Study of Properties of Cryolite Lithium Fluoride Melt …...through electrolysis. The objective of the thesis was to evaluate the physico-chemical properties of a cryolite-lithium

Study of Properties of Cryolite – Lithium Fluoride Melt containing Silica

by

Sridevi Mariyam Thomas

A thesis submitted in conformity with the requirements for the degree of Master of Applied Science

Materials Science and Engineering University of Toronto

© Copyright by Sridevi Mariyam Thomas 2012

ii

Study of Properties of Cryolite – Lithium Fluoride Melt containing

Silica

Sridevi Mariyam Thomas

Master of Applied Science

Materials Science and Engineering

University of Toronto

2012

Abstract

The ultimate goal of this study is to examine the feasibility of extracting silicon from silica

through electrolysis. The objective of the thesis was to evaluate the physico-chemical properties

of a cryolite-lithium fluoride mixture as an electrolyte for the electrolysis process. A study of

86.2wt%Cryolite and13.8wt%Lithium fluoride melt with silica concentration varying from 0-

4wt% and temperature range of 900-1000°C was done. Three properties were measured using

two sets of experiments: 1) Dissolution Behaviour Determination, to obtain a) solubility limit, b)

dissolution rate (mass transfer coefficient) and 2) density using Archimedes’ Principle. The study

concluded that solubility and dissolution rate increases with temperature and the addition of LiF

to cryolite decreases the solubility limit but increases the rate at which silica dissolves into the

melt. With addition of silica, the apparent density of electrolyte first increases up to 2-3wt% and

the drops.

iii

Acknowledgments

My sincerest thanks go to Professor Barati for giving me the opportunity of working on this

thesis and his invaluable guidance and support. I am most grateful for the freedom with which he

allowed me to work on this project letting me learn a lot more about my subject than I would

have otherwise done.

I would like to thank the Materials Science and Engineering Department for the financial

assistance and the close-knit community within the department for their support. My gratitude to

Professors Argyropoulos and Utigard as well for first giving me the opportunity to work on

metallurgy related projects in my undergraduate studies. I would also like to thank Sal Boccia of

the MSE department and George Kretchmann of the Geology department for their advice and

assistance in various analyses.

I am indebted to my parents for pushing me all my life to achieve my best and doing whatever

they could to aid me. I am thankful for the support they gave me when I made the most important

decision to date; choosing to switch to materials science. Had I not, I would not have discovered

my love for metallurgy and for this I am always grateful.

I really appreciate the support of the Sustainable Materials Processing Research Group especially

Karim and Samira. A lot of the work done for this research would not have been possible without

Karim’s advice and Samira’s help.

Finally, my express thanks to my friend Michal Fulmyk, for his criticisms and lending ear to the

many discussions about the experimental procedure and results obtained. His insight as an

engineer of a different discipline helped me shape this thesis.

iv

Table of Contents

Acknowledgments .......................................................................................................................... iii

Table of Contents ........................................................................................................................... iv

List of Tables ................................................................................................................................. vi

List of Figures ............................................................................................................................... vii

List of Abbreviations and Symbols ................................................................................................ ix

Chapter 1 Introduction .................................................................................................................... 1

Chapter 2 Literature Review ........................................................................................................... 5

2.1 Solar Power and the Role of Silicon ................................................................................... 5

2.2 Current Methods of Extraction and Purification of Silicon .............................................. 10

2.2.1 Siemens Process .................................................................................................... 11

2.2.2 Zone Refining ....................................................................................................... 13

2.2.3 Other Methods ...................................................................................................... 14

2.3 Generation of Si by Electrolysis ....................................................................................... 17

2.4 Operating Temperatures .................................................................................................... 20

2.5 Choice of Electrolyte ........................................................................................................ 24

2.6 Measuring Properties of Electrolyte ................................................................................. 25

2.6.1 Determining Dissolution Behaviour ..................................................................... 25

2.6.2 Measuring density ................................................................................................. 30

Chapter 3 Experimental ................................................................................................................ 34

3.1 Dissolution behaviour ....................................................................................................... 35

3.1.1 Examined Approaches .......................................................................................... 35

3.1.2 Selected Method .................................................................................................... 41

3.2 Density Measurement ....................................................................................................... 47

3.3 Conductivity Measurement ............................................................................................... 51

Chapter 4 Results and Discussions ............................................................................................... 56

4.1 Dissolution Behaviour ...................................................................................................... 56

4.2 Density .............................................................................................................................. 62

Chapter 5 Conclusions and Future Work ...................................................................................... 69

5.1 Summary and Conclusions ............................................................................................... 69

5.2 Future Works .................................................................................................................... 71

v

References ..................................................................................................................................... 72

Appendices .................................................................................................................................... 78

Appendix A – Phase Diagrams generated by FactSageTM

....................................................... 79

Appendix B – Derivation of Equation 7 .................................................................................. 84

Appendix C – Effect of considering V and h as constants on k and CS ................................... 89

Appendix D – XRD analysis of white deposit on silica rod in dissolution behaviour

experiment ................................................................................................................................ 91

vi

List of Tables

Table 1 Impurity ranges for MG-Si and 11.5% efficient Si solar cell [9] ...................................... 8

Table 2 Material requirements for an electric arc furnace producing silica [10] .......................... 11

Table 3 Typical analysis of acid leached MG-Si [9] .................................................................... 15

Table 4 Voltages required for electrolysis of Ta2O5 and Nb2O5 [6] ............................................. 18

Table 5 Impurity levels in ppm for different silicon sources and products [26] ........................... 19

Table 6 Additive effect on cryolite-alumina electrolyte [33] ....................................................... 22

Table 7 Conditions of various studies on electrolysis of silicon .................................................. 32

Table 8 Effect of time on the calculated k and Cs ........................................................................ 59

Table 9 Molar volumes of the various constituents of the melt .................................................... 65

vii

List of Figures

Figure 1 Price make-up of solar cells [3] ........................................................................................ 2

Figure 2 Percentage contributions from each energy source [8] .................................................... 6

Figure 3 Estimates of available energy resources using today's technology [8]............................. 6

Figure 4 Efficiency drop with impurity content [2] ........................................................................ 8

Figure 5 Correlation between purity and cost of silicon [9] ........................................................... 9

Figure 6 Schematic representation of the Siemens process [1] .................................................... 12

Figure 7 Alumina solubility versus cryolite ratio ......................................................................... 23

Figure 8 Schematic for measuring diffusion coefficients by rotating disc method [51] .............. 29

Figure 9 Modified schematic to observe dissolution behaviour for this study ............................. 29

Figure 10 Density of the electrolyte with alumina and silica additions to cryolite [45] ............... 30

Figure 11 a) SEM Image of a 6wt% sample b) Si EDS of sample ............................................... 36

Figure 12 Sample size of 2g containing 6 wt% silica ................................................................... 36

Figure 13 Anticipated diffusion profile across a silica tube with CR+LiF melt ........................... 38

Figure 14 Diffusion Experiment Sample ...................................................................................... 39

Figure 15 SiO2 profile across a 2mm tube held in CR+LiF melt for 30 minutes ......................... 39

Figure 16 Effect of convection flow on the concentration profile ................................................ 40

Figure 17 Capillary reservoir to measure diffusivity .................................................................... 40

Figure 18 Schematic of the experimental setup for dissolution behaviour study ......................... 42

Figure 19 Rods after the experiment ............................................................................................. 43

Figure 20 Immersed portion of rod ............................................................................................... 43

Figure 21 Reproducibility of dissolution behaviour experiment .................................................. 45

Figure 22 Set up of the sink to be immersed in the CR+LiF melt. ............................................... 49

Figure 23 Schematic of the experimental setup for density measurement ................................... 50

Figure 24 Schematic of the experimental setup for conductivity measurement ........................... 51

Figure 25 Typical Nyquist plot for impedance spectroscopy ....................................................... 52

Figure 26 Two arrangements for four probe electrode ................................................................. 53

Figure 27 Comparing results from different experimental setup .................................................. 55

Figure 28 Typical mass loss versus time graph (1000C) ............................................................ 56

Figure 29 k versus temperature ..................................................................................................... 57

Figure 30 CS versus temperature ................................................................................................... 58

viii

Figure 31 Comparison of variation of mass loss for different temperatures ................................ 61

Figure 32 Density versus silica concentration for five temperatures ............................................ 63

Figure 33 Density comparison with and without LiF at ~1000°C ................................................ 64

Figure 34 Density comparison of silica versus alumina addition at 1000°C ................................ 64

Figure 35 Comparison of effect of LiF addition with different oxides ......................................... 65

Figure 36 Density versus temperature for 1wt% SiO2 electrolyte ................................................ 66

Figure 37 Variations of density with silica concentration versus temperature ............................. 67

Figure 38 Density compared to liquid aluminium ........................................................................ 68

ix

List of Abbreviations and Symbols

A area (cm2)

CB bulk concentration (g/cm3)

CR+LiF cryolite and lithium fluoride

CS saturation concentration (g/cm3)

D diffusivity

DC direct current

EDS energy-dispersive X-ray spectroscopy

EG-Si electronic grade silicon

EIS electrochemical impedance spectroscopy

EPMA electron probe microanalyzer

ICP inductively coupled plasma

k mass transfer coefficient (cm/h)

m mass (g)

MG-Si metallurgical grade silicon

ppma parts per million atoms

PV photovoltaic

SEM scanning electron microscope

SoG-Si solar grade silicon

t time (h)

XRD X-ray diffraction

XRF X-ray fluorescence

ΔG Gibb's free energy

ΔT change in temperature

Δρ change in density

#N number of nines of purity e.g. 2N = 0.99

1

Chapter 1 Introduction

This thesis is in aid of exploring the feasibility of electrowinning solar grade silicon (henceforth

referred to as SoG-Si) from a cryolite, lithium fluoride and silicon dioxide electrolyte bath.

Hence the purpose of this research was to conduct experiments for measuring properties of a

cryolite and lithium fluoride mixture (hereafter referred to as CR+LiF) with varying silica

concentrations and at temperatures between 900 to 1000°C. The end goal is to obtain silicon of

solar grade purity for solar cell applications.

Solar energy is available in most inhabited areas of the earth on a daily basis, more accessible

than hydro and geothermal sources and more reliable than wind and tidal as it is more consistent.

The technology for harnessing this energy is well developed by using solar cells which is

typically made of silicon, usually 10-20% efficient. Over 90% of the today’s photovoltaic (PV)

systems use silicon as the conversion material [2]. There are four main reasons why silicon is the

preferred material for solar cells:

1. Abundance: common in its oxide form quartzite SiO2, silicon constitutes 25.7% of the earth’s

crust by mass [2]. This makes it relatively cheap compared to other semiconductor materials.

2. Non-toxicity: silicon is inherently stable. However the current method of purification

(Siemens method) is toxic and efforts are underway to find a different route to extract and

purify silicon.

3. Acceptable band gap: silicon has a band gap of 1.11 eV which is very close to the optimum

band gap of 1.4eV [2]. The optimised value is determined by a balance between the electrical

energy created; a function of voltage created when electrons are excited and energy lost; a

function of current of the same. Gallium (III) arsenide, as another semiconductor material has

a more favourable band gap (1.43eV) but is not as abundant as silicon and is more toxic.

4. Good conversion efficiency: usually >10% using a simple device; the solar cell. Solar cell

devices are projected to have a 30 year operating life and their manufacture is a mature

technology.

Pertinent to the development of the solar grade silicon is a balance between its purity and cost

both of which are determined by the extraction and purification process. Although numerous

methods have been proposed for the generation of SoG-Si the problem appears to be low

2

production rate and high cost of the product. The economy of solar power depends on 1) the

making of the device and the technology involved 2) the efficiency of the product and 3) the

extraction method of material from ore. The cost of material is primarily dictated by the

production rate and energy consumption of the extraction processes involved. It has been

estimated that the silicon material accounts for 45% of the final price of the solar cell [3]. Figure

1 shows the proportions of cost with respect to semiconductor material, processing of the module

and the assembly of the module [3]. It implies that a reduction in the materials cost results in

substantial decrease in the device price. Therefore, a process capable of producing low cost SoG-

Si will contribute to lower cost PV electricity.

Figure 1 Price make-up of solar cells [3]

The current technology for generation of high grade silicon is a two stage process 1) extraction

of crude silicon through carbothermic reduction of silica and 2) purification of silicon involving

chemical vapour deposition (Siemens’ process) followed by directional solidification. The

method is essentially similar to the process of making electronic grade (semiconductor) silicon

that requires much higher purity (9N) than that of SoG-Si (6-8N). The disadvantages of such

method are low production rate and high energy consumption which lead to prohibitive costs for

solar uses. The ultimate goal of this project is to investigate the possibility of using electrolysis

as a less studied but promising technique for producing low-cost SoG-Si. Electrolysis has the

potential to combine extraction and purification into a one step process to obtain solar grade or

near solar grade purity.

Electrolysis has been explored as a feasible process because of the similarity of SiO2 to Al2O3

and Si to Al. Alumina is reduced by electrolysis in cryolite at 1000°C in liquid state (voltage of

Module Assembly

24%

Process 31%

Silicon 45%

3

4V, current density 1A/cm2, efficiency 0.9) [4] to produce nearly all the aluminium that is made

today. The resulting aluminium is denser than the electrolyte so it sinks to the bottom of the bath

and is prevented from oxidation by the atmosphere. Al ion requires 3 electrons to be reduced to

metal, while Si which is next to Al on the periodic table, needs 4 electrons. This higher energy

requirement is offset due to silica being available in high purity at a relatively low cost (unlike

alumina which requires the Bayer process for refining bauxite) making the end product a

comparable price to aluminium.

Purity of the silicon plays an important part as impurities will reduce the efficiency of the solar

cells. Purification inherent to electrolysis of Si is carried out due to the different deposition

voltage of elements. Elements with a lower deposition voltage than Si will deposit out first so a

low voltage can be applied to sacrificially remove them. Elements with higher deposition voltage

will stay in the solution. This means the electrolyte will need to be purified (using electrolysis)

after a while to prevent build up else these impurities will deposit out due to saturation of the

solution [4-6].

Carbothermic reduction involves an arc furnace and a carbon source all which introduces

impurities into the silicon making it only 95-99% pure. The silicon that is obtained from

electrolysis has been shown to have a higher purity than MG-Si and so can be used as a precursor

to zone refining. For instance, in zone refining the number of repetitions can be reduced if

sufficient purity of Si has been achieved by electrodeposition.

Carbothermic reduction uses non-renewable coke and results in the emission of CO2. Electrolysis

uses electricity which can be obtained from a clean and/or renewable resource such as hydro or

solar, resulting in a process with a very small CO2 footprint compared to today’s technologies.

The reason the electrolysis of silicon is not as of yet applied on an industrial scale is because of

the high melting temperature of silicon, 1414°C, which means to keep it molten, high operating

temperatures must be maintained. High temperature electrolysis will be accompanied by such

problems as stability of the electrolyte and choice of electrodes and the furnace walls. Lowering

the temperature would assist solving these problems and also reduce energy consumption but this

would mean the silicon is deposited as a solid. Liquid deposition is favoured to solid because 1)

inclusions of impurity particles are reduced due to the liquid surface tension and 2) since solid-

liquid interface energy does not need to be overcome like in solid deposition where increasing

4

the current density to increase deposition rate leads to powdery solid. The recovery of such

powdery product from electrolyte will also be difficult and involve significant losses. The

method proposed by our group is to overcome both problems by depositing the silicon in a liquid

metal cathode resulting in an alloy of low melting temperature, thereby reducing operating

temperatures and increasing deposition rate. The alloy is then slow cooled to precipitate out

almost pure silicon dendrites from the alloy that are then removed by physical or chemical

separation methods. In addition to lower temperature, the method has the advantage of producing

an alloy heavier than Si that settles to the bottom and is protected against oxidation.

In order to identify a suitable electrolyte and characterise it, this fundamental study was carried

out. The literature review indicated that cryolite was preferred among previous authors. To lower

the working temperature, lithium fluoride was added to reduce the liquidus temperature. Three

properties were investigated between 900 and 1000°C:

1. Solubility limit of silica in the CR+LiF electrolyte.

2. Dissolution rate of silica in the CR+LiF electrolyte by calculating a mass transfer

coefficient.

3. Density of the CR+LiF electrolyte with varying silica concentrations between 0-4wt%.

Having knowledge of these (and other) properties is essential in determining whether electrolysis

of the CR+LiF melt containing silica to obtain silicon in its pure or alloy form is a viable option.

Chapter 2 of this thesis will provide an overview of the previous works in this area; the various

types of electrolytes, the temperature of deposition; liquid silicon versus solid silicon deposition

and the additives used to improve electrolyte properties. Since the aim of this thesis is to look at

the properties of CR+LiF bath the literature review will also cover techniques on how to measure

solubility, mass transfer coefficient and density. The review will explore the viability of using

electrolysis of silica to extract silicon and explain why this thesis has chosen cryolite as the base

for the electrolyte. Chapter 3 on materials and methods will describe the experimental procedures

used and the way the data was collected and processed to obtain said properties. Chapter 4 on

results and discussion will examine the results and the conclusions that may be drawn from the

obtained data and how this helps to realise the method in an industrial setting. Finally, the major

conclusions of the study will be presented in Chapter 5.

5

Chapter 2 Literature Review

2.1 Solar Power and the Role of Silicon

A renewable source of energy is an absolute must because of our world’s dependency on non-

renewable sources in the form of chemical energy i.e. fossil fuel resources; which as the name

suggests cannot be replaced, certainly not at the current rate of consumption. At the present rate

of usage, it is estimated that the earth resources will be depleted of oil in 46 years, natural gas in

100 years and coal in 168 years [7]. Fossil fuels also have a further disadvantage in that their use

creates greenhouse gases (GHG), mainly CO2, which are thought to be the primary reason for

global warming. Recording of the average temperature over ten-year periods have shown that the

mean temperature has increased over the decades.

