Threshold ionization mass spectrometry in the presence of excited silane radicals

T Moiseev¹, D Chrastina¹, G Isella¹ and C Cavallotti²

¹ Politecnico di Milano, Reg. di Como, L-NESS, Via Anzani 42, Como, 22100 (CO), Italy
² Politecnico di Milano, Dipt. Chimica, Materiali e Ing. Chimica, `G. Natta', Via Mancinelli 7, Milano, 20131 (MI), Italy

E-mail: tamara.moiseev@como.polimi.it

Received 28 November 2008, in final form 19 January 2009
Published 13 March 2009

1. Introduction

Threshold ionization mass spectrometry (TIMS) can be an important tool for optimization of plasma-assisted chemical vapour deposition (PECVD) processes in the transition zone from amorphous to crystalline layers for nano-crystalline silicon (nc-Si) deposition in photovoltaic applications.

PECVD processes generate silane radicals on electronically or ro-vibrationally excited states, such species having much larger ionization cross-sections with lower thresholds [1]. While the densities of excited radicals are generally low, their increased ionization cross-sections lead to overestimations in TIMS analysis. An improved TIMS method is proposed in order to discriminate the excited radical signal from the ground-state one when the analysis indicates the presence of excited silane radicals.

This study observes TIMS relative densities of ground-state silane radicals and their evolution with hydrogen input in the low-energy plasma-enhanced chemical vapour deposition (LEPECVD) [2, 3] plasma used for the deposition of nc-Si layers at low temperature (200–250 °C) and fast growth rates (1.3–3.5 nm s^–1) required in solar cell applications. The improved analysis allows the evaluation of silane radicals (SiH_x; x < 4) over an extended range of hydrogen flow rates, revealing that their density trends in the process region for nc-Si deposition [4] have a turning point where the most abundant radical is SiH instead of SiH₃.

2. Experimental set-up

The LEPECVD process is based on a low energy hot filament dc arc discharge (50 A) at 20–30 V in argon–silane–hydrogen at total pressures of 0.01–0.04 mbar. The argon gas inlet and the hot filament are situated at the base of the reactor. The filament heated by 130 A ac emits electrons confined by an axial magnetic field (B ≤ 1.2 mT) generated by coils surrounding the reactor and reach the wafer region where silane and hydrogen gas are added by a ring-shaped inlet. The wafer is biased at floating potential (–12 V to –14 V) and its temperature is regulated at 200 °C. Reactor base pressures are in the UHV range (10^–9 mbar).

A HIDEN analytical EQP300 quadrupole mass spectrometer has been used for TIMS analysis. The electrostatic quadrupole plasma (EQP) probe operated at floating potential is positioned at the wafer level. TIMS data have been acquired at 3 cm from the wafer edge. A pressure below 8 × 10^–7 mbar is maintained inside the spectrometer by a Œrlikon TMP361 turbo pump. The measurements were performed using a 300 µm sampling orifice, time-integration (dwell time) 3000 ms, RGA filament current of 80 µA and scanning electron energy in the range from 7 to 20 eV. The 80 µA filament emission current has been chosen below the values leading to the formation of space-charge [5] inside the ionizer (which here occurs for emission currents starting at 100 µA). The electron energy scale calibration has been performed by the linear extrapolation method using a pure Ar discharge. An off-set of about  + 0.5 eV has been observed and the energy scale has been corrected accordingly.

3. The TIMS method

TIMS analysis of radical densities is based on selective ionization of each radical using electron energies between the ionization potential (IP) of the given radical (threshold E_I) and the threshold E_DI for dissociative ionization (DI) towards the same radical starting from silane. The relative density of species can be determined from the formula [6, 7]:

where I, ΔI are the number of counts at each energy (above the DI threshold and, respectively, between the I and DI thresholds) σ_I and σ_DI the cross-sections for ionization of the given neutral radical (A) and DI of parent molecule (AB) towards A. The energy interval for integration is chosen so that E_I < E '₁ – E''₁ < E_DI. The net radical signal from the plasma is

where I_ON and I_OFF are the signals during plasma on and off and I(A ⁺ )_ions is the ion background signal given by ions generated by the discharge (during plasma on).

