Investigation of the Thermal Decomposition of Triethylgallium Using in situ Raman Spectroscopy and DFT Calculations

Transcription

1 / The Electrochemical Society Investigation of the Thermal Decomposition of Triethylgallium Using in situ Raman Spectroscopy and DFT Calculations Jooyoung Lee, Young Seok Kim and Tim Anderson Department of Chemical Engineering, University of Florida, Gainesville, FL 32611, USA Pathways for homogeneous thermal decomposition of (C 2 H 5 ) 3 Ga (TEGa) were followed using in situ Raman spectroscopy measurements in an up-flow, cold-wall CVD reactor. The results of Density Functional Theory (DFT) calculations were used to assign measured Raman bands to the decomposition products (Et) 3 Ga, (DEGa) 2, (Et)GaH Ga(Et) 2 and (Et)GaH GaH 2. The results of this study are consistent with both hydride elimination and homolysis pathways active. DFT calculations using the B3LYP/LanL2DZ model chemistry were also performed to screen possible routes and estimate their reaction rate parameters. Introduction Gallium-based compound semiconductors have played an important role in many electronic and optoelectronic devices, and MOCVD is often the preferred method of film deposition. Although several Ga precursors have been explored, TEGa ((C 2 H 5 ) 3 Ga, Triethylgallium) and TMGa ((CH 3 ) 3 Ga, Trimethylgallium) are most commonly used as Ga sources for MOCVD of Ga-V compound semiconductors (1-3). The limited reports on gas phase decomposition mechanisms for TEGa suggest either simple homolysis of the C-Ga bond or hydride elimination as the initial step (4-6). As summarized in Table I the reported experimental activation energy for removal of the first ethyl group ranges from 22.6 to 47.2 kcal/mol, and frequency factors range from 11.3 to 36.1 in natural logarithm scale (4-7,8). These deviations may in part be attributed to chemical effects on the decomposition reactions arising from selection of carrier gas (i.e., toluene vs. hydrogen) and reaction conditions that may favor one mechanism over another. In addition to these values derived from experimental data, the activation energy values 44.0 kcal/mol for hydride elimination and 60.0 kcal/mol for bond homolysis were calculated using ab initio calculations (restricted Hartree-Fock MO method) by Tsuda et al. (9). Our DFT calculation results suggest 40.3 kcal/mol for hydride elimination and 65.0 kcal/mol for homolysis as activation energies and 30.6 and 52.2 as frequency factors, respectively. Unfortunately, this wide range in the values of the rate parameters produces large uncertainties in reactor modeling results. TABLE I. Experimental and Calculated Rate Parameters for First Ethyl Group Dissociation T(K) ln k 0 E a (kcal/mol) Method Ref 610 ~ ~ ~ ~ ~ ~ ~ ~ Flow system, toluene carrier Desorption on GaAs Static system Desorption on GaAs Decomposition on GaAs(100) Decomposition on GaAs(100) Hartree-Fock (homolysis) Hartree-Fock ( hydride elimination) DFT (homolysis) DFT( -hydride elimination) (4) (7) (7) (10) (11) (11) (9) (9) This work This work

2 42 In this study, the gas phase decomposition kinetics of TEGa weree investigated by in situ Raman spectroscopy in a vertical up-flow, cold-wall CVD reactorr and the results are compared with numerical calculations. Likely reaction pathways were evaluated with DFT calculations, which were also used to assist in the assignment of the observed bands. Experimental and Theoretical Methods The homogeneous decomposition of TEGa in N 2 was investigated in a CVD reactor interfaced with an in situ Raman spectrometer (Ramanor U-1000, Jobin Yvon) ), which includes a double additive monochromator. The nm line of a Nd:YAG solid-state laser was used as the light source. As schematically shown in Figure 1, the instrumented CVD reactor is an up-flow, impinging-jet reactor designedd to produce a stable 2-D flow condition while isolating the reactants from the reactor walls to prevent deposition. The entire CVD reactor assembly could be translated x-y-z inlets tubes are incorporated into to allow measurement of temperature and composition profiles. Three concentric the design with the reactant input into the center tube. Each tube is packed with 3 mm glass beads to provide an equal velocity, parallel-flow inlet boundary condition. Figure 1. Schematic of the experimental reactor for in situ Raman spectroscopic measurements.