Current options for renewable resources are solar, hydro, wind, biomass and geothermal energy.

Technology to capture tidal and wave energy are still in the infant stage but the technologies for

the other sources have grown due to concern about climate change and the depletion of fossil

fuels. About a fifth of the world’s electricity supply is already generated by renewable energy as

shown in Figure 2 [8] most of which is provided by hydro power. Hydro has the advantage that it

has a high energy density per volume but as observed in Figure 3 [8] when compared to the

world energy use hydro cannot meet the demand. While wind and geothermal energy can

potentially meet current world energy use, this may not be the case in the future since as

population and industrial growth continues, energy use increases. Solar energy, on the other

hand, can more than 3 times fulfill the current energy need which is why it is important to

develop technologies for harnessing this energy source.

6

Figure 2 Percentage contributions from each energy source [8]

Figure 3 Estimates of available energy resources using today's technology [8]

Coal/Peat 40.3

Gas 21.4

Oil 5.1

Nuclear 13.4

Hydro 16.5

Solar 0.1 Geothermal 0.3

Wind 1.4

Biofuels 1.4

Others 0.1

Other

Renewables 3.3

477

>1600

600 500

>250

50 <1

0

500

1000

1500

2000

En

ergy F

low

(ex

ajo

ule

s p

er y

ear)

World Solar Wind Geo- Biomass Hydro- Ocean

Energy thermal power

Use

7

Current global consumption of energy is 16.3TW. Energy available from sun is estimated at

86,000TW. With a typical 12% efficient solar cell at least 136TW of the sun’s energy need to be

captured to satisfy the world’s energy demand. On average an area of 1m2 can produce 100W of

DC power in direct sunlight (if efficiency is 12%) [2]. To produce 136TW an area of

1,360,000km2, about a seventh of the land area of Canada, is required. Theoretically it is possible

to implement a totally solar power dependent planet with storage methods implemented.

However it may not be practical due to the transient nature of solar power, the need for chemical

energy in many application, and cost. Nevertheless there is a consensus that it can play a big role

in our energy supply.

A solar cell is a device that absorbs solar energy and uses it to promote an electron from the

valence band to the conduction band of the semiconductor material used (leaving a hole in the

valence band) and conducts electricity. Solar radiation is a spectrum of electromagnetic waves

(EM waves) of different energies. The percentage of solar energy that is utilised depends on the

energy gap between the conduction and valence bands. Only EM waves with energies above the

energy of the gap can be absorbed to promote the electron and any extraneous energy is used to

heat up the cell material.

As discussed in Chapter 1, silicon is a suitable choice for semiconductors in solar cells because it

has a band gap close to the optimal band gap that would make maximum use of the EM waves

with minimal energy loss. To do this at an efficiency of 10% the silicon used in solar cells need

at least a purity of 6N but it is crucial to remove the dopant impurities such as P, B and As which

degrade the solar cell performance.

The purity of metallurgical grade silicon, obtained from carbothermic reduction of silicon

dioxide, usually only reaches a maximum of 2N. Impurity atoms can act as electron scatterers

that impede the mobility of electrons; they lose energy and fall back into the valence band and

recombine with a hole which results in the destruction of charge carriers. Figure 4 shows the

efficiency drop off with impurity content for silicon cells [2]. Table 1 shows impurity ranges for

certain elements in MG-Si given in the first column. The second column shows how much they

need to be reduced to make solar cells of 11.5% efficiency (to make 100W/m2). Also included

are segregation coefficients which are important for zone refining [9].

8

Table 1 Impurity ranges for MG-Si and 11.5% efficient Si solar cell [9]

Impurity Element Conc. Range in

MG-Si (ppm)

Conc. required for 11.5%

efficient Si solar cell (ppm)

Segregation

coefficient

Al 1000-4000 25 2.0 x 10-3

B 40-60 0.01 0.8

P 20-45 0.01 0.35

Cr 40-220 <5 1.1 x 10-5

Fe 1500-6000 <5 8.0 x 10-6

Cu 15-40 <5 4.0 x 10-4

Mn 10-80 <5 1.3 x 10-5

Ni 10-95 <5 1.0 x 10-4

Ti 120-275 <5 2.0 x 10-6

V 50-250 <5 4.0 x 10-6

C 1000-3000 <5 0.05

Zr 15-25 <5 1.6 x 10-8

Figure 4 Efficiency drop with impurity content [2]

9

The hardest impurities to remove are phosphorus and boron because of their relatively high

solubility limit and segregation coefficient in silicon. It is essential that they be removed since

these elements are used for doping the Si to make a semiconductor. Since overdoping can lead to

a degenerate semiconductor (essentially a useless device which conducts electricity without

promoting electrons from valence to conduction band) the dopant impurities need to be

maintained at a low and well controlled concentration. Consequently it is necessary to implement

a purification process, such as the Siemens, after the initial extraction process of carbothermic

reduction. The Siemens method, which is currently used in industry, is very time consuming and

energy intensive driving the cost of the product up.

As Figure 5 depicts, there is an inverse relationship between the impurity content and price for

various grades of silicon. Due to the lower purity requirement for SoG-Si, there seems to be a

window of opportunity to generate this material with a cost of <$20/kg. It has been discussed

that this in fact matches a target price of $10-15/kg that is believed to enable solar power, that is,

grid parity with fossil fuel and nuclear based electricity. In the past few years the market price of

SoG-Si has been fluctuating between $25 and $400/kg depending on the supply-and-demand

situation. Achieving a price in the range of $10-15/kg will make solar power mainstream which

will be when it can be a major contributor to the energy supply.

Figure 5 Correlation between purity and cost of silicon [9]

Alloy grade

Metallurigical

grade

Semiconductor

grade

Detector grade

1.E-11

1.E-10

1.E-09

1.E-08

1.E-07

1.E-06

1.E-05

1.E-04

1.E-03

1.E-02

1.E-01

1.E+00

0.1 1 10 100

Imp

uri

ty C

on

ten

t

Silicon Cost ($/kg)

SOLAR

GRADE

10

2.2 Current Methods of Extraction and Purification of Silicon

The main source of silicon is from the mineral quartzite SiO2 which exists in very high purities

and has less than 1% metallic impurities. The other source is as a by-product of phosphate

fertiliser production: fluorosilicic acid H2SiF6 which can come with a purity of 98-99%. It can be

neutralised to form K2SiF6 which can be used in electrolysis to extract Si. A third and recently

looked at source is rice hulls which contain 95-99% pure silica. The US production from silica

content in rice hulls alone could supply 100,000 tons of Si/year. [4]

Crude Si is primarily produced by carbothermic reduction of silica SiO2 at 1900-2100°C in an

electric arc furnace. The product is known as metallurgical grade silicon (MG-Si) with purity

ranging from 95% to 99% [9]. The overall reaction is as follows:

SiO2 (l) + 2C (s) = Si (l) + 2CO (g) (1)

However, there are a lot of side reactions that happen at the high temperatures and side products

are made such as gaseous SiO, which react with O2 in the air to make fine dust particles of SiO2,

which along with SiC (from reactions with the graphite electrode) clog up the furnace. Apart

from loss of yield, handling these unwanted materials is also an issue. The CO gas in Reaction

(1) is burnt to form CO2 to partially recover chemical energy and the gaseous products are

currently not collected to be sequestered. The carbon for reduction is obtained from coke or coal

and wood chips which is the main source of impurities. About 1 million tons of MG-Si is

produced each year using an electric arc furnace. Typical operating parameters for this process

are shown in Table 2 [10]; for 1 ton of silicon produced, about 4 tons of non-renewable carbon is

required in the carbothermic reduction reaction. This will result in a very large CO2 footprint for

the process. The advantage of this method, however, is that production of silicon is fast and the

liquid silicon is easily siphoned off and cast into ingots.

11

Table 2 Material requirements for an electric arc furnace producing silica [10]

Requirement Quantity

Quartz 2.9-3.1 ton

Coke 1.2-1.4 ton

Charcoal 1.7-2.5 ton

Graphite 0.12-0.14 ton (electrode consumption)

Electricity 12.5-14 MWh

= 1 ton of Silicon

To help improve the end product purity level, some authors suggest starting with purer starter

materials (such as quartz deposits from Arkansas, USA and BC, Canada with B<3ppm,

P<10ppm, Al<40ppm, and Fe<120ppm [9]) and carbon black as reductant. However this

approach may drive the cost prohibitively high, and still not yield a Si with high purity due to

contamination from electrodes, refractory, and other parts of the equipment.

Two main methods are used in industry to improve purity of the Si in photovoltaic and solar

applications: 1) Siemens’ process and 2) Zone refining, however alternatives have been

investigated.

2.2.1 Siemens Process

The first purification step for MG-Si is by the Siemens’ method. MG-Si is reacted with hydrogen

chloride gas to form volatile chlorosilanes:

Si (impure) + 4HCl (g) = SiCl4 (g) + 2H2(g) (2)

Or trichlorosilanes:

Si (impure) + 4HCl (g) = SiHCl3 (g) + H2(g) + ½ Cl2(g) (3)

The impurity chlorides can then be separated by fractional distillation due to their different

boiling points. The chlorosilanes are then reduced with hydrogen and Si atoms deposit on a pure

Si rod at 1150°C [1, 2, 9]:

SiHCl3 (pure) + H2 (g) = Si (pure) + 3HCl(g) (4)

The schematic in Figure 6 shows a simplified flowsheet of the Siemens’ process; the different

stages, the inputs and outputs [1].

12

A high temperature is required to make sure the chlorosilane decomposes onto the pure silicon

rod (EG-Si) rather than the other parts of the chamber. This is achieved by electrically heating

the Si rods to serve as the reaction site, while keeping the chamber walls at low temperature. The

silicon made from this method is polycrystalline and the deposition rate is very slow; to make a

ton of silicon, 12 Siemens reactors need to be operated for 10 days [1]. The other disadvantage to

this method is the use of SiHCl3 and H2 gas which are explosive. Furthermore AsH3 and PH3 are

collected as by products which are poisonous gases, and HCl is produced which is very corrosive

as a gas, thus handling and disposing of them in a safe manner is very difficult. The advantage of

the method, however, is that it produces 9N silicon which has purity higher than the SoG-Si

requirements.

Figure 6 Schematic representation of the Siemens process [1]

HCl MG-Si

SiHCl3

Energy Source

H2

SiHCl3 Feed

Fractional Distillation

Vent

EG-Si

Chemical Vapour Deposition Chamber

Fluidised Bed Reactor

13

2.2.2 Zone Refining

Zone refining involves heating a silicon rod from one end with a heating element so that the

silicon melts and slowly moving the heating zone along the rod. The melted portion of the rod

resolidifies to make a purer form of the silicon as the heating element moves away while

segregating impurities in the liquid. This is possible due to phase equilibria; the impurities tend

to stay in liquid rather than solidify into the Si crystal.

The segregation coefficient k quantifies how much of an impurity can be isolated out of the

crystal.

k = CS / CL (5)

CS is the concentration of the impurity in the solid silicon and CL is the concentration in the

liquid silicon. As can be seen from Equation (5) the lower the segregation coefficient the more

likely the impurity will stay in liquid. The last portion of the rod with all the impurities is

recycled.

Zone refining has a disadvantage in that it is a slow process that needs to be repeated several

times if EG-Si is to be produced. The heating element moves at a rate of 0.1 to 10mm per minute

to create polycrystalline silicon and 0.1 to 10mm per hour to create monocrystalline silicon. The

process is usually repeated about 10-15 times [1] and is only efficient if the segregation

coefficient of the impurity is low enough (10-5

to 10-1

). However boron and phosphorus have

segregation close to 1 and so are very hard to remove [2, 11].

Another disadvantage is that only small rods can be used; approximately 10cm in diameter. But

the purity level is much higher (with resistivity of 100Ωm) than silicon produced via directional

solidification (Czochralski method). Furthermore a monocrystalline silicon boule can be formed

through zone refining; this is important for semiconductors because grain boundaries,

microcracks and dislocations (defects in crystals) scatter electrons reducing efficiency.

To address the problem of production volume an alternative method is employed which creates

monocrystalline silicon, the Czochralski method. It requires a single crystal seed which is

lowered onto the surface of liquid silicon and pulled upwards in a controlled manner while

allowing silicon to solidify onto it. This method allows much larger boules to be created with 10-

15cm diameter, 1-2m length, 50-100kg. About 80% of monocrystalline silicon is grown via

14

directional solidification [1]. On the other hand, the downside is that this method produces

silicon of a lower purity (resistivity of only 20Ωm) compared to zone refined silicon because a

silica crucible is required thereby allowing introduction of oxygen impurity when silica dissolves

into the liquid silicon [1].

There are other minor methods that have been looked at which are not as time consuming or

difficult as Zone and Siemens’ process. However they are either energetically expensive or not as

effective at removing crucial impurities. Section 2.2.3 describes several methods that have

achieved some success at removing specific impurities.

2.2.3 Other Methods

1. Acid Leaching [9]

This is a purification process that digests impurity precipitates. Impurities with low segregation

coefficients tend to precipitate at grain boundaries making the area near the grain boundaries

more brittle. If the polycrystalline silicon is crushed the cracks will tend to form along the grain

boundaries exposing the impurities. To ensure most of the impurities are leached away the

silicon has to be crushed to a size less than or equal to its grain size and then the exposed

impurities can react with an acid. Silicon is very stable in most acids except a combination of

hydrofluoric acid and an oxidising acid so the silicon remains unreacted while the impurity

products react and stay in solution. Mineral acids such as HCl, HF, H2SO4 and HNO3 are

normally used in a combination to leach out most if not all the different impurity precipitates.

Table 3 shows a typical analysis of crushed MG-Si before and after leaching with aqua regia

(HCl and HNO3 mix), HCl and HF at an elevated temperature to get total impurity level less than

400ppm [9].

A purity of up to 4N can be achieved but this can vary depending on the particle size the silicon

has been crushed to, the leaching temperature and length of time the leaching was done. Another

factor to take into consideration is the impurity atoms that get trapped in the crystal structure as

opposed to precipitating to the grain boundaries which cannot be leached out. This method has

the disadvantage in that crushing requires time and mechanical energy and material handling

issues arises when dealing with powdery material. It is also not very successful at removing B, P

and Cu and so cannot be used alone for purification.

15

Table 3 Typical analysis of acid leached MG-Si [9]

Impurity Impurity Concentration (ppm)

Removal (%) Before After

Al 2080 150 93

Cr 180 5 97

Fe 2310 60 97

Mn 180 5 97

Ni 80 5 94

Ti 200 10 95

V 180 6 97

B 20 14 30

Cu 22 13 40

P 50 40 20

2. Reactive gas blowing [9, 11]

This method requires gases of Cl2, O2, SiCl4, wet H2 and CO2 (or combinations of) to be bubbled

through molten silicon resulting in the reaction of the impurities to make oxides or chlorides that

either bubble off because of volatility at high temperatures (>1450°C) or collect as dross on top

of the liquid silicon. This method is effective at removing Al, Ca, C, Mg, B, P and Ti because

they are more thermodynamically likely to react with the gasses than silicon. Silicon losses are

minimal and a purity level of at least 99.99% can be achieved. However other elements like Cr,

V, Cu, Mn, Ni and Fe will stay in the silicon and cannot be removed without sacrificing a

substantial amount of silicon since oxides and chlorides for those impurities are

thermodynamically less favourable than for silicon.

3. Slagging [9]

This process involves adding a variety of oxides (CaCO3, BaO, MgO, Al/SiO2, CaO/SiO2 and

CaF2/SiO2) to create slags which will favourably dissolve the impurity but not the silicon and

will not dissolve in the silicon. It has been shown that this method can remove Ti, Mn, V and Al

by one order of magnitude.

16

4. Solvent Refining [11-14]

This method uses a “getter” metal to alloy with the silicon to favourably dissolve impurities.

Some metals have such low solubility in silicon that when trying to alloy it with silicon the end

product simply forms two phases; one of metal-Si alloy and one of pure silicon. If this metal has

a more favourable solubility for impurities than silicon then it is referred to as a getter metal and

the impurities leave the silicon phase to be dissolved in the metal phase. A few getter metals that

have been researched are aluminium, copper, iron and tin. The alloy can then be cooled and

crushed to separate silicon from the alloy by chemical or physical separation methods.

5. Plasma Refining [9]

This method requires the silicon be in the molten state. Electromagnetic stirring transports the

impurities to the surface of the molten silicon and a plasma torch consisting of oxygen and/or

hydrogen is blown at the surface allowing the gases in the plasma to react with the impurities.

The gases are effective at removing boron but the method is very energy intensive.

17

2.3 Generation of Si by Electrolysis

The silicon that is deposited by electrolysis can have three microstructures; monocrystalline,

polycrystalline or amorphous. Amorphous silicon is usually deposited using organic electrolytes

and typically has a small amount of hydrogen dissolved in the silicon [15-17], though this can be

released by heat treatment. Amorphous silicon is usually used for coating for wear resistance.

They also have applications in solar cells for calculators and other devices that do not require

much power, but generally it has a lower efficiency than crystalline silicon [17]. Monocrystalline

silicon growth is more time consuming to grow because of low current density to prevent the

nucleation of new grains. In most cases, for solid silicon deposition, polycrystalline silicon can

be expected.

