1

Cost Reduction in Polysilicon

manufacturing for Photovoltaics.

By: H.S.Gopala Krishna Murthy, Ph.D.,

Director, ShanGo Technologies Private Limited, Bengaluru 560085, India. Mobile: 091 9449850875.

e-mail: hsgkmurthy@gmail.com, hsgkm@rediffmail.com.

Today, polysilicon price is drastically coming down. What was being sold at more than US $ 400 per kg is today readily available at less than US $ 60 per kg. There is every possibility of the price still coming down to US $ 35 to 50 per kg in the near future. The steep fall in price of polysilicon is due to the market supply-demand. A year ago, it was in short supply and hence, the price was very high. Customers had to pay hefty advances to manufacturers and wait for supply of the material. However, the recent recession, reduction in governmental programs and subsidies and almost complete collapse of the banking systems in America and Europe resulting in no finances for new projects, regular manufacturers and new entrants coming with new and higher capacities turned the table and led to the collapse of the price. Though fall in the price of polysilicon and its ready availability is good for the PV Industry, it is a serious problem for the manufacturers particularly those who have entered in to the manufacturing recently. Most of the old manufacturers have well established processes and plants, fully integrated facilities where the by-products of Polysilicon manufacture like silicon tetrachloride could be profitably used for making other products and disposal of wastes generated is well established. Further, they have increased their capacities by using funds advanced by customers and hence, the high capital cost of polysilicon manufacturing does not really affect them financially. However, the new entrants do not have these advantages. For most of them, the cost of manufacture of polysilicon is high of the order of about US $ 60 to 100. With the current low pricing for polysilicon they are facing serious financial problems and unless they evolve cost cutting methods for silicon manufacture, they may become bankrupt. There are many reasons why polysilicon is expensive to make.

1. The technology for manufacture of polysilicon is not readily available. 2. The capital cost for a polysilicon plant is very high. 3. The capacity of the plant is very small compared to the capital required. 4. Production is basically batch process and hence labour intensive. 5. Large quantity of electricity is required for manufacture. 6. The purity of the material produced is more than required for solar grade. 7. Considerable quantities of by-products and wastes are generated

2

We shall discuss these points and suggest ways of reducing the costs.

Polysilicon manufacture.

The process of polysilicon manufacture by the Siemens process is now well known and will not be described in detail here. Briefly, this consists of producing trichlorosilane (TCS) in a fluidized bed reactor using metallurgical silicon and hydrogen chloride gas. TCS is purified by fractional distillation. It is vapourised and fed along with hydrogen gas in to

a special chemical vapour deposition (CVD) reactor in which hot inverted U shaped filaments are kept at about 1100 C by resistive heating with electrical energy. Silicon deposits on the hot filaments which grow from a small diameter of about 8 mm to larger sizes of 125 mm and higher. When sufficient deposition has taken place, the process is stopped, and the material deposited is taken out. The effluents from the reactor consisting of unreacted TCS and hydrogen along with by-products silicon tetrachloride (STC), dichlorosilane (DCS), HCl and small quantities of polysilanes is subjected to

POLYSILICON PROCESS

Met. Si + HCl TCS (SiHCl3) + STC (SiCl4)

DISTILLATION PURIFICATION OF TCS

POLYSILICON REACTOR

ELECTRICITY

REACTOR EFFLUENT RECYCLING

WASTE HCl / SiO2 DISPOSAL

STC HYDRO

GENATION

H2 RECYCLE

FRESH H2

3

recovery and recycle of the feeds and byproducts. One would notice that manufacturing of polysilicon is a chemical process and involves many of the unit operations of the chemical industry. The only unique feature of the process is the CVD reactor with its special design. Sometimes, a Converter is also used to convert STC produced during deposition to TCS. While these two are unique to polysilicon manufacture all the others are regular equipment used in chemical processes. Fluidized bed reactor for trichlorosilane manufacture cannot be considered to be unique because fluidized bed technology is well known and is being extensively used in the chemical industry. Hence, if the capital cost of polysilicon plant is high it should be mostly because of these two equipment. Hence, one needs to look in to why these two capital equipment are expensive and how their costs could be reduced. One other reason why silicon from the Siemens Polysilicon process is expensive is historical. Polysilicon was originally developed for the semiconductor industry which needed extremely high purity silicon with impurities present at sub part per billion levels. Moreover, the cost of silicon in the final product namely the electronic device like a chip is very small—less than 1% of the overall cost. Hence, the prime goal of the polysilicon manufacture was purity rather than the cost. Therefore, processes and equipment to meet this important criterion were developed. Further, because of the limited market for polysilicon and because of the peculiar market situation, there were only a handful of manufacturers who were more concerned with purity than cost. Also, there was no real competition in marketing the product. In the same way, equipment fabricators were also limited and were able to get high prices. Hence, from the beginning till recently there have been no serious attempts to reduce the cost of manufacture or cost of equipment. Since for photovoltaics, rejects from semiconductors were available at low prices, there was no pressure on the cost of silicon from this sector. Though many new entries have been made in the last four years in technology licensing, capital equipment manufacture, and production facilities, because of the great demand for silicon and the high price prevailing, there was a rush in to the arena by many new entrants some of whom knew very little of the business or technology and were ready to invest huge capital for new plants as the cost of the final product was very lucrative. The order booking with the equipment supplier was high and delivery times were extended. Hence, the equipment suppliers could negotiate high prices for capital equipment. Now with the Photovoltaic industry blooming, alternatives for silicon showing their teeth and market transformed from sellers to buyers leading to the price tumbling down like nine pins an urgent need to look in to all aspects of the cost of polysilicon manufacture has come up. The same situation prevails in the down stream operations of ingot casting and wafering. These two sectors will be dealt with separately elsewhere. The quality of silicon required for PV is now clearly known. The high purity material made using the Siemens process is not really required for PV. Particularly, the levels of boron and phosphorus, the two crucial impurities can be at about 0.1 to 1 ppm which is about three to four orders higher than needed for semiconductor applications. Hence, there are many possibilities of reduction in the cost of manufacture and also new ways in which silicon could be made. Realisation that silicon required for PV could have purities levels of six 9’s particularly with boron and phosphorus being at less than 1 ppm and

4

metal impurities at even higher levels a detailed review of the Siemens process and possible ways to reduce the cost of manufacture is essential. We shall study these possibilities further.

