2016 年 57 巻 11 号 p. 1930-1935
For many years the production of solar-grade silicon remained a costly process resulting in a large amount of carbon gas being emitted, and so the process still requires improvement to suppress carbon emission. The starting point of the processes is to produce raw silicon materials from a natural resource via mostly carbothermic reduction. However, this process is very complicated and SiO and SiC form as by-products. Further improvement of the carbothermic reduction process requires an understanding in real time of the reactions occurring and the weight change during heating. In particular, the behavior of the SiO by-product plays a major role in the production of silicon because the loss of Si is caused by the escape of SiO gas. In this study, we developed an in-situ weight measuring system for our induction heating furnace, and successfully suppressed much of the error in weight by improving the crucible configuration. The real-time monitoring of the crucible weight loss during the reaction may assist in understanding of the carbothermic reduction process in a more detailed fashion.
The carbothermic reduction process using an electric furnace is still the most popular way to produce solar-grade silicon, and this process was commercialized 100 years ago. Silica from gravel and stones and carbon from charcoal, wood chips, coal, and coke, were used as raw materials in the silicon production process which requires high temperatures and much energy. The by-products of reaction process determine the yields of the reduction1). The ability of reduction materials to react with SiO gas, an intermediate product in the silicon process, has been a subject of attention during the last few decades. This SiO-reactivity affects the loss of SiO, that is, the silicon yield, and the specific energy consumption and the productivity of the furnace. Several attempts have been made to develop relevant methods to measure the influence of the reduction materials on the performance of the silicon process. The development of the SINTEF SiO-reactivity test was reported2–4). Also, petrographic analysis has been carried out to understand how the microstructure of coal and coke influences the SiO-reactivity, among others5).
For the understanding of the phase diagram of carbothermic silica reduction process, estimation of SiO and CO gas partial pressure is inevitable. The in-situ analysis of the high-temperature gasses containing CO, CO2, H2O and SiO was reported6). On the other hand, thermogravimetric analysis was used to study the kinetics of reactions in pellets, which were made of carbon black and silica7). However, the configurations of the crucibles used in the previous works were far from that of the real reduction process of Si. The real reduction furnace is designed as semi-closed crucible because the completely open crucible cannot increase the partial pressure of SiO gas. Also, a too small crucible cannot make the silicon because the crucible wall strongly absorbs the SiO gas. The most interesting information to understand is the gas configuration related to the reaction in semi-closed crucibles. Therefore, the thermogravimetric analysis in the reducing furnace which can make silicon serves the understanding of the complicated silica reduction process.
In this study, we developed a real-time measuring system to monitor the weight loss with the quadrupole mass (Q-mass) analyzer for the CO gas on the realistic reduction process. This system enables us to estimate the behavior of SiO gas.
The heating process was carried out by an induction heating furnace of 30 kW power (Toei Scientific Industrial Co., Ltd) as shown in Fig. 1. The evacuation system of the apparatus is composed of a rotary pump and a diffusion pump, which can reach to 10−3 Pa in a vacuum. During the heating experiment, the top temperature of the crucible was monitored by a high sensitive color type infrared thermometer with a temperature range from 650℃ (923 K) to 3500℃ (3773 K) through the top glass window of the chamber, and a quadrupole mass spectrometer analyzed the chamber gas. The induction coil heated the crucible with a frequency of 30 kHz under an open-loop control, and a quartz tube was placed in between the crucible and the coil for the protection of the crucible.
Image of the induction heating furnace used for the reduction process.
The eqs. (1)–(5) below, show the overall reaction in silicon production, which requires knowledge about the reaction at high-temperature zones8).
