JOURNAL OF ANALYTICAL ATOMIC SPECTROMETRY NOVEMBER 1994 VOL. 9 1299 Quantitative Analysis of Electroniclgrade Anhydrous Hydrogen Chloride by Sealed Inductively Coupled Plasma Atomic Emission Spectroscopy* Tracey Jacksier Air Liquide Chicago Research Center 5230 S. East Avenue Countryside IL 605253787 USA Ramon M. Barnes Department of Chemistry Lederle Graduate Research Center University of Massachusetts Box 345 70 Amherst MA 01 003-45 10 USA Anhydrous hydrogen chloride at concentrations of up to 100% has been introduced into a sealed inductively coupled plasma system for quantitative spectrochemical analysis. Vapour-phase sampling of monobutyltin trichloride was developed to calibrate tin and carbon. The addition of chlorine as a modifier gas was required in a ratio of 1 :1 to maintain stable and reproducible emission signals.Under flowing conditions detection limits for tin and carbon were 49 and 271 (ng g - ' ) respectively in a 16% v/v HCI-CI argon plasma. Impurities identified qualitatively included Al C Ca Cr Fe Ni and Sn. In addition the bulk temperature of the plasma was determined to be 10500+1500 K. Keywords Sealed discharge; gas analysis; inductively coupled plasma; hydrogen chloride; emission spectroscopy A trend in very large scale integration (VLSI) technology requires ultra high purity chemicals in fabrication process steps. Chemical vapour deposition (CVD) and reactive ion etching (RIE) are particularly sensitive to impurities present in the process gases and impurities on the silicon surface from prior process steps. These impurities diffuse into the bulk semiconductor and lead to the formation of various types of defects.Therefore these impurities have an impact on product quality and reliabilit~.'-~ For example contaminants such as Fe Ni and Cu can generate defects that are responsible for yield losses caused by leakage current^.^ Anhydrous HCl is used to etch the silicon wafer surface prior to epitaxial crystal g r ~ w t h ~ and the trace element content is therefore of critical importance. Gases used in VLSI fabrication are specified with metal contamination in the lower ng g-' range. The production of ultra-pure chemicals therefore requires precise measurement of impurities at or below this level. The analysis of trace metal contamination in HCl includes various chemical preparation techniques direct analysis and preconcentration methods to improve sensitivity in the determination of trace impurities.For most metal determination techniques the metal impurit- ies must first be transferred into a liquid phase by bubbling HCl through aqueous solutions (hydrolysis) to preconcentrate the sample regardless of whether the impurities are distributed in the gaseous liquid or solid phase. The maximum concen- tration is determined by the solubility of HCl(g) in water (ix. solubility of HCl at 60 "C is 56.1 g per 100 g H20).6 The concentration of HCl dissolved in the aqueous solution can be determined by volumetric titration. Dilution of the HCl(1) is necessary before inductively coupled plasma (ICP) analysis owing to the change in physical properties of the sample solution and the relative sensitivity for the analyte (i.e.signal suppression). Additionally high concentrations of HCl may decrease the plasma excitation temperature resulting in loss of analyte ~ensitivity.~ Reducing the sensitivity loss by dilution can be overcome with evaporation of the sample. However many sample handling steps increase the risk of Contamination and elemental loss. The direct analysis of concentrated HCl using inductively coupled plasma atomic emission spectroscopy (ICP-AES) or * Presented in part at 1993 Pittsburg Conference and Exposition Atlanta GA USA March 8-12 1993 electrothermal atomic absorption spectroscopy (ETAAS) has been used to eliminate elemental loss during evaporation.8 However the direct analysis of HC1 by ICP-AES suffers from signal suppression of analytes such as As and P resulting in lower sensitivity.Lower analytical sensitivity is also observed in the analysis of concentrated acids with ETAAS. Additionally corrosion of parts of the analyser occurs in ETAAS. An analytical