Basic features of the Siemens Polysilicon reactor. In order to understand the possible means of reducing the costs of the Siemens reactor, it is essential to know the basic features of the polysilicon reactor. Further, as mentioned earlier, the polysilicon reactor is unique to the process and all the factors contributing to the high cost of polysilicon could be attributed to the reactor design and operation. A

Figure 1 schematic of the Siemens reactor is provided in Figure 1. The Siemens reactor consists of a pressure vessel made with a base plate and a bell jar. These two are hermitically sealed and can operate at high pressures generally about 6 bar. The base plate is provided

5

with ports for feed gas and exhaust gas and electrical feed-throughs. Inverted U shaped hairpins of silicon slim rods are mounted on the electrical feed-throughs their number depending on the size of the reactor. Electrical power source is provided to heat the slim rods to the working temperature of about 1100 C by resistive heating. The bell jar is provided with view ports for measuring the temperature of the hairpins using non-contact temperature gauges. Both the base plate and the bell jar are provided with jackets through which cold water is circulated to remove the radiant heat from the hairpins. The base plate is placed on a firm pedestal. After assembling the hairpins, the bell jar is covered over them and using suitable gaskets the bell jar is sealed on the base plate. The hairpins are brought to the working temperature using complicated and sophisticated electrical power supply. A mixture of trichlorosilane and hydrogen gas is fed through the inlet feed nozzle. The exhaust gases are removed through the exhaust port. Typical polysilicon reactors have diameters of about 0.8 to 1.5 meter and heights of about 1 to 2.5 meter and accommodate 9, 18, 24 or higher number of hairpins. Production capacities of the reactors could vary from 25 to 250 tons of polysilicon per annum. During deposition, the reactor will have high temperature environment with corrosive species like HCl, TCS and STC. The materials of construction of the various parts of the reactor are selected such that they are kept cool by water-cooling or even if they are heated, they do not impart any impurities in to the deposited silicon. As highly hazardous chemicals like hydrogen and trichlorosilane are present inside the reactor at high pressure and temperature, the design of the reactor should take in to consideration, the safety aspects of the process. The reduction of TCS takes place by a complex thermodynamically and kinetically controlled reaction and silicon deposits on the hot hairpins which grow in diameter. The design of the reactor should be such that the deposition is smooth and uniform over the entire length of the hairpins and is free of cracks, pores, popcorn structures and other defects. The polysilicon reactor can be subdivided broadly in to three parts, namely,

• The reactor;

• Power supply; and

• Process control. Let us study each of these and examine the scope of reduction in cost.

Polysilicon Bell Jar Reactor. As stated earlier, the Siemens reactor is a pressure vessel operating at high pressures of up to 30 bars typical operating pressure being about 6 bar. As the base plate is flat and the reactor diameter exceeds one meter, to withstand the operating pressure, the base plate has to be very thick which could be more than 100 mm for a diameter of one meter. As the inner surface of the base plate is exposed to direct radiation from the hairpins, it will have a high skin temperature. This will lead to several problems. The first one is possible reaction of the base plate material with the reactive environment of HCl and chlorosilane leading to incorporation of impurities in to the deposited silicon. For the

6

same reason, the surface reacts with the chlorosilane to produce silicides of the metals in the base plate. On opening the reactor, the silicides layers will peal off due to differential thermal expansion as the silicides are brittle. Thus, the base plate is corroded in each run and may have to be replaced soon. To avoid this, extensive cooling paths have to be provided in the base plate which makes it mechanically weak. Even then since there is a thermal gradient from the top of the base plate to the bottom, thermal stresses are generated which lead to deformation of the base plate. To overcome these problems special grades of steel have to be used for the fabrication of the base plate. The cost of the material and the cost of fabrication are both high leading to high cost of the base plate. Even then, the base plate has to be replaced sooner or later. This problem can be solved in a simple manner. Any one who has done mechanical design of pressure vessel will use a dish end to seal a cylindrical portion of a vessel. In fact in the bell jar, the top end is a dish closure. If the same dish could be used in the base plate, then the thickness required for withstanding the operating pressure will drastically come down. Indeed, the thickness required for operating a one meter diameter reactor at 6 bar will be hardly 6 mm! For safety reasons, a 10 or 12 mm thick dish could be used. Forming such dish ends is routine in pressure vessel fabrication. Moreover since thin plates are needed for fabrication, better material of construction can be used. We had made such dish bottoms for the polysilicon reactors in the early nineties using Nickel 201 as the material. This is a much better material compared to say stainless steel of 304 or 316 grades. Ports required for the inlets and outlets and electrode feed-through ports can be easily made in the dish using fabrication standards like ASTM or others. Figure 2 depicts a reactor with

Figure 2 a dish bottom. Suitable modifications will have to be made to fabricate electrode holders and graphite holders taking in to account the dish shape of the base plate. We have operated such dish base plates for more than a decade and the performance has been much superior compared to the conventional flat base plates. The cost of manufacture comes down drastically which is the icing on the cake! The top bell jar is fabricated like a jacketed vessel and does not need any special features other than the normal fabricating procedures. With this knowledge, a polysilicon manufacturer can get a polysilicon

7

reactor from a capable pressure vessel fabricator at a price which would be considerably lower than from the present vendors.

