Boron removal from silicon melt by gas blowing technique

3.1 Refining experiments

In this research, all the refining experiments were carried out in a vacuum induction furnace with the setup configuration presented in Figure 3. As shown in the figure, Si was melted in graphite (high density, with the properties presented by Hoseinpur and Safarian [53]) or alumina sintered (ALSINT) crucibles. The crucible used for holding material was put in a bigger graphite crucible and a thermocouple type C (W – 6% Re, W – 26% Re, protected by an alumina sheath) was placed in-between the crucibles to measure the temperature of the process. The preliminary experiments with two thermocouples, one in the inner crucible and the second one in-between the two crucibles, indicated that there is only a 2–4°C temperature difference, and hence the gas refining experiments were carried out with the thermocouple placed in-between the two crucibles. The inner crucible was charged by 213 g of Si, with a mixture of 50 wt% of polysilicon (FBR^®, 8N purity) and 50% Silgrain^® (HQ – micron cut; 0.04 wt% Fe, 0.09 wt% Al, 0.013 wt% Ca, 0.001 wt% Ti, 0.085 wt% C, 25 ppmw P, and 30 ppmw B). This mixture provides about 15–20 ppmw B impurity in the initial melt. Before the experiments, the chamber was vacuumed down to 5–7 Pa and flushed by Argon (6N) or Helium (6N) for 3 times. Subsequently, the power was switched on and after the material was melted, a sample was taken from the melt to record the initial composition of the melt. Then, the refining process was started by blowing the refining gas over the Si melt surface, as shown in Figure 3. Table 2 presents the experimental conditions applied for various experiments in this research. The refining gas flow was adjusted by mass flow controller during the experiment and the gas was blown over the melt surface through a quartz lance with a 2 mm nozzle and the nozzle distance to melt surface was kept as 30 mm in all the experiments. In those experiments that humidified hydrogen, which was used as the refining gas, the hydrogen flow was redirected to a gas humidifier unit and then was humidified with 3% H₂O. In order to study the effect of bulk atmosphere in the furnace, in one experiment the chamber was filled with He to compare the results with the experiments where Ar was used to fill the chamber. In addition, in another experiment, the gas refining in vacuum conditions was also studied by blowing the refining gas over the melt surface while the chamber was being vacuumed continuously. In this special experiment, the pressure in the chamber was almost 5 mbar while carrying out the gas refining. Then, the gas blowing was started and several samples were taken from the melt during the refining process to track the B concertation change over time. These samples were taken by quartz tubes and later were digested in a mixture of HF and HNO₃ acids, subsequently characterized by inductively coupled plasma mass spectrometry (ICP-MS; Agilent 8800 Triple Quad). When the experiments were done, we shut down the power and let the crucible to cool down by itself. Then, some samples were taken from the fumes settled on the chamber’s wall to be characterized by scanning electron microscopy (SEM).

Figure 3

The schematic of the furnace and gas refining set up.

3.2 Molecular beam mass spectrometry (MBMS) characterizations

Hot gas analysis in this study was conducted using MBMS. A detailed description of the system used in this study is given by Wolf et al. [54]. For all the MBMS measurements in this study, the MBMS system has been coupled to a high-temperature reactor shown schematically in Figure 4. A sample boat made of graphite, alumina, or silica containing 2 g of a Si–B (350 ppmw) was attached to the end of an alumina rod and inserted into a tubular alumina reactor with an inner diameter of 21 mm, which was housed in a high-temperature furnace. Before running the experiment, the reactor chamber was flushed with Helium gas for 10 min to reduce the oxygen potential in the chamber, and then the furnace was switched on. The He flow to the chamber was maintained during the experiment. The furnace was maintained at a constant temperature of 1,500°C. The reactor was coupled to the sampling orifice of the MBMS device, to sample the high-temperature gases. The orifice was protruded into the furnace to maintain an elevated temperature to prevent condensation of gas phase species on the tip of the orifice. At the beginning of each experiment, the sample boat was held in the cooled zone of the reactor and a background spectrum was acquired for about 1 min. While the MBMS was kept in a constant scanning mode, the sample boat was inserted into the heated region of the reactor and the evaporated species were monitored over time. During experiments, 5% H₂ in He flowed through the reactor at a flow rate of 4 NL·min⁻¹. The residence time of released vapors in the reactor before sampling was about 0.1 s. Water steam was added after a few minutes via a vaporizer achieving humidity concentrations of 3–5% in the gas stream flowing to the reactor.

