected at the converging part of the air outlet section.

Selected Test Conditions
Four RH conditions (= 55%, 65%, 80% and 90%) were selected for conducting TCS combustion experiments to simulate possible humidity conditions of the storage area year round. After being pretreated by the air cleaning and conditioning system, the RH conditions used in the present study were 57.3%, 65.5%, 79.8% and 89.8%, respectively (with relative standard deviations (RSDs) < 1.01%). The air flow feeding rate was specified at 0.042 m 3 /min by a mass flow controller in the present study to simulate the air change per hour (ACH = 6) commonly found in the indoor gas yard. The resultant chamber air velocity (0.0017 m/s) was also comparable to the indoor work environment. To assess the uniformity of air velocities occurred at the TCS burning section, four cross-sections were chosen from the length of the burning section. For each cross-section, it was divided into 16 equal areas (~0.035 m 2 for each area) and air velocities measurements were conducted at the center of each area. The total measured air velocities were ranging from 0.0017 to 0.0019 m/s (RSD = 1.20%) indicating the uniformity of the test chamber was quite acceptable.

Sample Collection and Analysis
For each selected test conditions, three repeated particle size segregating samplings were conducted at the converging part of the air outlet section of the test chamber using a Micro-Orifice Uniform Deposit Impactor (MOUDI TM model 110, MSP Corp., Minneapolis, MN, USA), and followed by Nano Micro-Orifice Uniform Deposit Impactor (Nano-MOUDI TM model 115, MSP Corp., Minneapolis, MN, USA) for collecting the generated HCl p . The whole sampling train consisted of an inlet stage (with a 50% cutoff aerodynamic diameter (d 50% ) of 18 m), thirteen impaction stages (with d 50%s of 10.0, 5.6, 3.2, 1.8, 1.00, 0.56, 0.32, 0.18, 0.10, 0.056, 0.032, 0.018, and 0.010 m, respectively), and a 37 mm PTFE back-up filter (pore size = 2.0 m, Zeflour TM Pall Corp., Ann Arbor, MI, USA). The summation of HCl p concentrations of all impaction stage and the back-up filter was regarded as the total HCl p concentration. The sampling flow rates of MOUDI TM and Nano-MOUDI TM were set at 30.0 and 10.0 L/min, respectively and were checked periodically throughout the entire sampling period. Yet, it is true that many instruments can be used to conduct aerosol samplings with particle sizes covering both micron and submicron ranges (Li et al., 2009;Kim et al., 2010;Liu et al., 2010;Intra et al., 2011), the use of MOUDI and Nano-MOUDI in the present study has the advantage in their continuous collected particle size ranges.
Beside HCl p , three repeated HCl g and Cl 2 samples were simultaneously collected using a silica gel tube (Cat. No. 226-10-03, SKC Inc., Eighty Four, PA, USA) and a 25 mm sliver membrane filter (pore size = 0.45 m, Cat. No. 225- 1802, SKC Inc., Eighty Four, PA, USA) per the NIOSH method 7903 and 6011, respectively. Both sampling flow rates were specified at 0.3 L/min and were also checked periodically throughout the entire sampling period.
All collected HCl p , HCl g and Cl 2 samples were analyzed by using the Ion Chromatography with Electron Capture Detection (DX-100 model, Dionex Corp., Sunnyvale, CA). This study yields method of detection limit (MDL) of 3.09 g/ L for chloride (Cl -).