Cathodes used in silicon deposition can determine the macrostructure which can vary from

powdery to dendritic to coherent layers. Monnier et al. [18] managed to deposit solid silicon of

99.99% purity using a graphite cathode but the silicon tended to react with the graphite and

disintegrate it and the rest of the silicon was in a powdery form which broke off and dispersed in

the electrolyte and hence was difficult to collect. Work done by Elwell et al. [19-22] using Flinak

(eutectic LiF/NaF/KF) as the electrolyte determined that silver had minimal reaction with the

deposited solid silicon at 750°C. After adjusting operating parameters such as concentration of

the silicon source (K2SiF6), current density and deposition voltage, they finally succeeded in

getting coherent growth of silicon layers on graphite as well. Pure silicon in solid or liquid form

can have unwanted reactions with the cathode material. Sometimes however the reactions are

deliberate if an alloy or a coating is the intended end product. In either case cathodes need to be

chosen carefully.

Purification of the silicon obtained, for the purpose of this thesis, would have to ensure the end

product can be used in solar cells. Solar cells, as mentioned before, require a purity of 6N or

higher [9] so it is essential the silicon be deposited with high purity or post processing succeed in

lowering impurity levels. So far there are Siemens process and zone refining in place for post

processing MG-Si but since Siemens process is energy intensive and involves toxic gases and

zone refining is time consuming it would be preferable if a low cost processing method was

available.

18

In electrolytic production of metals, nobility of the ions determine the voltage at which the ions

can deposit out and this is the main principle applied in “automatic” purification in the process.

This is also the reason why aqueous solutions with Si4+

ions cannot be used because hydrogen

will be produced at the cathode instead. This can already be observed to some extent in

electrolysis of organic solutions as mentioned above. Hence most of the electrolysis experiments

are done with molten salts instead. It is possible to separate out ions even if the difference in

theoretical deposition voltages between them is quite small [23, 24] as shown in Table 4. The

difference between the experimental voltages that are required to decompose and deposit

tantalum and niobium is only 0.27V.

Table 4 Voltages required for electrolysis of Ta2O5 and Nb2O5 [6]

Oxide Experimental

(V)

Theoretical without anodic

phenomenon (V)

Theoretical with anodic

phenomenon (V)

Ta2O5 0.91 1.55 0.45

Nb2O5 0.64 1.40 0.30

In electrolysis a “pre-electrolysis” processing can be employed as Elwell et al. [19, 21] have

done, which runs the cell in the same manner as one would with electrolysis but with a much

higher deposition potential and current density to deposit out the impurities with a lower

deposition potential than silicon. Some silicon will be sacrificially lost in this process. It should

be noted that in aluminium electrolysis where the electrolyte is cryolite, Na3AlF6, and alumina,

Al2O3, the deposition voltage of Na+ ions is less than Al

3+ ions. Na

+ ions do not however deposit

out first; the reason for this is yet unknown [6]. Theoretically the impurities with higher

potential than the one used in pre-electrolysis should stay in solution when depositing silicon but

this is not what is observed in practice. With co-deposition of the higher potential impurities with

silicon the Gibb’s free energy, ΔG, of the system is reduced [19]. Even so the researchers

managed to get a purity of 99.98-99.999%. Even with electrorefining similar principles can be

applied; the impurity ions that are more electronegative remain at the anode while those more

electropositive dissolve and stay in solution but do not deposit at the cathode until their

concentration reaches a certain level in the solution (then the electrolyte needs to be “cleaned”).

This is the principle used when producing copper of 99.99% purity using electrolytic refining.

An electrolyte material must be carefully chosen to avoid ions that co-deposit with silicon.

Sometimes a starting material with as high purity as possible is utilised, for instance Elwell et al.

19

started off with electrolyte and potassium fluorosilicate of 99.95% purity or higher. Also both

Monnier and Elwell have stated that equipment itself can lend to introduction of impurities into

the silicon [6, 19].

As mentioned in the Chapter 1, silicon can have several sources. Silica is abundant in its quartz

form with a purity of 99.5-99.8% with a price of 14c/kg [5]. However the difficulty of obtaining

the quartz, its purity and any processing done to purify it can vary the price, for instance the

quartz crystal grown by hydrothermal method for fused silica crucibles go at $1/kg [5]. It is

interesting to point out here that natural SiO2 tend to have lower B and P concentrations than

MG-Si [25] which is a huge advantage as these two elements are considered the most difficult to

remove. Fluorosilic acid H2SiF6 which is a byproduct of making P2O5 for fertiliser is produced at

20-40kg/ton of P2O5 which in turn is made at 40 million ton/year [19]. Fluorosilic acid is

neutralised to make K2SiF6 (lowest volatility) in a reaction with KF (in turn making dangerous

HF) or KOH (making H2O) and can be 98-99% pure at 6c/kg [5]. And finally a recent source for

silica comes from rice hulls, also a byproduct, at 20c/kg [5] where the impurities are

“comparable to mineral silica”, stating that it has a purity of 95-99%. Table 5 shows the

concentration of impurities for various sources and a comparison to what SoG-Si should be.

Please note Table 2 is not in agreement with the values given for SoG-Si in Table 1 and the

efficiency for SoG-Si in Table 2 was not stated in the reference [26]. Different references have

various ranges for specific elements but generally agree on the total impurity content.

Table 5 Impurity levels in ppm for different silicon sources and products [26]

Impurity Natural SiO2 Fluorosilic

acid Rice hulls MG-Si SoG-Si

B 0.3-0.5 1 2 5.4 <0.1-0.3

P 0.3-0.6 33 130 48 <0.1

Al 185 8 10 450 <0.1

Fe 56 13 20 3000 <0.1

Ti 210 - 3 130 <0.1

The choice of electrolyte, source of silicon and materials to construct the electrolysis cell can

determine the operating temperatures. This is an important parameter because this usually

dictates how much energy is consumed for the process of silicon extraction.

20

2.4 Operating Temperatures

The methods to electrodeposit silicon can be categorised into three groups based on the operating

temperature.

1) Producing Si from a molten salt electrolyte at temperatures above melting point of Si

2) Molten salt electrolysis at a temperature range of 400-1150°C

3) Low temperature electrolysis at temperatures below 100°C

The melting temperature of silica is >1700°C so it is necessary to dissolve it in a low melting

temperature electrolyte for electrolysis. A binary or even ternary system is required. The authors

generally preferred fluorides of alkali and alkaline earth metals for their high decomposition

voltage, low melting points, high conductivity and low viscosity [6]. If the base electrolytes (in

cases where oxides were used) did not have these properties then fluoride additives were

introduced.

An attempt was made at gathering and summarizing the basic parameters in most studies made to

obtain Si or its alloys through electrolysis. Table 7 provided at the end of this chapter shows the

different methods separated into the three categories described above. The information thought

most useful were the electrolyte used; which will indicate how easy it is to obtain said

electrolyte, the operating temperatures; which will influence the materials used for the equipment

and energy consumption, the cathode material, the physical state of the deposited silicon (solid or

liquid) and the purity of the silicon; which will determine if further processing is needed.

Liquid deposition is more favourable to solid deposition since the rate of solid deposition would

have to be slow, on the order of 10’s of μm per hour for plane growth [19], to prevent protrusions

which trap inclusions. In liquid form the surface tension would exclude most of the solid

impurities and deposition rate can be increased by increasing the current density. However the

melting temperature of silicon is 1414°C [27] which means that to deposit in liquid form silicon

will have to be deposited at temperatures above 1414°C. When comparing to aluminium

deposition, the operating temperatures are between 940 and 1000°C [28]. Because of the

electrolyte composition and since the melting temperature of aluminium is 660°C [29] the

aluminium deposits as a liquid and can be siphoned off; an easy separation from the electrolyte.

Operating temperature depends on a couple of factors:

21

The physical state of the deposited silicon: pure solid or liquid or alloy solid or liquid.

The melting temperature of the electrolyte mix.

To reduce operating temperatures, the silicon can be deposited as a liquid alloy by tailoring the

melting temperature of the alloy by changing its composition. For instance the silicon can be

deposited in a liquid copper-silicon cathode. As shown in the Cu-Si phase diagram in Appendix

A, the liquid alloy can start with approximately 5wt% silicon (lowest composition allowed while

still remaining liquid at 900°C) and can increase to about 20wt% silicon while still in the liquid

phase. In some cases alloy formation is done intentionally as the copper can also act as a getter

metal (i.e. solvent refining described in Section 2.2.3); the impurities that may have also co-

deposited with silicon preferably remain in the copper-rich phase rather than silicon during the

subsequent solidification. Two phases are created; a Cu3Si phase with dissolved impurities and a

pure phase of silicon. The alloy can then be crushed into particles the same size or smaller than

the grains of the pure silicon and can be separated from the rest of the alloy by flotation using

heavy media [13]. Which particular metal is preferred for alloying depends on the operating

temperatures, how well it works as a getter metal and in one case: if creation of shallow and deep

level recombination centres in the silicon can be avoided; the reason why tin is preferred to

copper or aluminium [4]. The ideal method would be to recycle the getter metal for further

silicon purification but the getter metal itself must be purified to prevent buildup of impurities

else it will not act as an effective getter metal. It has also been suggested that in cases where the

getter metal was Al, there is a distinct possibility that Al has substitution reactions with SiO2 to

make Si and Al2O3 and electrolysis does not play a part [25]. In this case there is a loss of getter

metal although a higher than expected silicon recovery occurs. However, if the silicon is to be

deposited in a solid state then the operating temperature is determined by the electrolyte

composition.

Regardless of Si being deposited as liquid or solid, lowering the operating temperatures is

beneficial from several aspects:

Reduced vapourisation losses of electrolyte

Lower energy consumption

Easier operation and wider choice of construction materials

The electrolyte must remain liquid at the operating temperature and if necessary, its composition

be adjusted for this purpose. For instance cryolite melts at 1012°C±2 [30] but when 10wt%

22

alumina is dissolved in it the melting temperature drops to ~980°C, eutectic being at 20wt%

alumina at 960°C [31]. Temperature can be further reduced by adding aluminium fluoride AlF3.

Additives like aluminium fluoride and alkaline and alkaline earth metal halides are also added to

decrease viscosity and increase conductivity. When silica is added to cryolite the melting

temperature does not decrease significantly; preliminary experiments showed the eutectic to be

around 998°C, so additives should be added to reduce operating temperatures. According to the

cryolite-lithium fluoride phase diagram [32] shown in Appendix A, 13.8wt% of LiF dissolved in

cryolite can reduce the melting temperature of the whole composition to 900°C; one reason why

it was chosen as an additive for this thesis. Table 6 below shows the effect of a number of alkali

and alkaline earth fluorides on melt properties when electrolysing alumina in cryolite to form

liquid aluminium [33]. What must be determined is if the same table is applicable to cryolite

melts with silica; although silicon is next to aluminium on the periodic table, silicon oxide has a

slightly more acidic nature than aluminium oxide, silicon ions needs 4 electrons for reduction to

metal compared to the 3 for aluminium and silica reactions with cryolite leads to more phases

being formed than when alumina is added to cryolite.

Table 6 Additive effect on cryolite-alumina electrolyte [33]

Additive Conductivity Density Viscosity Liquidus

Temperature

Al Metal

Solubility

Surface

Tension

Vapour

Pressure

CaF2 ↓ ↑ ↑ ↓ ↓ ↑ ↓

AlF3 ↓ ↓ ↓ ↓ ↓ ↓ ↑

MgF2 ↓ ↑ ↑ ↓ ↓ ↑ ↓

NaCl ↑ ↓ ↓ ↓ ↓ ↓ ?

LiF ↑ ↓ ↓ ↓ ↓ ↑ ↓

Ideal

Additive ↑ ↓ ↓ ↓ ↓ ↓ ↓

Temperature

increase ↑ ↓ ↓ --- ↑ ↓ ↑

Additives should increase conductivity of the electrolyte to reduce energy consumption, decrease

density to allow floatation on top of deposit, decrease viscosity to increase mobility of the ions

that are to deposit, decrease liquidus temperature to decrease operating temperature and hence

23

energy consumption, decrease deposit solubility in melt to decrease loss of yield, decrease

surface tension to improve wetting of the melt with the electrodes and reducing polarisation

resistance, and decrease vapour pressure to avoid loss of electrolyte. As can be seen LiF meets

all but one of the criterion and hence an ideal choice as an additive.

It is also important to mention that all the additives listed in the table decreases the solubility

limit of alumina in cryolite. This can have a negative consequence in that the anode effect will

happen sooner if the alumina concentration in the melt is not monitored. A study done by Zhang

et al. [34] looked at the possible complexes formed by alumina to explain how the solubility

changes with cryolite ratio and with additives. Cryolite ratio is defined as the mole ratio

NaF/AlF3; 3 is considered a neutral melt, ratio below 3 is acidic and above is basic. This affects

the complexes formed like so:

In acidic melt: 1/3 Al2O3 (s) + 2 NaF (l) +

4/3 AlF3 (s) = Na2Al2OF6 (l)

In neutral melt: 2/3 Al2O3 (s) + 2 NaF (l) +

2/3 AlF3 (s) = Na2Al2O2F4 (l)

In basic melt: 1/3 Al2O3 (s) + 2 NaF (l) +

1/3 AlF3 (s) =

1/2 Na4Al2O2F6 (l)

As can be seen from the reactions above, when the ratio between NaF:AlF3 increases from acidic

to neutral the solubility of alumina increases, but when the melt becomes more basic the

solubility decreases, illustrated in Figure 7 showing experimental results.

Figure 7 Alumina solubility versus cryolite ratio

0.04

0.05

0.06

0.07

0.08

0.09

0 2 4 6 8 10 12 14

Dis

solv

ed A

l 2O

3 (

mola

r fr

act

ion

s)

Cryolite Ratio (NaF/AlF3)

24

The points in red indicate cryolite ratio at 3 and near 8 (explained in Chapter 4). The same

authors [34] found that LiF is a basic additive so it can be expected to decrease the solubility

limit of alumina in cryolite, as is observed in experiment [28].

2.5 Choice of Electrolyte

The Hall-Héroult process has been using cryolite as the main electrolyte in the electrolysis of

alumina to obtain aluminium. Numerous studies have been done on the cryolite; its physical

properties, chemical properties and handling of the material. Cryolite itself is described as an

“uncommon mineral” and the largest deposit in Ivigtût, Greenland has been depleted. However it

can be manufactured through various routes, three of which are shown below [35, 36]:

12HF + Al2O3.6NaOH 2(AlF3.3NaF) + 9H2O

2AlF3.3HF + 6NaOH 2(AlF3.3NaF) + 6H2O

AlF3 + 3NH4F + 3NaCl Na3AlF6 + 3NH4Cl

Cryolite and its additives are known to produce toxic HF gas but the infrastructure to treat the

gas has already been put in place at aluminium industries. With so much physical and chemical

information about the electrolyte already available, groundwork for the use of this chemical for

electrolysis is already laid out, and with the ease of manufacture of the electrolyte, one can

conclude it would be convenient and worthwhile to investigate the viability of using cryolite as a

base electrolyte in the electrolysis of silica to obtain silicon. Electrolysis work has already been

done using cryolite and solid silicon with upto 3-5N purity has been obtained (Table 7) therefore

cryolite has been deemed a practical electrolyte worth further investigation [6, 37, 38].

The additive chosen for this thesis is lithium fluoride and a composition, based on the Na3AlF6-

LiF phase diagram (Appendix A), was chosen to reduce the melting temperature of the CR+LiF

mixture to 900°C. Firstly, the melting temperature of pure cryolite has been reported as 1012°C ±

2 [30] depending on the impurity concentration of the cryolite used in the various studies.

Secondly the Si-Cu phase diagram (found in Appendix A) shows that the minimum temperature

at which Si-Cu alloy will remain liquid between 10-30 mol% of Si at is 860°C. Therefore a

working temperature between 900-1000°C keeps the alloy liquid and the reaction kinetics

relatively fast, while not having to work at overly high temperatures thus saving energy. Hence a

25

melt composition of 13.8wt% LiF with 86.2wt% cryolite was used as the basis for the

electrolyte.

A special mention should be given to Ming et al. [39] who have already deposited silicon in the

solid state from a cryolite and lithium fluoride melt at a purity of 99.5 to 99.97% thereby

demonstrating that the electrolyte used in this study is functional. However fundamental

properties of the CR+LiF and silica melt cannot be found in literature and this thesis can

contribute knowledge about three properties for this melt and the way they have been measured.

2.6 Measuring Properties of Electrolyte

Pertaining to electrolyte production of metals, the following properties of the system are of

interest:

Density

Electrical conductivity

Electronic and ionic conductivities

Viscosity

Solubility and diffusivity of species being deposited in electrolyte.

In the present study, density of CR-LiF-SiO2 melts together with dissolution behaviour of SiO2 in

CR+LiF melt is investigated experimentally.

2.6.1 Determining Dissolution Behaviour

Operating parameters have an effect on the macrostructure. The goal of the work done by Elwell

et al. [19, 21] was to get coherent deposition of layers of silicon on the cathode. For low

concentration of silicon ions and high current density they got powdery or dendritic growth. For

low concentrations, the silicon ions become depleted at the cathode surface and by the time new

ions diffuse through to the cathode they start a new grain leading to powdery deposition or even

deposition of unwanted cations (although this can be avoided using pulsed electrolysis to allow

for transport of silicon ions to cathode surface). For high current density the ions are quickly

placed in the easiest to grow crystal direction which leads to dendritic growth. Increasing the

concentration of the silicon ions for a higher current density or decreasing the current density for

a lower silicon ion concentration gave the researchers coherent layers on a silver cathode.