Figure 4

Setup used for vaporization experiments, from Wolf et al. [54].

Due to the relatively high gas flow necessary to minimize ambient air leaking into the reactor at the connection between furnace and MBMS, vaporization is unlikely to reach equilibrium. Therefore, the gas flow was stopped for about 20 s in some measurements to locally increase the concentration of vapor species above the sample boat. After switching on the gas again, high intensity peaks for qualitative analysis could be recorded. Because of this procedure, the present results are of a rather qualitative nature and therefore, released species were not quantified.

The mass-to-charge ratio (m/z) range of 5–100 was subjected to a preliminary scan via MBMS to determine the major compounds. The ions of interest and their corresponding m/z ratio are all mentioned in Table 3. It should be mentioned that either ions originate from gas molecules or fragmentation within the ionization region of the MS. For example, B⁺ can originate from any B containing gas molecule. Unfortunately, not all masses could be properly recorded due to superimposing of species with the same m/z originating from background or small amounts of ambient air. For example, N 2 + (m/z 28) superimposes BOH⁺ and Si⁺ on the same m/z. Furthermore, m/z with too high signal intensity, e.g., 44 (SiO⁺ and CO 2 + from background) had to be excluded to prevent an overload of the multiplier.

4.1 Rate of B removal in gas refining experiments

The B concertation in liquid Si was measured by ICP-MS and all the results are presented in Table A1 (in appendix section). To study the rate of B removal under experimental conditions, the first-order kinetic model was applied, presented here as follows:

where t denotes time in seconds, the [ B wt% ] 0 and [ B wt% ] t are the B concentrations in liquid Si at initial time and at time t , respectively. A/V is the surface to volume ratio of melt in (m⁻¹), and k B is the overall mass transfer coefficient of B removal in the experiments. The calculated k B for various experimental conditions are presented in Table 4.

4.2 MBMS measurements

The results from the MBMS measurements are all presented in Figure 5. This figure represents the gaseous species that were detected in the gas phase when having liquid Si in quartz, graphite, and alumina boats. In Figure 5, the intensity of the detected species in each sample is normalized based on the sharpest peak. Figure 5(a) shows the B species in He – 5% H₂ gas stream without any humidity added to the gas. As mentioned before, the sample was inserted into the chamber after 10 min of He – 5% H₂ flushing and hence, it is expected to have oxygen partially present in the chamber. From Figure 5(a) it is clear that the major B species detected in all the samples are BH _x compounds. However, when comparing the graphite and quartz boats, it is clear that there are more B _x O _y compounds with higher intensities in the case of quartz boat. As can be seen in Figure 5(a), in the case of quartz boat, the BO⁺ ₂ compound had the second highest intensity after BH₂. From Figure 5(a) it is obvious that when alumina boat is applied, the new AlBO⁺ compound is detected by MBMS, which indicates the positive role of Si melt interaction with alumina leading to the formation of new volatile B compounds. In addition to all the B _x H _z , B _x H _y O _z , and B _x O _y compounds, the B⁺ ion is obvious in Figure 5(a). It is worth mentioning that B has a very low vapor pressure [31] and the direct evaporation of B from Si is not assumable. The authors have already studied the vacuum evaporation for Si having P and B concertation of about 10–15 ppmw in the initial melt, and they never detected any B evaporation even in vacuum conditions. Figure 5 shows that the BH _x compounds have the highest intensities, while the thermodynamic calculations indicated that these compounds have higher Gibbs free energy than the other B containing species. Hence, we believe that the B⁺ and BH _x ⁺ ions detected in all cases are mainly the results of fragmentation of bigger molecules in the ionization chamber of MBMS. In addition, Figure 5(b) depicts the detected gaseous species in the gas phase when humidity (3–5%) was added to the gas stream. As it is obvious from this figure that many of the B _x O _y peaks (in case of quartz boat) and the AlBO⁺ peak (in case of the alumina boat) have vanished or have lost their intensities. In the experiment with the graphite and quartz boats, it can be seen that when the humidity was added, the intensity of the compound HBOH⁺ was increased in both cases, but the HBOH⁺ compound was detected with higher intensities in the alumina case. HBOH⁺ is detected as a new compound in this study and previously was proved only to exist by theoretical calculations [55]. It is previously discussed and shown that the concertation of O ̲ is lower in graphite crucibles compared to alumina crucibles due to the formation of CO_(g), and this can explain the higher intensities of the HBOH⁺ compound detected when quartz boat was used [31,56].