The ion current has been measured at both a zero emission filament current (I_f  =  0 µA) and at I_f  =  1 µA and subtracted according to (2). The measurements at I_f  =  1 µA serve as a reference level in this work as they provide an active control over the filament temperature (and emission) during ion background measurements and there are negligible differences from values with I_f  =  0 µA in the energy region below the DI thresholds. No other background signal was observed for discharge off, filament off (I_f  =  0 µA) while any noise signal during plasma on and I_f  =  0 µA (or I_f  =  1 µA) is included in the measured ion background.

The coefficient C, which describes the residence time of silane inside the spectrometer [8], was estimated for a pumping speed S  =  300 ± 30 l s^–1, 300 K gas temperature and a distance l  =  4.5 cm between the ionizer and the sampling orifice, at C  =  10 ± 2. Lower pumping speeds, higher gas temperatures or kinetic energies of the radicals will lead to higher C values, especially for lower mass radicals. To avoid overestimations at lower mass radicals we have considered C  =  10 throughout the calculations.

The cross-sections proposed as analytical fit functions by Janev and Reiter [9] have been used as they provide partial cross-sections with corrections for the energy dependent branching ratios between reaction thresholds and errors within 20% accuracy of the experimental data.

4. Improved TIMS analysis

The main non-systematic errors of the TIMS method come from the choice of the E '₁ – E''₁ range limits for integration. The upper limit can be chosen at the DI threshold (as the DI cross-section at the threshold is too low to give significant counts) while the choice of the lower limit is related to background signal discrimination.

Counting ions produced inside the ionizer at electron energies close to the IP of the radical, relating low counting rates to very low cross-sections, leads to large errors in the estimation of radical densities. These can be avoided by choosing E '₁ at a higher energy where ionization cross-sections have larger values. The signal lost by shortening the energy interval corresponds to regions with low ionization cross-sections leading to a low and random signal that is not a relevant quantity.

A significant source of errors is the presence of excited radicals (molecular species on vibrational/rotational or electronically excited levels) generated mainly through electron impact excitation or ion–electron recombination, some leading to highly excited (Rydberg states) long-lived molecules [10, 11]. Although the presence of excited molecules in process plasmas has been observed and their influence on the evolution of process parameters is expected to be significant, there is a lack of data concerning their cross-sections [12] and recent studies show that such species have much larger ionization cross-sections with thresholds shifted towards lower energies.

Typical electron energy scans are shown in figures 1 and 2, where the signals for plasma on (ON, I_f  =  1 µA) plasma off (OFF) corrected for silane depletion (OFF corrected) and ion background (ON, I_f  =  1 µA) are indicated on each graph. The ion background signal is shown for a filament current of I_f  =  1 µA instead of I_f  =  0 to emphasize the excess signal between the total radical signal for plasma on and the measured ion background. A strong signal above the ion background but below the IP thresholds (IP  =  8.01 eV for SiH₃ and IP  =  7.91 eV for SiH) [9] can be observed, this situation being common for all radicals and gas mixtures. Such a signal can be ascribed to excited radicals that have high ionization cross-sections with thresholds at much lower energies. Integrating the signal between the ion background (ON, I_f  =  1 µA) and the one for `plasma on' (ON I_f  =  80 µA), as required by formula (2), leads to radical density overestimation of orders of magnitude, exceeding total silane density.

The semi-log signal for plasma on (figures 1 and 2) shows a clear change in slope situated 1–1.5 eV above the IP thresholds indicating the occurrence (dominance) of another process (the ionization of the ground-state radicals) superimposed on the signal given by ionization of excited radicals that reaches saturation (the constant signal level below the IP threshold). This observation provides a means to set the lower integration limit for the estimation of ground-state radicals.

In this work the influence of excited species on the overall count is minimized by considering the ion background level as the count level at the above change in the signal slope, thus including the excited radical signal in the ion background. This is justified as cross-sections of excited radicals are much larger than those of ground-state radicals and the number of counts that can be ascribed to excited species corresponds to a low density of such species.