26

However the concentration of silicon ions had to be much higher to get coherent deposition on

graphite cathodes. The current density determines how fast layers of silicon can grow and in the

work done by Elwell they managed to grow a 3mm thick silicon layer on an area of

approximately 2cm by 1cm silver plate in 4 days. Because the deposition took so long, the

vapours from the electrolyte reacted with the Inconel tube the graphite crucible was set up in. Fe

and Cr vapour compounds dissolved in electrolyte and were co-deposited with silicon

compromising the purity of the deposit.

The importance of the solubility limit is highlighted above; the researchers could only grow a

coherent deposit after a certain concentration of silicon ions in the melt was met. Also the anode

effect must be avoided; when the ions to be deposited at the cathode are depleted, an insulating

gas layer on the anode is formed and the cell voltage increases and the current flow decreases. In

the aluminium electrolysis cell the gases are fluorocarbons (in addition to COx) and have to be

dealt with [28]. Knowing solubility limits help avoid this; anode effects were a common

occurrence at the beginning of the era of electrolysis of alumina but with saturation limits used to

advantage a cell can run a whole day without an anode effect. It is interesting to note that

Grjotheim et al. discovered that the solubility of silica in cryolite increases significantly with the

addition of alumina with a maximum at 69wt% SiO2 for 14wt% Al2O3 [40]. Of course this

increases the probability of aluminium co-depositing with silicon. But also crucial to avoid is

oversaturation; too much silica could lead to silica particles sinking in the melt form inclusions in

the liquid deposit.

The importance of the property of diffusivity or mass transfer coefficient is also stressed; the

higher the diffusivity of an ion the less likely it is to be depleted at the cathode and the more

likely a coherent layer will grow. The k values can also be used to determine the limiting current

that can be used to avoid the anode effect. A secondary significance of mass transfer was shown

in a couple of studies [41, 42] where the dissolution rate of alumina particles into the cryolite

melt determined whether the particles would completely dissolve before it reached the bottom of

the crucible or sink into the molten aluminium at the bottom of the crucible and become an

inclusion or be lost in the sludge.

Diffusivity, D, and mass transfer coefficient, k, are related as [43] with a constant of

proportionality called the effective film thickness, δ, which needs to be calculated based on k and

27

D values. Depending on the difficulty of handling the material and equipment at hand, one

property may be favoured over the other for measurement. Both properties are related to the

dissolution rate of a solute into a solvent. With studies of alumina in cryolite several researchers

found that some additives such as LiF, CaF2, MgF2, AlF3 and NaCl slowed the dissolution rate

while others like NaF increased it [28]. It has been found that silica dissolution into cryolite was

very slow indeed [44] but no quantitative measurements were made. This will be one of the areas

this thesis will be exploring along with solubility limits.

2.6.1.1 Measuring solubility

Although Grjotheim et al. mention that the maximum solubility of silica in cryolite is 5wt% at

1000°C [45] they have not mentioned what experimental procedure they had used. For alumina

solubility in cryolite Zhang et al. [46] added excess alumina to cryolite and the mixture was

stirred with a platinum rod at regular intervals of 20 minutes. A small sample was then quenched

onto an alumina rod which was then digested (although the solvent they used is not mentioned)

and analysed using inductively coupled plasma (ICP). The total aluminium ion content was

compared to a “blank” (a pure cryolite sample) and the additional alumina in the sample was

calculated which gave an indication of the solubility. It is important to note here that AlCl3 is a

well-known solvent for dissolving cryolite and it does not dissolve alumina [47]. The problem

with applying this method to silica is that when silica dissolves in cryolite it reacts to form a

number of compound that do not dissolve in AlCl3 solution. When using AlCl3 solution a gel like

substance was created which was believed to be silica gel. This does not give a true

representation of the solubility of silica in cryolite.

Jentoftsen et al. [48-50] found the solubility of FeO, FeAl2O4, NiO and NiAl2O4 in cryolite by: 1)

dissolving an excess amount of each oxide in cryolite and then quenching the sample, 2)

carbothermically reducing all excess oxides, 3) digesting the sample with hydrochloric acid to

dissolve Fe or Ni metal and the using ICP to determine Fe or Ni content in the sample. This is

not feasible for silica in cryolite because after dissolving in cryolite silica starts reacting with it.

Then they found the solubility of TiO2 in cryolite by: 1) dissolving an excess amount of each

oxide in cryolite and then waiting 90 minutes for the TiO2 powder particles to settle in the melt

and then quenching the sample, 2) using X-ray fluorescence (XRF) which does an elemental

analysis to find titanium content. Solubility of TiO2 in cryolite at 1020°C is 5.2wt%. This method

28

is clearly not very accurate as settling of very fine particles can be very slow. In particular, in the

cryolite-silica system, the difference in density is small, making the settling even slower. Finally

Jentoftsen et al. found the solubility of CuO and Cu2O in cryolite by: 1) dissolving an excess

amount of each oxide in cryolite and then quenching the sample, 2) digesting the sample and

analysed using ICP. This method is again not applicable for silica.

2.6.1.2 Measuring diffusivity/mass transfer coefficient

The importance of the need to know diffusivity and/or mass transfer coefficient (D or k) has

already been emphasised above. There is an extensive review on different methods of measuring

diffusivity [51] and a few of them were tried to see if they were applicable to cryolite melts. One

of the main problems with cryolite is that it is very corrosive and so far only graphite has been

found not to react with it. Many researchers have mentioned using platinum crucibles to contain

it (for very small samples) and rods to stir it.

Capillary reservoir methods involve having a reservoir which contains the melt and the ion

whose diffusivity needs to be measured, and a capillary tube with only the melt [51]. The

capillary is left in the reservoir and the ion in question diffuses into the tube because of a

concentration gradient. The tube is subsequently quenched and the concentration of the ion is

measured. Some authors [52, 53] used a radioactive isotope to track the movement of an ion

through the melt by gamma-ray attenuation, however there is no silicon isotope with gamma

decay. Chronopotentiometry is another method used to measure diffusivity but because silicon

ion mobility is much slower than sodium and fluorine especially since it creates complexes in the

melt it does not contribute significantly to the measurements therefore is hidden in the variation

of the results. A final method described by Kubíček [51] was forced convection dissolution of

rotating discs and measuring the loss of weight with time as shown in Figure 8. Knowledge of

viscosity and angular velocity will allow determination of diffusion coefficient. This method was

modified to a stationary rod with natural convection and used for this thesis, Figure 9. The mass

loss with time can be fitted to the general mass transfer equation to determine k.

m = mass of silica rod dissolving in melt (g) (only considering the portion inserted in the melt)

t = time of dissolution (h)

29

k = mass transfer coefficient (cm/h)

A = area of silica rod exposed to the melt (cm2)

CS = saturation concentration of silica in melt at the temperature of experiment (g/cm3)

CB = concentration of silica in the bulk at time t (g/cm3)

Figure 8 Schematic for measuring diffusion coefficients by rotating disc method [51]

Figure 9 Modified schematic to observe dissolution behaviour for this study

30

2.6.2 Measuring density

Once the silicon is deposited, it should be protected against the atmosphere to prevent its

oxidation. When aluminium is deposited as liquid, it sinks to the bottom and the floating

electrolyte prevents it from oxidising. It is necessary that either the liquid silicon (or alloy of) has

a higher density than the electrolyte or that the surrounding atmosphere is inert. Work done by

Grjotheim et al. [45] is shown in Figure 10; the density variation with alumina and silica

additions to pure cryolite. For both instances increase in the oxide concentration, the researchers

measured a decrease in density.

As mentioned, Grjotheim et al. [45] determined the solubility of silica in cryolite at 1000°C to be

5wt% which is why the additions for silica stop at 6wt%. Both oxides decrease density but silica

slightly less than alumina, as shown in Figure 10. The decrease in density for alumina-cryolite

melts is explained by the aluminium complexes (with oxygen and fluorine) that form in the melt

which has a large radius [33]. However for silica addition to cryolite Sokhanvaran et al. [54]

found an increase in density.

Figure 10 Density of the electrolyte with alumina and silica additions to cryolite [45]

2.03

2.04

2.05

2.06

2.07

2.08

2.09

2.10

2.11

0 2 4 6 8 10 12

Den

sity

(g/c

m3)

Oxide Concentration (wt%)

Silica

Alumina

31

For comparison purposes the density of liquid aluminium in the rage of 933 – 1190K (660 –

917°C) is:

where Tref is the melting temperature of aluminium at 933.47K [29]. This indicates that the

molten aluminium will be denser than the electrolyte. The reason why aluminium is considered

here is because of all the considered alloy metals, aluminium has the lowest density so if the melt

has lower density than liquid aluminium then the other liquid metals like copper and tin will not

cause a problem. The bigger the difference between the densities the better because the less

likely the silicon liquid globules that may form due to disturbance of the bath remain suspended.

One author suggested that if the liquid silicon or silicon alloy is less dense then the cell can be

kept in an inert environment or use Down’s cell similar to the one used for sodium electrolysis

[4]. Another way to make sure that oxidation is at a minimum is to deposit the silicon as a solid

(or alloy of) at the cathode which can be positioned so that the electrolyte is floating on top. It

should be noted that in most instances in Table 7 given below solid alloying of silicon was

unintentional; the silicon reacted with the cathode to form an intermetallic silicide. In some cases

[15, 55] the silicide formation was used advantageously to create a hard coating e.g. iron silicide.

The most common way of measuring density is using Archimedean principle. Grjotheim et al.

[45] measured the density of Na3AlF6-SiO2-Al2O3 using the buoyancy method with Pt-10Rh

sphere. This method was modified for this thesis by using a nickel ball for the sink instead of

platinum; nickel reaction with the CR+LiF melt is negligible.

32

Table 7 Conditions of various studies on electrolysis of silicon

Year Authors Electrolyte Silica Source Cathode Silicon form Working

Temperature

Silicon

Purity

Molten Salt Electrolysis 400-1150°C

Pure Silicon (or attempt)

1854 [56] Deville AlCl+NaCl

Accidental silicon

impurity in the

melt

Graphite Crystalline silicon ? 10.3%

1854 [56] Deville NaF + KF + SiO2 SiO2 Platinum Platinum silicide (intermetallic;

melts and disappears into the bath) ? -

1865 [57] Ullik KF+K2SiF6 K2SiF6 - Metallic K and Si (immiscible

alloy?) ? -

1884 [58] Gore K2CO3+K2SiF6 K2SiF6 Platinum Platinum silicide ? -

1888 [59] Hampe NaSiO3 NaSiO3 Carbon

Amorphous silicon (extracted by

reacting with hydrochloric acid and water)

? ?

1934,39 [60, 61]

Dodero

Alkali and alkaline earth metal

silicates with fluoride additives to

lower temp and viscosity.

From electrolyte Iron

Cathode reacts with deposited Si.

Some alkali metal co deposited as

well.

800-1250°C 72%

1951 [62] Wartenberg et al. NaCl + K2SiF6 K2SiF6 Nickel Brown needles with 0.1%Fe,

0.05%Ni, 0.1% Na & K 900°C >99%

1959 [63] Stern at al. Alkali chloride (or mixture) +

K2SiF6 (electrorefining) SiC anode Steel

Solid crystalline (easily broken

away from cathode). 400-1000°C 92-99.3%

1959 [64] Olstowski Alkali and/or alkaline earth fluoride

or AlF3 mixtures + Na2SiF6 Na2SiF6 Iron or copper

Initially intermetallic silicide and

after saturation some silicon 450-800°C ?

1957-64 Monnier et al. Na3AlF6 + SiO2 (electrorefining) SiO2 MG-Si at anode to deposit Si

at cathode Solid silicon 1000°C 99.99%

1968 [65] Jorgensen

Thin beds of solid SiO2 (obtained

from oxidation of p-type silicon wafers – this experiment was done

to obtain reduction voltages.)

SiO2 Silicon Polycrystalline silicon 800-880°C Depends on

purity of

silicon wafer

1971 [55] Lyakhovich et al. Na2O + SiO2 SiO2 Armco iron or steel Intermetallic of iron and silicon 1050-1150°C -

1975-77

[66-68] Cohen Alkaline fluorides + K2SiF6

Silicon cathode

5N purity Monocrystalline Silicon Epitaxial monocrystalline silicon 600-850°C ?

1980-81 [19,

21]

Rao, Elwell &

Feigelson

Flinak-Eutectic LiF+NaF+KF and

K2SiF6 or

LiF+KF and K2SiF6

K2SiF6 Silver or graphite Polycrystalline silicon ~745°C 99.98-

99.999%

1981 [19,

69]

Rauh and separately

De Mattei High temperature fluorides K2SiF6 Silicon Crystalline silicon 700°C ?

1981 [20] Elwell NaF + CaF2 + SiO2 SiO2 Graphite Discrete silicon particles near

cathode. 1150°C ?

33

Table 7 (Cont’d) Conditions of various studies on electrolysis of silicon

Year Authors Electrolyte Silica Source Cathode Silicon form Working

Temperature

Silicon

Purity

1981, 83

[70, 71] Olson & Carleton LiF + KF + K2SiF6 (electrorefining) MG-Si-Cu alloy

anode Si with Cu addition at anode Solid polycrystalline silicon 750°C 99.999%

1981 [72] Olson & Kibbler Na3AlF6 + LiF + AlF3 + BaF2 SiO2 Liquid Sn Si phase separate in Sn with Al, Ba

and Sn impurities. ? ?

1996

[73] Stubergh and Liu

Bytownite (mixture of CaAl2Si2O8

and NaAlSi3O8) + Na3AlF6 Bytownite Graphite

Solid polycrystalline silicon with Fe and Si3Fe impurities. Fe from

bytownite carbothermically reduced

by graphite.

970°C 99.79-

99.98%

2005

[74] Yasuda et al. LiCl + KCl + CaCl2 SiO2 Silicon

Solid amorphous and microcrystalline silicon; sponge-

like particles at 50nm

500°C ??

Alloy of silicon

1891 [75] Minet NaCl+Na3AlF6+(Al2O3 or FexOy) +

SiO2 SiO2 Carbon

Al-Si alloy (liquid?) or solid Fe-Si

alloy 700-1000°C -

1893 [76] Warren Alcohol solution + SiF4 SiF4 Mercury Silicon amalgam ? -

1933 [77] Batashev & Zhurdin Na3AlF6 + Al2O3 + (Al2Si2O5(OH)4

or SiO2)

Al2Si2O5(OH)4 or

SiO2 Al-Si alloy ?

27% of silicon in

alloy

1957-64 [18,

23, 24, 78-

81]

Monnier et al.

Mixture of NaF, LiF or KF and

K2SiF6 (electrowinning). Other

electrolytes were tried; please see paper for full detail.

K2SiF6. Other

sources were used

like SiO2 in

different

experiments

Liquid Cu Si-Cu alloy 1000°C 4-25%

silicon

1970-75 [45,

82-89]

Grjotheim, Matiasovsky, Fellner

& Bøe (et al.)

Mainly Na3AlF6 + Al2O3 + SiO2; properties of and deposition of

alloys of Silicon

SiO2 Various: e.g. graphite, copper, aluminium and

nickel

Liquid alloys of silicon. Solid

silicon deposited on graphite. >1000°C -

Low temperature electrolysis below 100°C

1966 [90] Zyazev & Ezrielev propylene

glycol or pyridine + KI, NH4Cl SiF4 or SiCl4 Platinum Amorphous silicon ? ?

1976 [15] Austin

Aprotic dipolar organic solvent one

of: propylene carbonate, tetrahydrofuran, acetonitrile,

dimethyl sulfite,

tetramethylenesulfone, ethylenecarbonate, N,N-

dimethylacetamide, and

dimethylformamide

SiX4 (X = Br, Cl

or I) or HnSiX4-n For electroplating purposes

Amorphous Si:H (Other electrolytic works in organic

mixtures have been done but all

obtain a-Si:H which cannot be used for high power solar cells because

of lower efficiency.)

20-100°C -

Melting temperature above Silicon >1400°C

1981 [22] De Mattei, Elwell &

Feigelson BaO + SiO2 + BaF2 SiO2 (99.5% pure) Graphite

Liquid silicon (NB: efficiency was

only ~40%) >1450°C 99.97%

34

Chapter 3 Experimental

This chapter explains the experimental procedures, apparatus and materials used to obtain three

properties with two types of experiments:

Dissolution behaviour:

Solubility limit of silica in the CR+LiF electrolyte at temperatures between 900 and

1000°C.

Dissolution rate of silica in the CR+LiF electrolyte at temperatures between 900 and

1000°C.

Density measurements:

Density of the CR+LiF electrolyte with varying silica concentrations at temperatures

between 900 and 1000°C, using Archimedean principle.

Each experimental technique will be presented in two main sections. For each property

measurement, the most adaptable method was employed based on the salt that was being used

and the equipment at hand. For instance the corrosive nature of the cryolite melt and the working

temperature had to be taken into account when designing experimental methods. Under the

dissolution behaviour section various approaches that were examined in the present study will

also be discussed to highlight the difficulty of working with cryolite.

A further section is added to explain an attempt at measuring conductivity of the CR+LiF melt

with varying silica concentration. The Van der Pauw method was unsuccessfully employed and

what transpired and why it did has been described in this section.

35

3.1 Dissolution behaviour

The experiment explained in this section was used to quantify the dissolution behaviour and

solubility limit of silica into the CR+LiF electrolyte at various temperatures.