Figure 5

The measured species in MBMS results: (a) He – 5% H₂, (b) He – 5% H₂ humidified with 3–5% H₂O. (Detected oxide compounds in (a) is due to partial pressures of oxygen remaining in the chamber).

The B removal from Si melt takes place by formation of HBO compound. We can assume the formation of HBOH in gas refining through the following reaction:

However, the thermodynamics calculations presented in Figure 1 indicated that there are other molecules than HBO and HBOH having considerable negative values of Gibbs energy, such as B(OH)₃, HB(OH)₂, and B₂O₃, but none of these compounds were detected in the MBMS measurements. This could be due to the need for several elements to reach together at the melt surface and form the aforementioned molecules, which reduces the formation chance of these molecules.

Considering the discussions presented in Section 1 and here, the mechanisms of B removal from liquid Si with H₂–H₂O gases is schematically summarized in Figure 6. This figure shows that an important step in the process is the dissolution of the H ̲ and O ̲ in the liquid Si from gas phase. When quartz and alumina crucibles are applied, O ̲ and Al ̲ can also be dissolved from the crucibles, while in case of graphite crucibles, C ̲ will be dissolved from graphite crucible. Formation of solid silicon oxide on melt surface is also obvious from this illustration.

Figure 6

An illustration of the gas refining process in the quartz/alumina (left) and graphite crucibles (right), summarizing the B removal mechanisms.

4.3 Effect of crucible interactions with melt on B removal

A comparison between the experiments (1) with (2) shows that when dry hydrogen was used as the refining gas, almost, no B removal happened from the liquid Si. However, with the addition of humidity to hydrogen (H₂–3% H₂O), the rate of B removal increased from 0.9 to 13 µm·s⁻¹ which indicates the important role of oxidation reactions on B removal from liquid Si. This indicates that the B removal mainly takes place through the formation of B _x H _z O _y species and not through the BH _z compounds, and these results are in good agreement with the findings of Nordstrand and Tangstad [25] and Sortland and Tangstad [13].

From Table 4, it is obvious that the k _B in the experiment (5) is greater than the experiment (1). Both experiments were carried out at 1,500°C and with dry H_2(g), but in alumina and graphite crucibles, respectively. These results are in good agreement with the MBMS measurements where we showed that in the case of applying alumina crucibles new volatile compounds of B like AlBO⁺ evaporate from the melt surface, and hence the kinetics of the refining process could be accelerated in the alumina crucibles. When refining in graphite crucibles and with hydrogen gas, we can assume the B removal with BH _z compounds, and when doing the refining process in the alumina crucibles, we can assume the removal in the form of B _x O _z , B _x O _z H _y , and AlBO _x compounds.

The effect of the temperature on B removal in the alumina crucibles could also be studied by comparing the results obtained from the experiments (5–7). It is obvious that an increase in temperature leads to an increased rate of the B removal and the value of k _B increases from 1.64 to 4.15 µm·s⁻¹ when the temperature is increased from 1,450 to 1,500°C, which is 2.5 times. However, when the temperature is increased to 1,600°C, the k _B equals 4.96 µm·s⁻¹. Then, beyond 1,500°C the temperature rise is no more effective for B removal. The effect of temperature is already discussed by Safarian et al. [30] indicating that when temperature increases beyond 1,500°C, silicon oxidation becomes more favorable than B oxidation reactions, leading to consumption of all the dissolved oxygen in the melt to form SiO_(g). In experiment (7), the humidified hydrogen (H₂–3% H₂O) was applied as the refining gas and the k _B value increased to 15.3 µm·s⁻¹. A comparison of the experiment (7) with experiment (8) makes it clear that when humidity is added to the refining gas, the rate of B removal has increased almost three times. This indicates that although the alumina crucibles can supply the dissolved Al ̲ and O ̲ to the melt, the oxygen dissolved from alumina is not enough, and an exterior oxygen source is required to perform the B removal from the liquid Si.