3 43 Based on a previous study (13) that reported a validated steady-state, twodimensional reactor model using data from CH 4 tracer experiments, reaction conditions were identified that emphasized the extent of homogeneous decomposition and avoided the occurrence of recirculation flows in the reactor. The experimental system and methodology are described in greater detail elsewhere (12,13). In a preliminary experiment, the relative Raman cross-section of TEGa was measured in the reactor operating at room temperature and atmospheric pressure. In this experiment, the center tube carried a gas mixture of 5 mol % TEGa in N 2 carrier at 2.5 cm/s steady velocity. Pure N 2 gas was delivered through the annulus and sweep lines such that the exit flow velocity (2.5 cm/s) was identical at all three inlets. Sufficient time was provided to allow for the flow to reach steady state before the Raman scattering experiments were initiated. The Nd:YAG solid-state laser line (1.5 W laser output power) was used to excite the TEGa/N 2 source and the intensity of the Ga C 3 vibrational Raman excitation line (490 cm -1 line) of TEGa was recorded (Figure 2). 120 Intensity [a.u.] Raman shift [cm 1 ] Figure 2. Recorded Raman intensity of the Ga C 3 vibration (490 cm -1 ). Although the scattered intensity of TEGa is considerably weaker than most other group II and III metal alkyls (see Table II), the signal to noise ratio was sufficiently large that the TEGa concentration could be quantified from the measured intensity of the 490 cm -1 excitation line. With 10 repeated measurements and long integration time (3 s per 0.5 cm -1 ) at room temperature, the relative Raman TEGa cross-section was estimated to be TABLE II. Reported Relative Raman Cross-Section for selected Group II and III Metal-organic Precursors Precursor DMCd TMGa TMIn DEZn TEGa Relative Raman Cross-section a b 22.3 c 4.2 a 2.69 d (a: ref. (14), b: ref. (15), c: ref. (16), d: this work)

4 44 In addition to the band at 490 cm -1, several other bands were observed. To assist in the assignment of these additional Raman bands as well as to probe the energetics of possible reaction paths, DFT calculations were performed using the Gaussian 03 program suite (17). For all bond dissociation, activation energy, and frequency calculations, the DFT level with the spin polarized hybrid density functional B3LYP was combined with the LanL2DZ for all elements. Experimental values of the Ga C 3 bond length and Ga C C bond angle obtained from X-ray diffraction patterns from TEGa single crystal samples (18) were used for selection of the model chemistry. After discreet comparison of a number of cross-combinations of methods (HF, B3LYP, MP2, MP4) and basis sets (6-31G*, 6-31G**, 6-311G, 6-311G, 6-311G**, SDD, LanL2DZ) as well as the split basis set of B3LYP/(LanL2DZ for Ga, 6-311G for other elements), the B3LYP/LanL2DZ model chemistry best represented the experimental Ga C 3 average bond length (1.966 ~ Å experimental vs Å calculated) and the Ga C C bond angle (113.3 ~ experimental vs calculated). DFT Calculations Results and Discussion Two initial reactions have been suggested for homogeneous thermal decomposition of TEGa (Ga-C bond homolysis and hydride elimination) as well as dissociative desorption on semiconductor surfaces such as GaAs or Si (1,4-6,11,19). The detection of the direct products DEGa (homolysis) and (Et) 2 GaH ( hydride elimination) is decisive evidence for a specific mechanism. Although detection of the indirect decomposition products butane (homolysis) or ethylene ( hydride elimination) is often used to differentiate these two mechanisms (6), they are not unequivocal. The observation of other likely reaction intermediates, although not exclusive, is stronger supporting evidence. To this end DFT calculations were performed for 39 potential reaction steps. An initial screen was made on the basis of H and G values. Using the results shown in Figure 3, the reaction enthalpies for both homolysis and hydride elimination initial reactions are summarized in Figure 4. Figure 3. Reaction enthalpies [kcal/mol] for selected reactions.