3.1.1 Examined Approaches

1. SEM and EDS of quenched CR+LiF+Silica samples

An initial study was carried out to determine solubility limit of silica in the CR+LiF sample. The

concept was to dissolve different amounts of silica in the CR+LiF melt and quench the crucible

at various temperature, then observe under the SEM the different phases that would be formed if

the sample was oversaturated with silica. EDS with point analysis could be used to determine

concentration of silica in each of the phases including the saturated liquid. However even though

the sample size was varied from 2 to 15 grams, water quenching was not quick enough as the

subsequent analysis revealed. The crucible used was thick graphite; to contain the corrosive

cryolite and prevent being completely consumed by oxygen. SEM and EPMA analysis showed

that the silicon is concentrated into globular phases within a matrix as shown in Figure 11. The

SEM image shows a sample containing 6wt% silica quenched at 1000°C and on the right

elemental mapping of silicon. EPMA analysis of the sample showed the globular phases within a

matrix had high concentrations of silica, 20-40wt%, whereas the matrix itself had 1-2wt% silica

content. Previous authors had concluded that cryolite, when oversaturated, will react with the

silica to form combinations of sodium, aluminium and silicon oxides, (Na.Al.Si)Ox, called albite,

jadeite and nepheline [91] which are high in silica content. Which particular mineral is formed is

time dependent and it was concluded that the cooling rate was too slow allowing the formation of

the albite/jadeite/nepheline phases.

Visual observations of the sample showed a “shiny” phase which could be observed at 4-6wt%

silica depending on the sample size as shown in Figure 12. This value depends on the size of the

sample, therefore the smaller the sample the more accurate the method. However it can be only

estimated to the nearest whole number and this may lead to the same value for a range of

temperatures, therefore this method was deemed too imprecise to predict solubility limit.

36

a) b)

Figure 11 a) SEM Image of a 6wt% sample b) Si EDS of sample

Figure 12 Sample size of 2g containing 6 wt% silica

“Shiny” phase believed to

be high in silica content

37

2. ICP Analysis of oversaturated CR+LiF+Silica samples

Another attempt at measuring solubility limit was made using a chemical approach; 10wt% of

silica was added to CR+LiF melt to make 10g samples. Blank CR+LiF samples with no silica

were also made for comparison. These were quenched at different temperatures, crushed to

powder and leached with 15wt% AlCl3 solution which can dissolve cryolite. The rationale was

that any silica dissolved in the cryolite will be able to stay in solution whereas the undissolved

silica will remain as a solid and can be filtered out. However the silica reacted with cryolite to

form oxide compounds of albite/jadeite/nepheline phases which reacted with the AlCl3 solution

to form silica gel and blocked the filter paper corrupting the results.

The mass of the residue from filtering out the powder and solution indicated a decrease in

solubility with increasing temperature which is counterintuitive. Furthermore ICP of the filtrate

showed very little silica content and when a mass balance was conducted, the masses of the

residue and that calculated from the filtrate did not add up to the initial mass of silica in the

sample. The main reason for this is the loss of silica content as silica gel. Moreover the excess

silica is not believed to remain unreacted in the CR+LiF melt; it is supposed that it will further

react with cryolite to form the aforementioned albite/jadeite/nepheline phases. Therefore a true

solubility limit cannot be determined and this method was deemed impracticable.

3. Diffusivity profile in silica tubes

A small silica tube was inserted in a molten CR+LiF mixture to draw molten salt up the tube due

to capillary action and slowly dissolve into the cryolite. Ideally, a diffusion profile along with a

saturation concentration can be obtained if the CR+LiF inside the tube is analysed with

EDS/EPMA, Figure 13. In that case, the method could give both diffusion coefficients and

solubility limits. Different tube inner diameters and different immersion times were

experimented to obtain a reasonable profile.

38

Figure 13 Anticipated diffusion profile across a silica tube with CR+LiF melt

The whole crucible had to be quenched since extracting only the silica tube would disturb the

profile as some of the liquid would flow out the bottom. Unsuccessful attempts were made to

seal the end of the silica tube but quenching the whole crucible was deemed safer and more

accurate. The graphite crucible had outer diameter of 6cm, inner diameter near the top of the

crucible of 5cm (inside crucible is conical in shape), height of 8cm. Again due to slow cooling

there were phase separation and EPMA point analysis did not produce any meaningful results.

However with EDS average area analysis, Figure 14, a reasonable profile of silica concentration

across the tube was obtained, Figure 15. The black squares in Figure 14 indicate the areas that

were analysed for silicon (which was then converted to silica concentration) and the centre of the

box was taken as the distance from the edge. Area analysis ensured that the phase separation due

to albite/jadeite/nepheline formation was averaged out and Figure 15 shows a reasonable value

for saturation concentration of 5-7wt% (as opposed to 20-40wt% in the mineral oxide formed).

However as can be seen from the graph it is not a parabolic curve as was expected.

8cm

6cm

39

Figure 14 Diffusion Experiment Sample

Figure 15 SiO2 profile across a 2mm tube held in CR+LiF melt for 30 minutes

The silica walls dissolve away creating circular flow motion of the melt near the walls which

disturbs the diffusion profile as shown in Figure 16. There is a significant problem involved in

this experimental method; natural convection due to ΔT (temperature) and Δρ (density). ΔT can

be eliminated if the sample is small enough and placed near the centre of the furnace. Δρ ,

0

1

2

3

4

5

6

7

0 500 1000 1500 2000

SiO

2 (

wt%

)

Distance (μm)

40

created due to silica dissolving into the CR+LiF melt and making a denser composition, can be

eliminated by using the capillary reservoir methods.

Figure 16 Effect of convection flow on the concentration profile

4. Capillary reservoir methods

This approach involves having a reservoir and a capillary tube made of graphite, shown in Figure

17. The reservoir contains a CR+LiF and silica melt, whereas the capillary tube is filled with

CR+LiF only. The idea is that the silica from the reservoir will diffuse into the capillary tube; the

highest concentration at the opening of the tube and a concentration profile into the tube. This set

up allows for the higher density silica+electrolyte melt to be near the bottom.

Figure 17 Capillary reservoir to measure diffusivity

Capillary tube:

melt (no silica)

Reservoir: melt

with silica

41

However this method was attempted without success; one of the main problems was that graphite

floats on cryolite so the capillary tube cannot be dropped into the melt and expect diffusion to

happen. There are other drawbacks to this method:

1. Convection due to temperature and density gradient. These can be overcome if the setup

is small enough and so it is all at one temperature and if the higher density solution is

created at the bottom of the capillary tube so as to prevent fluid motion.

2. Boundary conditions are uncertain even though at the entrance [CSi] is defined as the Si

concentrations in the reservoir.

3. Wall effects that create a drag on the diffusion.

4. During solidification the volume contraction can change the distribution affecting results.

3.1.2 Selected Method

Quartz rods were immersed in the electrolyte and the mass loss and change in radius were

measured against the immersion time. The results were treated to yield solubility limit and mass

transfer coefficient of silica ions. Cryolite is a very corrosive compound, thus graphite crucible

was used. However the graphite at elevated temperatures will react with the oxygen inside the

furnace so argon gas purging was necessary to keep the atmosphere inside the furnace as inert as

possible. Quartz rods (impurity less than 25ppm) 2.5cm in diameter and 10 cm in height were

inserted in the molten sample size of 200g CR+LiF through a hole in the crucible cap (also of

graphite). The schematic in Figure 18 shows the overall setup used to observe dissolution

behaviour. The quartz rod was left resting at the bottom of the crucible for various periods of

time. The fume hood is necessary for this set up since silica reacts with the CR+LiF melt

producing SiF4 gas.

It should be noted that a study was done by Gairola et al. [92] on how the dissolution behaviour

differs when a nickel cylinder is suspended in liquid aluminium versus resting it at the bottom of

the crucible. Resting the cylinder at the bottom of the crucible interrupts the free convection flow

leading to different dissolution rates; the suspended cylinder dissolves quicker. In this study it is

believed that by resting the cylinder at the bottom of the crucible accelerated dissolution around

the edges of the cylinder is eliminated hence the dissolution is more even around the cylinder.

42

Mass loss and reduction in radius of the rods were measured after immersion. Mass

measurements are preferred over radius measurements because as Figure 20 shows 1) radius can

vary along the length of the immersed portion of the rod; the biggest change in radius is near the

surface of the melt where the velocity of the melt is highest and 2) the immersed portion did not

have a smooth surface. One experiment consisted of keeping the temperature constant and was

run for a total of 8 hours, with different dissolution time periods for 6 rods: 15min, 15min,

30min, 1h, 2h, 4h. Figure 19 shows the rods after immersion; from left to right the cumulative

immersion time increases. After removing from the crucible, the rods were cleaned with AlCl3

solution by immersing them into the solution and using a hot plate and magnetic stirrer for 8

hours to dissolve any solidified melt on the surface of the rods. This procedure was done for five

temperatures: 920, 940, 960, 980 and 1000°C and repeated three times.

Argon Gas

Inlet

Furnace

Silica

Rod

Electrolyte

Figure 18 Schematic of the experimental setup for dissolution behaviour study

43

Figure 19 Rods after the experiment

Figure 20 Immersed portion of rod

0.25 h 0.5 h 1 h 2 h 4 h 8 h

44

For each temperature, the cumulative mass loss was plotted against time and the curve fitted to

the mass transfer equation described below to find the values for mass transfer coefficient and

solubility limit.

To calculate the mass transfer coefficient k and solubility limit CS from mass loss-time data, the

general mass transfer expression, Equation (6), was integrated to get an expression where mass

of the silica rod, m, was expressed in terms of time, t.

General mass transfer equation:

(6)

where:

m = mass of silica rod dissolving in melt (g) (only considering the portion inserted in the melt)

t = time of dissolution (h)

k = mass transfer coefficient (cm/h)

A = area of silica rod exposed to the melt (cm2)

CS = saturation concentration of silica in melt at the temperature of experiment (g/cm3)

CB = concentration of silica in the bulk at time t (g/cm3)

The area A and bulk concentration CB were expressed in terms of m and the equation was

integrated over time (full derivation is found in Appendix B) to yield Equation (7).

(7)

Where C1, C2 and C3 are constants defined as:

|

√ √

√ √ |

√

( √

)

V = volume of melt is assumed to be constant with dissolution of silica because the maximum

amount of silica dissolved is less than 9g into a melt of 200g. Also at each time (15min, 30min,

1, 2, 4, 8hrs) a fresh rod of initial radius of 1.25cm is inserted and the change of radius is less

than 0.06cm.

45

h = height of melt after the rod is inserted into melt, also assumed constant for reasons given

above (associated change in height due to change in volume is less than 0.3cm).

Others constants: m0, ρ, π, CS, and k.

m0 = initial mass of silica rod inserted in the melt

ρ = density of silica rod

In order to fit the data into Equation (7), the MS Excel Solver program was used. CS and k were

set to vary, while the sum of square of errors, defined as the difference between calculated and

measured masses, was minimised. It should be emphasized that the cumulative mass loss at 8

hours was not used as the saturation limit (CS) because it was assumed at 4 hours it had already

reached saturation and further dissolution was in fact reaction of silica with the cryolite. The k

and CS were also calculated using data from one experiment that ran for 16 hours and then using

data points up to 4 hours to see how this affected the calculated k and CS values.

In order to test the reproducibility of the results, dissolution of quartz in CR+LiF melt was

repeated three times at 1000°C. At 1000°C one can expect the highest velocity of fluid causing

the most amount of uneven dissolution. Figure 21 showed initially that the experiment was

indeed reproducible.

Figure 21 Reproducibility of dissolution behaviour experiment

0

1

2

3

4

5

6

7

8

9

0 2 4 6 8

Cu

mu

lati

ve

Mass

Loss

(g)

Time (h)

exp1

exp2

exp3

46

During the experiment, SiF4 can be released from the reaction shown in Equation (8) [91]:

SiO2 + Na3AlF6 = xNa2O.yAl2O3.zSiO2 + NaF + SiF4↑

(8)

The values for x, y and z indicate which of albite, jadeite or nepheline is formed. Since it is

thought that jadeite is formed initially and with time decomposes to albite and nepheline, all

three may be present in the melt.

The SiF4 is thought to further react with water vapour as shown in Equation (9):

SiF4 + H2O SiO2 + HF

(9)

which tends to shift to the right at temperatures above 800°C (HSC Chemistry 6.1) when the

Gibbs’ free energy becomes negative. The SiO2 deposits as a white solid that was observed on

the top part of the silica rod. The deposit could not be dissolved away with AlCl3 solution so it

cannot be cryolite. XRD analysis (Appendix C) on the deposit showed silica of cristobalite β

structure which is different from the quartz crystal structure the silica rods are made of indicating

the silica formed condensed onto the silica rod. This deposit will either increase the weight or

make it so brittle that pieces fall off decreasing the weight affecting the results either way.

Sometimes the mass loss and gain may cancel each other out because some deposits stay intact

and some fall off. It was found that if the hole in the lid has a diameter very close to the diameter

of the rod (so as to contain the SiF4 within the crucible) this reaction is reduced leading to more

accurate results.

Errors in this methods are due to:

Variations in experimental measurements; deduced from repetition.

The quartz at the bottom of the rod sometimes chips off because of cracks due to

temperature differences when the silica is removed from the high temperature furnace. A

loss of 0.05 – 0.1g can be expected.

SiF4 reactions can affect the mass measurements especially at low temperatures where

mass loss is low and a difference of 0.1g can have a high percentage change.

47

3.2 Density Measurement

The experiment explained in this section was used to measure density of the CR+LiF electrolytes

with varying silica concentrations and temperatures, based on the Archimedean principle.

A graphite crucible was used to contain 300g of CR+LiF melt. The reason for the larger melt size

is that the height of the melt was enough to accommodate an entirely immersed sink without the

walls of the crucible affecting the measured results. Again the fume hood is necessary for this set

up since silica is added to the CR+LiF melt and produces SiF4. The furnace was also purged with

argon between measurements. One whole experiment consisted of a given silica concentration in

the CR+LiF melt while temperature was varied from 920 to 1000°C in 20° intervals and density

measured at each temperature. For each new experiment the concentration of silica (99.5%

purity) was increased in 0.5wt% increments from 0 - 4wt%. The melt was stirred and the

temperature was measured before each density measurement.

A sink, made of nickel ball and held up by a Ni-Cr wire, was used to measure the density of the

molten CR+LiF melt. To avoid errors made from fluctuations in mass due to natural convection

currents the sink was weighed down using a steel rod threaded through the wire. The manner in

which the ball was attached to the wire also ensured that the entire set-up was buoyed as opposed

to only the nickel ball, shown in Figure 22 on page 49.

Before each new experiment the mass of the sink was measured. An estimate was made of how

much of the sink (ball and wire) would be immersed in the CR+LiF melt and this same volume

was immersed in distilled water at room temperature and the apparent mass was measured. The

difference in masses and the known density of distilled water (0.9956 g/cm3 at 24°C) allowed

calculating the density of the sink that would be immersed in the melt according to Archimedes’

principle in Equation (10).

(10)

Density of water was used to calculate density of sink which is assumed to be entirely pure

nickel (even the wires). Volume of the immersed sink at ambient temperature was calculated

48

using its measured density and mass. This volume in conjunction with the linear thermal

expansion coefficient of nickel (α = 13.4m.m-1

.K-1

) were used to calculate the volume of the

immersed sink at different temperatures, Equation (11), and hence the densities at each of these

temperatures. This is then used to calculate the density of the CR+LiF melt using Equation (10).

(11)

The sink was suspended from a mass balance which was connected to the computer to record

values every 5 seconds for 30 readings at a given temperature. Figure 23 shows the schematic of

the experimental set up for density measurements. The sink was allowed to come to equilibrium

at the temperature in the melt so as to avoid error due to melt solidification on the sink and

expansion of the sink.

The first density measurement was done at 920°C and the temperature was increased in

increments of 20° from 900 to 1000°C each time a density measurement made. Then the

temperature was decreased in decrements of 20° back to 920°C and again density measurement

was made at each temperature. This procedure was repeated 3 times for each melt composition.

The reason for measuring the density while decreasing the temperature was because of the silica-

cryolite reaction described in Equation (8), which changes the composition of the melt during the

experiment. This means that the density measured while heating the melt might not necessarily

be the same while cooling the melt, since this reaction is both time and temperature dependent.

Furthermore this might lead to variation in the results when the experiment is repeated that may

be mistaken as errors. The results will be further discussed in Chapter 4.

As can be seen in the photograph in Figure 23 the furnace is open and allows for the escape of

SiF4 gas. Using Equation (10) the density of the melt was calculated for varying temperatures

and varying silica concentrations.

49

The Ni-Cr wire was twisted around the nickel ball as

to avoid any independent buoyancy of the ball as

opposed to the whole sink; ball, wire and steel rod.

The photograph shows the the sink used for this

method.

Figure 22 Set up of the sink to be immersed in the

CR+LiF melt.

50

Errors in this method are due to:

Fluctuations of the mass readings due to natural convection currents in the melt varying

between 0.1-0.4g per 70-80g measured.

Solidification of the melt onto the nickel ball after saturation concentration of silica has

been reached at a particular temperature thereby giving false results (explained in the

results and discussion section).

Slight calculation errors due to considering Ni-Cr wire has the same density and

expansion coefficient as Nickel ball; this error is not as significant as the variations due to

the change in composition and therefore not taken into account. The average mass of the

ball is 21.65g and the average mass of the wire that was wrapped around the ball was

0.75g which is 3% of the whole mass. Calculation error in assuming the density as the

same for both ball and wire will contribute 3% error at most. Whereas the density error

due to compositional change was on average 0.23g/cm3 for 2.25g/cm

3 which is over 9%.

Figure 23 Schematic of the experimental setup for density measurement

Furnace

Mass Balance

Argon Gas Inlet

Electrolyte

Sink

51

3.3 Conductivity Measurement

An attempt was made at measuring conductivity of the CR+LiF melt with varying concentration

and temperatures. The four probe electrode had already been used successfully to measure

conductivity for cryolite-silica melts and oxide melts [93, 94]. The set up for this method is

shown in Figure 24.