method was developed by Bridenne et aL8 in which metallic impurities were determined by ICP-AES after hydrolysis acid evaporation and reconstitution in aqua regia. Approximately 50 ng cm-3 were reported for Cr Cu Fe and Ni in HCI with an average relative standard deviation (RSD) of ~ 7 % . However the many sample handling steps necessary in this procedure increased the risk of sample contamination. An on-line extraction method was developed by Schramg to introduce gaseous HC1 directly into the ICP.The gaseous sample was passed into a mixing chamber positioned under the aerosol tube of the ICP torch. The sample gas flow was regulated by a peristaltic pump. Feeding the gas directly into the plasma reduced the risk of sample contamination since the gas reached the plasma directly without further chemical or physical proces~.~ Two major difficulties were identified; matrix effects and calibration. The direct calibration of impurities was impossible because many metals did not form sufficiently volatile compounds to matrix match standards. Total reflection X-ray fluorescence (TXRF) spectrometry has been used for the analysis of HF HC1 HNO H,SO and NH3.2 However this method was developed for aqueous acids and sample evaporation was required to remove the solvent.Despite efforts current state-of-the-art techniques are inad- equate to determine ultratrace metal impurities in gaseous HC1. At present HC1 gas purity is established indirectly from the quality of the epitaxy obtained.' Thus additional effort to improve methodologies and sensitivity is warranted. The sealed inductively coupled plasma (SICP) has been developed as an analytical spectrochemical ~ o u r c e . ~ ~ - ~ ~ It exhibits a number of advantages which suggest its suitability for application to the analysis of HCl. Low flow rates (< 70 cm3 min-l) result in limited sample consumption and dilution. No sample preparation is required because anhydrous HC1 is a gas at room temperature. Finally the enclosed source makes the safe introduction of HCl contamination free.The SICP analysis of HCl is presented in this paper.1300 JOURNAL OF ANALYTICAL ATOMIC SPECTROMETRY NOVEMBER 1994 VOL. 9 Qualitative detection of metal impurities in HCl as well as quantitative results for the determination of C and Sn are reported. Additionally the temperature of an HCl-Cl,-Ar discharge is estimated. Experiment a1 Instrumentation The instrumentation used was described previous1y.lo3l2 The gas handling system was modified (Fig. 1) to allow simul- taneous mixing of Cl anhydrous HCl and a standard into the Ar gas stream. Hydrogen chloride (Alphagaz Walnut Creek CA USA) was introduced through an electronic mass flow controller F4 (0-100cm3 min-l Model 1159B MKS Instruments Andover MA USA) C1 (Alphagaz) through F3 (0-100 cm3 min-l Model 1159B MKS Instruments) Ar through F1 (0-1000cm3 min-l Model 1159B MKS Instruments) and the standard through F6 (0-20 cm3 min-' Model 1159B MKS Instruments).To eliminate the possibility that the impurities were the result of particle generation by valves or the gas handling system a 0.01 pm in-line particle filter (Membralox Model 28235G Alcoa Separations Technology Pittsburgh PA USA) was installed. After the filter was in place the gas handling system was purged with Ar for 48 h to remove entrained air. Before experimental use the filter was conditioned for 48 h by passing 10 cm3 min-l of HCl through it. The discharge was generated at 40.68 MHz (RF Plasma Products Model HFS 5000D 0-5 kW Marlton NJ USA) by using a 2 turn induction coil (66.5 mm diameter 3.2 mm 0.d.copper tubing) around the 65 mm SICP ~0ntainer.l~ The end- on view of the centre of the discharge was imaged through a 100mm focal length quartz lens (2cm diameter Oriel) onto the entrance slit (50 pm) of a 0.35 m Czerny-Turner mono- chromator (Heath Model EU-700/E) with 1180 grooves mm-' grating blazed for 250 nm. Emission signals were detected with an RCA 4832 photomultiplier tube operated with an anode voltage of 1100 V. The output from the photomultiplier tube was amplified with a programmable picoammeter (Kiethley Model 18000-20 Kiethley Instruments Cleveland OH USA). A