The considered interval for signal integration is indicated by the hashed area having as limits the energy corresponding to the first change in slope (lower limit), the DI threshold towards the given radical from SiH₄ (upper limit) and as ion background the count level at the slope change (indicated as `considered ion level'). The lower limit of integration changes with hydrogen input (within 0.5–1.5 eV above IP threshold) depending on the density ratio of excited to ground-state species and had to be assessed for each case.

Starting the integration at higher energies is advantageous in lowering the error levels from both the discussed points of view: the signal is counted at energies well above the IP threshold where ionization cross-sections are higher and a steady increasing signal above the one from the excited radicals allows for discrimination.

Detection efficiency is affected by the surface reactivity of silane radicals. Corrections involving sticking coefficients lead to an increase in the estimated relative densities mainly for SiH and Si radicals (with sticking coefficients close to 100%). Such corrections have not been applied here explicitly. Choosing silane as a reference gas and calculating its density from the count levels of each radical includes corrections for residence times and sticking coefficients which should be similar when monitoring the same radical during either discharge on or off as required by the method.

5. Results and discussion

TIMS measurements have been performed for SiH_x (x < 4) at two silane flow rates (10 and 20 sccm) and a range of hydrogen flow rates (0–50 sccm) leading to a silane dilution in hydrogen in the range from 1 to 0.17. At each silane–hydrogen gas mixture TIMS analysis was performed for the silane radicals considering only the main DI reactions starting from SiH₄, using the reduced electron energy intervals for each radical as shown previously to improve the error level for measurements in the presence of excited radicals. The obtained relative density trends of SiH_x (x < 4) radicals with the addition of hydrogen are shown in figures 3 and 4 for silane flow rates of 10 and 20 sccm. The error bars correspond to the relative difference between radical densities reported to the silane level (during off time) calculated from the DI signal at 17 eV and at 20 eV. If the σ_DI energy dependence is accurate, the deviation between radical densities calculated with readings at two energy values estimates the error due to measurement.

The observed density of Si neutral radical is very small, below 0.005% for all situations, with large errors due to low counting rates as a result of its fast recombination rate in the gas phase and its high reactivity to surfaces, significant loss being expected as deposition inside the spectrometer before reaching the ionizer.

At low hydrogen input (0–20 sccm) increasing density trends of the SiH_x (1 < x < 4) radicals can be observed for both 20 sccm and 10 sccm silane input. These can be ascribed to a combined effect of electron impact dissociations and hydrogen–silane gas-phase reactions under the increasing density of free hydrogen atoms with the addition of hydrogen. Above 20 sccm hydrogen input the two charts show significant differences in trends, the SiH radical becoming the dominant radical for the 10 sccm silane flow rate (figure 4).

While TIMS analysis reported by other authors often did not allow for direct estimations of SiH density, the presence of SiH in rf PECVD plasmas is known and used for OES monitoring of the μc-Si deposition [13–16]. As observed by Perrin [17], the high OES signal obtained for SiH^* makes this radical a non-negligible participant in the deposition process.

Previously reported [4] Raman crystallinity of nc-Si samples grown by LEPECVD show a strong increase in crystalline fraction with increase in silane dilution, a transition zone between low and high crystallinity being obtained for at d  =  0.3–0.25, corresponding to the observed (figure 4) turning point in radical relative densities, where SiH is dominant.

The formation of lower silane radicals (SiH, Si) can take place at low silane input (allowing for high silane depletion) due to either electron impact dissociation of the higher silane radicals (SiH₃, SiH₂) into lower ones (SiH, Si) or by H abstraction from SiH_x leading to SiH_x–1 (0 < x < 4) at high hydrogen input. Therefore, the requirements for high silane depletion and high silane dilution observed in the optimization of micro-crystalline silicon (μc-Si) growth [18–21] are also met here, leading to the depletion of higher silane radicals towards the lower ones.

6. Conclusions

The improved TIMS method allows for the observation of the four mono-silane radicals (SiH₃, SiH₂, SiH and Si) and reveals the trends in their relative densities with the addition of hydrogen in argon–silane–hydrogen plasmas.