Figure 24 Schematic of the experimental setup for conductivity measurement

52

The resistance of the melt Rmelt is described by equation

Rmelt = (l/A) = (1/σ) (l/A) (12)

Where ρ is resistivity and σ is conductivity. The cell constant is described as l/A and calibration

is needed with a standard solution whose conductivity is already known to determine this

constant. The resistance is measured using electrochemical impedance spectroscopy (EIS) using

a potentiostat and a frequency analyser. An AC voltage is applied at varying frequencies and the

responding current measured. From these values a Nyquist plot is made and the resistance value

Rmeasured is read off the graph where it intersects the real axis (resistance with units of ohm).

Figure 25 shows a typical Nyquist plot when measuring impedance and the inset shows a close

up of where the curve intersects the real axis.

Figure 25 Typical Nyquist plot for impedance spectroscopy

53

However R measured = R melt + Rleads + R electrode therefore the Rleads + R electrode should be measured

by short circuiting the electrode with a copper foil and using EIS.

There are drawbacks to this method [95]:

The cell constant of the electrode cell is determined using a solution or melt whose

conductivity is already established in literature. However the cell constant is a function of

temperature and the conductivity of the liquid. To avoid cell constant change due to

expansion of the electrode at higher temperatures, measuring it at working temperatures

using a molten salt is recommended. However to avoid the latter of these errors the

unknown conductivity of the new liquid needs to be similar to that of the liquid used for

calibration.

There are fringe effects when the electrodes are immersed in the melt.

To avoid these errors the Van der Pauw method was attempted [96]. The arrangement of the

electrodes varies as shown in Figure 26. The schematic on the left is the ordinary four probe

method used by Sokhanvaran et al. [93]. The drawing on the right is the Van der Pauw

arrangement.

Figure 26 Two arrangements for four probe electrode

The method uses four electrodes arranged in a square. Again EIS is used and voltage V1 is

applied and current I1 is measured at height H1 in the melt. To make sure any discrepancies are

eliminated due to imperfect square arrangement voltage V2 and I2 are also measured at H1 and

an average resistance RH1 is calculated. This procedure is repeated at another height H2 in the

melt to get RH2; a total of 8 measurements are made for one conductivity value. This is time

54

consuming but eliminates fringe effects and the need for cell calibration. The resistivity is

calculated by Equation (11).

(13)

The conductivity, σ, is the inverse of resistivity σ = 1/ρ. However this method also has its

drawbacks in that results are affected by the crucible walls, depth of immersion and the diameter

of the individual probes [96].

This method had been previously used [97] to measure the conductivity of cryolite using

platinum electrodes. Platinum leads to the platinisation effect whereby the values obtained are

inaccurate because of catalytic reactions on the platinum surface leading to solvent

decomposition [98] and increasing apparent conductivity.

The method was modified for the present work for pure cryolite by using molybdenum rods as

electrodes and a voltage of 50mV was applied and the current measured. However an unforeseen

problem was encountered; one of the current electrodes was dissolving up to the length that was

immersed in the melt at such a quick rate that the electrode would lose the immersed length at

the end of one experiment. The only difference between the Van der Pauw method used in this

study and the four probe method used by Sokhanvaran [93] was the arrangement of the

electrodes which leads to conclusion that the electric field created by the electrode arrangement

is the reason for the dissolution of the molybdenum. The following reaction is believed to happen

at the positive counter electrode:

Mo Moz+

+ ze-

The Mo ion goes into the melt. However a deposition of the Mo ion is not seen at any of the

other electrodes. This may be because the electrons at the negative electrode are used to reduce

some other ion in the melt or the deposition of Mo is not noticeable because the electrodes are

slightly oxidising during the experiment and MoO2 created masks any deposition.

This method had to be abandoned because the measurements could not be trusted; the shape and

size of the electrodes were changing during the experiment. The initial results obtained from pure

cryolite are reported below and compared to that obtained by Sokhanvaran et al. in Figure 27.

55

There are other results for pure cryolite available but the cryolite used in this study is from the

same batch as used by Sokhanvaran. The measurements stop at 1080°C because the electrode

was completely dissolved and the experiment was not repeated.

Figure 27 Comparing results from different experimental setup

Figure 27 shows that the Van der Pauw results are higher; this could be due to the dimensions of

the electrode changing thus affecting the results.

1.5

2.0

2.5

3.0

3.5

4.0

4.5

1000 1020 1040 1060 1080 1100

Con

du

tivit

y (

S/c

m)

Temperature (°C)

This study - Van der Pauw

Sokhanvaran - four wire [89]

56

Chapter 4 Results and Discussions

4.1 Dissolution Behaviour

A typical mass loss versus time graph looks like Figure 28. The graph shows the experimental

values obtained at 1000°C and the theoretical curve using calculated k and CS values. As will be

explained later k and CS values were calculated based on data points up to the 4 hour experiment

time, hence the last data point in Figure 28 does not coincide with the model.

Figure 28 Typical mass loss versus time graph (1000C)

The experimental values were used to fit the data to Equation (7) and minimise sum of errors to

obtain values for mass transfer coefficient k and solubility limit CS. According to Equation (14)

the initial slope of the graph is a function of k, CS and the initial area A of silica rod exposed to

the CR+LiF melt (since CB = 0 initially).

(14)

0

1

2

3

4

5

6

7

8

9

0 1 2 3 4 5 6 7 8

Cu

mu

lati

ve

Mass

Loss

(g)

Time (h)

experimental

model (best fit)

57

This is expected since the initial change in mass would increase by increasing the surface area

exposed to the melt. CS, being the “driving force” for the solute to dissolve into the solvent, the

higher the CS value the more mass that can be dissolved. As time tends to infinity the graph tends

to a horizontal asymptote, i.e. dm/dt 0 and Equation (6) would indicate that either CB CS or

A = 0. Since observation indicates that the whole rod is not dissolved then A cannot equal 0

which means that CB approaches CS. Figure 29 shows the results calculated for mass transfer

coefficient k for experiments done at various temperatures and it increases with increasing

temperature.

Figure 29 k versus temperature

The values of k were fitted to an exponential curve because k is a function of D (diffusivity):

and D is an exponential function of temperature

. n can vary between 0.5

and 1 depending on the chemicals. n = 0.5 when the diffusion coefficient or time of exposure of

solute in solvent is small; fluid comes in contact with the solid, mixes and then moves away. n =

1 when diffusion coefficient or time of exposure is large and transfer of solid is through a

diffusion layer. In realistic experiments n is a value between 0.5 and 1 and for this study it is

more likely that n is closer to 0.5 because of the slow dissolution rate (hence k value) and small

time of exposure due to high convection currents at high temperatures.

0

1

2

3

4

5

6

7

8

9

10

900 920 940 960 980 1000 1020

Mass

Tra

nsf

er C

oef

fici

ent

(cm

/h)

Temperature (°C)

58

As is expected from the relation between k and D, the mass transfer coefficient increases with

increasing temperature. Diffusivity is difficult to measure directly through the standard

electrochemical process because silicon is speculated to form a large complex molecule in the

melt and the mobility is very small affecting the measurements. Attempts at measuring

diffusivity were abandoned to study dissolution behaviour. However it is necessary to point out

that viscosity affects dissolution rate and therefore the apparent mass transfer coefficient. Since

viscosity changes with temperature, usually decreasing as temperature increases, and

composition as the silica dissolves into the melt, viscosity should be measured to determine the

exact relation with mass transfer coefficient values.

Figure 30 shows the results obtained for solubility limit CS (also calculated with data points up to

4 hour result) which increases with temperature and has a linear relationship between 920 and

1000°C. The line was constructed using least squares regression. The use of the solubility limit is

twofold; 1) to avoid the anode effect from lack of silica or too many inclusions in the deposit

from too much silica, 2) to construct part of a ternary phase diagram; the liquidus line.

Figure 30 CS versus temperature

1.0

1.5

2.0

2.5

3.0

3.5

4.0

900 920 940 960 980 1000 1020

Solu

bil

ity L

imit

(w

t%)

Temperature (C)

59

The dissolution behaviour experiment was conducted for pure cryolite to validate the

experimental procedure. The solubility limit in pure cryolite was calculated, using data points up

to the 8 hour result, and found to be varying between 9-15wt% which is significantly higher than

the value deduced by Grjotheim et al. [38], 5wt% at 1020°C. As explained in Chapter 3 the

EPMA analysis indicated a formation of an oxide phase containing silicon, aluminium and

sodium. After the silica dissolves in the cryolite, once it has reached saturation it most likely

continues to react with the cryolite to form the high silica concentration phases. This may

explain why, when the dissolution behaviour experiment is conducted, the solubility values are

high; the melt becomes saturated with silicon and then compounds of albite, jadeite and

nepheline (based on the duration of the reaction) are formed and precipitate out.

To determine how the time affected the calculated k and CS values, both sets of data were

calculated based on all 6 data points (up to 8 hours) and only 5 data points (up to 4 hours) as

shown in Table 8.

Table 8 Effect of time on the calculated k and Cs

With 8 hour point With 4 hour point

Temperature

(°C)

k

(cm/h)

CS

(wt%)

k

(cm/h)

CS

(wt%)

CR+

13.4wt%LiF

920 2.7 2.2 2.7 2.2

940 4.0 2.2 4.3 2.0

960 4.8 2.8 5.0 2.7

980 5.3 3.2 6.5 2.8

1000 7.7 3.8 9.1 3.5

Pure cryolite 1025 2.3 11.1 4.5 7.5

Generally the percentage change of k and CS values are not significant for the melts containing

lithium fluoride whereas for the pure cryolite melt the percentage change is significant. The

solubility limit has decreased from 11.1 to 7.5wt% which closer reflects the value found by

Grjotheim et al. This seems to indicate that after a while the silicon reacts with the cryolite as

opposed to dissolve in it and the lithium fluoride acts as an inhibitor to this reaction because it

decreases the overall solubility of silica in cryolite. This is a reasonable assumption as lithium

60

fluoride makes up 56 mol% of the melt and might be attributed to silicon complex formation

affected by LiF addition. This was also the reason why when calculating k and CS values only

data points up to the 4 hour result was used; the longer the experiment the more likely the silica

is reacting with the cryolite as opposed to dissolving in it.

It is reported that the addition of LiF decreases the solubility limit of alumina in cryolite and

reduces the dissolution rate [28]. A lower solubility limit might explain why the dissolution

decreases; there is a lower driving force for the alumina to dissolve in the cryolite. Section 2.4 in

Chapter 2 tried to explain why the solubility of alumina would decrease with addition of LiF;

increasing basicity of the melt decreases solubility limit of alumina in cryolite. In this study

addition of LiF decreases the solubility limit of silica in cryolite but increases the dissolution

rate. It is likely that the same mechanism for alumina would explain why the solubility of silica

is reduced with addition of LiF, however the increased dissolution rate might be explained by

examining the mass transfer coefficients.

When comparing mass transfer coefficients the result for pure cryolite was 4.5cm/h at 1025°C

and for CR+LiF melt it was 9.1cm/h at 1000°C. Either this increase in mass transfer is due to the

increased superheat for the CR+LiF melt (1000°C compared to the 15°C for pure cryolite) or it is

because the lithium fluoride increases dissolution rate of silica into the melt. Although at 920°C

the dissolution rate is only 2.7 for the CR+LiF melt this could be due to the lower working

temperature range thus affecting the dissolution kinetics of silica. LiF is reported in literature as

decreasing viscosity of melts; for pure cryolite at 1025°C the viscosity is reported as 2.16mPa.s

whereas at 15wt%LiF in cryolite at 1000°C the viscosity is about 1.7mPa.s, [33]. This decrease

in viscosity will might allow more convection flow in the melt and hence more dissolution. The

effect of viscosity and temperature go together and cannot be distinguished without further

experiments. It is also likely that the Si-O chains are terminated with Li+ ions with more

thermodynamic favourability and hence encourage the dissolution of SiO2 into the melt.

Furthermore if a reaction happens between the solute and solvent, as is applicable to this case,

then the apparent mass transfer coefficient is increased. For now it can be concluded that lithium

fluoride increases the apparent mass transfer coefficient.

61

It should be noted that the experiments with the largest variation when looking at the mass loss

versus time curves were those done at lower temperatures as shown in Figure 31.

Figure 31 Comparison of variation of mass loss for different temperatures

At lower temperatures the natural convection flows are slower and might not average out over

time as it would for higher temperatures hence the dissolution of the silica can be uneven leading

to an uneven dissolution curve. Also as the silica reacts with the melt changing the composition

and hence the melting temperature of the now changed melt, this effect may affect the results

more at lower temperatures. From this it can be concluded that this method is more accurate the

higher the superheat of the melt.

0

1

2

3

4

5

6

7

8

9

0 2 4 6 8

Cu

mu

lati

ve

Ma

ss L

oss

(g

)

Time (h)

1000 - exp 1

1000 - exp 2

1000 - exp 3

920 - exp 1

920 - exp 2

920 - exp 3

62

4.2 Density

The results for density measurements for each temperature with varying silica concentrations are

presented in Figure 32. The data showed a considerable scatter particularly with respect to

temperature. However, for each temperature, with addition of silica the density first increased,

and then when the silica content reached ≈ 3wt%, began to decline.

The density increase in the first part of each curve may be related to the increased silica

concentration because silica has a higher density than the silica-free electrolyte. The decline in

the second part is most likely because the electrolyte has reached saturation and when the nickel

ball is immersed there is precipitation of solid phases on the ball which gives false results. The

actual volume of the sink will be larger than the assumed “clean” sink, resulting in a smaller

calculated density. A similar trend was seen in a parallel work on pure cryolite [54]. The value at

which this phenomenon is observed corresponds to the solubility limits obtained in Section 4.1

for the dissolution experiments.

LiF addition results in a decrease in silica solubility and generally increases density of the melt

with silica addition compared to pure cryolite as can be seen in Figure 33. This graph shows one

set of data conducted at 1010C for pure cryolite, study done by Sokhanvaran [54], and 1000C

for CR+LiF used in this study. This is the opposite behaviour to alumina addition as shown in

Table 6; LiF addition to cryolite – alumina melt decreases density. A comparison of varying

alumina concentration in a fixed ratio of 13.8wt%LiF-Cryolite [28] is made with the melt

composition in this study in Figure 34. Figure 35 (page 65) shows a compilation of all the data of

pure cryolite with oxide and CR+LiF with oxide. With increasing alumina content the density

decreases but with increasing silica content the density increases. Addition of 14wt% LiF

decreases the overall density of cryolite-alumina melt whereas it increases the overall density of

cryolite-silica melt. This indicates one cannot make assumptions for silica using alumina data

and hence further studies on other properties are necessary.

63

Figure 32 Density versus silica concentration for five temperatures

1.5

1.7

1.9

2.1

2.3

2.5

2.7

0 1 2 3 4

Den

sity

(g

/cm

3)

Silica Concentration (wt%)

920°C

940°C

960°C

980°C

1000°C

1.5

1.7

1.9

2.1

2.3

2.5

2.7

0 1 2 3 4

Den

sity

(g/c

m3)

Silica Concentration (wt%)

920°C

1.5

1.7

1.9

2.1

2.3

2.5

2.7

0 1 2 3 4

Den

sity

(g/c

m3)

Silica Concentration (wt%)

940°C

1.5

1.7

1.9

2.1

2.3

2.5

2.7

0 1 2 3 4

Den

sity

(g

/cm

3)

Silica Concentration (wt%)

960°C

1.5

1.7

1.9

2.1

2.3

2.5

2.7

0 1 2 3 4

Den

sity

(g

/cm

3)

Silica Concentration (wt%)

980°C

1.5

1.7

1.9

2.1

2.3

2.5

2.7

0 1 2 3 4

Den

sity

(g/c

m3)

Silica Concentration (wt%)

1000°C

64

Figure 33 Density comparison with and without LiF at ~1000°C

Figure 34 Density comparison of silica versus alumina addition at 1000°C

Molar volume was calculated from the density of the various constituents of the melt (as shown

in Table 9). The molar volume of LiF is lower than that of cryolite therefore if it made up 13wt%

of the melt then the density should increase when compared to pure cryolite. The difference in

the molar volume of alumina and silica is very small so the reason why when LiF is added to the

melt it would have opposite effects on the densities would be down to the complexes formed in

the melt.

1.9

2

2.1

2.2

2.3

2.4

2.5

0 1 2 3 4 5 6 7

Den

sity

(g/c

m3)

Silica Concentration (wt%)

Sokhanvaran (1010°C)

without LiF [54]

This study (1000°C)

with LiF

1.9

2.0

2.1

2.2

2.3

2.4

2.5

0 1 2 3 4 5 6 7 8 9

Den

sity

(g/c

m3)

Oxide Concentration (wt%)

Matiasovsky (1000°C)

with Alumina [28]

This study (1000°C)

with Silica

65

Table 9 Molar volumes of the various constituents of the melt

Molecular Mass

(g/moles)

Density

(g/cm3)

Molar volume

(cm3/mole)

Na3AlF6 209.94 2.04 [54] 102.91

LiF 25.94 1.74 [99] 14.95

SiO2 60.08 2.20* 27.31

Al2O3 101.96 3.69** 27.63

Aluminium is reported to form Al2OFx y-

complex ions in the melt [28, 33]. Addition of

aluminium oxide would provide the melt with a source of aluminium to form these large

complexes. It was also concluded from previous studies that increasing the cryolite ratio would

increase the coordination number of the complexes [33]. Addition of LiF would effectively

increase the mole ratio (from 3 to 8 with 13.8wt%LiF addition) as it could be redefined as

(NaF+LiF)/AlF3. Both these factors might explain why in the cryolite-alumina melt addition of

lithium fluoride would decrease the density. And an increase in density of the cryolite-silica melt

would suggest the opposite behaviour hence a detailed study of silica complex formation in

cryolite and CR+LiF melts need to be conducted.