In addition, the Al dissolved from alumina crucibles was also measured and is shown in Figure 7. The following reaction can be suggested for the dissolution of Al from alumina crucible:

Figure 7 indicates that the rate of Al dissolution in the liquid Si increases with temperature and proves that when melting Si in alumina boats and crucibles, there is enough Al ̲ in the liquid to form the AlBO _x compounds.

Figure 7

Aluminum concertation in liquid Si over time of gas refining.

4.4 Effect of chamber gas atmosphere

By comparing the results of experiments (2–4), we can study the chamber bulk gas’ effect on B removal kinetics. From Table 4 it is evident that when the chamber bulk gas is changed to He, the kinetics of B removal has accelerated and the value of k _B has increased from 13 to 17.3 µm·s⁻¹, at the same temperature accounting for 33% increase in the process rate. The positive effect of He was already reported by Safarian et al. [28] and they showed that when H₂ – 4% H₂O is mixed with He, the mass transfer coefficient of the B removal process is higher than that when mixed with Ar. He has smaller molecules compared to Ar, with He and Ar having atomic radius of 0.49 and 0.88 Å, respectively. Then, assuming the same velocity for Ar and He, the momentum of Ar molecules will go higher. The Ar and He molecules will collide with the evaporated B species from the melt surface and the higher the momentum of the foreign molecule (Ar or He), the higher the chance for bouncing the B species molecules back to the melt surface. In addition, even when the B species are successfully evaporated, they should diffuse in the gas phase to take distance from melt surface and find their way out of the crucible, unless they may return to melt through a back reaction, and this slows down the overall process kinetics for B removal. Obtaining the diffusion coefficient of the gaseous B species in the gas phase is beyond the scope of this study, but by considering the diffusion coefficient relation for gas molecules presented by Chapman and Cowling [57], we can obtain a general view about the differences between He and Ar on the diffusion of the B species in the gas phase.

where D denotes the diffusion coefficient, suffixes 1 and 2 indicate gas molecule 1 and gas molecule 2, m is the mass of the molecules, and σ is the average radii of the species, σ ₁₂ = 0.5(σ ₁ + σ ₂). By assuming Ar and He as the molecule 1, and any gaseous B compound as molecule 2, then from equation (6) it is obvious that the higher the mass and diameter of the gas molecules, the lower the diffusion of the B species in the gas phase. Therefore, it is completely expectable for the same B species under study to have a higher diffusion in He than Ar. In order to accelerate the diffusion of the B species in the gas phase we carried out experiment (4); however, Table 4 indicates that when carrying out the gas refining in vacuum condition, the mass transfer coefficient for B removal has reduced to 2.56 µm·s⁻¹, which was totally opposite to our expectations. In this experiment, the chamber bulk gas was continuously vacuumed during the gas refining experiment. Figure 8 compares the melt surface in experiments (3 and 5). It is obvious from Figure 8(b) and (d) that the surface of the melt has fully impinged in the case of gas blowing in vacuum condition; however, when the chamber was in atmospheric pressure, there was no significant impinging effect on the melt surface (Figure 8[a] and [c]). Figure 8(b) and (d) also indicates that when doing the gas refining in vacuum condition there are less amounts of condensates settled on the lance and crucible compared to Ar atmosphere, and this will further be discussed in the next section.

Figure 8

Photographs of the crucible during the (vacuum) refining experiments. (a) and (c) Gas refining in Ar atmosphere. (b) and (d) Gas refining in vacuum.