Reactor Power Supply In the Siemens reactor, silicon slim rods are generally used for making hair-pin assemblies. While the material produced using these slim rods is of excellent quality, the cost of manufacture of the slim rods, the cost of slim rod pullers/sawing equipment for this purpose will add to the cost of capital and manufacturing. Further assembling these slim rods in to hairpins is cumbersome considering that the slim rods would be long and are very fragile and they have to be bridged at the top using special silicon strips with suitable jointing means. Apart from these mechanical difficulties in using silicon slim rods, there is a greater disadvantage. Silicon is a bad conductor at room temperature and will not pass electrical current easily. To bring the slim rod to the working temperature it is to be heated by means of electrical energy. For this purpose, there are two possibilities: one is to apply very high voltage so that the high resistance of the slim rods assembly is overcome and the rods are heated. Since silicon has a high negative coefficient or resistance, as the temperature of the rods increases, the resistance of the slim rod assembly comes down drastically. The voltage applied should be reduced very rapidly to ensure that the rod temperature does not go above the melting point of silicon and the slim rod melts. At the working temperature, the resistance of the assembly will be low and hence low voltage power supply would be sufficient. Further, as deposition takes place, the resistance comes down further but as the surface area from which heat is radiated also increases, increased current has to be supplied to sustain the working temperature. Thus, the power supply needed would have to provide initially very high voltage of the order of a few thousand volts, which should be quickly reduced to a low voltage and then the current is increased from less than 100 amps to more than 2000 to 3000 amps depending on the diameter reached. This cannot be achieved by a single power supply; there will be two supplies, one for initial heating and the other for the growth phase of the silicon deposited. The electrical feed-throughs should be designed to withstand the high voltage applied and also to ensure that there is no arcing between two slim rods or between the slim rods and the base plate. With the limitation in availability of materials of construction, this will pose serious challenges for the designer of the reactor power supply. The general practice therefore is to reduce the initial voltage required for heating by preheating the slim rods by suitable means to about 400 to 600 C so that the voltage required for forcing the current is brought down considerably. However, the slim rod assembly is inside a water cooled bell jar reactor, heating from external heaters is not possible. Different techniques have been developed for this purpose. In one such, a quartz enveloped heater is introduced from the top of the bell jar through an appropriate flanged opening. The heater is switched on with an inert gas atmosphere in side the bell jar. When the slim rods are heated to reasonable temperature, the initiation of heating is started with the high voltage power supply. When the slim rods attain the working

8

temperature, the heater is removed, the flange closed and the inert gas replaced by hydrogen and polysilicon deposition is started. This step is obviously cumbersome. In a different technique, heaters of graphite are provided at suitable positions in the base plate and these are heated by a simple power supply and the slim rods are heated by receiving radiant heat from the graphite heaters. In yet another technique, plasma heating is done by using suitable plasma electrodes located in side the bell jar. The power supply will therefore have three parts; one is the supply required for the external heater or plasma torch, the second one is a high voltage supply for triggering the slim rod; and the third one is for regular operations. As the resistance, voltage and current of the slim rod assembly are highly sensitive to the temperature of the slim rods, the controls should be capable of fast switching from one to another and controlling the temperature has to be very carefully monitored. Hence, the power supply would be highly sophisticated and hence, very expensive. For producing semiconductor grade silicon, necessarily silicon slim rods have to be used to ensure the purity of the produced silicon. However, as we have seen earlier, this purity is not required for photovoltaic silicon. Hence, there is possibility of using other slim rods for deposition which will very much simplify the starting process of the reactor and the power supply to the reactor will also be come much simpler and of much lower cost. This is not a new concept. In fact, Theurer of Bell Telephone Labs had deposited silicon using a tantalum strip. Rogers had taken patents on using tungsten rods as slim rods. Many polysilicon manufacturers used to deposit polysilicon on graphite tubes and other shapes to produce silicon ware for diffusion furnaces. However, such efforts were not exploited because silicon at that time was being produced for semiconductors and the purity of silicon deposited with tungsten was not really good enough. Moreover, when tungsten or other material is used, the produced silicon has to be necessarily broken to smaller pieces to remove the core of the metal which means long rods of silicon required for the float zone crystal growth could not be made. Further, removal of the core needed labour and hence, this was not favoured. In today’s changed circumstances, this procedure would be acceptable for reducing the capital cost. In countries like China and India where labour is cheap, the cost of removal of the core would not be an important issue. Moreover, for directional solidification even if some core material is present in the silicon, this will not matter as the metal impurity will be thrown away in to the discarded portion. The present author has operated such polysilicon reactors for more than a decade and has produced polysilicon which has been regularly used for growing single crystal of silicon and further wafered for making solar cells. The efficiency of solar cells made with such material has been found to be no different from silicon produced in the conventional way. An extensive study on life time of minority carriers in the wafers has proved that the quality of the wafers is excellent. Several advantages of the metal hair pins become apparent:

• The power supply will become very simple. As the metals are good conductors, they can be heated with low voltage supply. Hence, the complications of pre\-

9

heating, high voltage triggering, changeover to regular power supply will not be required. The cost of the power supply will thus be much smaller.

• Making of the hairpins and assembly will be easy. As tungsten is sufficiently strong, there is no fear of breakage.

• The number of electrical feed-throughs could be drastically reduced by interlinking many hairpins inside the reactor itself thus improving the integrity of the base plate. Since high voltage is not required, insulation of the feed-through can also be simple.

• If there is a power failure in the middle of a process, it is possible to restart the reactor because of the conducting core. Such a possibility cannot be thought in the case of silicon slim rod assemblies as bringing the grown rods to working temperature again would be almost impossible.

• No investments are needed for slim rod pullers or diamond wire saws for cutting slim rods. Thus, considerable savings in capital costs as well as labour can be achieved.

Process control: Thermodynamic Factors The CVD process taking place in the Siemens reactor is complicated. A large number of species are available in the high temperature–high pressure process even though hydrogen and trichlorosilane are the only two feeds to the reactor. The prominent among them are TCS, STC, DCS, HCl, H2 and SiCl2 at high temperature. Extensive studies have been made on the thermodynamic equilibrium in the Si-H-Cl system by many groups. Though such studies are theoretical and are based on minimization of the free energy of the system under the given conditions of pressure, temperature and composition, they throw valuable light on the actual process. From such studies, the optimum temperature, pressure and feed composition at which the maximum deposition of silicon can take place can be found out. In Figure 3 Si deposited is plotted against temperature for 3 different mole ratios of H2:TCS. It will be noticed that the maximum deposition takes place at about 1160 C when the mole ratio of H2:TCS is 1:1. This increases to 1180 C when the mole ratio is 1:0.5. However when the mole ratio is 1:0.1, the deposited quantity increases with temperature. Further, the graph shows that even at as low a temperature as 400 C some silicon is deposited. This is not observed in practice because of other factors.