Figure 35 Comparison of effect of LiF addition with different oxides

1.90

2.00

2.10

2.20

2.30

2.40

2.50

0 2 4 6 8 10 12

Den

sity

(g/c

m3)

Oxide Concentration (wt%)

Grjotheim (1000°C) without LiF with Alumina [45]

Sokhanvaran (1010°C) without LiF with Silica [54]

Matiasovsky (1000°C) with LiF with Alumina [28]

This study (1000°C) with LiF with Silica

* http://www.alfa.com/en/gp100w.pgm?dsstk=013024&rnd=394767240 ** http://accuratus.com/alumox.html

66

When looking at the density behaviour with temperature as shown in Figure 36, the trend is not

as expected i.e. a decrease in density with an increase in temperature. In fact, while increasing

temperature from 920 to 1000°C the density increases and then decreases. Conversely when

temperature is reduced from 1000 to 920°C the density increases with decreasing temperature

which is expected.

Figure 36 Density versus temperature for 1wt% SiO2 electrolyte

This behaviour is due to the reaction as described in Equation (8) (stated earlier):

SiO2 + Na3AlF6 = xNa2O.yAl2O3.zSiO2 + NaF + SiF4↑

implying that the melt composition is changing during the experiment. The composition is a

function of time and temperature. It is also an interesting fact that according to Snow et al. [91]

even the x, y, and z values of the compound is dependent on time, forming at first jadeite, then

disproportioning into albite and nepheline. While heating the melt the reaction is continually

happening, the rate of reaction depends on the temperature and the extent to which the reaction

has happened depends on the time. Hence when repeating the experiment, reproducible results

cannot be achieved. While cooling the melt the reaction is in equilibrium and the expected

behaviour of density increase with decreasing temperature is observed. What these results do

2.0

2.1

2.2

2.3

2.4

2.5

2.6

2.7

2.8

900 920 940 960 980 1000

Den

isty

(g/c

m3)

Temperature (°C)

1% silica - exp 1 - heating

1% silica - exp 2 - heating

1% silica - exp 3 - heating

1% silica - exp 4 - cooling

67

indicate is the maximum density that can be expected at a given temperature and silica

concentration. The electrolysis process can then be designed to accommodate for this. An

additional implication of the result is that the molten (Na, Si, Al)xOy compound is denser than

CR+LiF and further addition of silica will change the composition of the melt to such an extent

that cryolite may be entirely tied up in forming (Na, Si, Al)xOy compounds.

Figure 37 Variations of density with silica concentration versus temperature

2.0

2.1

2.2

2.3

2.4

2.5

900 920 940 960 980 1000 1020

Den

sity

(g/c

m3)

Temperature (°C)

0%

0.50%

1%

1.50%

2%

2.50%

3%

68

Due to a combination of the silica-cryolite reaction and varying silica concentration, the effect of

temperature is not observed in Figure 37 and average density is established for varying silica

concentration. If aluminium was used as a liquid cathode then the cathode would start with about

2.3g/cm3. As shown in Figure 38, the solid line indicates the density of liquid aluminium. The

dotted line indicates the maximum concentration of silica allowed, to make sure the melt will

float on top of the aluminium cathode. As the silicon dissolves into the cathode the density will

increase [100], it is possible to use up to 2.5wt% silica in the melt and make sure the melt will

float on top of the aluminium cathode. Other molten cathode materials like copper and tin have

high enough densities to stay below the molten CR+LiF melt.

Figure 38 Density compared to liquid aluminium

1.50

1.70

1.90

2.10

2.30

2.50

2.70

0 1 2 3 4

Den

sity

(g/c

m3)

Silica Concentration (wt%)

920°C

940°C

960°C

980°C

1000°C

69

Chapter 5 Conclusions and Future Work

5.1 Summary and Conclusions

Three properties were measured for a Cryolite-13.8wt%LiF melt with varying concentrations of

silica at varying temperatures; solubility limit, dissolution rate and density. Two experiments

were conducted a) dissolution behaviour which gave solubility limit and dissolution rate and b)

density measurement using Archimedes’ principle. Silica concentrations were varied at 0.5%

intervals between 0 and 4% and the temperature at 20°C intervals between 900 and 1000°C.

The dissolution behaviour experiment results indicate that solubility is in the range of 2-4wt%

and increases with a linear trend with increasing temperature in the range of 900-1000°C. The

mass transfer coefficient increases exponentially with temperature, ranging from 3-9cm/h. The

solubility values are used to a) prevent depletion of silicon ions at the cathode and creating the

anode effect where a film of gas forms around the anode preventing reduction of silicon ions and

b) avoid adding too much silica which will settle to the bottom be inclusions in the liquid

cathode. The solubility limits can also be used to construct a phase diagram with fixed cryolite to

lithium fluoride ratio. This study concluded that lithium fluoride increases the dissolution rate

but decreases the solubility limit of silica into cryolite. Also once the silica saturates the melt it

continues to react with the melt and lithium fluoride hinders this reaction.

The Archimedes’ experiments illustrated that the density of the melt does not show the common

trend of decrease in density with increase in temperature. When heating from 920 to 1000°C the

density increases and then decreases, a maximum appearing around 960°C. This indicates a

reaction is taking place which changes the composition of the melt. There is a release of SiF4

when the silica reacts with the cryolite-lithium fluoride melt. However after the reaction has

reached equilibrium the density increases with decreasing temperature (cooling from 1000 to

920°C). The reaction and silica concentration counters each other such that the effect of

temperature is not observed.

70

The density values indicate if the liquid alloy of cathode and silicon will float on top or sink

below the electrolyte melt. Up to 2.5wt% of silica dissolved in the electrolyte melt can ensure

that an aluminium-silicon alloy will sink below the electrolyte. Addition of lithium fluoride to

the cryolite-silica melt increases the density; the opposite behaviour to cryolite-alumina melt.

Study should be conducted on the complex formation of silica in cryolite and lithium fluoride

melts to gain a deeper and more fundamental understanding of the melt structure. The decrease

in density for the Al2O3-Cryolite melt is reasoned with alumina forming complexes with oxygen

and fluoride ions in the cryolite melt [33]. Sokhanvaran [54] has found the cryolite-silica melt

increases with silica content; addition of LiF increases the density further which can be attributed

to the lower molar volume. Certainly this study has shown that information gleaned from

alumina-cryolite-lithium fluoride melts (summarised in [33]) cannot always be extrapolated to

make assumptions about silica-cryolite-lithium fluoride melts. The table below summarises how

the oxides affect the CR+LiF melt differently (or to put it another way; how a fixed wt% addition

of LiF affects the CR+oxide melt).

CR+LiF Melt Alumina (Al2O3) Silica (SiO2)

Dissolution Rate

Addition of LiF to cryolite

decreases the dissolution of

alumina into the melt.

Addition of LiF to cryolite

increases the dissolution of silica

into the melt.

Viscosity

Although addition of LiF

decreases viscosity of cryolite, it

increases with increasing

alumina concentration.

Unknown

Solubility Limit

Addition of LiF to cryolite

decreases the solubility of

alumina in the melt.

Addition of LiF to cryolite

decreases the solubility of silica

in the melt.

Density

Increasing alumina concentration

decreases the density. Addition

of LiF decreases the density.

Increasing silica concentration

increases the density. Addition of

LiF increases the density.

71

5.2 Future Works

There are several studies that can be done to continue the investigation on the cryolite-lithium

fluoride-silica melts.

Investigating further the dissolution behaviour; measure mass loss at shorter time intervals

between the 2 and 8 hour points, i.e. measure every hour or even half hour to see if a cross

over between dissolution of silica and reaction of silica with cryolite can be identified.

Improve upon density measurements by developing a method to continuously measure

density to examine how it varies with time at a specific temperature and repeat this

procedure for different temperature to get a comprehensive set of results.

Use the four electrode method to measure conductivity; this is a procedure that has already

been used to measure conductivity of cryolite – silica melts [93].

Measure other properties such as viscosity and surface tension; several methods are

available and further studies will need to be done to determine which method will be most

suitable. The final method will depend on the materials the equipment will have to be made

of to withstand working temperatures and the corrosive cryolite.

Raman spectroscopy to study complex formation for deeper understanding of melt structure.

Attempt actual electrolysis to obtain silicon alloy.

72

References

[1] H. J. Moller, Semiconductors for Solar Cells Applications: Pergamon Press, 1991.

[2] M. G. Mauk, "Silicon Solar Cells: Physical Metallurgy Principles," Journal of Materials,

vol. 55, pp. 38-42, 2003.

[3] GreenZu, "Breakdown of Solar Panels Cost." vol. 2012, 2012.

[4] D. Elwell and G. M. Rao, "Electrolytic Production of Silicon," Journal of Applied

Electrochemistry, vol. 18, pp. 15-22, 1988.

[5] D. Elwell and R. S. Feigelson, "Electrodeposition of Solar Silicon," Solar Energy

Materials, vol. 6, pp. 123-145, 1981.

[6] R. Monnier, "L’obtention et le Raffinage du Silicium par Voie Electrochimique," Chimia,

vol. 37, pp. 109-124, 1983.

[7] U. S. EIA, "Energy Explained, Your Guide to Understanding Energy," 2012.

[8] C. Flavin, "Low-Carbon Energy: A Roadmap," Worldwatch Institute, Washington,

DC2008.

[9] B. R. Bathey and M. C. Cretella, "Review Solar-grade Silicon," Journal of Materials

Science, vol. 17, pp. 3077-3096, 1982.

[10] G. Flamant, V. Kurtcuoglu, J. Murray, and A. Sieinfeld, "Purification of Metallurgical

Grade Silicon by a Solar Process," Solar Energy Materials and Solar Cells, vol. 90, pp.

2099-2106, 2006.

[11] K. Morita and T. Miki, "Thermodynamics of Solar-grade Silicon Refining,"

Intermetallics, vol. 11, pp. 1111–1117, 2003.

[12] J. L. Gumaste, B. C. Mohanty, R. K. Galgali, U. Syamaprasad, B. B. Nayak, S. K. Singh,

and P. K. Jena, "Solvent Refining of Metallurgical Grade Silicon," Solar Energy

Materials, vol. 16, pp. 289-296, 1987.

[13] S. Esfahani, "Solvent Refining of Metallurgical Grade Silicon Using Iron," in Materials

Science and Engineering. vol. MASc: University of Toronto, 2010.

[14] T. Y. K. Morita, "Refining of Silicon during its Solidification from a Si–Al Melt,"

Journal of Crystal Growth, vol. 311, pp. 776-779, 2009.

[15] A. E. Austin, "Silicon Electrodeposition." vol. US Patent 3990953 USA: Battelle

Development Corporation, 1976.

[16] B. G. Gribov and K. V. Zinov'ev, "Preparation of High-Purity Silicon for Solar Cells,"

Inorganic Materials, vol. 39, pp. 653-662, 2003.

[17] Q. Ma, W. Liu, B. Wang, and Q. Meng, "Electrodeposition of Silicon in Organic Solvent

containing Silicon Chloride," Advanced Materials Research, vol. 79-82, pp. 1635-8,

2009.

[18] R. Monnier and D. Barakat, "Dual Cell Refining of Silicon and Germanium." vol. US

Patent 3219561 USA: Gen Trustee Company Inc 1962.

[19] G. M. Rao, D. Elwell, and R. S. Feigelson, "Electrowinning of Silicon from K2SiF6-

Molten Fluoride Systems," Journal of Electrochemical Society: Electrochemical Science

and Technology, vol. 127, pp. 1940-1944, 1980.

[20] D. Elwell, "Electrowinning of Silicon from Solutions of Silica in Alkali Metal

Fluoride/Alkaline Earth Fluoride Eutectics," Solar Energy Materials, vol. 5, pp. 205-210,

1981.

73

[21] G. M. Rao, D. Elwell, R. S. Feigelson, "Electrodeposition of Silicon onto Graphite,"

Journal of Electrochemical Society: Electrochemical Science and Technology, vol. 128,

pp. 1708-1711, 1981.

[22] R. C. De Mattei, D. Elwell, and R. S. Feigelson, "Electrodeposition of Silicon at

Temperatures above its Melting Point," Journal of Electrochemical Society:

Electrochemical Science and Technology, vol. 128, pp. 1712-1714, 1981.

[23] R. Monnier, P. Grandjean, J. Zahler, J. L. Blanc, and T. Tola, "Recherches sur

l’obtention, la séparation et la purification du tantale et du niobium, notamment par voie

électrolytique III. Obtention du tantale, du niobium et du vanadium par électrolyse des

solutions cryolithiques de leurs pentoxydes," Helvetica Chimica Acta, vol. 44, pp. 1862-

1867, 1961.

[24] R. Monnier, P. Grandjean, and J. Zahler, "Recherches sur l’obtention, la séparation et la

purification du tantale et du niobium, notamment par voie électrolytique IV. Essais de

séparation électrolytique du niobium et du tantale," Helvetica Chimica Acta, vol. 46, pp.

2966-2970, 1963.

[25] T. Oishi, M. Watanabe, K. Koyama, M. Tanaka, and K. Saegusa, "Process for Solar

Grade Silicon Production by Molten Salt Electrolysis Using Aluminum-Silicon Liquid

Alloy," Journal of The Electrochemical Society, vol. 158, pp. E93-E99, 2011.

[26] R. H. Hopkins, J. R. Davis, A. Rohatgi, R. B. Campbell, P. D. Blais, P. Rai-Choudhury,

R. E. Stapleton, H. C. Mollenkopf, and J. R. McCormick, "Effect of Impurities and

Processing on Silicon Solar Cells " Westinghouse Research & Development Center and

Dow Coming Corporation1980.

[27] P. D. Desai, "Thermodynamic Properties of Iron and Silicon " Journal of Physical and

Chemical Reference Data, vol. 15, pp. 967-983, 1986.

[28] K. Grjotheim, C. Krohn, M. Malinovsky, K. Matiašovský, and J. Thonstad, Aluminium

Electrolysis Fundimentals of Hall-Heroult Process, 2nd ed. Dusseldorf: Aluminium

Verlag Dusseldorf, 1982.

[29] M. J. Assael, K. Kakosimos, R. M. Banish, J. Brillo, I. Egry, R. Brooks, P. N. Quested,

K. C. Mills, A. Nagashima, Y. Sato, and W. A. Wakeham, "Reference Data for the

Density and Viscosity of Liquid Aluminum and Liquid Iron," Journal of Physical and

Chemical Reference Data, vol. 35, pp. 285-300, 2006.

[30] P. A. Foster, "Melting Point of Cryolite," Journal of the American Ceramic Society, vol.

51, pp. 107-109, 1968.

[31] J. Brynestad, K. Grjotheim, F. Gronvold, J. L. Holm, and S. Urnes, "Cryolite + Alumina

Phase Diagram and the Constitution of Melt in this System," Discussions of the Faraday

Society, vol. 32, pp. 90-96, 1961.

[32] J. L. Holm and B. J. Holm, "Phase Relations and Thermodynamic Properties in the

Ternary Reciprocal System LiF-NaF-Na3AlF6-Li3AlF6," Thermochimica Acta, vol. 6, pp.

375-398, 1973.

[33] G. Mamantov and R. Marassi, Molten Salt Chemistry: An Introduction and Selected

Applications: D. Reidel, 1987.

[34] Y. Zhang and R. A. Rapp, "Modeling the Dependence of Alumina Solubility on

Temperature and Melt Composition in Cryolite- Based Melts," Metallurgical and

Materials Transactions B, vol. 35, pp. 509-515, 2004.

[35] K. Hellberg and J. Massonne, "Process for making Cryolite ". vol. US Patent 3875291

USA: Kali-Chemie Aktiengesellschaft (Hannover, DT) 1975.

74

[36] J. Aigueperse and e. al., "Fluorine Compounds, Inorganic," in Ullman's Encyclopedia of

Industrial Compounds. vol. 15, W. Gerhartz and e. al, Eds.: VCH Publishers, 1985.

[37] K. Grjotheim, K. Matiašovský, and P. Fellner, "The Electrodeposition of Silicon from

Cryolite Melts," in Light Metals 1982 Dallas, Texas, 1982.

[38] R. C. DeMattei and R. S. Feigelson, "Electrochemical Deposition of Semiconductors," in

Electrochemistry of Semiconductors and Electronics, J. M. F. Ludwig, Ed.: William

Andrew Publishing/Noyes, 1992.

[39] J. Ming, L. Yan-Qing, T. Zhong-Liang, L. Fang-Yang, L. Jie, X. Peng-Fei, and L. Ye-

Xiang, "Electrodeposition Behavior of Silicon from Na3AlF6-LiF Melts," Acta Physico-

Chimica Sinica, vol. 27, pp. 1108-1115, 2011.

[40] D. F. Weill and W. J. Fyfe, "The 1010° and 800°C Isothermal Sections in the System

Na3AIF6-AI2O3-SiO2," Journal of Electrochemical Society, vol. 111, pp. 582-585, 1964.

[41] R. G. Haverkamp and B. J. Welch, "Modelling the dissolution of alumina powder in

cryolite," Chemical Engineering and Processing, vol. 37, pp. 177–187, 1998.

[42] N. V. Vasyunina, I. P. Vasyunina, Y. G. Mikhalev, and A. M. Vinogradov, "The

Solubility and Dissolution Rate of Alumina in Acidic Cryolite Aluminous Melts,"

Russian Journal of Non-Ferrous Metals, vol. 50, pp. 338–342, 2009.