It is worth mentioning that when carrying out the gas refining in vacuum conditions, there is almost no condensate on the lance and crucible edge. The formation of the condensates on the cold parts of the crucibles and gas lance provides practical challenges in the gas refining of Si. For example, as shown in Figure 8(a), after 60 min of the refining process, the condensates are grown from the crucible edge toward the center of the crucible, which leads to clogging the gas path toward out of the crucible, affecting the rate of the process. The fluid dynamics for melt impinging by gas blowing is already discussed and numerically simulated in refs. [58–60] and Figure 8(c) and (d) illustrates the gas fluid pattern when blowing in in gas and vacuum conditions, respectively. As shown in Figure 8(c), when the gas lance is blowing in atmospheric pressure conditions, the gas stream spreads over the melt surface. The exact fluid dynamic of the gas blowing under the experimental condition of experiment (3) is already simulated by Safarian et al. [30], and the Figure 8(c) and (d) is regenerated after their simulation results. Figure 8(d) shows the fluid pattern of gas blowing in the vacuum condition and indicates that the gas jet makes a fully impinged point on the melt surface. In the case of vacuum condition, there is less resistance due to the low pressure of the bulk gas in the chamber and this makes the velocity of the gas jet increase, leading to impinging the melt surface. As shown in Figure 8(d), when the gas jet impinges the melt surface, it splits and bounces back and then there is no further contact with the melt surface. As it is obvious in Figure 8(b), the surface area of the impinged region seems to be considerably smaller than the whole melt surface area and this means that under the conditions of experiment (5), the contact area of gas and melt is smaller than that in the other experiments. From the melt surface photograph presented in Figure 8(b), the radius of impinged point (cavity) is determined as r _cav = 0.00658 m. The impinging of melt surface with gas jets is already modeled in refs. [61,62] and here we can apply the following equations to calculate the depth of the cavity formed on the melt surface.

where r cav is the radius of the cavity or impinged area formed on the melt surface, and h lance is the distance of the lance tip to the melt surface. The M ̇ d , m ̇ n , M ̇ h , and m ̇ t are all dimensionless mass flow rates of the gas stream blowing out of the nozzle, and a full explanation for them could be found in the study by Koria and Lange [63]. n lance is the number of the nozzles on the gas lance tip, and it is one in this study, θ is the inclination angle of the nozzle with the lance axis, which is zero in our case, and ρ Si is the density of the liquid Si. All these parameters are shown schematically in Figure 9. By inserting the measured r _cav = 0.00658 m from Figure 8(b) in equation (7), M ̇ d will be obtained, then m ̇ t , m ̇ n , and M ̇ h will be obtained from equations (8–10), respectively. Finally, from equation (11), h cav will be calculated as 0.0298 m. The cavity with the obtained scaled dimensions is shown in Figure 9, and as seen it has a half ellipse shape. Having the dimensions of the ellipse, we can calculate the impinged area’s surface ( S impinged = 0.5 π h cav r cav ) as 2.17 × 10⁻⁴ m². Consequently, the effective A/V ratio is calculated as 2.713 m⁻¹. By correcting the A/V ratio in the first-order kinetic model for the experiment (4) carried out in vacuum condition, the effective k _B value is obtained as 23.3 µm·s⁻¹, which is 1.79 times of that when Ar was in the chamber, and 1.34 times of the case when He was in the chamber. Therefore, the effective k _B in vacuum condition shows that when the gas atmosphere in the chamber is removed by vacuuming, the rate for the vacuum refining process increases intensively. This indicates the important role of gas phase and is in good agreement with the result of the experiment (3), where He was used as chamber gas, providing higher diffusivities for the gas molecules.

Figure 9

The calculated geometry of the impinged point on melt surface in vacuum condition, the dimensions are presented in a right scale.