10

Si Deposited Vs Temperature C

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

0.20

400 600 800 1000 1200 1400 1600

Temperature C

Si

De

po

sit

ed

kg

mo

l

H2: TCS 1:1 kg mol

H2: TCS1:0.5 kg mol

H2: TCS1:0.1 kg mol

1160 C

1180 C

Figure 3

In practical deposition situations, the optimum operating temperature range has been fixed between 1100 and 1150 C. In Figure 4 the effect of increase in pressure against efficiency of deposition is depicted. It may be noted that as the pressure increases, conversion efficiency decreases. However, for safe operation of the reactors and also facilitate treatment of effluents from the reactors, generally, the reactors are operated at about 5 to 6 bar. One important effect of pressure which is not clear from the above figure is that with increased pressure, the nutrient present near the filaments increases linearly with pressure so that though there is a marginal decrease in efficiency of deposition, since the amount of TCS present is higher, the deposition rate increases with pressure. It may be noted that when the SiemensC process was developed, the reactor made with fused quartz was operating at atmospheric pressure. When the size of the reactors had to be increased, the metal bell jar concept was introduced. As the metal bell jar could be operated at high pressure, it was noticed that the deposition rate increased tremendously reducing the energy require for deposition and increasing the capacities of the running plants several folds.

Effect of Pressure on Si Deposition Efficiency

SiHCl3:H2=1:1, 1100 C

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0 10 20 30 40 50

Pressure bar

Sp

ecie

s k

g m

ol

SiCl4 SiHCl3

Efficiency

HCl

SiH2Cl2

SiCl2

Figure 4

11

Conversion efficiency above 6 bar pressure will not increase much but the pressure rating of the reactor and the downstream recovery equipment will have to be increased from 150 ASA to 300 ASA or higher rating which leads to increased cost of equipment not commensurate with the marginally increased deposition rate. In addition to predicting the conversion efficiency of the process for a given feed condition thermodynamic equilibrium studies will also throw light on how much of TCS is converted to STC which is a by-product of the process. Though in principle TCS fed alone to the reactor can produce silicon as shown in the Figure 5 below, this is not done because feeding vapourised TCS in to the reactor is not easy.

Equilibrium Concentrations for decomposition of

TCS Alone at 6 bar pressure

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

400 600 800 1000 1200 1400 1600

Temperature C

kg

mo

ls o

f s

pe

cie

s.

SiCl4

H2

SiHCl3

Si

SiCl2

HCl

TCS/STC

SiH2Cl2

Figure 5 Further, extensive formation of STC and dangerous polysilanes by polymerization of SiCl2 will result. Thus, always TCS is fed along with hydrogen. How much of each is fed depends on the site conditions. A composition of 1:1 TCS: H2 will produce more STC as shown in Figure. 6

TCS:H2=1:1, Pressure 6 bar

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600

Temperature

Sp

ecie

s K

g m

ol SiHCl3

SiCl4

Si

HClSiCl2

SiH2Cl2

Figure 6

12

However, recovery efficiency in the down stream operations will be better as will be explained further. Higher hydrogen content increases conversion efficiency, reduces formation of STC and ensures smooth deposition as shown in Figure 7.

Effect of increased H2 on Equilibrium and

efficiency

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1.40

1.60

0 10 20 30 40 50

Hydrogen Kg mol

Specie

s k

g m

ol

HCl

EfficiencySiHCl3

SiCl4SiCl2 SiH2Cl2

Figure 7 However, the rate of deposition will decrease, the sizing of the down stream recovery section will increase and the recovery efficiency reduces. This is for three reasons. The first is the active silicon precursor namely TCS concentration decreases in the feed. The feed gas is now richer in H2 and the actual quantity of TCS available for deposition at any given time and space is small. Secondly space velocity in the reactor increases for a given feed rate of TCS. This reduces conversion efficiency though the quality of the deposit improves. Lastly, as hydrogen is a non condensable gas, when the effluent gas is compressed and cooled to low temperature, it does not condense while the chlorosilanes condense to the extent of their equilibrium vapour pressures at the process condition. The remaining un-condensed chlorosilanes escape along with hydrogen. Depending on the subsequent treatment of this stream, the chlorosilanes may end up as silica gel or fumed silica. Thus, the overall requirement of TCS for silicon deposition increases with increase in the hydrogen concentration. If STC finds use for producing pyrogenic fumed silica, then the formation of STC may not be an issue. However if STC has to be disposed of or has to be converted back to TCS by hydrogenation (which is an expensive process) then one has to carefully optimise process parameters depending on the site conditions.

Process Control: Kinetic Factors Though thermodynamic studies can predict the efficiency of conversion of trichlorosilane to silicon, this is what obtains under equilibrium condition. However, the reactor is operating in a dynamic condition of flow of species and the rate of deposition will depend on the kinetics of the process. Since CVD of silicon is a heterogeneous process involving deposition of solid silicon from a gaseous phase, the rate of deposition depends very much on the fluid dynamics in the reactor for a given composition of the feed. Silicon

13

atoms to be deposited have to move from the gas phase to the solid surface. The reaction takes place on the surface of the solid. Gas will be moving over the surface of the stationary solid surface. The layer of gas in immediate contact with the surface will also be stationary or moves at very low velocity compared to the bulk gas. This layer of gas on the surface is called the boundary layer. Silicon molecules from the bulk will move through this layer by diffusion and react on the surface. The gaseous products of the reaction also have to move from the surface through the boundary layer to the bulk. Hence, the rate of deposition will depend on the thickness of this boundary layer. The boundary layer thickness can be reduced by increasing the turbulence in the gas phase. Turbulence is dependent on many factors like:

• Size of the reactor.

• Feed rate which depends on the total design of the plant, number of reactors, the capacity of the down stream exhaust handling system, etc.

• Design, numbers and location of the feed and exhaust nozzles.

• Temperature gradient between the wall of the reactor and the slim rods.

• The numbers and relative location of the hairpins in the reactor.

• The size of the deposited rods.

• Final Diameter of the rods.

Let us consider the influence of these factors on the performance of the reactor.