5 45 Figure 4. Calculated TEGa decomposition energetics via simple Ga-C bond homolysis and hydride elimination with reaction enthalpies [kcal/mol] listed. Raman Scattering Experiments The thermal decomposition of TEGa was observed using a susceptor set point temperature 750 C. The relative mole fraction of each observed Ga species was obtained from the ratio of the integrated peak intensity of the primary peak [Ga C 3 ] at 490 cm -1 to that of the N N vibration at 2331 cm -1.

6 46 Figure 5. Raman spectra of gas-phase TEGa as a function distance below the heated susceptor along the reactor centerline: (a) 1 mm (b) 2 mm (c) 4 mm (d) 6 mm (e) 8 mm (f) 10 mm and (g) 11 mm. Figure 5 shows Raman scattering spectra obtain at different positions along the reactor centerline over 3 ranges of wavenumber. It is seen that in addition to the strong band [Ga C 3 ] at 490 cm -1, peaks at 517, 537, 555, and 785 cm -1 are also detected, likely associated with reaction intermediates. In addition, Raman shifts around 3100 cm -1, probably representative of hydrocarbon products, were observed, however, assignment of these bands to specific motions was not made. The temperature at selected positions along the reactor centerline below the heated susceptor (750 C set point) was calculated from analysis of the N 2 rotational bands and the concentration of TEGa along the centerline was computed using the measured Raman cross-section and scattering intensities. Figure 6 shows the measured values of gas phase temperature and TEGa concentration along with predicted profiles using the 2- dimensional axisymmetric reactor model. This model uses the finite element Galerkin method (FEM) to solve the equations of change for heat, momentum and species simultaneously. It is noted that this model accounts for both axial and transverse dispersion. Figure 6. Simulated and experimental TEGa concentration and gas phase temperature profiles as a function of distance from the heated susceptor (750 C). The rate constants in the simulated concentration profile are estimated from DFT calculations. A more complete description of the model along with its validation is presented elsewhere (13,20). It is seen that the computed temperature profile (solid line) agrees well with the measured temperatures plotted in this figure. The error bar associated with each datum was calculated from the standard deviation of repeated experiments of the measured cross-section and scattering intensity. Figure 6 also compares the profile

7 47 calculated using the 2-D reactor model and computed first-order rate parameters estimated from DFT calculations as listed in Table I. Comparison the calculated profiles assuming only homolysis and only hydride elimination indicates decomposition occurs primarily by bond homolysis under the conditions of this experiment. The calculated results combining both reactions agree reasonably well with the experimental results. gives TEGa profile close Comparison of DFT Calculations with Experimental Results The B3LYP/LanL2DZ model chemistry suggests the band at 474 cm -1 was produced for symmetrical [Ga C 3 ] stretching. Comparison of the DFT calculations with the experimental results yielded a scaling factor of Table III shows the calculated and corrected Raman frequencies assigned to Ga C symmetric motions in species selected from comparison of 39 potential reactions. As shown in Table III, the three measured Raman shifts (517 cm -1, 537 cm -1, 555 cm -1 ) agree well with the computed shifts for three of the screened molecules. In particular the two corrected frequencies for (DEGa) 2 match very well with the measured values. In addition, the computed frequency for (Et)GaH GaH 2 (555 cm -1 ) is very close the computed value (554.7 cm -1 ). Since this band is also positioned close to the value for (DEGa) 2, it is difficult to conclude that both (DEGa) 2 and (Et)GaH GaH 2 are present since the relative Raman cross-sections of the two molecules are not known. The peak located at 537 cm -1 is very close to the calculated frequency for (Et)GaH Ga(Et) 2 (534.3 cm -1 ). In summary these results suggest the presence of (DEGa) 2 and (Et)GaH Ga(Et) 2 and possibly (Et)GaH Ga(Et) 2. Given that the dimerization of DEGa yields (DEGa) 2, and DEGa is produced by homolysis of TEGa, it appears the homolysis pathway is operative in spite of the larger energy barrier for homolysis as compared to that of the hydride elimination route. Moreover, the presence of (Et)GaH Ga(Et) 2, and possibly (Et)GaH GaH 2, supports hydride elimination is also occurring. These results substantiate but do not confirm that both homolysis and hydride elimination reactions take place during the thermal decomposition of TEGa under the conditions of this study, which are typical of CVD conditions. TABLE III. Calculated and Corrected (scaling factor of ) Raman Active Ga C Stretching Frequencies for Symmetric Motions Calculated Experimental Corrected TEGa DEGa MEGa (Et) 2 GaH (Et)GaH 2 (Et)GaH Ga(Et) 2 (Et)GaH GaH 2 DEGa CH 3 (DEGa) s s m m m m m w m m m s s (MEGa) 3 (MEGa) 4 (Lowercases of s, m and w attached to the calculated frequencies mean strong, medium and weak intensity, respectively)