4.5 Silica fume formation in gas refining

During the gas refining experiments with humidified hydrogen, SiO_(g) forms as a product of the silicon oxidation process. The SiO_(g) can then react with the humidity to produce small solid particles of SiO₂ and create white dust on the furnace chamber, known as silica fumes. The formation of silica fumes from the SiO gas is well discussed in the literature [26], and it could be described through the overall reaction:

The morphology of the silica fumes settled on the chamber and lance surfaces in various experiments with chamber atmosphere of Ar, He, and vacuum conditions were studied by SEM and are presented in Figures 10 and 11. Figure 10 indicates a huge difference between the sizes of the fume particles in three different experimental conditions. When He and Ar were used as the chamber bulk gas, the fume particles, settled on the chamber wall, had spherical morphology, consisting of separate spheres or several spheres attached. In addition, it is obvious from Figure 10 that the fume particles are much bigger when the chamber was filled with He gas compared to Ar and vacuum condition. The fumes settled on the chamber wall in Ar and vacuum conditions have relatively smaller sizes than in the case of He. In the case of vacuum conditions, it can be seen that some of the fume particles have grown like a comet tail. Silica fume has applications in concrete production, and the change in the morphology and size of the particles could be of interest for further study.

Figure 10

The SEM micrographs of the fume settled on the chamber wall: (a) He gas in the chamber, (b) Ar gas in the chamber, and (c) chamber vacuumed during the gas refining process.

Figure 11

The SEM micrographs of the condensates settled on the gas lance: (a) and (b) He gas in the chamber, (c) and (d) Ar gas in the chamber, and (e) and (f) vacuum condition.

Figure 11(a and b) shows that the morphology of the fumes settled on the gas lance in case of the He gas is spherical but compared to the fume settled on chamber walls (Figure 10a) have considerably smaller particle sizes. However, in the case of Ar gas, some tubular morphologies could be seen among the other spheres. In the case of the vacuum condition, however, the morphology of the fumes settled on the lance is totally different, and the fume is grown in the form of whiskers and columnar morphologies. We did not find any spherical particle in the sample collected from the lance of the experiment with vacuum condition, while the fume settled on the chamber wall was spherical.

Figure 12 represents the various mechanisms for the formation of silica fumes. As it is obvious from this figure, silica fumes could form in the gas phase without any preferential nucleation site or on the body of the lance, with a preferential growth direction. When forming in gas phase, small seeds could be formed in the gas and then growing equiaxially leading to the formation of spheres. However, if the silica fume forms by initiation on a preferential nucleation site, like the lance body, then a directional growth will form the columnar morphologies and the whisker.

Figure 12

The schematic illustration of the silica fume formation in gas refining of Si. (a) The equiaxed growth and (b) nucleation on surface.

Considering the nucleation and growth mechanisms shown in Figure 12, the differences in the morphologies detected in the fumes could be explained. The larger sizes of the spherical particles detected in the case of He gas (in the sample collected from the chamber wall) are in good agreement with the previous discussion about the higher diffusivities of gaseous species in He compared to Ar. Having higher diffusivity, the gaseous species (SiO and H₂O) will reach the surface of the seeds faster. This leads the formation of SiO₂ seeds shown in Figure 10(a) to grow larger, before settling on the chamber wall. In the vacuum condition, however, there is lower gas density above the melt, since the chamber is being vacuumed continuously and the pressure is in the range of 5–25 mbar. Then, the seeds formed in the gas phase on top of the melt will immediately reach the chamber wall, where they settle down. Then, similar to the case of Ar atmosphere, the silica fumes in vacuum condition will have smaller sizes. In this case, further growth on the spherical particles settled on the chamber can take place, leading to the comet tail morphologies detected in Figure 10, and schematically shown in Figure 12. In addition, when doing the vacuum refining in the vacuum condition, the velocity of the gas jet flowing out of the nozzle increases intensively (Figure 8(d)). When the gas jet impinges the melt surface and bounces back, it still has high velocity and hence will carry all the silica seeds away from the melt surface toward the chamber wall. However, the continuous gas stream over the outer surface of the gas lance provides the required gaseous reactants (SiO and H₂O) for the formation of the silica whiskers and the columns on the gas lance. The photograph of the fumes settled on the chamber wall in the two conditions, vacuum and Ar atmosphere, are shown Figure 13. As shown on this figure, the fumes collected from the experiment carried out in vacuum condition are fluffy, while in the case of Ar the fume is a fine powder.

Figure 13

The photograph of the fumes collected from the chamber after gas refining. (a) In vacuum condition and (b) chamber filled with Ar gas.