Size of the reactor: As the size of the reactor increases, the diameter and height increase. Thus the cross section through which the gas flows increases. To keep a certain minimum space velocity which will generate sufficient turbulence, the feed rate has to be increased. This is dealt with in the next paragraph.

Feed rate: Since the reactor is of a finite size, the space velocity in the reactor is dependent on the feed rate. Increased feed rate increases the space velocity. Thus, at any given point, higher feed rate ensures that higher concentration of TCS molecules are available at the surface for reaction leading to higher deposition rates. However the feed rate cannot be increased to any extent because of physical limitations in the plant. Basically, the plant should be capable of handling increased throughputs from the polysilicon reactors which means the auxiliary systems of the poly-reactors (which include the feed system consisting of H2 feed, liquid TCS pumps and vapourisers and the mixing chambers for H2 and TCS) and the down-stream reactor effluent handling system (which comprises of the exhaust cooler, compressors, deep refrigeration units for condensing the chlorosilanes, scrubbers and dryers for hydrogen gas, distillation and hydrogenation reactors) should be able to handle higher feed throughputs. Another aspect of the feed rate increase is that the conversion efficiency decreases with increased feed as there is less time for the gas to take part in the chemical reaction and bulk of the gas comes out without any reaction taking place. This leads to increased load on the downstream recovery units and distillation columns which separate TCS and purify it. Generally, at the beginning of the cycle, feed rate will be low, but is gradually increased as the

14

diameter of the slim rods increases because as the diameter increases, surface area available for reaction increases in geometric proportions. Hence, feed rate can be programmed to satisfy this condition.

Feed and exhaust nozzle: As the reactor is a cylindrical body in order to keep the feed uniformity, generally the feed nozzle is kept at the center of the base plate or placed symmetrically around. If more than one nozzle is provided, it is important to ensure that all the nozzles have equal feed rates. If this is not achieved, then the gas dynamics in the reactor will be affected. Exhaust ports in the reactor will not drastically affect the turbulence of the system. However, they should be located such that the symmetry of the gas flow is not affected. The exhaust gas will be at considerably high temperatures. It may therefore be economical to tap this to heat the feed gas by means of a heat exchanger which should be of proper material of construction such that impurities are not drawn in to the feed gas.

Temperature gradient: Generally the slim rods are maintained at about 1000 to 1180C. The bell-jar and the base plate are water cooled. Hence, the bell surface would be at about 50 to 90C depending on the temperature of the cooling water and its flow rate. Thus, there is a steep temperature gradient between the slim rods and the reactor walls. Hence, thermal convection develops in the reactor. This is increased by the central feed nozzle from which gas would be jetting out at high velocity. Because of this, the turbulence in side the reactor would be automatically high and the deposition would be uniform. However, if the reactor walls are kept at higher temperatures by circulating thermic fluids for tapping the heat for down stream processes, then this gradient decreases. In such a situation, special arrangements have to be made to increase the turbulence.

Number and location of hairpins: With increased demand for polysilicon, the size of the poly reactors is increasing. Hence, these days, reactors with 24 or more hairpins are being made though reactors with 9 and 24 hairpins are also being operated. One important benefit of increasing the number of hairpins in the reactor is reduction in electrical energy consumption. When the number of hairpins is more, the hot filaments will see more number of equally hot rods around them which reduces the loss of energy by radiation. The hairpins are located in symmetrical fashion so that the gas dynamics is uniform around them and feed gas concentration is uniform throughout. Hence, there is limited scope for deciding the location of the hairpins in the reactors.

Size of the hairpins: The size of the hair pins should be as high as to occupy the maximum space in the bell jar reactor. This will not only ensure that the reactor is used to the optimum, but also, the conversion efficiency in the reactor is also high. The gap between two rods of the hair-pin is decided by the final size of the rods that are grown in the reactor. At the final stage of growth the gap between two rod surfaces decreases which results in higher temperature in these regions. This will lead to poor quality of the deposit in these regions. Hence, the hair pin widths

15

should be sufficiently large to get uniform deposition through the circumference of the rods.

Final Diameter of the rods: In the beginning of deposition the diameter of the rods will be small—about 3 to 10 mm depending on the type of slim rods used. Hence, the surface area available for reaction will be small. The deposition rate will be high at this stage as the concentration of the nutrients in the reactor would be much higher than what is required at this diameter. However, as the diameter increases, the rate of deposition comes down. In order to keep higher deposition rates, the feed rate has to be increased in geometric proportion of the diameter as explained above. As the weight of silicon deposited depends on the square of diameter, it is desirable to increase the diameter of the slim rods to as high a value as possible. However, limitations in the power supply and the distance between to adjacent rods limits this.

Thus, one can notice that there are various process parameters which process engineers can change to optimise the deposition process. However, since electrical energy consumption is the most important parameter in the manufacture of polysilicon, utmost attention should be paid in reducing this. Thus, other parameters like per pass conversion efficiency and down-stream recovery processes could be compromised to achieve this target.

How to reduce cost of manufacture: The most important cost factor in silicon manufacture is electrical energy. Different reactor designs lead to different energy consumptions starting from 60 units to more than 200 units for producing 1 kg of polysilicon. Various ways in which the energy consumption in silicon production in the Siemens reactor can be reduced are considered now. We have already discussed the effect of the size of the reactor on energy consumption. Larger the reactor less is the energy consumption. We have also considered ways and means of increasing the rate of deposition of silicon which has a direct impact on energy consumption. Faster the deposition rate, shorter the time required for completion of a run and hence, net energy consumption would be lower. Both can be optimised for lowering the energy consumption. There are limitations on both the above techniques. Hence, other ways of energy reduction should be explored. As we have seen, the high energy consumption in the Siemens reactor is because the rods at about 1100 C radiate out heat to the cooled metal bell jar wall. Hence, if we can somehow reduce this energy loss by radiation, then we can conserve energy. One way is to provide good reflecting surface on the metal bell jar. This can be achieved by electro polishing the inside of the bell jar which makes the surface smooth and highly