For the LEPECVD process, a turning point in SiH_x (x < 4) relative densities is obtained in the process region for nc-Si deposition, where the most abundant radical is no longer SiH₃ but the SiH radical, the latter maintaining the highest densities with further increase in hydrogen input.

This study shows that significant changes in plasma composition occur at high silane dilution in hydrogen, leading to increased generation of SiH which becomes the dominant radical, suggesting that SiH is a main precursor in the growth of nc-Si layers by LEPECVD and that the changes in plasma composition are the main factor leading to nc-Si growth.

Acknowledgments

This work has been funded by the European Commission, Nanophoto Project Contract no 013944. The authors are grateful to Professor Hans von Känel from ETH Zürich for providing new lab equipment for use on this project.

[1] 

Cristophorou L G and Olthoff J K 2004 Fundamental Electron Interactions with Plasma Processing Gases (Berlin: Springer)  
CrossRef

[2] 

Kummer M, Rosenblad C, Dommann A, Hackbarth T, Höck G, Zeuner M, Muller E and von Känel H 2002 Mater. Sci. Eng. B 89 288–95 
CrossRef

[3] 

Pizzini S et al 2006 Mater. Sci. Eng. B 134 118 
CrossRef

[4] 

Le Donne A, Binetti S, Isella G, Pichaud B, Texier M, Acciarri M and Pizzini S 2008 Appl. Surf. Sci. 254 2804–8 
CrossRef

[5] 

Agarwal S, Quax G W W, van de Sanden M C M, Maroudas D and Aydil E S 2004 J. Vac. Sci. Technol. A 22.1 71–81 
CrossRef

[6] 

Robertson R, Hills D, Chatam H and Gallagher A 1983 Appl. Phys. Lett. 43 544–6 
CrossRef

[7] 

Kae-Nune P, Perrin J, Guillon J and Jolly J 1994 Japan. J. Appl. Phys. 33 4303–7 
CrossRef

[8] 

Kae-Nune P, Perrin J, Guillon J and Jolly J 1995 Plasma Sources Sci. Technol. 4 250–9 
IOPscience

[9] 

Janev R K and Reiter D 2003 Contrib. Plasma Phys. 43 401–17 
CrossRef

[10] 

1983 Rydberg States of Atoms and Molecules edStebbings R FDunning F B (Cambridge University Press: Cambridge)  

[11] 

Nagesha K and Pinnaduwage L A 1999 Chem. Phys. Lett. 312 19–27 
CrossRef

[12] 

Cristophorou L G and Olthoff J K 2002 Appl. Surf. Sci. 192 309–26 
CrossRef

[13] 

Amanatides E, Mataras D, Rapakoulias D, van den Donker M N and Rech B 2005 Sol. Energy Mater. Sol. Cells 87 765–805 
CrossRef

[14] 

Howling A A, Strahm B, P. Colsters P, Sansonnens L and Hollenstein Ch 2007 Plasma Sources Sci. Technol. 16 679–96 
IOPscience

[15] 

Wu Z, Sun J, Lei Q, Zhao Y, Geng X and Xi L 2006 Physica E 33 125–9 
CrossRef

[16] 

Strahm B, Howling A A and Hollenstein Ch 2007 Plasma Phys. Control. Fusion 49 B411–8 
IOPscience

[17] 

Perrin J, Leroy O and Bordage M C 1996 Contrib. Plasma Phys. 36 3–49 
CrossRef

[18] 

Strahm B, Howling A A, Sansonnens L and Hollenstein Ch 2007 J. Vac. Sci. Technol. A 25 1198–202 
CrossRef

[19] 

Strahm B, Howling A A, Sansonnens L and Hollenstein Ch 2007 Plasma Sources Sci. Technol. 16 80–9 
IOPscience

[20] 

Kondo M, Fukawa M, Guo M L and Matsuda A 2000 J. Non-Cryst. Solids 266–269 84–9 
CrossRef

[21] 

Matsuda A 2004 J. Non-Cryst. Solids 338–340 1–12 
CrossRef