low pass filter and offset circuit were installed to condition the signal. Data were collected with a data acquisition board (DT 2905 Data Translation Marlboro MA USA) and IBM PS/2 Model 80 computer (International Business Machines Armonk NY USA).Operating Procedure An Ar plasma is formed under flowing conditions (typically with 100 cm3 min-l) at approximately 1.0 kW. Immediately after plasma initiation HCl is added to the Ar stream. The HC1 concentration is then increased by simultaneously increas- ing the HC1 flow and decreasing the Ar flow. Once the desired volume ratio of HCl is achieved the applied r.f. power is increased to 1.4 kW. As the flow of Ar is decreased a slight change in tuning is required to minimize the reflected power. The flowing plasma is maintained for approximately 30 min to stabilize the discharge. The sealed ICP operates in two modes gas flowing and non-flowing (static) at atmospheric pressure. For static oper- ation the discharge is isolated with valves V13 and V5 (Fig. 1) from the gas handling system.Temperature Determination The excitation temperature is one of the main physical param- eters that characterizes the excitation mechanism in equilib- rium plasmas. The temperature was obtained by plotting a graph of log (1A/gA) as a function of the excitation energy E,, of the upper level of the transition. If E is expressed in cm-l the resultant Boltzmann plot slope represents -l/(kT). In this expression I is the intensity of the line at wavelength A from the excited level of energy E,, g is the statistical weight of this level and A is the transition probability . l5 Neutral iron (FeI) is generally selected because its emission lines possess desirable characteristics for precise temperature I I V13 f Standard container Vent $.- Sealed ICP Fig. 1 Gas handling system used for the introduction of standards and samples into the sealed ICP Fl-F6 mass flow controllers; V1-V19 shut off valves; and CV check valveJOURNAL OF ANALYTICAL ATOMIC SPECTROMETRY determinations and large transition probability data sets are available.16 The criteria used in selecting suitable Fe I lines for the temperature measurement were (i) no significant spectral overlap of the lines should occur (ii) line intensities should fall within the dynamic range of the detector (iii) transitions to lower energy non-resonant ground state levels to avoid self-absorption and (iv) maximal spread in upper energy levels to minimize error in measurements of the slope of the tempera- ture p l ~ t .~ ' ' ~ The National Institute & Standards and Technology (NIST) transition probability and energy level values were used for the temperature measurements." The transition probabilities given by O'Brian et al. also were used.lg Calibration Diffusion tubes had been used previously with the SICP for the calibration of As and P in silane."*'' However diffusion tubes were not available as calibration standards for C or Sn. Therefore vapour-phase sampling was investigated as an alternative source for calibration. Mono but yl- tin trichloride ( Aldrich Chemical Milwaukee WI USA) MBTTC a liquid with a room temperature vapour pressure of 5.3 Pa is neither pyrophoric nor toxic and was used as both the Sn and C standard.The MBTTC was pipetted into a 316 1 stainless-steel sample cylinder (Whitey 316LSS 150 cm3 Model 316L-HDF4-150) fitted with an outage tube adapter with a standard 26.4cm long tube stub (hereafter referred to as a bubbler). A known concentration of the MBTTC standard was introduced into the gas stream just prior to entering the discharge container. The MBTTC was carried by the argon flow in the tube stub as it bubbled through the sample and out through the outage tube adapter. The bubbler connection to the handling system is illustrated in Fig. 1. A mass flow controller (F6) located before the bubbler was used to meter the standard flow through the discharge. A 2.3 kPa crack pressure Teflon coated check valve (Model SS-4C-KZ-TS-1/3 Nupro Willoughby OH USA) was placed immediately before the stream merging point with the plasma gas.The calibration standard was added from the bubbler to the HCl-containing gas