16

reflective. However, the reflectivity of the bell jar wall which in most cases is stainless steel is low and hence, one can expect only a modest reduction in energy consumption. Further, the surface will become tarnished with use because of deposition of thin films of silicon oxide and /or because of corrosion of the surface by hydrochloric acid liberated by slight traces of chlorosilanes/polysilanes present at the end of the process. Thus, provision should be made to regularly re-polish the inner surface of the bell jar. One important way of increasing the reflectivity of the wall of the bell jar is to use silver. As is well known, silver has the highest reflectivity among all materials for radiation. Hence, silver lining on the inside surface of the water cooled metal bell jar would greatly reduce energy consumption. The present author had used a small bell jar made with silver and found that the energy consumption was reduced by almost half compared to stainless steel bell jar. However, for large size reactors, using silver is not possible. One way of providing silver lining is to produce explosion-claded sheets of steel with thin silver layer and fabricate the bell jar inside wall with such explosion cladded sheets. Wacker had patented an atomic hydrogen torch technique for brazing silver on steel surface. One other way is to electroplate silver on the steel surface. However, since silver plating baths contain phosphorus bearing chemicals, there is a possibility of phosphorus contamination in the produced silicon. Further, the integrity of the plated film is also doubtful. Hence, best results could be expected from mechanically coated silver on steel. The other obvious way of reducing the radiation loss is to increase the temperature of the bell jar wall. This can be done by circulating water at higher temperature say 150 to 200 C. This calls for high pressure circulation of water to keep it liquid at these temperatures. This could actually allow tapping of some heat from the polysilicon reactor for generation of steam from the hot water flowing through the jacket. This means the shell has to be designed to take care of the higher operating pressure of the jacket and not just the pressure inside the bell jar. The outer jacket also has to be designed for the higher pressure. Water pumping has to be done at high pressures. While all these are possible, being carried out regularly in steam boilers, the amount of saving in energy may not be commensurate with the efforts needed. Alternatively, a thermic fluid could be circulated through the jacket. By choosing appropriate thermic fluid, the whole operation could be conducted at atmospheric pressure of the jacket. The hot thermic fluid could be used for boiling chlorosilanes during distillation purification. We should note that thermic fluids can be used up to about 300 C with acceptable degradation of the fluid. However, the energy saved by increasing the wall temperature is depicted in Figure 8. In this graph, an

17

Saving in energy with increase in bell jar wall

temperature

0

0.2

0.4

0.6

0.8

1

1.2

0 200 400 600 800 1000 1200

Temperature C

Rela

tive E

nerg

y S

aved

Figure 8 approximate reduction in energy consumption with respect to energy consumption when the wall temperature is 25 C has been made without going in to details of the radiation conditions. The graph can be taken for guidance only. One notes that energy reduction is very small at less than 10% below wall temperatures of 600 C. Further, the value of the energy thus recovered is much smaller than the value of the electrical energy put in to the reactor. Thus, reducing energy by increasing the wall temperature is not really worth the efforts and capital required. From the graph, one finds that at about a wall temperature of 850 C, the energy reduction is about 30 to 35 % which is substantial. The present author achieved this by introducing a fused silica bell jar between the hair-pins and the metal bell jar. Figure 9 shows a schematic of such a reactor. As the deposition proceeds, the fused silica wall temperature

Figure 9

rises and thus energy consumption reduces. About 30 % reduction in energy was achieved. Further, since fused silica is fairly pure, it contributes little impurity to the deposited silicon. It is interesting to note that the original Siemens C reactor had quartz envelop and worked at atmospheric pressure. Radiation shield outside the quartz bell jar was provided to reduce energy consumption. Unfortunately, fused silica bell jars have limited sizes and therefore, cannot be made to suit the big polysilicon reactors presently

18

being operated. Moreover, they are expensive and will crack due to thermal stresses and need frequent replacements. Other materials with have better mechanical strength than fused silica can be considered for this purpose. One should bear in mind the following points: (a) the material used should be sufficiently pure and should also be inert to the gases present in the reactor at the high temperatures. (b). There should not be any deposition of silicon on such insulating material. If silicon deposition takes place, there will be damage to the insulator and also wastage of valuable silicon. Fortunately, suitable materials are available and the process conditions can be so adjusted to prevent deposition of silicon. A polysilicon reactor with such insulation incorporated is shown in Figure 10. Here, a suitable insulator covers the water cooled metal bell jar. A sacrificial inner liner has been suggested to take care of any deposition of polysilicon so that only this sacrificial liner can be replaced instead of the entire insulation.

Figure 10

One can immediately see that depending on how good the insulation is, energy loss is reduced. Thus, if the insulation is sufficiently thick and good, considerable energy loss can be avoided. This can be done with the existing operating reactors and benefits of lower energy consumption could be achieved.

19

In principle, if the surface temperature of the insulation material approaches the deposition temperature, there could be deposition of silicon on the surface. In this case, the heat is coming from the silicon hairpins. One can think of a situation when heaters are provided from the insulation surface to heat a tubular body to the working temperature of about 1100 C. When TCS/H2 mixture is passed through such a reactor, silicon will deposit on the inside of the tubular body and in course of time, sufficient build-up of silicon can take place. This is the general principle of tubular reactors for producing polysilicon. Deposition of silicon inside a tube is well-known. In the initial stages of development of polysilicon production processes, many workers had used externally heated tubes for deposition of silicon. However these did not come to prominence at that time as there were no suitable materials for the tubes. Moreover, the purity of the deposited silicon was affected by the material of the tube. The tube was also found to be either used up in each run or was breaking after deposition. Such tubular reactor designs have been proposed from almost the beginning of polysilicon production during the early 60’s. Several organizations have obtained patents on such tubular reactors. We give a schematic of a tubular polysilicon reactor in Figure 11.