| \[{\rm SiO}_2({\rm s}) + {\rm C}({\rm s}) = {\rm SiO}({\rm g}) + {\rm CO}({\rm g})\] | (1) |
| \[\Delta G^\circ = 668.07 - 0.3288T({\rm kJ})(1273\,{\rm K} \le T \le 2273\,{\rm K})\] |
| \[2{\rm SiO}_2({\rm s}) + {\rm SiC}({\rm s}) = 3{\rm SiO}({\rm g}) + {\rm CO}({\rm g})\] | (2) |
| \[\Delta G^\circ = 1415 - 0.6586T({\rm kJ})(1273\,{\rm K} \le T \le 2273\,{\rm K})\] |
| \[{\rm SiO}({\rm g}) + 2{\rm C}({\rm s}) = {\rm SiC}({\rm s}) + {\rm CO}({\rm g})\] | (3) |
| \[\Delta G^\circ = -78.89 + 0.0010T({\rm kJ})(1273\,{\rm K} \le T \le 2273\,{\rm K})\] |
| \[{\rm SiO}({\rm g}) + {\rm SiC}({\rm s}) = 2{\rm Si}({\rm s}) + {\rm CO}({\rm g})\] | (4) |
| \[\Delta G^\circ = 165.56 - 0.0751T({\rm kJ})(1673\,{\rm K} \le T \le 2273\,{\rm K})\] |
| \[{\rm SiO}_2({\rm l}) + {\rm Si}({\rm l}) = 2{\rm SiO}({\rm g})\] | (5) |
| \[\Delta G^\circ = 616.37 - 0.2875T({\rm kJ})(1673\,{\rm K} \le T \le 2273\,{\rm K})\] |
In these reactions, the gas phases are only SiO and CO gasses. Q-mass spectrometer can detect only CO gas because of the mass spectrometry analysis in this study was performed at room temperature which is an impossible condition to detect SiO gas phase due to its low stability below 1300℃ (1573 K)9). The gas phase of SiO cannot reach the Q-mass analyzer because the Q-mass analyzer and the chamber are connected through a long metallic tube (inner diameter 0.711 mm) with an orifice.
The atmosphere during the heating process was pure argon gas (99.999%) with a pressure of 0.07 MPa to avoid the leak of the lethal carbon monoxide gas. The evacuation of the chamber to a higher vacuum order is required before the filing of the argon gas because the mass peaks between nitrogen and carbon monoxide are overlapped completely. The total pressure inside the chamber was recorded manually from the Bourdon gauge connected to the chamber.
2.2 In-situ weight measuring systemThe original (conventional) high-purity graphite crucible (inner diameter 40 mm, height 70 mm) setup as shown in Fig. 2 was used as a basic design for the development of the weight sensing system. The primary configuration (A) based on the conversion of the electric signal variation given by a strain gauge (PW4M-500g) into weight. However, the direct connection of the crucible and the sensor caused substantial perturbations on the weight sensing system. Two phenomena, the thermal radiation from the crucible and the electromagnetic effect generated by the coil, can be listed as the origins of the perturbations causing the error on the real-time weight estimation. To suppress these perturbations, an interconnection between the crucible holder and the sensor is required via a light material to avoid damaging the sensor sensitivity. Two light materials with different electric conductivity, carbon (2.8 × 104 S/m) as shown in Fig. 2 setup (B1) and aluminum (3.5 × 107 S/m) as for setup (B2) were used as an interconnection.
The schematic figures of the crucible design, shape and material difference between the original setup, the primary (setup A), the improved one (B1) and the actual used (setup B2).
The blank test comparison of the weight variation between the first design of the sensing system (setup A) and the improved design (setup B2) showed a large suppression of the error (by ten times) on the weight changing measurement as illustrated in Fig. 3. This improvement is due to the addition of the interconnection aluminum which played the role of a shield (magnetic barrier) against the electromagnetic waves from the coil in setup (B2), comparable to the carbon in setup (B1) which showed a significant margin of error. Figure 4 shows the weight error as a function of the temperature on the sensor measured by a thermocouple. This curve was obtained by a blank test with the variation of the sensor temperature using a heat gun. In the case of the setup (A), the temperature on the weight sensor increased to 120℃ (375 K) during the heating, then the thermal effect was estimated roughly to be 9.34 g from Fig. 4. The ratio of the thermal effect of 27% was calculated from the ratio of 9.34 g/34 g, while the electromagnetic effect causes the remaining 73%.
(a) Temperature measured by the infrared thermometer during the blank tests process. Below the 650℃, the infrared thermometer cannot measure the temperature. The heating starts from the room temperature, (b) the error comparison on the blank tests between the primary setup (A) and the actual setup (B2).