stream. The concentration of the standard was varied by changing the total gas flow of the Ar gas stream. Additionally since the discharge container has a volume of 100cm3 information could be obtained for both absolute concentration ng and ppb (m/m). A calibration function was obtained by varying the concentration of Sn from 431 (2.3 pg Sn per g HC1) to 2777 ng (15 pg Sn per g HC1) and C from 174 (0.94 pg Sn per g HCl) to 1123 ng (6.1 pg Sn per g HCl). The lower concentration range was determined by the flow of the available mass flow controllers. Results are reported as the average of three replicate experiments per Sn (or C) concentration.Analyte equilibration time was also evaluated by introducing 414 ng of Sn (as MBTTC) into the discharge with a total gas flow rate of 61 cm3 min-'. The net Sn emission was monitored for 6 min. Equilibrium was reached after 2.5 rnin and remained constant for 2.5 min. A 4 min sample introduction time was adopted. The analyte purge time was evaluated by introducing 414 ng of Sn into the plasma gas stream with a total gas flow of 61 cm3 min-'. The net Sn emission was recorded for a total of 25 min. The Sn emission was decreased after 15 min. Therefore a 20 min purge time between samples was adopted. The calibration standard was allowed to flow through the discharge container for 4 min before emission intensities were recorded at either C I 247.857 nm or Sn I 326.234 nm lines.All experimental determinations were made in triplicate. Results and Discussion Maximum Hydrogen Chloride Addition Hydrogen chloride concentrations of less than 22% v/v in the discharge can be maintained with a minimum of 400 W for NOVEMBER 1994 VOL. 9 1301 100 l o 0 i 400 600 800 1000 1200 1400 1600 1800 2000 PowerAiV Fig. 2 Maximum HCl content as a function of net r.f. applied power. Data obtained by introducing the largest tolerable concentration of HCl into the discharge with a fixed r.f. power both flowing and static operating modes. Increasing the HC1 content to 38% v/v is possible by increasing the forward applied power to greater than 500 W. Formation of a 100% v/v HCl plasma at powers of > 1.6 kW can be accomplished in a 65 mm discharge container.However plasma stability for concentrations > 80% v/v is limited to approximately 20 min. Hydrogen appears to diffuse through the quartz container and causes arcing to the induction coil. The relation between r.f. power and maximum obtainable HCl is illustrated in Fig. 2. Impurity Identification The discharge was generated initially with 5% HCl in Ar (v/v) at 1.0kW. As the concentration of HC1 in the plasma was increased the power was increased to maintain the stability and diameter of the plasma. A concentration of 26% v/v HC1 at 1.5 kW was obtained. Although higher HC1 concentrations can be sustained the discharge requires powers of approxi- mately 2.0 kW to maintain a constant plasma size. Low power however results in a longer discharge container lifetime.The monochromator wavelength was scanned from 200-900nm and impurities identified in the ICP spectra by verifying the presence of at least three prominent lines for each element when possible except C for which only the C I 247.857 nm line was used. Impurities identified included Al As B C Ca Cr Cu Fe Ni and Sn (Figs. 3 and 4). Owing Si I 1 0 0 0 ~ fl .- a 700 750 I 650 ' I I I 250 255 260 265 Wavelengt h/n m Fig.3 Emission spectrum of Si I and Fe I1 (250-265nm) in 26% HCl v/v in Ar at 1.5 kW in a 65 mm discharge container. The peak at 257 nm is attributed to Clz molecular emissionI21302 JOURNAL OF ANALYTICAL ATOMIC SPECTROMETRY NOVEMBER 1994 VOL. 9 - C cn E Lo 0 u! Iz 650 625 600 305.00 310.00 315.00 320.00 325.00 330.00 Wavelengt h/n m - I I I - Fig.4 Emission spectrum of A1 I Sn 1 and Cu I (305-330 nm) in HCl 26% v/v in Ar at 1.5 kW in a 65 mm discharge container. The broad molecular emission has been attributed to chlorineI2 to the previous transfer of arsine and boron trichloride in the gas handling system As and B emission appear in all spectra. Since the coil is in contact with the container wall Cu from the induction coil is believed to be diffusing through the quartz