Figure 11

Basically the tube reactor resembles a Siemens reactor in that it has a bell jar and base plate. The main difference is instead of numerous hairpins which are fixed to electrodes at the base plate, a single tube is placed in the bell jar on the base plate. The tube is surrounded by an electrically heated heater which is well insulated from the water cooled bell jar by suitable insulation. The tube is brought to the working temperature of deposition by means of the external heater. A mixture of trichlorosilane and hydrogen is fed to the reactor in the usual manner. Silicon deposition takes place on the inside of the tube. With time, the thickness of the tube increases. However, there is no need to change the power supply to the heater as heat is conserved and the heat loss that normally takes place and heat removed by the gas is to be made up at all the times. Thus, the tubular reactor needs a very simple power supply. Further, as the heater is insulated the heat loss to the water cooled bell jar is very much reduced. Thus, the tube reactor will enable production of silicon with low electrical energy. The estimated energy requirement for polysilicon deposition in the tubular reactor could be as low as 30 % of the Siemens reactor. Thus, the tubular reactor would lead to low cost production of silicon. The

20

advantages of the tubular reactor over the Siemens reactor are given in the table below. It also states the limitations of the two.

Features Tubular Reactor Siemens Reactor

Power supply Fairly simple. If the heater element is made of material like molybdenum or Kanthal, the voltage and current are small. Power control is also very simple and mostly constant power would be needed with very little variation

Power supply would be complex. In order to maintain the temperature of the surface high current would be needed. Power control becomes complex as power has to be gradually increased from the initial to final stage.

Energy consumption Will be small since the entire reactor would be thermally insulated and hence heat loss by radiation would be small. Heat loss due to heating the input gas is only to be supplied plus the normal losses that take place in a tubular resistance furnace (for example a diffusion furnace).

Energy requirement would be This is because the heated hairpins are facing a water cooled bell jar and hence, the radiation loss has to be compensated.

Mechanical design Similar to a Siemens reactor but does not require numerous feed-throughs for electrode connection.

Need numerous electrical feed-throughs

High pressure operation Can be carried out Can be carried out

Gas feed and outlet Concentric at the centre of the base. Heat exchange between the two is easily achieved by a double pipe arrangement.

Feed and outlets will be separate and have to be piped to a double pipe heat exchanger.

Uniformity of deposition For obtaining good surface smoothness, the design of the inlet nozzle has to be carefully made as there is no thermal convection inside and turbulence has to be created by the energy of the feed gas.

Deposition would be better and smoother as there would be considerable thermal convection as a result of the temperature difference between the hairpins and the water cooled bell jar surface. However, as the feed and outlets are at different

21

Features Tubular Reactor Siemens Reactor

locations, in order to achieve uniformity in deposition, feed may have to be located in more than one place.

Assembly of the reactor Fairly easy as the tube has to be fixed to the central location. The electrical heater need not be dismantled and may be a part of the bell jar assembly. Simple external electrical connections need to be made.

More complicated assembly procedure as the hairpins have to be carefully connected and electrical contacts with buss bars have to be made.

Rate of deposition Depends on the extent of turbulence created inside the tube.

Can be better because of natural thermal convection.

Conversion of an existing Siemens reactor to the new design

Possible. High Power supply has to be replaced with a simple power supply

Not applicable

Quantity of silicon deposited Depends on the initial diameter of the tube. Large diameter tube is required to start with for getting more deposition per cycle as the inner diameter decreases with time.

Depends on the final diameter of the deposited rods which is determined by the design and power supply.

Cost of the reactor Low High

Operation Easy Needs more attention

Maintenance Simple More complicated.

Thus, the tubular reactor has several important advantages over the conventional Siemens reactor. Particular note should be made of the low cost of the reactor and ease of operation and low energy consumption. Hence, it is desirable to use a tubular reactor. However, till now we are not aware of any body using tubular reactors for actual production. The main reasons for this are:

1. Suitable tube materials have not been available. Materials like fused quartz while being good, is not acceptable because the quartz tube will shatter at the end of the run and hence, for each run, a new fused quartz tube has to be used. Even

22

accepting this, there is no possibility of making large fused quartz bell like tubes required for the purpose. GT Solar have proposed to use silicon tubes made by the EFG technique. This would be an excellent solution. However, presently the size of tubes made by this technique appears to be not more than 200 mm and hence, the polysilicon reactors of only small size could be operated. Graphite could be another material of the tube. Making silicon tubes by depositing silicon on graphite tubes/rods have been commercialized and available. However, these reactors had the graphite tubes heated by using them as resistive heaters. At the end, the graphite was removed by machining and/or burning. This resulted in high cost of the produced silicon. However, what is important to note is in today’s context of low cost poly and low purity poly, the concept could be suitably adopted for commercial production of silicon.

2. The second difficulty has been how to isolate the heaters from the chlorosilanes. If chlorosilanes come in contact with the heaters, silicon will deposit on them and spoil them. This problem can be overcome by using appropriate designs of the reactors and using special process conditions.

3. Because of the large size of the tubular reactor, the silicon deposited may not be dense and rough and pop-corn structures are likely to be observed. Earlier, such surface qualities were not acceptable when the material was used for semiconductors. However today, with silicon casting as the major crystallization technique, such defects could be accepted.

4. Tubular reactor will continue to be a batch reactor and hence, will have all the difficulties of batch operations. However, there is a good scope for making this reactor at least semi-continuous. The present author has made a design of the reactor in which, after some quantity of silicon is deposited, it is tapped out by melting and collecting the melt in suitable receptacle which can be programmed to produce DSS crystals. After tapping silicon, the poly deposition could be again started. This concept will bring down the cost of silicon ingot manufacture to low levels.

Thus, there is considerable justification for introducing the tubular reactor in to production of solar grade silicon. It will not only bring down the cost of manufacture, it is easy to construct, assemble and operate. The present Siemen’s reactors can be modified in to tubular reactor so that the benefits of this can be reaped. It is believed that in the near future, this concept will become popular.

Optimisation of use of trichlorosilane. The next important cost in the manufacture of silicon is trichlorosilane consumption. The reason for this is that during silicon deposition, roughly about 1/3rd is used for deposition and almost 1/3rd to ½ is converted to silicon tetrachloride. Even at the time of production of trichlorosilane by the fluidized bed reactor process, about 10 to 30 % of silicon tetrachloride is produced. This is a waste product for the polysilicon manufacture. One way of using this silicon tetrachloride is to burn it in a hydrogen flame to produce pyrogenic silica also called fumed silica. This fetches a good price for the manufacturer.