Effect of the sensor temperature variation on the error of the weight sensor.
A mixture of silica (diameter 20～100 um, Taiheiyo Cement Corporation Japan) and glassy carbon (diameter 20 um, Tokai Carbon, Ltd), with an additional amount of silicon carbide were loaded alternatively as layers into the crucible setup (B2). The amount of the raw materials used is shown in Table 1, and was optimized based on the ideal chemical reduction reaction of silica with carbon, while silicon carbide was added to increase the reaction speed to get silicon. Therefore, the mixture molar ratio between silica, carbon and silicon carbide used was SiO2:C:SiC = 1.5:1:0.749). In this study, we focused on the generation of SiO gas reactions described in eqs. (1) and (2), on the viewpoint of low error on the weight sensor and small number of simple reactions. For the generation of silicon in the carbothermic reduction, the reaction temperature of more than 1500℃ (1773 K) and enough SiO gas are required.
| Weight | Si (mol) | |||
|---|---|---|---|---|
| (g) | mass% | |||
| Inputs | SiO2 | 12.3 | 68.71 | 0.205 |
| C | 1.6 | 8.93 | ||
| SiC | 4 | 22.34 | 0.1 | |
| Total Input | 17.9 | 100 | 0.305 | |
| Product | SiO2 | 6.62 | 71.95 | 0.11 |
| SiC | 2.57 | 28.05 | 0.064 | |
| Subtotal | 9.2 | 51 | 0.174 | |
| Lost Gas | SiO | 5.8 | 66.67 | 0.131 |
| CO | 2.9 | 33.33 | ||
| Subtotal | 8.7 | 49 | 0.131 | |
| Total Output | 17.9 | 100 | 0.305 | |
Phase stability diagrams for the reactions in eqs. (1) and (2) as illustrated in Fig. 5 (a) and (b) were made by the standard Gibbs energy of the two reactions taken from MALT210). The direction of the reactions in eqs. (1) and (2) depend on the sign of the Gibbs energy for each reaction. Silica reacts with carbon in eq. (1) over 1054℃ (1327 K) to generate SiO gas which is shown as the first gray color in Fig. 5 (a), while silicon carbide starts to react with carbon in eq. (2) at 1070℃ (1343 K) to generate SiO corresponding to the first gray color in Fig. 5 (b). Colorless area below 1054℃ (1327 K) corresponds to no reaction in Fig. 5 (a) and (b) due to the negligible partial pressure of SiO(g) and CO(g) due to the presence of solid silica and carbon11,12). In our system, the partial pressure ratio of SiO and CO gasses was between 0.01 to 0.1 with a total pressure PT = PSiO + PCO from 0.032 to 0.056 atm, the temperature across the ratio and the total pressure is shown as a dashed line in Fig. 5 (a) and (b) corresponding to 1320℃ (1593 K). Moreover, the change in the Gibbs energy of the reaction in eq. (1) suggests that most likely, SiO gas is generated through a solid-solid reaction in the presence of direct contact between silica and carbon13). Based on the phase diagram stability, the crucible was heated at 1320℃ (1593 K) for 25 min.
(a) Thermodynamic calculations of phase stability diagram of SiO/CO gas phases: (a) for reaction in eq. (1) and (b) for reaction in eq. (2). This diagram affected by temperature, partial pressure ratio of SiO/CO and total pressure PT = PSiO + PCO, calculated with data from MALT2.
As for the reactions in eqs. (3), (4) and (5), silicon carbide is generated via eq. (3) with the interaction between SiO gas and carbon, and thermodynamically the reaction proceeds even at low temperature but experimentally it is difficult because its required rich SiO gas pressure generated from the previous reactions14). Si is generated over 1500℃ (1773 K) via eq. (4) by the contact between SiO gas and SiC based on the standard Gibbs energy data of the reaction in eq. (4)15).