container and thereby producing significant emission.12 The molecular emission in Figs. 3 (251 nm maximum) and 4 has been identified as C1 molecular emission and has been dis- cussed e1~ewhere.l~ Additionally the analyte impurity emission was observed to decrease from flowing to static conditions. Therefore all further analyses were conducted in the flowing mode.Effect of HCl and Modifier on Net Sn Signal Owing to the physical changes in the size of the plasma as a function of HCl concentration HCI concentrations (16,20 and 24%) needed for standardization were investigated. Introducing more than 15% C1 into the 20% HCl-Ar plasma or more than 4% C1 into the 24% HC1-Ar plasma was not possible. Additions of up to 26% C1 were introduced into a 16% HC1-Ar plasma. When compared with the 20% and 24% HC1-Ar plasmas the l6Y0 HC1-Ar plasma was more robust and therefore chosen for further investigation. For a 16% v/v HCl-Ar plasma the concentration of C1 was varied from 6.5 to 26% for a constant Sn concentration of 414 ng. For C1 concentrations below 16% Sn appeared to deposit inside the container.The Sn sensitivity increased with increas- ing C1 concentration above 16%. To minimize the contri- bution of the impurities present in the C1 to the HC1 analysis the smallest concentration of C1 necessary to prevent depos- ition was adopted. Therefore standardization conditions were set at 16% HCl v/v 16% Cl v/v in Ar at 1.4 kW with a total flow of 61 cm3 min- through the discharge. Temperature Determination All the Fe I spectral lines from the 16% v/v HC1 with 16% C1 v/v in Ar discharge viewed 1 cm from the centre of the discharge fitting all the criteria are listed in Table 1. A line with an r2 value of 0.88 was obtained (Fig. 5 ) that yielded a temperature of 10 500 f 1500 K. With the O’Brian et al. trans- ition pr~babilities,’~ the calculated temperature was 11 800 2400 K (r2 = 0.77).Since no data exist to suggest that the SICP is in local thermal equilibrium (LTE) with 32% molecular gas in Ar the Fe I temperatures may not be a unique measure of the energy characteristics of the SICP. Table 1 for temperature determination Excitation energy,18 gAi8 and wavelengths for Fe I lines used A/nm 294.79 300.10 304.76 305.91 356.54 351.01 358.12 360.89 361.88 363.15 364.78 373.49 373.71 374.56 374.95 375.82 376.38 381.58 382.04 382.59 Eexclcm - 34329 34017 33507 33096 35768 35379 34844 35856 35612 35257 34782 33695 27 167 27395 34040 34329 34547 38175 33096 33507 €54 2.2 2.8 2.9 2.0 7.8 18.0 23.0 10.0 9.5 8.6 6.1 20.0 1.5 1.2 13.0 10.0 6.2 16.0 12.0 8.9 IFlowing 13 19 28 10 11 32 68 12 23 23 22 74 32 36 41 28 16 14 45 31 log(lA/gA) 3.24 3.31 3.47 3.18 2.70 2.80 3.02 2.64 2.94 2.99 3.12 3.14 3.90 4.05 3.07 3.02 2.99 2.52 3.16 3.12 5.0 1 1 4.5 F\ 4.0 5 3.5 3.0 07 2 2.5 Y 1.5 2.0 1 1.01 1 I I I I 1 1 26000 28000 30000 32000 34000 36000 38000 40000 E Jcm - ’ Fig.5 Temperature determination for 16% HC1 with 16% C12 in argon at 1.4 kW using Fe I lines viewed 1 cm from the centre of the discharge slope =0.0001362 -t 15%; intercept = 7.739 However the Fe I temperature reflects the ability of the discharge to populate excited levels in the energy range under ~0nsideration.l~ Cali bra tion A linear fit to the tin data gave a slope of 0.6803_+0.013 and an intercept of 0 with an r2 value of 0.9938 (Table2).The detection limit for Sn in the gas stream was 49 ng 8-l. The detection limit was calculated as 3 times the standard deviation of the background divided by the slope of the calibration func- tion.The background value was obtained from the least concentrated sample. For C a linear fit gave a slope of Table 2 Tin calibration 16% HCl-l6% C12 in Ar; 1.4 kW for Sn I 326.234 nm CSnl/ng CSnl/pg g-’ Average _+ SD/ng RSD (Yo) 43 1 2.3 376+45 12.1 1058 5.7 783 & 49 6.3 1659 9.0 1170+75 6.4 223 1 12.0 1487 +_ 77 5.2 2777 15.0 1853 k 79 4.3JOURNAL OF ANALYTICAL ATOMIC SPECTROMETRY NOVEMBER 1994 VOL. 9 1303 Table 3 Carbon calibration 16% HCI-16% Cl in argon; 1.4 kW for C I 247.857 nm CCl/ns CCl/PS g - Average -t SD/ng RSD (Yo) 174 0.94 23+4 16.6 428 2.3 54+4 8.3 67 1 3.6 78f4 5.1 903 4.9 112f2 1.9 1124 6.1 139+1 0.3 The advantages of