23

However, marketing fumed silica is not easy. It is used in small quantities of about 1 to 2 % in various formulations. However, even at this small level, its role is important in the formulations. Generally users do not want to change the source of supply for fear of the quality of the final product being affected by this small component. Hence, selling fumed silica is not easy. All the big names in polysilicon have facilities for manufacture of fumed silica and have good market share. They also have facilities for manufacture and formulation of silicones which use considerable quantity of fumed silica. New entrants find it almost impossible to penetrate this market. Hence, they are forced to different ways of disposing silicon tetrachloride which will not add value in the manufacture and thereby increase the cost of silicon produced. The most important way of use for silicon tetrachloride is to convert it to trichlorosilane by reaction with hydrogen gas at high pressures and temperatures in a hydrogenation reactor also called a converter. Here, a equi-molar mixture of silicon tetrachloride and hydrogen is fed in to the converter which consists of hot graphite heaters kept at about 1200 C by means of electrical energy. About 12 to 18 % of silicon tetrachloride is converted to trichlorosilane according to the following equation:

SiCl4 + H2 SiHCl3 + HCl

The hot exhaust gas from the converter is quenched to low temperatures to reduce the re-conversion of TCS to STC. The chlorosilanes are condensed and sent to distillation columns for separation of TCS from STC. HCl produced is sent to TCS production unit. This process consumes electrical energy to the extent of 10 to 40 units. This has to be added to the total energy consumption in the manufacture of silicon. A better process is to pass the equimolar mixture through a bed of metallurgical silicon which results in higher conversion rate for trichlorosilane and HCl produced is also converted to trichlorosilane. The chemical reaction can be described as follows:

Si + 3SiCl4 + 2H2 4 SiHCl3

This process was developed by Union Carbide and is presently owned by REC. The additional advantage of this process is that the gases can be heated using hydrocarbon fuels thus avoiding use of expensive electrical energy though exotic materials of construction are required in making the reactor. Both the above reactions are endothermic and need heat input for the reaction to proceed. The present hydrogenation converters that are available in the market have their own shortcomings. They use graphite heaters which are not only fragile but will not ensure uniform heating of the input gases to the required temperatures. Moreover, the heaters need to be replaced at regular intervals adding to maintenance costs. The present author has developed a much more rugged reactor design with heated beds of silicon or graphite which ensure uniform heating and avoid failure of heaters. Such reactors are simple to construct, need no sophisticated controls and can be used with metallurgical silicon bed to avoid formation of HCl.

24

Avoiding formation of silicon tetrachloride during polysilicon

production As we have seen silicon tetrachloride is produced during manufacture of trichlorosilane as well as during deposition of polysilicon in the Siemens reactor. We have seen how this silicon tetrachloride is handled in the plant. Cost of production of polysilicon is increased because of this handling of silicon tetrachloride. Therefore, if by some means silicon tetrachloride formation is suppressed, there could be great reduction in the cost of manufacture of polysilicon. We have seen that during silicon deposition in the Siemens reactor, formation of silicon tetrachloride is inevitable because of thermodynamic consideration. Attempts have been made to suppress this by adding the equivalent quantity of silicon tetrachloride to the starting feed. Though, this suppresses silicon tetrachloride formation, the rate of deposition of silicon is greatly reduced thereby increasing energy requirement for production and reducing the overall plant capacity. Hence, some different way of handling should be thought of. As is well known, production of trichlorosilane by reaction of anhydrous hydrogen chloride gas with metallurgical silicon is a highly exothermic reaction. Elaborate steps are used to remove the heat of reaction and to keep the fluidized bed at around 300 C to optimise the production of trichlorosilane. This is achieved by providing heat-exchangers inside the FBR. However, this is easily said than done since the heat exchanger inside the corrosive as well as abrasive fluid bed environment will fail. Attempts have been made to remove the heat of reaction by diluting the feed HCl with gases like hydrogen or inert gases. However, this will adversely affect the capacity of the FBR. Also, presence of an inert will lead to high refrigeration load for recovery of the chlorosilane. By injecting liquid silicon tetrachloride in to the fluid bed, this problem can be very elegantly tackled. Liquid silicon tetrachloride will extract heat for its own evaporation and to attain the reaction temperature of the fluid bed. Control of the temperature would be simple. Another advantage is that since silicon tetrachloride is condensable, the refrigeration load would be greatly reduced. One has to however, consider the requirement of energy for vapourising the condensed liquid now rich in silicon tetrachloride. However, this would be much smaller compared to when an inert gas is used for cooling. One important effect of using silicon tetrachloride for removal of heat of reaction needs greater recognition than has been hitherto given. Presence of silicon tetrachloride will not just suppress formation of silicon tetrachloride. More than this, some silicon tetrachloride is converted to trichlorosilane. This is seen in the Figure12 below. Here, effect of addition of equi-molar quantity of silicon tetrachloride to the feed HCl on equilibrium composition calculated is compared with when only HCl is fed. One can find that the amount of STC in the equilibrium composition is always less than 1 even at higher temperatures indicating that some STC has got converted to TCS. Hence, it is possible for the TCS production FBR to act as a hydrogenation reactor. No additional heat is required like in hydrogenation thus reducing the energy required.

25

Hydrochlorination Effect of STC in Feed

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

200 300 400 500 600 700 800 900 1000 1100 1200

Temperature C

Sp

ecie

s k

g m

ol

STC with dilution

TCS with dilution

TCS without dilution

STC without dilution

Figure 12

Thus, addition of STC to HCl in the FBR for TCS manufacture leads to dramatic changes in the process and capital and manufacturing cost. One word of caution is that the thermodynamic calculations should be taken for broad guidance as the actual composition may be different from thermodynamic predictions due to kinetic factors. One notices that in the above figure, at 300 C, the composition of the reactor exhaust without STC dilution shows only about 2/3rd of TCS formation. However, it is well known that the composition of TCS could be more than 90% under actual reactor operations. This is because of the kinetic factors in the FBR. Therefore, one has to use experience to optimise the process conditions. CONCLUSIONS; Attempt has been made to critically examine the limitations of the Siemens reactor for the production of polysilicon. Why production cost of the Siemens process is high has been explained. Several novel ways in which the limitations of the Siemens process could be overcome have been described. Emphasis has been made on using out-of the way solutions for cost reduction.