Table 1 represents the mass balance, including the amount of input raw material, the product output and the weight loss as gas phase. The loss of the product was 49% and was calculated subtracting the total weight of the product from the total weight of the reactants. The mass molar balance for each reactant and product was calculated by the weight fraction in mole. The temperature profile curves of the reduction are illustrated in Fig. 6(a) left axis; the thermometer cannot measure the target temperature below 650℃ (923 K). The crucible was heated at a constant heating rate of 171.2℃/min until reaching 1320℃ (1593 K) and was stabilized on it for 15 min. The sample starts to react at a rate of 0.16 g/min, 16% of weight loss was observed until 1250℃ (1523 K), above that temperature, the reaction rate increased quickly to 0.44 g/min given a 21% weight loss when reaching a temperature of 1300℃ (1573 K). From 1300℃ (1573 K) to 1320℃ (1593 K), a slow reaction rate of 0.2 g/min concludes to a weight loss of 12%. The total sample weight loss was about 49% as shown in the Fig. 6 (b) right axis, and no weight loss precipitation or adherence on the inside walls of the crucible was observed.
(a) The temperature curves of the reduction process, (b) the relative mass peak intensity (m/Z = 28) and the signal of the weight sensor during the heating process.
Figure 7 shows the x-ray diffraction pattern of the product. Several peaks mapped to silica and silicon carbide were obtained. No peak mapped to Si was observed. This result suggests that the reactions in eqs. (1) and (2) proceeded around 1320℃ (1593 K). Therefore, only the amounts of SiO2 and SiC in the product were shown in Table 1.
X-ray diffraction patterns of the product.
Figure 6 (b) left axis, shows the temporal change of the relative intensity m/Z = 28, which is mainly generated from the carbothermic reduction equation of silica. The changes of the both curves have started at almost the same time by the reaction between the silica and carbon. The weight loss caused by the escape of SiO and CO gasses was mostly proportional to the increase of the CO gas. The amount of the CO emission of the atmosphere in the chamber can be calculated by the ratio of CO and Ar gas and the total pressure. Then, the weight loss as SiO gas can be estimated, because the weight loss should consist only of SiO and CO gasses. In this study, the ratio of SiO and CO gasses was estimated to be 1.3:1 calculated using eqs. (6) and (7).
| \[M({\rm CO})_{\rm mol} = P_{\rm chamber} * k * V_{\rm chamber}/R * T\] | (6) |
| \[M({\rm SiO})_{\rm mol} = (W_{\rm Loss} - M({\rm CO})_{\rm mol} * 28)/44\] | (7) |
In the reaction of eq. (1), SiO and CO gasses should be generated with the molar ratio of SiO:CO = 1:1, while, in the reaction of eq. (2), they should be generated with the molar ratio of SiO:CO = 3:1, while our obtained ratio of SiO:CO = 1.3:1 is placed in between the two previous ratios. On the viewpoint of this, our estimation of mass balance is reasonable.
Figure 8 shows the comparison of the quadrupole mass spectra before and after the heating. In the case of m/Z = 28 which refers to N2 or CO, before heating its correspond to N2 gas, a substantial increase in the intensity after heating because of CO gas which is mainly generated from the carbothermic reactions. As for m/Z = 20, m/Z = 36, m/Z = 38 and m/Z = 40 which correspond to Ar gas, their intensities were quite stable because of the chamber enclosure. In the case of m/Z = 17, m/Z = 18 which map to H2O and m/Z = 32 to O2 gas, a slight decrease in their intensities due to the absorption of water or oxygen gas on the surface of raw material powder, crucible walls, and chamber walls.
The mass spectra of the chamber gasses before and after heating.
Thus, the analysis of the CO gas emission and the weight loss of the reaction crucible must be quite a powerful method in order to understand the veiled reactions in the crucibles under the carbothermic reduction of silica.
We successfully integrated and developed the in situ real-time monitoring system for our induction heating furnace. The development required the changing of the crucible setup design, shape and materials which lead us to minimize the error on the real-time weight change. The experimental test of the weight sensing system showed a reasonable relationship between SiO and CO gasses on the silicon production process. Further error-suppression of the weight monitoring system is required for understanding all of the complicated reactions of the carbothermic silica reduction.
This research was supported by JST-JICA, SATREPS. R. Benioub, A. Boucetta, A. Chahtou and M. Heddadj gratefully acknowledge the scholarship from Ministry of Education, Culture, Sport, Science, and Technology (MEXT) of Japan.