[43] D. R. Poirier and G. H. Geiger, Transport Phenomena in Materials Processing: Wiley,

1998.

[44] C. Nicollin, J. Lenoir, and C. Eyraud, Bulletin de la Société Chimique de France, vol. 2,

1967.

[45] K. Grjotheim, K. Matiašovský, P. Fellner, and A. Silný, "Electrolytic Deposition of

Silicon and of Silicon alloys Part I: Physiochemical Properties of the Na3AlF6-Al2O3-

SiO2 Mixtures," Canadian Metallurgical Quarterly, vol. 10, pp. 79-82, 1971.

[46] Y. Zhang, X. Wu, and R. A. Rapp, "Solubility of Alumina in Cryolite Melts:

Measurements and Modeling at 1300 K," Metallurgical and Materials Transactions B,

vol. 34B, pp. 235-242, 2003.

[47] F. J. Frere, "Equilibrium in Fluoride Systems. I. Solubility of Cryolite in Aqueous

Solutions of Iron and Aluminum Salts at 25°C," Journal of the American Chemical

Society, vol. 58, pp. 1695–1697, 1936.

[48] T. E. Jentoftsen, O. A. Lorensten, E. W. Dewing, G. M. Haarberg, and J. Thonstad,

"Solubility of Some Transition Metal Oxides in Cryolite-Alumina Melts: Part I.

Solubility of FeO, FeAl2O4, NiO, and NiAl2O4," Metallurgical and Materials

Transactions B, vol. 33B, pp. 901-908, 2002.

[49] T. E. Jentoftsen, O. A. Lorensten, E. W. Dewing, G. M. Haarberg, and J. Thonstad,

"Solubility of Some Transition Metal Oxides in Cryolite-Alumina Melts: Part II.

Solubility of TiO2," Metallurgical and Materials Transactions B, vol. 33B, pp. 909-913,

2002.

[50] T. E. Jentoftsen, O. A. Lorensten, E. W. Dewing, G. M. Haarberg, and J. Thonstad, "The

Solubility of Some Transition Metal Oxides in Cryolite-Alumina Melts: Part III.

Solubility of CuO and Cu2O," Metallurgical and Materials Transactions B, vol. 38B, pp.

833-839, 2007.

[51] P. Kubíček and T. Pepřica, "Diffusion in Molten Metals and Melts: Application to

Diffusion in Molten Iron," International Metals Reviews, vol. 28, pp. 131-157, 1983.

[52] K. Grjotheim, T. Naterstad, K. E. Torklep, and H. A. Øye, "Tracer Diffusion Coefficients

Molten Salts I. Methods," Electrochimica Acta, vol. 23, 1978.

75

[53] R. A. Khairulin, S. V. Stankus, R. N. Abdullaev, Y. A. Plevachuk, and K. Y. Shunyaev,

"The Density and the Binary Diffusion Coefficients of Silver-Tin Melts," Thermophysics

and Aeromechanics, vol. 17, pp. 391-396, 2010.

[54] S. Sokhanvaran, M. Barati, S. Thomas, "Physiochemical Properties of Cryolite-Silica

Melts," in COM 2011 Montreal, QC, MetSoc, 2011, pp. 25-34.

[55] E. N. Lyakhovich, E. D. Shcherbakov, L. G. Voroshrim, and G. G. Panich, in Vestsi

Akademii Navuk, Belarus S. S. R., 1971.

[56] S. Deville, Comptes rendus des séances de l'Académie des sciences, vol. 39, p. 323, 1854.

[57] F. Ullik, Sitzungsberichte der Kaiserlichen Akademie der Wissenschaften in Wien, vol.

52, p. 115, 1865.

[58] G. Gore, "On the Electrodeposition of Carbon," Chemical News and Journal of Industrial

Science, vol. 50, pp. 113-115, 1884.

[59] W. Hampe, Chemiker Zeitung, vol. 12, p. 115, 1888.

[60] M. Dodero, "Electrolytic Preparation of Calcium Silicide," Comptes rendus des séances

de l'Académie des sciences, vol. 109, p. 566, 1934.

[61] M. Dodero, "Electrolysis of Molten Alkali and Alkaline Earth Silicates," Bulletin de la

Société Chimique de France, vol. 6, p. 209, 1939.

[62] H. V. Wartenberg, "Über Silicium," Zeitschrift für anorganische und allgemeine Chemie,

vol. 265, pp. 186-200, 1951.

[63] D. R. Stern and Q. H. McKenna, "Production of Pure Elemental Silicon." vol. US Patent

2892763 USA: American Potash & Chem Corp, 1959.

[64] F. Olstowski, "Preparation of Silicon." vol. US Patent 3022233 USA: DOW Chemical

Co. , 1962.

[65] P. J. Jorgensen, "Electrolysis of SiO2 on Silicon," The Journal of Chemical Physics, vol.

49, pp. 1594-1598, 1968.

[66] U. Cohen, "Epitaxial Growth of Silicon or Germanium by Electrodeposition from Molten

Salts ". vol. US 3983012 USA: The Board of Trustees of Leland Stanford Junior

University, 1975.

[67] U. Cohen and R. A. Huggins, "Silicon Epitaxial Growth by Electrodeposition from

Molten Fluorides," Journal of Electrochemical Society: Electrochemical Science and

Technology, vol. 123, 1976.

[68] U. Cohen, "Some Prospective Applications of Silicon Electrodeposition from Molten

Fluorides to Solar Cell Fabrication," Journal of Electronic Materials, vol. 6, pp. 607-643,

1977.

[69] R. D. Rauh, T. L. Rose, S. H. White, R. A. Boudreau, T. O. Hoover, D. L. Natwig, "Final

Report," EIC Corporation, USA1981.

[70] J. M. Olson and K. L. Carleton, "A Semipermeable Anode for Silicon Electrorefining,"

Journal of The Electrochemical Society, vol. 128, pp. 2698-2699, 1981.

[71] K. L. Carleton, J. M. Olson, and A. Kibbler, "Electrochemical Nucleation and Growth of

Silicon in Molten Fluorides," Journal of Electrochemical Society: Electrochemical

Science and Technology, vol. 130, pp. 782-786, 1983.

[72] J. M. Olson and A. Kibbler, "Electrowinning of Silicon Using a Molten Tin Cathode," in

Electrochemical Society, 160th Fall Meeting, Extended Abstracts, USA, 1981, p. 1130.

[73] J. R. Stubergh and Z. Liu, "Preparation of Pure Silicon by Electrowinning in a Bytownite-

Cryolite Melt," Metallurgical and Materials Transactions, vol. 27B, pp. 895-900, 1996.

76

[74] K. Yasuda, T. Nohira, Y. H. Ogata, and Y. Ito, "Direct Electrolytic Reduction of Solid

Silicon Dioxide in Molten LiCl–KCl–CaCl2 at 773 K," Journal of The Electrochemical

Society, vol. 152, pp. D208-D212, 2005.

[75] A. Minet, "Électrolyse par fusion ignée des sels de bore et du silicium," Comptes rendus

des séances de l'Académie des sciences, vol. 112, p. 1215, 1891.

[76] H. N. Warren, Chemical News, vol. 67, p. 303, 1893.

[77] K. P. Batashev and A. Y. Zhurdin, Metallurg, vol. 2, pp. 66-77, 1933.

[78] R. Monnier and D. Barakat, "Contribution à l’étude du Comportement de la Silice dans

les Bains de Cryolithe Fondue," Helvetica Chimica Acta, vol. 40, pp. 2041-2045, 1957.

[79] R. Monnier and P. Grandjean, "Recherches sur l’obtention, la séparation et la purification

du tantale et du niobium, notamment par voie électrolytique II. Etude d’une méthode de

détermination des tensions effectives d’éctrolyse et des tensions pratiques de

décomposition; application de cette méthode à des solutions cryolithiques d’oxydes,"

Helvetica Chimica Acta, vol. 43, pp. 2163-2172, 1960.

[80] R. Monnier, D. Barakat, and J. C. Giacometti, "Refining of Silicon and Germanium ".

vol. US Patent 3254010: Gen Trustee Company Inc., 1962.

[81] R. Monnier and J. C. Giacometti, "Recherches sur le raffinage éctrolytique du silicium,"

Helvetica Chimica Acta, vol. 47, pp. 345-353, 1964.

[82] K. Grjotheim and K. Matiašovský, "Preparation of Si-Al Alloys by Electrolysis of Molten

Cryolite," Hutnické Listy, vol. 25, pp. 876-882, 1970.

[83] K. Grjotheim, K. Matiašovský, and M. Malinovsky, "Influence of NaCl and LiF on the

Aluminium Electrolyte," Electrochimica Acta, vol. 15, pp. 259-269, 1970.

[84] G. Bøe, K. Grjotheim, K. Matiašovský, and P. Fellner, "Electrolytic Deposition of Silicon

and of Silicon Alloys Part II: Decomposition Voltages of Components and Current

Efficiency in the Electrolysis of the Na3AlF6-Al2O3-SiO2 Mixtures," Canadian

Metallurgical Quarterly, vol. 10, pp. 179-183, 1971.

[85] G. Bøe, K. Grjotheim, K. Matiašovský, P. Fellner, "Electrolytic Deposition of Silicon and

of Silicon Alloys Part III: Deposition of Silicon and Aluminum using a Copper Cathode,"

Canadian Metallurgical Quarterly, vol. 10, pp. 281-285, 1971.

[86] G. Bøe, K. Grjotheim, K. Matiašovský, P. Fellner, "Electrolytic Deposition of Silicon and

of Silicon Alloys Part IV: Preparation of Alloys with a High Content of Silicon, and

Silicon Refining," Canadian Metallurgical Quarterly, vol. 11, pp. 463-467, 1972.

[87] K. Grjotheim, K. Matiašovský, and P. Fellner, "Preparation of Silicon Alloys by

Electrolysis of Cryolite- Silica Melts," Revue Roumaine de Chimie, vol. 17, pp. 819-829,

1972.

[88] K. Grjotheim, K. Matiašovský, and P. Fellner, "Some aspects of the electrolytic

preparation of silicon alloys," Chemické Zvesti, vol. 27, pp. 165-171, 1973.

[89] P. Fellner and K. Matiašovský, "Electrolytic Silicide Coating in Fused Salts,"

Electrodeposition and Surface Treatment, vol. 3, pp. 235-244, 1975.

[90] Y. A. Zyazev and A. I. Ezrielev, Siberian Trudy Agronomi Fizik, vol. 13, p. 32, 1966.

[91] R. J. Snow, B. J. Welch, "Reactions in the Cryolite-Silica System," Proceedings -

Australasian Institute of Mining and Metallurgy, vol. 241, pp. 81-86, 1972.

[92] P. K. Gairola, R. K. Tiwari, and A. Ghosh, "Rates of Dissolution of a Vertical Nickel

Cylinder in Liquid Aluminum Under Free Convection," Metallurgical Transactions, vol.

2, pp. 2123-2126, 1971.

77

[93] S. Sokhanvaran, S. Thomas, and M. Barati, "Charge Transport Properties of Cryolite–

Silica Melts," Electrochimica Acta, vol. 66, pp. 239-244, 2012.

[94] M. Barati and K. S. Coley, "Electrical and Electronic Conductivity of CaO-SiO2-FeOx

Slags at Various Oxygen Potentials: Part I. Experimental Results," Metallurgical and

Materials Transactions B, vol. 37B, pp. 41-49, 2006.

[95] S. L. Schiefelbein, "A New Technique to Measure the Electrical Properties of Molten

Oxides," in Metallurgical Engineering and Materials Science. vol. PhD: Carnegie Mellon

University, 1989.

[96] Y. Ohta, A. Miyanaga, K. Moringa, and T. Yanagase, "Study on the Measuring of the

Electrical Conductivity of Ionic Solutions and Melts," Journal of the Japan Institute of

Metals, vol. 45, pp. 1036-1043, 1981.

[97] A. Silný and B. Haugsdal, "Electrical conductivity measurements of corrosive liquids at

high temperatures," Review of Scientific Instruments, vol. 64, pp. 532-537, 1993.

[98] R. Tamamushi and K. Takahashi, "Instrumental study of Electrolytic Conductance

Measurements using Four-Electrode Cells," Journal of Electroanalytical Chemistry and

Interfacial Electrochemistry, vol. 50, pp. 1036–1043, 1974.

[99] G. J. Janz, "Thermodynamic and Transport Properties of Molten Salts: Correlation

Equations for Critically Evaluated Density, Surface Tension, Electrical Conductance, and

Viscosity Data," Journal of Physical and Chemical Reference Data, vol. 17, 1988.

[100] T. Magnusson and L. Arnberg, "Density and Solidification Shrinkage of Hypoeutectic

Aluminum-Silicon Alloys," Metallurgical and Materials Transactions A, vol. 32, pp.

2605-2613, 2001.

78

Appendices

79

Appendix A – Phase Diagrams generated by FactSageTM

Phase Diagram of Silicon – Copper

80

Phase Diagram of Silicon – Aluminium

81

Phase Diagram of Silicon – Tin

82

Phase Diagram of Alumina – Cryolite

83

Phase diagram of Lithium Fluoride – Cryolite

84

Appendix B – Derivation of Equation 7

Deriving the equation for mass transfer coefficient

General mass transfer equation:

m = mass of silica rod dissolving in melt (g) (only considering the portion inserted in the melt)

t = time of dissolution (h)

k = mass transfer coefficient (cm/h)

A = area of silica rod exposed to the melt (cm2)

CS = saturation concentration of silica in melt at the temperature of experiment (g/cm3)

CB = concentration of silica in the bulk at time t (g/cm3)

Mass of rod inserted in the melt:

r = radius of silica rod

h = length of rod inserted in melt

ρ = density of silica rod

Radius of silica rod inserted in melt in terms of mass:

√

Area of silica rod exposed:

(Area should be A = πr2 + 2πrh but πr

2 is negligible because the silica rod rests on the bottom of

the crucible)

Area in terms of mass:

√

√

√

Bulk concentration:

m0 = initial mass of silica rod inserted in melt

85

V = volume of melt

Some numbers that can be assumed constant:

V = volume of melt does not change much with dissolution of silica because maximum amount

of silica dissolved is less than 9g into a melt of 200g. Also at each time (15min, 30min, 1, 2, 4,

8hrs) a rod of initial radius of 1.25cm is inserted and the change of radius is less than 0.06cm. To

make derivation easier V is assumed constant.

h = height of melt after the rod is inserted into melt. Also assumed constant for reasons given

above.

Others constants: m0, ρ, π, CS, and k.

When using MS Excel to optimise equation, k and CS will be solved.

Some constants are defined in terms of a single letter to make equations more legible:

√

Deriving the equation:

√ ([

] )

Defining more constants:

Continuing with derivation:

√ (

)

∫

√ ( )

∫

Using substitution integration:

u = √ u2 = m

2 u du = dm

∫

(

)

86

∫

Using integral identity:

∫

|

|

∫

∫

|

|

|

|

Since u = √ , using the experimental values obtained, √ > √ .

Define a = C1V, x = u

√ |

√ √

√ √ |

@t = 0, m = m0

Solving for C

√ |

√ √

√ √ |

Hence:

√

|

√ √

√ √ | |

√ √

√ √ |

Rearranging the equation to get mass in terms of time:

First defining more constants:

|√ √

√ √ |

√

Continuing:

|√ √

√ √ |

√ √

√ √

87

√ √ (√ √ )

√ √

√ √

@ t = 0

Since:

|

√ √

√ √ |

|

√ √

√ √ |

√ √

√ √

Also:

√

√

Therefore:

Just looking at:

√

√

√ √

√ √

√

√

√ √

√ √

( √ )(√ √ ) ( √ )(√ √ ) √ √ ( √ )

( √ )(√ √ )

( √ )(√ √ ) ( √ )(√ √ ) √ √ ( √ )

( √ )(√ √ )

88

( √ √ √ √ √ √ )

( √ √ √ √ √ √ )

( √ √ √ √ √ √ )

( √ √ √ √ √ √ )

( √ √ √ √ √ √ )

( √ √ √ √ √ √ )

√ √

√ √

√ √

(√ √ )

Returning to original equation:

@ t = ∞

CSV = the total mass of silica dissolved in the melt.

89

Appendix C – Effect of considering V and h as constants on k and CS

This section will explain why considering volume of the melt, V, and the height of the melt, h,

does not considerably change the values calculated for k and CS calculated for Equation 7. To

simplify calculations only limiting values will be used:

Maximum measured amount of silica dissolved into CR+LiF melt = 9g

Maximum measured change in radius of silica rod: 0.06cm

Initial mass of melt = 200g

Maximum density measured of CR+LiF with silica addition = 2.5g/cm3

Minimum density measured of CR+LiF without silica = 2.0g/cm3

Calculate initial volume of melt: V = m/ρ = 200/2.0 = 100 cm3

Calculate final volume of melt: V = m/ρ = 209/2.5 = 83.6 cm3

Associated change in height = 0.5cm

The values of k and CS are within experimental error of each other.

Plotting mass transfer coefficient:

0

1

2

3

4

5

6

7

8

9

10

910 930 950 970 990 1010

Mass

Tra

nsf

er C

oef

fici

ent

(cm

/h)

Temperature (°C)

k with volume considered constant

k with volume change

90

Plotting solubility limit:

1.7

1.9

2.1

2.3

2.5

2.7

2.9

3.1

3.3

3.5

3.7

910 930 950 970 990 1010

Solu

bil

ity L

imit

(w

t%)

Temperature (°C)

Cs with volume considered constant

Cs with volume change

91

Appendix D – XRD analysis of white deposit on silica rod in dissolution behaviour experiment