Data Analysis
In this study, the descriptive statistics was used to describe the concentrations of HCl p , HCl g and Cl 2 (i.e., C HClp , C HClg , and C Cl2 ) obtained from the four selected RH conditions. The differences among the above resultant concentrations were examined by using the Kruskal-Wallis test. For each test condition, the particle size distribution was obtained by averaging the three collected sizesegregating samples. Both the mass median aerodynamic diameter (MMAD) and geometric standard deviation (GSD) was used to describe the resultant particle size distribution. Here, GSD can be estimated either by d 50% /d 16% or d 84% /d 50% , where d n% represents the aerodynamic diameter at d ae with a n% cumulative fraction for the given particle size distribution (Ramachandran, 2005). The resultant size distribution data was further used to estimate mass concentrations of the inhalable (C inh ), thoracic (C thor ), and respirable (C resp ) fractions of C HClp based on the three particle size-selective sampling conventions adopted by the International Standards Organization (ISO), the Comité Européen Normalization (CEN), and American Conference of Governmental Industrial Hygienists (ACGIH) (ISO, 1995;CEN, 1993;ACGIH, 1984). Here, inhalable, thoracic, and respirable aerosols represent the fraction of particles which is aspirated through the nose and /or mouth during breathing, the fraction of inhaled particles which passes into the lung below the larynx, and the fraction of inhaled particles which passes down to the alveolar (or gas exchanging) region of the lung, respectively. Based on the above definitions, the concentrations of HCl p reaching the head region (C head = C inh -C thor ), tracheobronchial region (C tb = C thor -C resp ), and alveolar region (C alv = C resp ) under the four selected RH conditions could be estimated (Tsai et al., 1995(Tsai et al., , 1996Vincent, 2005;Shih et al., 2009). In addition, fractions of inhaled HCl p reaching the head region (F head = C head /C inh ), tracheobronchial region (F tb = C tb /C inh ), and alveolar region (F alv = C alv /C inh ) were also determined. Table 1 shows means and their corresponding RSDs of C HClp , C HClg , and C Cl2 emitted from TCS burning processes under the four selected RH conditions. For all test conditions, we found that the C HClp (mean = 1.30 × 10 5 -1.46 × 10 5 mg/m 3 ; RSD = 6.90-22.0%) was the most dominant by-product, followed by the C HClg (mean = 9.03 × 10 3 -11.4 × 10 3 mg/m 3 ; RSD = 8.90-37.1%) and C Cl2 (mean = 1.91 × 10 3 -2.18 × 10 3 mg/m 3 ; RSD = 4.00-11.0%). Here, it should be noted that no significant difference was found among C HClp obtained from the four selected RH conditions (Kruskal-Wallis test, p > 0.05). The same trend can also be seen in C HClg and C Cl2 . The above results indicate that RH didn't have a significant effect on the composition of the by-products emitted from the TCS burning process.

C HClp , C HClg , and C Cl2 Emitted from TCS Burning Processes
The emitted HCl concentrations (i.e., C HClp + C HClg ,) were much higher than Cl 2 concentrations (i.e, C Cl2 ). Apparently, it was contradictory to the theoretical predictions based on eq. (1) (i.e., C Cl2 should be ~14 times in magnitude higher than that of the summation of C HClp and C HClg ). The above inconsistency might be explained by Eq. (2) (i.e., most of the generated Cl 2 might further react with H 2 O to form HCl). The above inference is consistent with results found by Chow et al. (1994). They found that ~90% of chloride containing in the coal was converted to HCl during the coal combustion process. Therefore, it could be concluded that the existence of water vapor during TCS burning process would result in less acute inhalatory effect because of converting the emitted Cl 2 into HCl p . However, more serious deep lung irritation and inflammation might occur due to higher resultant concentrations in HCl p . In the present study, we also found that C HClp was much higher than that of C HClg . The above result was not so surprising because (1) the generated HCl g might condensate on existing SiO 2 particles to form HCl p via the heterogeneous condensation process (Friedlander, 2000), (2) the generated HCl g might dissolve into H 2 O coated on the surface of SiO 2 particle, and (3) the generated Cl 2 might further react with H 2 O to form HCl p .
Finally, our study also indicated that humidity didn't have a significant effect on the compositions of TCS burning products. By examining all test conditions, ~0.655-0.962 g/min humidity was originally containing in the inlet air (RH = 57.3-89.8 %). In addition, there was ~0.440-0.498 g/min water vapor emitted from TCS burning processes based on the TCS consumption rates (CR = 8.28-9.38 g/min) obtained from this study. Theoretically, the required water vapor importing rate to maintain the saturation condition of the test chamber was 1.07 g/min (at 25°C). Considering the total water vapor importing rates (i.e., water vapor containing in the inlet air + water vapor emitted from the TCS burning process) (= 1.12-1.46 g/min) were consistently higher than 1.07 g/min indicating that the test chamber was always on a water saturated condition for all TCS burning experiments. As a result, the effect of humidity on the compositions of TCS burning products becomes insignificant. Table 2 shows particle size distributions of HCl p obtained from the four selected RH conditions. It can be seen that the emitted particles (MMADs = 0.808-1.04 m; GSDs = 2.13-3.50) fell to the range of the accumulation mode (0.1-2.5 m). For illustration, Fig. 3 shows the particle size distribution obtained from the test condition of RH = 65.5%. As described in the previous section, HCl p could be generated by condensation of HCl g on the emitted SiO 2 particles via the heterogeneous condensation process, or via the absorption of HCl g onto H 2 O coated on SiO 2 particle surface. Based on the above description, the emitted HCl p particles might fell to the range of the nuclei mode (i.e., 0.1 m). However, it should be noted that the particle sizes of these primary HCl p particles might further increase to the accumulation mode because of the effects caused by both the Brownian and turbulent coagulation (Friedlander, 2000). In addition, our results were also consistent with the results obtained by Schenkel and Schaber (1995).