8 48 Conclusions Using Raman spectroscopy to follow the decomposition kinetics of TEGa in N 2 in a custom up-flow, cold-wall CVD reactor, shifts of 490 cm -1, 517 cm -1, 537 cm -1, and 555 cm -1 were observed and tentatively assigned to specific vibrational modes. The results are consistent but not confirming that both hydride elimination and homolysis of the Ga-C bond are accessible initial reaction steps. This conclusion was experimentally supported by evidence of Raman bands assigned to (DEGa) 2 and (Et)GaH Ga(Et) 2, and possibly (Et)GaH GaH 2. In addition reaction enthalpy estimates from DFT calculations using the B3LYP/LanL2DZ chemistry level are consistent with this conclusion. The concentration profile for TEGa and the temperature distribution along the reactor centerline were measured and compared to those from 2-D simulations using the finite element Galerkin method. The consistency between experiment, simulation, and DFT calculations suggests this methodology is useful for study of thermal decomposition of metal-organic precursors. References 1. J. H. Hendricks, M. I. Aquino, J. E. Maslar and M. R. Zachariah, Chem. Mater., 10, 2221 (1998). 2. M. S. Tomar, R. Rutherford, C. New and K. A. Kuenhold,. Sol. Energy Mater. Sol. Cells, 63, 437 (2000). 3. C. A. Wang, J. Cryst. Growth, 272, 664 (2004). 4. M. C. Paputa and S. J. W. Price, Can. J. Chem., 57, 3178 (1979). 5. Z. Mahmood, I. Hussain, R. E. Linney and D. K. Russel, J. Anal. Appl. Pyrolysis, 44, 29 (1997). 6. D. K. Russell, G. P. Mills, J. B. Raynor and A. D. Workman, Chem. Vap. Deposition, 4, 61 (1998). 7. V. A. Yablokov and N. V. Yablokova, Russ. Chem. Rev., 64, 953 (1995). 8. A. S. Grady, R. E. Linney, R. D. Markwellt and D. K. Russell, J. Mater. Chem., 3, 483 (1993). 9. M. T. Tsuda, S. Oikawa, M. Morishita and M. Mashita, Jpn. J. Appl. Phys., 26, L564 (1987). 10. J. A. McCaulley, R. J. Shul and V. M. Donnelly, J. Vac. Sci. Technol. A 9, 2872 (1991). 11. K. C. Wong, B. G. McBurnett, R. D. Culp, A. H. Cowley and J. G. Ekerdt, Surf. Sci., 416, 480 (1998). 12. Y. S. Kim, Y. S. Won, H. Hagelin-Weaver, N. Omenetto and T. J. Anderson, J. Phys. Chem. A, 112, 4246 (2008). 13. J. Y. Hwang, C. H. Park, M. Huang and T. J. Anderson, J. Cryst. Growth, 279, 521 (2005). 14. Y. S. Kim, Ph. D. Thesis, University of Florida (2007). 15. M. Huang, Ph. D. Thesis, University of Florida (2003). 16. J. Y. Hwang, Ph. D. Thesis, University of Florida (2004). 17. M. J. Frisch, G. W. Trucks, H. B. Schlegel et al. Gaussian 03, Revision B.04 (2004). 18. N. W. Mitzel, C. Lustig, R. J. F. Berger and N. Runeberg, Angew. Chem. Int. Ed., 41, 2519 (2002). 19. R. Lin, T. R. Gow, A. L. Backman, L. A. Cadwell, F. Lee and R. I. Masel, J. Vac.

9 49 Sci. Technol., B, 7, 725 (1989). 20. C. H. Park, J. Y. Hwang,M. Huang and T. J. Anderson, Thin Solid Films, 409, 88 (2002).