the SICP have been extended to the analysis of reactive gases.Additionally detection limits for C and Sn have been demonstrated to be competitive with conven- tional ICP-AES. 0.1228+0.003 with an intercept of 0.2368k2.55 and an r2 value of 0.9983 (Table 3). The detection limit for C was calculated as 271 ng g-'. These SICP detection limits are comparable with those of the conventional ICP for C (0.01 ppm m/v) and tin (30 PPb m/vh Conclusion With the SICP the direct analysis of 16% HCl was accomplished with minimal sample dilution and no sample preparation. Although HC1 is a corrosive gas the SICP provided an enclosed system for safe analysis. The direct analysis of HCl is an on-going research area and experiments are planned to compare data with hydrolysis experiments.' An equal concentration of Clz to HCl was required to prevent deposition of Sn and C standards in the discharge.Doping the HCl with Sn and C resulted in a detection limit of 49 and 271 ppb respectively in a discharge containing 16% hydrogen chloride v/v. Additionally absolute detection limits of 50.2 ng and 9.1 ng were obtained for C and Sn respectively. Extrapolation of these detection limits to a discharge contain- ing pure HCl results in detection limits of 306 ppb m/m and 1694 ppb m/m for Sn and C respectively. The repeatability of these data suggests that vapour phase sampling is an alternative to diffusion tubes for calibration. One limitation of this approach is the unavailability of vapour pressure data of suitable compounds. Additionally many com- pounds do not have an ideal vapour pressure.This would require heating and or cooling of suitable compounds and a more complex delivery system would be needed. However the use of diffusion tubes may provide better repeatability owing to a decreased dependency on temperature variation. The use of diffusion tubes for elements that do not form hydrides is under investigation. Iron impurities present in the HC1 allowed the temperature of the discharge to be estimated at 10 500+ 1500 K. The large deviation for the temperature measurement is due to exper- imental error. The value is somewhat lower than a similar discharge in C1,.20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 This research was supported by Air Liquide. Experiments were performed at the University of Massachusetts.References Hardwick S . J. Lorenz R. G and Weber D. K. Solid State Technol. 1988 10 93. Prange A. Kramer K. and Reus U. Spectrochim. Acta Part B 1991 46 1385. Faix W. G. Schramm W. Vix F. Weichbrodt G. and Hendelman R. Fresenius 2. Anal. Chem. 1988 329 847. Roth H. J. and Neunteufel P. Proceedings of the Satellite Symposium to ESSDERC 1989 Berlin. Pivonka D. E. Appl. Spectrosc. 1991 45 597. Handbook of Chemistry and Physics ed. R. C. Weast The Chemical Rubber Company Cleveland OH 61st edn. 1980. Thompson M. and Barnes R. M. in Inductively Coupled Plasmas in Analytical Atomic Spectrometry eds. Montasser A. and Golightly D. W. VCH Publishers New York 1992 ch. 5. Bridenne M. Carre M. Coffre E. Marot Y. and Simondet F. presented at the 1992 Winter Conference on Plasma Spectrochemistry San Diego CA January 1992 poster WP35. Schram J. Fresenius. Z . Anal. Chem. 1992 343 727. Jahl M. J. Jacksier T and Barnes R. M. J . Anal. At. Spectrum. 1992 7 653. Jahl M. J. and Barnes R. M. J. Anal. At. Spectrom. 1992,7 833. Jacksier T. and Barnes R. M. J. Anal. At. Spectrum. 1992,7,839. Jacksier T. and Barnes R. M. Appl. Spectrosc. 1994 48 382. Jacksier T. and Barnes R. M. Spectrochim. Acta Part B 1993 48 941. Jarusz J. Mermet J.-M. and Robin J. P. Spectrochim. Acta Part B 1978 33 55. Kubota A. Fijishiro Y. and Ishida R. Spectrochim. Acta Part B 1981 36 697. Blades W. M. and Caughlin B. L. Spectrochim. Acta Part B 1985 40 579. Corliss C. H. Bozman W. R. Experimental Transition Probabilities for Spectral Lines of Seventy Elements NBS Monograph 53 U.S. Government Printing Office 1962. O'Brian T. R. Wickliffe M. E. Lawler J. E. Whaling W. and Brault J. R. J . Opt. SOC. Am. B 1991 8 1185. Jacksier T. and Barnes R. M. Spectrochim. Acta Part B in the press. Paper 4/0005 71 Received January 5 1994 Accepted July 18 1994