Particle Size Distributions of HCl p
Yet, it is true that high moisture content in the air might lead to an increase in the particle size according to the Kelvin effect theorem (Friedlander, 2000;Hu et al., 2010). In this study, the four selected RH conditions did not affect the emitted particle sizes of HCl p which might because the test chamber was always on a water saturated condition for all TCS burning experiments (as described earlier). Fig. 4 shows the fractions of inhaled HCl p reaching the alveolar region (F alv ), tracheobronchial region (F tb ), and head region (F head ) under the four selected RH conditions during the TCS burning process. For all test conditions, a consistent trend of F alv (mean = 85.8-88.8 %; RSD = 1.50-7.45 %) > F tb (mean = 6.53-8.80 %; RSD = 0.50-4.90 %) > F head (mean = 4.67-5.40 %; RSD = 0.60-2.62 %) was found in the present study. The highest fraction found in the alveolar region indicates that most HCl p emitted from  TCS burning processes could penetrate into the deep lung of the respiratory tract. It is known that HCl g could only reach the upper respiratory tract due to its high water solubility. In particular, as described in the previous section, the emitted C HClp (= 1.30 × 10 5 -1.46 × 10 5 mg/m 3 ) were consistently much higher than C HClg (= 9.03 × 10 3 -11.4 × 10 3 mg/m 3 ) in all experimental campaigns. Our results suggest more serious damage might occur at the deep lung region than other regions of the respiratory tract as workers exposed to contaminants emitted from TCS burning processes. Table 3 shows the consumption rates of TCS and the measured total chloride emitted concentrations and their corresponding theoretical values under the four selected RH conditions. No significant difference can be found in consumption rates of TCS among the four selected RH conditions (mean = 8.28-9.39 g/min; RSD = 6.56-19.0 %) (Kruskal-Wallis test, p > 0.05). Therefore, it could be expected that no significant difference can be found in the theoretical total chloride emitted concentrations (mean = 1.55 × 10 5 -1.75 × 10 5 g/min; RSD = 6.56-19.0 %) (Kruskal-Wallis test, p > 0.05). On the other hand, we found that there were slight differences between theoretical and their corresponding measured total chloride emitted concentrations (mean = 1.40 × 10 5 -1.60 × 10 5 mg/m 3 ; RSD = 8.19-26.1%). Based on the above values, the resultant recoveries (= (measured values/theoretical values) × 100%) fell to the range of 90.1-91.0 %. The above discrepancies were because the parts of HCl were not sampled in the present study, including those were deposited on the chamber walls, or being absorbed by the H 2 O coated on the chamber wall. However, the above results also suggest that our results can fairly characterize the emissions of contaminants from the TCS burning processes.

CONCLUSIONS
We found that the HCl p was consistently as the most dominant contaminant, followed by the HCl g and Cl 2 , emitted from the TCS burning processes. We also found that humidity didn't have a significant effect on the composition of the above three emitted contaminants under the four RH test conditions. Therefore, lower RH test conditions are suggested for further research in the future, in particular for TCS could be stored in areas with much lower RH conditions. The emitted HCl p fell to the range of the accumulation mode, and the fraction of HCl p deposited on the alveolar region was consistently higher than those on tracheobronchial region and head region for the selected RH test range. Therefore, more serious ill-health effect on the deep lung region of the respiratory tract can be expect as workers exposed to contaminants emitted from TCS burning processes. The results obtained from this study will provide information for semiconductor and optoelectronic industries to initiate an appropriate TCS emergency response plan to prevent workers from the illhealth effects associated with toxics emitted from TCS burning processes.